, ‘ J5
“WW?
h

~\
.
I!
A

".
4,. .Q'
.1 .g
‘ .1.
3 -:.

.a
if ..
I‘g ‘—
,

7c
293-.-

‘32.

-.-.-..

u

”32-.

du-

.1

.. 5"!“-
i'.‘

-w
(

‘ ’1'
"‘51?
.a““""
,

m' up
flail: ’o‘.‘

:2!»

4. ; , 3; .g v ..
iiifufiééldigfiifii ‘* EM.-

 

 

N: ”E

l‘

a
J

K

This is to certify that the
dissertation entitled

LUCILIA SERICATA DEVELOPMENT: PLASTICITY,
POPULATION DIFFERENCES, AND GENE EXPRESSION

presented by

Aaron Michael Tarone

has been accepted towards fulfillment
of the requirements for the

PhD. degree in Zoology

 

 

 

'Major Profes’sor’s Signature

i/7/07

Date

MSU is an afiinnative-action, equal-opportunity employer

 

UBQARY
MlCl’iith". State
Universriy

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:/CIRC/DateDue.indd-p.1

 

 

LUCILIA SERICATA DEVELOPMENT: PLASTICITY, POPULATION
DIFFERENCES, AND GENE EXPRESSION

By

Aaron Michael Tarone

A DISSERTATION

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

DOCTOR OF PHILOSOPHY
Department of Zoology

2007

ABSTRACT

Lucilia sericata Development: Plasticity, Population Differences, and Gene Expression

By

Aaron Michael Tarone

Blow flies (Diptera: Calliphoridae) are parasitoids, parasites, and primary
successional species on carrion. A detailed understanding of their development can
impact medical, veterinary, and forensic sciences as they spread disease, debride wounds,
spoil food, parasitize humans and livestock, and predictably develop on carrion.
Comparisons of blow flies to Drosophila melanogaster and other higher flies can also
reveal important evolutionary effects on a locus or trait of interest. The research detailed
herein was based on the application of quantitative and molecular genetic methodology to
questions regarding the development of the globally distributed blow fly Lucilia sericata
(Diptera: Calliphoridae) (Meigen).

The main focus of this research was to improve the precision of postmortem
interval (PMI) estimates produced with entomological evidence, based on the inclusion of
developmentally variable gene expression data into the process of predicting blow fly
age. The expression profiles of 9 genes, in conjunction with generalized additive models
(GAMs), was useful in more precisely predicting the ages of immature L. sericata. Some
genes were also useful in identifying individuals in a state of arrested development,
which will further decrease error in entomology based PMI estimates.

The principles of quantitative genetics were also applied to the problem of

understanding developmental variation in L. sericata. Complimentary studies were

undertaken to understand how the environment and genetics affect body size and
development time. Components of developmental plasticity were identified, providing an
improved understanding of how laboratory rearing conditions affect blow fly
development time and enabling the identification of a rearing protocol that mimics
growth on carrion. Likewise, the effects of genotype and temperature on variation in the
development time and body sizes of L. sericata from California, Michigan, and West
Virginia were determined, revealing significant temperature and strain effects on these
phenotypes.

The results of these projects indicate that molecular and quantitative genetics are
usefiJl tools in understanding blow fly development, especially for predicting a PMI. The
approach used herein demonstrated a quantifiably improved ability to predict blow fly
age, addressing a key shortcoming in forensic entomology as interpreted by the Daubert
standard of scientific evidence. However, the data obtained will also be useful in
devising methodologies for controlling blow fly populations and in understanding the
evolution of development. Several of the genes studied interact with insecticides;
consequently, profiles of the expression levels of these genes can be useful for controlling
blow fly pests. Similarly, knowledge about the factors affecting developmental plasticity
may be used to better manage facilities affected by blow flies. Likewise, the high-
resolution gene expression data produced in this research will aid in understanding the
evolution of development in the Diptera. By studying gene expression, and other traits,
throughout the development of the blow fly Lucilia sericata, valuable knowledge

affecting a variety of biological studies has been acquired.

This dissertation is dedicated to my mother Cindy Tarone, who died without the
opportunity to watch her son grow up, and to my father Lawrence Tarone, who made sure

that I did.

iv

 

ACKNOWLEDGEMENTS

I greatly appreciated the advice received from all of my committee members:
David F oran, Rich Merritt, David Arnosti, Will Kopachik, and Ned Walker. I would
especially like to thank my advisor David Foran for the willingness to support a highly
interdisciplinary project in his laboratory and Rich Merritt for his role in my professional
development as a forensic entomologist. The proofreading skills of Missy Foran were
also appreciated.

Thanks to the members of the F oran, Merritt, and Walker laboratories and to the
Forensic Science graduate students. I am grateful for their friendship, companionship
during coffee runs, assistance, and tolerance of my research organism (especially the
unappealing odor associated with it). Thanks to Kim Jennings and Erin Lenz for help
with the genetic identification of fly strains. Thanks to Jeff Dake, Dave and Anne
Szymanski, Michiel Kramer, Lisa Ramos, Stephanie Kremer, Angela Donatelle, Mike
Gehring, Cory Michaud, Ken Eilert, Christina Rauzi, and Shane “Sunshine” Hoffmann
for minimizing the loss of sanity associated with the process of researching and writing a
doctoral dissertation. Thanks to Ryan Kimbirauskas, Mollie McIntosh, and Eric Benbow
for their assistance, advice, beers, and for their hospitality in the Merritt lab.

Various Michigan State University staff assisted the implementation of this
research. Thanks to the staff of the Michigan State University Research Technology
Support Facility, especially Annette Thelen, Jeff Landgraff, and Shari Tjigum-I-Iolland,
as they were instrumental in assisting with experimental design, the use of robotics, RNA
isolation, and quantitative PCR. Randall Shoemaker and Constance Crew from Michigan

State University Laboratory Animal Resources provided ethics advice and the rats for this

 

research. Brian McGill provided valuable insight into the world of multivariate statistical
analyses. Conversations regarding variation in fly development with Alex Shingleton
were appreciated, as was the microscope time. The validation experiment was possible
due to the excellent effort of Trevor McLean during his summer internship at Michigan
State University.

Many forensic entomologists provided useful insight and/or helpful assistance
regarding experimental design, data analysis, and species identification. These include
Ryan Kimbiraskaus at Michigan State University, Jeff Tomberlin at Texas A&M, Jeff
Wells as West Virginia University, Bob Kimsey at the University of California at Davis,
Neal Haskell at Saint Joseph’s University, and Gail Anderson at Simon Frasier
University.

To my father and grandparents: Your contributions to my childhood and
education have always been greatly appreciated. Thanks to you, a child with an unsure
future made it to adulthood intact and with a moderately good head on his shoulders. I
love you all and want you to know that your efforts will never be forgotten. This degree
would not have been possible without you.

I owe more to Paula Tarone than I could ever hope to list. Thank you for your
love and support, demonstrated by your willingness to many a first-year PhD student in
December and then move from California wine country to East Lansing (in January!) to
be with him. You were a source of strength through all of the stressful times. Thank you

for sharing me with the flies.

vi

Thanks to Dan Barbash and Sergey Nuzhdin for your mentorship at UC Davis.
You taught me many of the skills that are required to successfully design, implement, and

complete a doctoral degree.

The funds provided by the MSU Graduate School, the MSU Department of
Zoology, and the National Institute of Justice were integral to the completion of this

research.

Thanks to all of my friends and family.

vii

TABLE OF CONTENTS

LIST OF TABLES ................................................................................... xi
LIST OF FIGURES ................................................................................. xii
LIST OF APPENDICES ........................................................................... xxv
CHAPTER 1: INTRODUCTION AND BACKGROUND .................................... 1
The importance of blow flies ............................................................... 1
Blow fly development ....................................................................... 3
Evolution and ecology of blow flies ...................................................... 8
Forensic entomology ....................................................................... 12
The use of developmental plasticity and gene expression for blow fly pest
control ....................................................................................... 16
Research ..................................................................................... 17
CHAPTER 2: PLASTICITY IN BLOW FLY GROWTH .................................... 23
Introduction ................................................................................. 23
Materials and methods ..................................................................... 26
Fly collection and rearing ....................................................... 26
Statistical analyses ................................................................ 29
Results ....................................................................................... 30
Species identification ............................................................. 30
Developmental plasticity ......................................................... 31
Development on carrion ......................................................... 34
Discussion ................................................................................... 37
Environmental components of variation in the development of
L. sericata .......................................................................... 37
Other potential sources of variation .......................................................... 40
Optimal rearing condition using liver and growth on rat carrion ......... 41.
Applications to forensic entomology ........................................... 41

CHAPTER 3: THE USE OF GENERALIZED ADDITIVE MODELS IN ANALYZING

FORENSIC ENTOMOLOGICAL DATA ...................................................... 46
Introduction ................................................................................. 46
Materials and methods ..................................................................... 48

Species identification ............................................................. 48
Growth experiments ............................................................... 48
Statistical analyses ................................................................ 5 1
Results ....................................................................................... 54
Species identification ............................................................. 54
Immature development ............................................................ 54
Assessing statistical models ...................................................... 59

viii

 

GAM validation with independent data ........................................ 63

Discussion ................................................................................... 64
CHAPTER 4: GENE EXPRESSION IN BLOW FLY EGGS ............................... 72
Introduction ................................................................................. 72
Materials and methods ..................................................................... 74
Species identification and egg collection ....................................... 74

Gene sequencing and primer design ............................................ 74

cDNA synthesis and quantitative PCR .......................................... 75

Results ....................................................................................... 77
Discussion ................................................................................... 83

CHAPTER 5: PREDICTING BLOW FLY AGE WITH LARVAL AND PUPAL GENE

EXPRESSION ....................................................................................... 87
Introduction .................................................................................. 87
Materials and methods ..................................................................... 92

Species identification, fly rearing, and collections ............................ 92
Sequencing of L. sericata loci .................................................... 93
RNA extraction ..................................................................... 93
Reverse transcription and quantitative PC R .................................. 94
Statistical analyses ................................................................ 95
Results ....................................................................................... 97
Gene sequences .................................................................... 97
Quantitative PCR and statistical modeling .................................... 97
Discussion ................................................................................. 150

CHAPTER 6: VALIDATION OF GENE EXPRESSION BASED PREDICTIONS OF

BLOW FLY AGE WITH BLIND PREDICTIONS .......................................... 168
Introduction ................................................................................ l 68
Methods .................................................................................... 168

Strain collection and identification ............................................ 168
Collection of unknowns ......................................................... 168
Converting RNA later size measurements to live size measurements. . . I 69
Molecular biology and statistics ............................................... 170
Results ...................................................................................... 171
Discussion ................................................................................. 178

CHAPTER 7: GENE EXPRESSION ANALYSES OF NON-PUPATIN G

LARVAE ........................................................................................... 182
Introduction ................................................................................ 1 82
Methods .................................................................................... 1 83
Results ...................................................................................... 184
Discussion ................................................................................. 189

Potential causes of the “Peter Pan” state .................................... 189
Interpreting gene expression changes in “Peter Pan ” larvae ............. 191
Forensic analyses of non-pupating larvae .................................... 193

ix

CHAPTER 8: POPULATION AND TEMPERATURE EFFECTS ON BODY SIZE

AND DEVELOPMENT TIME .................................................................. 194
Introduction ................................................................................ 1 94
Materials and Methods ................................................................... 199
Results ...................................................................................... 200
Discussion ................................................................................. 210

Temperature efiects on boay size .............................................. 210
The evolution of age and size at metamorphosis ............................ 212
Year-to-year variation .......................................................... 213
Forensic entomology ............................................................ 214

CHAPTER 9: DISCUSSION .................................................................... 217

Improved predictions of blow fly age .................................................. 217
Introduction ....................................................................... 217

Improving accuracy and precision in forensic entomology ................ 218

Addressing the Daubert standards in forensic entomology ................ 219

Logistical improvements associated with molecular predictions of blow fly

Age ................................................................................. 221

Future research in forensic entomology ...................................... 222

Controlling blow fly populations with gene expression and plasticity data ...... 228
Conclusions ................................................................................ 229
LITERATURE CITED ........................................................................... 239

List of Tables

Table 2.1. Treatment types for the 37 liver-fed cohorts ....................................... 43

Table 2.2. Average development times :t standard deviations in hours and accumulated
degree-days* for significant rearing treatments of Lucilia sericata .......................... 44

Table 3.1. The 18 generalized additive models for predicting development percent

assessed in this experiment ........................................................................ 69
Table 3.2. Summaries of the estimated significance of each variable in a model ......... 70
Table 4.1. Primers used in this study ............................................................ 86

Table 5.1. Primers used for sequencing and qPCR for all loci in this experiment ....... 160
Table 5.2. Gene sequencing results ............................................................. 161
Table 5.3. Generalized additive models assessed in this experiment ...................... 162
Table 5.4. Estimated significance and degrees of freedom for variables in all GAMs..163
Table 6.1. Regressions of live length against preserved length for each stage ............ 181

Table 6.2. Regressions of live weight against preserved weight for each stage .......... 181

xi

List of Figures

Images in this dissertation are presented in color.

Figure 1.1. The lifecycle of a blow fly. Females lay a clutch of eggs on carrion (I),
typically either at an orifice or on a wound. First larval instars hatch from the eggs (2)
and feed until they molt into the second instars, which feed and grow until they molt into
third instars. Feeding third instars (3) can be identified by three slits in their posterior
spiracles, the visibility of blood in their crop, and by the fact that they are actively
feeding. Postfeeding third instars (4), which lack visible tissue in the crop, cease feeding
and leave the food source to form a puparium. In the puparium (5), the fly
metamorphoses and eventually ecloses as an adult by breaking the lid off of the puparium
with its ptilinum (top and middle of 5). After a few hours, the adult cuticle and wings
achieve a normal appearance and the fly is fully capable of flight (6) ........................ 5

Figure 1.2. A phylogeny of the Diptera. Taken from Yeates and Wiegmann (2005). The
Nematocera are the most primitive flies and include mosquitoes and crane flies.
Cyclorrhapha are the “higher flies”, which have three larval instars and form a puparium.
The Schizophora, which include the Calliphoridae and Drosophilidae, have a specialized
structure, a ptilinum, which is used to break the lid off of the puparium during eclosion.
The Drosophilidae belong to the Acalyptrata, as they lack calypters on their wings. The
Calliphoridae belong to the superfamily Oestroidea, which is part of the Calyptrata.
Evolutionarily, the Drosophilidae and Calliphoridae are closely related families within
the Diptera, which means that they are expected to share many molecular, developmental,
and morphological similarities including three larval instars, the formation of a puparium,
and the presence of a ptilinum ..................................................................... 11

Figure 1.3. Ecoregions of the United States. Map from work by Hargrove and Hoffman
(2004). The white dots indicate approximate origins of the CA, MI, and WV flies
studied. Regions of similar color have comparable elevations, soil, and climates. Both
the MI and WV strains were located in ecosystems that were mapped in shades of blue,
while the CA strain came from a yellow region ................................................. 21

Figure 2.1. Developmental variation among Lucilia sericata cohorts by treatment type.
Boxplots of total development time Grours) for each of the 37 liver-fed cohorts. The line
within the box represents the median develo ment hours, the box represents the
development times between the 25th and 75 percentiles, and the ‘whiskers’ (outer-most
lines) represent the 5th and 95‘h percentiles. 1a: fresh meat daily or no fresh meat daily
(F MD vs. NFMD); 1b: paper towel (moist paper towel placed under meat); 1c: transfer:
transfer of larvae to fresh substrate for pupation; 1d: destructive (removal of 12
individuals each day). Note: treatments were in combination with other treatment types
(Table 1) that had significant effects on development time. For instance, the two outliers
in the FMD boxplot (1a) are those that were also destructively sampled .................. 33

xii

Figure 2.2. Growth curves of Lucilia sericata on liver versus growth on rat carrion. Non-
linear quantile regression curves created from the lengths of maggots in daily collections
of each treatment type. 2a: meat added every 3rd day, no moist paper towel, larvae were
not transferred to fresh substrate to pupate; 2b: fresh meat daily, no moist paper towels,
larvae transferred fresh substrate to pupate; 2c: fresh meat daily, moist paper towel used,
larvae transferred to fresh substrate to pupate; 2d: the locally weighted sum of squares
curve of data estimated from Grassberger and Reiter (2001) plotted against larval growth
on rat carrion. Numbers of cohorts plotted for each treatment were 3, 4, 6, and 6 for Rat,
NF MD, FMD, and F MDPT respectively. The solid line on each curve is the 50th
percentile plot from cohorts raised on rats. Treatments are shown as dashed lines, with
the thicker dashed line representing the 50th percentile and the thinner lines representing
the 97.5th and 2.5“1 percentiles (95% confidence intervals). Confidence intervals for the
rat cohorts are present in 2d ....................................................................... 35

Figure 2.3. Linear growth of Lucilia sericata on liver versus growth on rats. Regression
lines of the same treatments displayed in Figure 2.2, for the first three days of growth—
the linear phase of development. Line types used to indicate treatments are the same as
in Figure 2.2 ......................................................................................... 36

Figure 2.4. Development times of Lucilia sericata cohorts raised on liver versus rat
canion. Comparison of total development hours produced by the 37 liver-fed cohorts in
this study to the development of the three rat-fed cohorts. Development time on rats was
much less variable than growth on liver, with a development time most similar to the
fastest growing liver-fed cohorts. Boxplot design is as in Figure 2.1 ........................ 36

Figure 3.1 . The lengths (mm) of 2559 immature L. sericata throughout immature
development (percent of development values are on a 0—] scale). Note the tight
distribution of sizes during the earlier, linear growth phase compared to the more variable
postfeeding third instar and pupal stages ......................................................... 56

Figure 3.2. A plot of the distribution of development percents for individuals at each
developmental stage. As development progressed, the proportion of the lifecycle spent in
a stage increased. 3rd indicates the feeding portion of the third instar; 3'dPF indicates the
postfeeding stage of the third instar ............................................................... 56

Figure 3.3. The lengths and weights of individuals throughout development from the 6
cohorts. Growth is compared by strain and by temperature. Solid lines represent the
average for all strains or both temperatures. a) Length (mm) plots for each strain. The
largest strain, denoted by the line with short bars and spaces, was CA, and the smallest
strain, designated by the line with short bars separated by dots, was WV. The M1 strain
was close to the average size and is represented by the spaced line with long bars and
short spaces. Less size variation existed during the feeding portion of the lifecycle (when
size was increasing) than in the postfeeding and pupal stages. b) Length plots comparing
grth at 20°C versus 335°C. Growth at 20°C is represented by the spaced line with
short bars separated by dots and 335°C is represented by the line with short bars and long
spaces. The higher temperature resulted in a growth curve that had a steeper slope during

xiii

the linear growth phase of development; individuals fiom these treatments peaked in
body size proportionally faster than cooler treatments, which resulted in smaller body
sizes as pupae. c) Weight (mg) plots for each strain. Comparisons among strains were as
in (a). d) Weight plots for the two temperature treatments, with similar results as in

(b) ..................................................................................................... 58

Figure 3.4. A comparison of diagnostic plots for model 10 (a, c, e, g) and model 18 (b, d,
f, h). Model 10 predicted age using length and stage, while model 18 used all data to
make age predictions, and explained the most deviance in the data. Panels a) and b) are
quantile-quantile plots, assessing the validity of each model’s assumptions. The line is
relatively straight and increasing for both models, indicating that the distributional
assumption of the model did not violate the other assumptions of the model, however
model 18 generated a smoother line. Panels c) and (I) show the distribution of residuals
throughout the lifecycle; in both models residuals are of equivalent size for all ages,
showing that there is no bias in residuals based on age. Panels e) and f) are the
distributions of residuals, which are normally distributed thus the gamma distribution was
acceptable to use with these data. Panels g) and h) are plots of true (response) versus
predicted (fitted) values for data used to construct the models. Fitted values accurately
predicted true age through linear (feeding) development (the first 25% of the lifecycle),
after which the first gap in prediction appeared. This indicates that an overestimation of
ages for postfeeding larvae between c. 19.1 and 30% developed was produced using
model 10. Model 18 was less likely to overestimate age for individuals at this point in
the lifecycle, but predictions were still biased toward overestimating age in young
postfeeding third instars. Neither model represents a noticeable improvement over using
stage alone for aging pupae. Predicting pupal age with just developmental stage resulted
in an age estimate between 43.2 and 100% of immature development. The most
predictive model (model 18) plots fitted values between 61.9 and 81.2 % (pupae). At
either end of this predicted range, true values could be between 43.2 and 100%. As with
all size/stage models, error increased with age .................................................. 61

Figure 3.5. A plot of predicted versus true ages of 252 individual flies raised on rats as
estimated with a GAM for stage and length (model 10). A plot of predicted versus true
development percents of 252 larvae raised on rats as estimated with a GAM for
developmental stage and length (model 10). Development is displayed to 50% because
length measurements ceased when pupation occurred. The solid line represents the Y
(predicted age) = X (true age) line. Dashed lines represent 95% confidence intervals for
the predictions based on the data in (a). Model 10 accurately (data span Y = X) and
precisely (ranging c. 5% above and below Y = X) predicted age for the feeding portion of
the lifecycle (<26% of development), when body size increased in a linear fashion. As
expected, precision decreased greatly once the postfeeding third instar was reached,
although overall error for the model was within predicted levels ............................ 64

Figure 4.1. Binary gene expression profile for cs at 32°C in L. sericata eggs, fi'om 0—]
through 8—9 hours of development. 0 indicates no detectable expression of the gene and
1 indicates detectable expression of cs. Expression was not detected from 0—2 hours.

xiv

 

 

Between 2 and 6 hours some eggs clusters expressed es and some did not. After 6 hours,
all egg clusters expressed cs ....................................................................... 79

Figure 4.2. Standardized expression of cs in L. sericata eggs, from 0—1 through 8—9
hours of development. The CT was standardized against the average of rp49 and fl
tubulin 5 6D CT values and plotted through time. The regression of cs CT over time was
also included. High expression levels are indicated by low CT values. cs was not
expressed from 0—2 hours and then its expression increased ................................. 80

Figure 4.3. Standardized transcript abundance of bed in L. sericata eggs, from 0—1
through 8—9 hours of development. Standardization was as in Figure 4.2. bcd gene
expression was highest from 0—2 hours, then transcripts decreased in abundance. . . . .....81

Figure 4.4. Standardized transcript abundance of SI] in L. sericata eggs, from 0—1
through 8—9 hours of development. Standardization was as in Figure 4.2. sll expression
was highest from 0—2 hours, then tended to be lower as eggs developed, though variance
in expression was high .............................................................................. 82

Figure 4.5. Predicted (fitted) versus true (response) ages for a generalized additive model
that made age predictions for the 33 egg masses that expressed bcd, sll, and cs.
Estimated ages were within 2 h of the true age in 30 of the 33 cases ........................ 83

Figure 5.1. The positive correlation between housekeeping genes. The CT scores
followed a line with a slope of 1 (after accounting for the average CT difference of 1.2)
with a correlation of 0.84 ........................................................................... 98

Figure 5.2. Variance in gene expression due to the standardization against housekeeping
genes. Since the average of both housekeeping genes was used for standardization, one
half of the difference between rp4 9 and the beta tubulin gene was plotted throughout
development. The variance in this plot was relatively constant and indicates that an
approximate difference in expression of up to 4 fold (2 CT units) could be explained by
variation in housekeeping gene expression ...................................................... 99

Figure 5.3. Standardized gene expression of scalloped wings throughout the immature
development of L. sericata. No apparent pattern of gene expression was observed
throughout development .......................................................................... 100

Figure 5.4. Standardized gene expression of Wingless throughout the immature
development of L. sericata. No informative pattern of gene expression was observed
throughout development .......................................................................... 101

Figure 5.5. Standardized gene expression of rhodopsin 3 throughout the immature

development of L. sericata. No apparent pattern of gene expression was observed
throughout development .......................................................................... 102

XV

Figure 5.6. Standardized chitin synthase CT scores plotted by stage. The postfeeding
third instar stands out as expressing significantly less cs than the other stages ........... 103

Figure 5.7. Standardized chitin synthase CT scores plotted by development percent for
(a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 335°C. The shift in the red line between (a) and (b) indicates a cluster
of individuals that did not express the gene at the time point where gene expression is at
its lowest levels for this locus .................................................................... 104

Figure 5.8. Standardized chitin synthase CT scores plotted by development percent at
335°C for (a) all individuals that expressed the gene and (b) in all 958 individuals, with
NA expression values assigned a CT of 50. The red line indicates gene expression in the
CA strain. Blue indicates gene expression in the M1 strain. The green line indicates gene
expression in the WV strain. The shift in the red line between (a) and (b) indicates a
cluster of individuals that did not express the gene at that time point ...................... 105

Figure 5.9. Standardized chitin synthase CT scores plotted by development percent at
20°C for (a) all individuals that expressed the gene and (b) in all 958 individuals, with
NA expression values assigned a CT of 50. The red line indicates gene expression in the
CA strain. Blue indicates gene expression in the M1 strain. Few individuals did not
express the gene at this temperature. The CA strain expressed more of cs than the M1
strain, though in the same pattern ............................................................... 106

Figure 5.10. Standardized heat shock protein 60 CT scores plotted by stage. Gene
expression was highest during the first two stages and lowest during the last two
stages ................................................................................................ 107

Figure 5.11. Standardized heat shock protein 60 CT scores plotted by development
percent for all individuals that expressed the gene. The black line is the lowess curve for
all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 335°C. The expression levels of this gene decrease from
hatching, until 60 percent development (early pupation), then increase until eclosion...108

Figure 5.12. Standardized heat shock protein 60 CT scores plotted by development
percent at 335°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The green
line indicates gene expression in the WV strain. The strains expressed the gene in the
same general pattern, though the CA strain had higher expression around 40 percent
development ........................................................................................ 1 09

Figure 5.13. Standardized heat shock protein 60 CT scores plotted by development

percent at 20°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The CA

xvi

strain expressed more of this gene than the M1 strain, at this temperature. Few
individuals did not express the gene at this temperature ..................................... 110

Figure 5.14. Standardized heat shock protein 90 CT scores plotted by stage. Gene
expression was lowest during the postfeeding third instar ................................... 11 1

Figure 5.15. Standardized heat shock protein 90 CT scores plotted by development
percent for all individuals that expressed the gene. The black line is the lowess curve for
all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 335°C. The expression levels of the gene decreased from
hatching, until approximately 25 percent development (when postfeeding larval
development is attained), then increased until eclosion when flies were raised at the high
temperature. At the low temperature, expression was higher than at high temperatures,
and expression was maintained at a steady level after 25 percent development .......... 112

Figure 5.16. Standardized heat shock protein 90 CT scores plotted by development
percent at 335°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The green
line indicates gene expression in the WV strain. Expression decreased until the
postfeeding third instar (around 25 percent) then increased until eclosion. The strains
expressed the gene in the same general pattern, though the time of minimum expression
occurred later in the CA strain ................................................................... 113

Figure 5.17. Standardized heat shock protein 90 CT scores plotted by development
percent at 20°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The CA
strain expressed more of this gene than the MI strain, at this temperature ................. 114

Figure 5.18. Standardized acetylcholine esterase CT scores plotted by stage. Gene
expression decreased through the third instar, then increased until eclosion .............. 1 15

Figure 5.19. Standardized acetylcholine esterase CT scores plotted by development
percent for (a) all individuals that expressed the gene and (b) in all 958 individuals, with
NA expression values assigned a CT of 50. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 335°C. The shift in the lines between (a) and (b) indicates a cluster of
individuals that do not express the gene at the time point where gene expression is at its
lowest levels for this locus. Many individuals did not express ace between approximately
20 and 50 percent development (late third instar through early pupation). This time
period mirrors the period of least ace expression at the high temperature treatment.
Expression increased through development in the low temperature treatments ........... l 16

Figure 5.20. Standardized acetylcholine esterase CT scores plotted by development
percent at 335°C for (a) all individuals that expressed the gene and (b) in all 958
individuals, with NA expression values assigned a CT of 50. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The green

xvii

line indicates gene expression in the WV strain. The shift in these lines between (a) and
(b) indicates a cluster of individuals that do not express the gene at that time point. . . ..1 17

Figure 5.21. Standardized acetylcholine esterase CT scores plotted by development
percent at 20°C for (a) all individuals that expressed the gene and (b) in all 958
individuals, with NA expression values assigned a CT of 50. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The CA
strain expressed more ace than the M1 strain (a). MI was also more likely to not express
ace ................................................................................................... 117

Figure 5.22. Standardized ecdysone receptor CT scores plotted by stage. Gene
expression increased through the postfeeding third instar, then decreased until
eclosion ............................................................................................. 118

Figure 5.23. Standardized ecdysone receptor CT scores plotted by development percent
for all individuals that expressed the gene. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 335°C. The expression levels of this gene increase from hatching, until
approximately 35 percent development (during the postfeeding third instar), then
decrease until eclosion ............................................................................ 119

Figure 5.24. Standardized ecdysone receptor CT scores plotted by development percent
at 335°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The green
line indicates gene expression in the WV strain. The strains expressed the gene in the
same general pattern, though the M1 strain expressed less ecr than the other strains. . ..l 19

Figure 5.25. Standardized ecdysone receptor CT scores plotted by development percent
at 20°C for all individuals that expressed the gene. The red line indicates gene expression
in the CA strain. Blue indicates gene expression in the M1 strain. The CA strain
expressed more of this gene than the M1 strain, at this temperature, though they converge
to the same maximum expression level during the postfeeding third instar ............... 120

Figure 5.26. Standardized resistance to organophosphate 1 CT scores plotted by stage.
Gene expression increased through the postfeeding third instar, then decreased until
eclosion ............................................................................................. 121

Figure 5.27. Standardized resistance to organophosphate 1 CT scores plotted by
development percent for all individuals that expressed the gene. The black line is the
lowess curve for all individuals, the blue line is the curve for expression at 20°C, and the
red line is the curve for expression at 335°C. The expression levels of this gene increase
from hatching until approximately 25 percent development at high temperatures and until
approximately 35 percent development at low temperatures, then gene transcript
abundance decrease until eclosion ............................................................ 122

xviii

Figure 5.28. Standardized resistance to organophosphate 1 CT scores plotted by
development percent at 335°C for all individuals that expressed the gene. The red line
indicates gene expression in the CA strain. Blue indicates gene expression in the MI
strain. The green line indicates gene expression in the WV strain. The strains expressed
the gene in the same general pattern ............................................................ 123

Figure 5.29. Standardized resistance to organophosphate 1 CT scores plotted by
development percent at 20°C for all individuals that expressed the gene. The red line
indicates gene expression in the CA strain. Blue indicates gene expression in the MI
strain. The CA strain expressed more of the gene than the M1 strain, at this temperature,
though both followed the same pattern ......................................................... 124

Figure 5.30. Standardized white CT scores plotted by stage. Gene expression increased
through the postfeeding third instar, then decreased until eclosion ......................... 125

Figure 5.31. Standardized white CT scores plotted by development percent for (a) all
individuals that expressed the gene and (b) in all 958 individuals, with NA expression
values assigned a CT of 50. The black line is the lowess curve for all individuals, the
blue line is the curve for expression at 20°C, and the red line is the curve for expression at
335°C. The shift in the lines between (a) and (b) indicates a cluster of individuals that
did not express the gene at the time point where gene expression is at its lowest levels for
this locus. The expression levels of the gene increased from hatching until approximately
25 percent development at high temperatures and until approximately 35 percent
development at low temperatures, then gene transcript abundance decrease until eclosion.
High temperature treatments were less likely to express the gene .......................... 126

Figure 5.32. Standardized white CT scores plotted by development percent at 335°C for
(a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the CA
strain. Blue indicates gene expression in the M1 strain. The green line indicates gene
expression in the WV strain. The shift in the lines between (a) and (b) indicates that CA
was most likely to not express w and MI was most likely to express the gene ............ 127

Figure 5.33. Standardized white CT scores plotted by development percent at 20°C for
(a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the CA
strain. Blue indicates gene expression in the M1 strain. The CA strain expressed more w
than the M1 strain (a). MI was also more likely to not express w during pupation (the last
half of development) .............................................................................. 128

Figure 5.34. Standardized ultraspiracle CT scores plotted by stage. Gene expression
increased through the postfeeding third instar, then decreased until eclosion ............. 129

Figure 5.35. Standardized ultraspiracle CT scores plotted by development percent for all

individuals that expressed the gene. The black line is the lowess curve for all individuals,
the blue line is the curve for expression at 20°C, and the red line is the curve for

xix

expression at 335°C. The expression levels of the gene increased from hatching until
approximately 35 percent development, then gene transcript abundance decreased until
eclosion ............................................................................................. 130

Figure 5.36. Standardized ultraspiracle CT scores plotted by development percent at
335°C for all individuals that expressed the gene. The red line indicates gene expression
in the CA strain. Blue indicates gene expression in the M1 strain. The green line
indicates gene expression in the WV strain. The strains expressed the gene in the same
general pattern ...................................................................................... 130

Figure 5.37. Standardized ultraspiracle CT scores plotted by development percent at
20°C for all individuals that expressed the gene. The red line indicates gene expression in
the CA strain. Blue indicates gene expression in the M1 strain. The CA strain expressed
more of the gene than the M1 strain, though both followed a similar pattern .............. 131

Figure 5.38. Standardized slalom CT scores plotted by stage. Gene expression increased
through the third instar, then decreased until eclosion ....................................... 132

Figure 5.39. Standardized slalom CT scores plotted by development percent for (a) all
individuals that expressed the gene and (b) in all 958 individuals, with NA expression
values assigned a CT of 50. The black line is the lowess curve for all individuals, the
blue line is the curve for expression at 20°C, and the red line is the curve for expression at
335°C. The shift in the lines between (a) and (b) indicates a cluster of individuals that
did not express the gene at the time point where gene expression was at its lowest levels.
The expression levels of the gene increased from hatching until approximately 25 percent
development, then gene transcript abundance decreased until approximately 50 percent,
and then increase again until eclosion. High temperature treatments were less likely to
express the gene .................................................................................... 133

Figure 5.40. Standardized slalom CT scores plotted by development percent at 335°C
for (a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the CA
strain. Blue indicates gene expression in the M1 strain. The green line indicates gene
expression in the WV strain. The shift in the lines between (a) and (b) indicates that CA
was most likely to not express sll and MI was most likely to express the gene ........... 134

Figure 5.41. Standardized slalom CT scores plotted by development percent at 20°C for
(a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the CA
strain. Blue indicates gene expression in the M1 strain. The strains expressed the gene in
a similar pattern (a). MI was more likely to not express sll ................................. 134

Figure 5.42. The combined standardized expression (CT score) patterns for all nine
genes throughout L. sericata immature development. The lowess curves for each gene
represent expression for all individuals that expressed the gene. The combined
expression patterns of the genes were used to predict blow fly age with GAMs ......... 136

XX

Figure 5.43. Diagnostic plot for Model 1, which used developmental stage to predict L.
sericata development percent (on a scale of 0—1). A gamma distribution was used with
this model. Descriptions for each panel type can be found in Figure 3.4. Predicted
(Fitted) values represent a range of true (Response) values. Error increased with age, and
gaps exist between predictions for each developmental stage ............................... 137

Figure 5.44. Diagnostic plot for Model 15, which used developmental stage, strain,
temperature, length, and weight to predict L. sericata development percent (on a scale of
0—1). A gamma distribution was used with this model. Descriptions for each panel type
can be found in Figure 3.4. Predicted (Fitted) values represent a range of true (Response)
values. Error increased with age, and a gap exists between predictions for postfeeding
third instar and pupal ages ........................................................................ 138

Figure 5.45. Diagnostic plot for Model 18, which used developmental stage, strain,
temperature, length, weight, binary gene expression for four genes, and quantitative gene
expression for nine genes, to predict L. sericata development percent (on a scale of 0—1).
A gamma distribution was used with this model. Descriptions for each panel type can be
found in Figure 3.4. Predicted (Fitted) values represent a range of true (Response)
values. The increase in error with age has diminished compared to other models and the
gap between predictions for postfeeding third instar and pupal ages has shrunk ......... 139

Figure 5.46. Diagnostic plot for Model 19, which used the same parameters as Model 18,
to predict L. sericata development percent (on a scale of 0—1 ), but a gaussian distribution
was used. Descriptions for each panel type can be found in Figure 3.4. Predicted (Fitted)
values represent a range of true (Response) values. The increase in error with age has
diminished compared to Model 15 and the gap between predictions for postfeeding third
instar and pupal ages has shrunk. However, compared to Model 18, this model exhibits
more error in predictions of younger individuals ............................................. 140

Figure 5.47. Diagnostic plot for Model 20, which used the same parameters as Model 18,
to predict L. sericata development percent (on a scale of 0—1 ), but a gaussian distribution
was used with this model and all genes expression scores were anchored against hsp60
expression. Descriptions for each panel type can be found in Figure 3.4. Predicted
(Fitted) values represent a range of true (Response) values. The increase in error with
age has diminished compared to Model 15 and the gap between predictions for
postfeeding third instar and pupal ages has been eliminated. However, compared to
Model 18, this model exhibits more error in predictions for younger individuals, though it
is an improvement over predictions made with Model 19 .................................... 141

Figure 5.48. Diagnostic plot for Model 21, which used the same parameters as Model 18,
to predict L. sericata development percent (on a scale of 0—1), but a gaussian distribution
was used with this model and all genes expression scores were anchored against length
measurements. Descriptions for each panel type can be found in Figure 3.4. Predicted
(Fitted) values represent a range of true (Response) values. The increase in error with
age has diminished compared to Model 15 and the gap between predictions for

xxi

postfeeding third instar and pupal ages has been eliminated. However, compared to
Model 18, this model exhibits more error in predictions for younger individuals, though it
is an improvement over predictions made with Model 19 .................................... 143

Figure 5.49. Diagnostic plot for Model 22, which used the same parameters as Model 21,
to predict L. sericata development percent (on a scale of 0—1), but parameters that were
non-significant predictors of age in Model 21 (length, weight, cs, w, strain) were
removed. Descriptions for each panel type can be found in Figure 3.4. Predicted (Fitted)
values represent a range of true (Response) values. The increase in error with age has
diminished compared to Model 15 and the gap between predictions for postfeeding third
instar and pupal ages has been eliminated. However, compared to Model 18 this model
exhibits more error in predictions for younger individuals, though it is an improvement
over predictions made with Model 19. Removing strain, cs, and w, had little effect on the
predictions made with the model compared to predictions made with Model 21. . . . . ....144

Figure 5.50. Diagnostic plot for Model 23, which used the same parameters as Model 22,
to predict L. sericata development percent (on a scale of 0—1 ), except that temperature
was removed from this model. Descriptions for each panel type can be found in Figure
3.4. Predicted (Fitted) values represent a range of true (Response) values. Removing
temperature had little effect on the predictions made with this model compared to
predictions made with Model 22 ................................................................. 145

Figure 5.51. Diagnostic plot for postfeeding third instar incorporating
stage/length/weight/ temp/strain. Descriptions for each panel type can be found in Figure
3.4. Note the extensive scatter in response vs. fitted values ................................. 147

Figure 5.52. Diagnostic plot for pupae incorporating stage/length/weight/ temp/strain.
Descriptions for each panel type can be found in Figure 3.4. Note the extensive scatter in
response vs. fitted values ......................................................................... 148

Figure 5.53. Diagnostic plot for postfeeding third instar incorporating
stage/length/weight/ temp/strain and expression data. Descriptions for each panel type
can be found in Figure 3.4. Note the tightening in response vs. fitted values over Figure
5.51 ................................................................................................... 149

Figure 5.54. Diagnostic plot for postfeeding third instar incorporating
stage/length/weight/ temp/strain and expression data. Descriptions for each panel type
can be found in Figure 3.4. Note the tightening in response vs. fitted values over Figure
5.52 ................................................................................................... 150

Figure 6.1. Temperature plots for the ambient temperature experiment. Temperatures
spiked in the afternoon as sunlight was directly on the terrarium ........................... 172

Figure 6.2. The regression of ADH (base 10) percents versus percent development in
hours. The values were strongly correlated, with an R-squared and slope almost 1.

xxii

However, the 1.2448 shift up the Y axis indicates that ADH is a better measure of
developmental progress ........................................................................... 172

Figure 6.3. Predicted versus true development percents for all 90 individuals in the blind
study. Predictions were made with GAMs that did not use gene expression information.
The four points in the gray bar represent the four larvae that failed to pupate, and were
eliminated from analyses in this section ........................................................ 174

Figure 6.4. Predicted versus true development percents for the 86 normally developing
individuals in the blind study. Predictions were made with GAMs that used unanchored
gene expression terms ............................................................................. 176

Figure 6.5. Predicted versus true development percents for 49 normally developing
individuals in the blind study. Predictions were made with GAMs that used unanchored
gene expression terms and were Type 2 models ............................................... 176

Figure 6.6. Predicted versus true development percents for 71 normally developing
individuals in the blind study. Predictions were made with GAMs that used length
anchored gene expression terms ................................................................. 177

Figure 6.7. Predicted versus true development percents. Predictions were made with
GAMs that used length anchored gene expression terms and were Type 2 models ...... 177

Figure 7.1. Standardized gene CT scores for hsp90, plotted by stage, for all individuals
from the previous section, for the 55 non-pupator individuals from this study, and the
individuals in the blind study. The left panel represents expression from a previous
section (Chapter 5), with results from the 55 non-pupators (NP) placed next to
postfeeding third instars (3rdPF). The right panel represents gene expression for the
individuals in the blind study (Chapter 6), including the 4 non-pupators. lSt = first instar.
2nd = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar. NP =
non-pupator or “Peter Pan” larvae. On average, hsp90 was expressed at higher levels in
non-pupators than in postfeeding third instars ................................................. 186

Figure 7.2. Standardized gene CT scores for rop-I, plotted by stage for all individuals
fi'om the previous section, for the 55 non-pupator individuals from this study, and the
individuals in the blind study. The left panel represents expression from a previous
section (Chapter 5), with results from the 55 non-pupators (NP) placed next to
postfeeding third instars (3rdPF). The right panel represents gene expression for the
individuals in the blind study (Chapter 6), including the 4 non-pupators. 1St = first instar.
2nd = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar. NP =
non-pupator. On average, rop-I was expressed at lower levels in non-pupators than in
postfeeding third instars, though this pattern was not apparent in the blind study ........ 187

Figure 7.3. Standardized gene CT scores for sll, plotted by stage, for all individuals from

the previous section, for the 55 non-pupator individuals from this study, and the
individuals in the blind study. The left panel represents expression from a previous

xxiii

section (Chapter 5), with results from the 55 non-pupators (NP) placed next to
postfeeding third instars (3rdPF). The right panel represents gene expression for the
individuals in the blind study (Chapter 6), including the 4 non-pupators. lSt = first instar.
2nd = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar. NP =
non-pupator. On average, sll was expressed at lower levels in non-pupators than in
postfeeding third instars ......................................................................... 188

Figure 7.4. Standardized gene CT scores for usp, plotted by stage, for all individuals
from the previous section, for the 55 non-pupator individuals from this study, and the
individuals in the blind study. The left panel represents expression from a previous
section, with results from the 55 non-pupators (NP) placed next to postfeeding third
instars (3rdPF). The right panel represents gene expression for the individuals in the blind
study, including the 4 non-pupators. 1St = first instar. 2"d = second instar. 3lrd = feeding
third instar. 3rdPF = postfeeding third instar. NP = non-pupator. On average, usp was
expressed at lower levels in non-pupators than in postfeeding third instars .............. 189

Figure 8.1. Boxplots of minimum development time comparisons among strains at
335°C in 2006. The CA strain developed more slowly compared to the other
strains ................................................................................................ 202

Figure 8.2. Boxplots of minimum development times comparisons among strains at 20°C
in 2006. The strains exhibited similar median and fastest development times, but CA had
a faster mean development time as compared to the other strains, which had some

minimum development times that were longer than 540 h ................................... 203

Figure 8.3. A comparison of average minimum development times for each strain in
2005 (a) and in 2006 (b). CA is red, MI is blue, WV is green. The M1 mean minimum
development time was consistently faster than WV, though their development times were
comparable. The CA mean was fastest at 20°C and slowest at 335°C in both years.
Development was faster at 335°C ............................................................... 204

Figure 8.4. Boxplots of average weights compared among the strains at 335°C in 2006.
On average, CA was larger than Ml, which was larger than WV ........................... 206

Figure 8.5. Boxplots of average weights compared among the strains at 20°C in 2006.
On average, CA was larger than MI, which was larger than WV ........................... 207

Figure 8.6. A comparison of average lengths for each strain in 2005 (a) and in 2006 (b).
CA is red, MI is blue, WV is green. The CA mean length was the largest in both years.
WV was the smallest in 2006, but MI was smallest at 335°C in 2005. Lengths were
greater for 20°C treatments ....................................................................... 208

Figure 8.7. A comparison of average weights for each strain in 2005 (a) and in 2006 (b).

CA is red, MI is blue, WV is green. The CA mean weight was the largest and WV was
the smallest. Weights were greater for 20°C treatments ..................................... 209

xxiv

LIST OF APPENDICES

Appendix 2.1. Treatments and duration of the immature lifecycle and individual stages
from all cohorts of Lucilia sericata .............................................................. 232

XXV

CHAPTER 1: INTRODUCTION AND BACKGROUND

The Importance of Blow Flies

Flies from the family Calliphoridae (Diptera), commonly known as blow flies, are
parasites, parasitoids, and primary successional species on canion, which regularly
interact with people in several key areas of human activity. Calliphorids are parasites of
sheep and cattle, resulting in financial damage that was significant enough, in one
instance, to warrant the development of a sterile male release program to eliminate the
primary screwwonn fly Cochliomyia hominivorax from Texas and Florida (Crystal and
Ramirez 1975). Blow flies also influence human health. Maggot therapy has been
employed at some hospitals to efficiently debride wounds (Sherman and My-Tien 1995).
More importantly, they detrimentally affect human health, as they can transmit diseases
(Faulde et al. 2001) and cause myiasis (parasitization) of humans (Sherman 2000).
Calliphorid flies are also useful to forensic scientists and medical examiners. The
immature forms of these flies are frequently found at death scenes in association with
bodies, providing valuable information for estimating a postmortem interval (PMI) in
death investigations (Haskell and Catts 1990; Greenberg and Kunich 2002). Due to these
interactions, a more detailed knowledge of blow fly development has the potential to
promote the beneficial interactions, and ameliorate negative interactions with humans, by
enabling the production of tools that can be used to better describe and control blow fly
development.

As genomic sciences progress, detailed knowledge of development in non-model

organisms will help to answer questions in the emerging field of comparative genomics.

Evolutionary biologists compare genomes to understand how selection acts on them and
to determine whether adaptations among similar organisms are due to gene sequence
evolution or differences in the regulation of those genes (Yeates and Wiegmann 2005,
Khaitovich et a1. 2006). Medical researchers also benefit from comparative genomics, as
they can learn how to apply results from model organism research to human medicine.
Both fields will increasingly rely on genomic comparisons to answer questions about how
a gene is regulated in similar species, and the consequences of differential regulation
(Collins et a1. 1998).

Drosophila melanogaster is a well-established model organism that is poised to
be a central pillar of comparative genomic research. In the near future, twelve
Drosophila genome sequences will be complete (www.flybase.org), allowing for multi-
species comparisons (Kulathinal and Hartl 2005). Such comparisons have already
revealed that the conservation of a seven-striped pattern (known as a pair-rule pattern) of
gene expression in the embryos of this genus is due to different variants of the same
regulatory mechanisms (Ludwig et al. 1998). Essentially, each species has been shown to
use the same set of gene expression-inducing and inhibiting proteins, in subtly different
ways, to achieve the same embryonic pattern.

Just as comparisons within the Drosophila genus are valuable, comparisons of
gene regulation between Drosophila and other higher flies (including blow flies) have
revealed insight into the nature of gene expression evolution, including bicoid (McGregor
2005), Wingless (Mellenthin et al. 2006), and slalom (Ali et a1. 2005). The sequences,
timing, and locations of Wingless and slalom expression are conserved between

Drosophila and blow flies. Similarly, bicoid serves the same anterior determination

function in blow flies as it does in Drosophila, though the blow fly genes cannot fully
rescue Drosophila mutations due to regulatory sequence evolution. Likewise, a
comparison of gene expression within the Calliphoridae has shown that acrostichal bristle
differences between Calliphora vicina and Protophormia terraenovae are due to
heterochrony in the expression of proneural genes, with larger bristles resulting from
earlier initiation of the genes (Skaer et a1. 2003). Given the advances that have already
arisen from comparisons of developmentally regulated genes in blow flies to homologous
genes in other species, it is likely that any additional knowledge of blow fly development,
at a molecular level, will further contribute to the understanding of gene expression and

phenotype evolution among higher flies, and thus among animals in general.

Blow Fly Development

The interactions of blow flies with humans occur primarily with immature
developmental stages (the one exception is the transmission of disease, which is due to
adult flies feeding on filth and human food sources). Also, it is necessary to have
knowledge of the development of an organism, before it is possible to understand how
that process evolved. Given the potential importance of immature blow flies to forensic,
medical, veterinary, and evolution sciences, a detailed understanding of blow fly
development is required.

The lifecycle of a blow fly progresses predictably (Catts and Haskell 1990) as
shown below (Figure 1.1). Female blow flies lay eggs around orifices or wounds. The
eggs hatch into larvae, which feed on a body and grow through three larval instars. Each

instar is separated by a molt of the cuticle that enables further larval growth. The

appearances of the anterior and posterior spiracles differ among instars, making them
easily distinguishable from each other. During the third instar, larvae cease feeding and
(usually) leave the body to form a hardened cuticular structure known as a puparium.
Within the puparium, the fly experiences metamorphosis and eventually ecloses as an
adult by breaking the lid off of the puparium with an extendable sac called a ptilinum.

An immature lifecycle with three instars is a hallmark of cyclorrhaphan or “higher
flies”, like the Drosophilidae and Calliphoridae, which pupate within a puparium
(McAlpine 1989, Yeates and Wiegmann 2005). Similarities in development among
cyclorrhaphans can be used to infer molecular processes that drive developmental
changes in blow flies. As all flies develop, they must incorporate information from
physiological and environmental signals to determine when it is appropriate to molt or
metamorphose. They are interpreted in the brain where neurosecretory cells respond to
information provided by the insulin pathway (Shingleton 2005), temperature, and
photoperiod (Flatt et a1. 2005); these stimuli ultimately cause the cells to induce
prothroacitropic hormone (PTTH) secretions from the corpora allota (Kalthoff2001).
The release of PTTH directs the prothoracic glands to produce a steroid hormone,

ecdysone, and release it into the hemolymph.

 

Figure 1.1. The lifecycle of a blow fly. Females lay a clutch of eggs on carrion (I),
typically either at an orifice or on a wound. First larval instars hatch from the eggs (2)
and feed until they molt into the second instars, which feed and grow until they molt into
third instars. Feeding third instars (3) can be identified by three slits in their posterior
spiracles, the visibility of blood in their crop, and by the fact that they are actively
feeding. Postfeeding third instars (4), which lack visible tissue in the crop, cease feeding
and leave the food source to form a puparium. In the puparium (5), the fly
metamorphoses and eventually ecloses as an adult by breaking the lid off of the puparium
with its ptilinum (top and middle of 5). After a few hours, the adult cuticle and wings
achieve a normal appearance and the fly is fully capable of flight (6).

The release of ecdysone initiates larval molt or metamorphosis by activating a
series of gene expression changes that were first observed in Drosophila. Experiments
by Ashbumer (1974) demonstrated that when polytene chromosomes were treated with
ecdysone, they developed thickened regions, called “puffs”, which appeared in a
temporally specific order. Early puffs were shown to respond directly to ecdysone

treatments, while late puffs require the expression of genes (Kalthoff 2001 ). Since the

early Drosophila experiments, there have been significant discoveries as to how ecdysone
affects development at a molecular level and many of the genes expressed within puffs
are currently known (Kalthoff 2001). For instance, the ecdysone receptor (ecr) and
ultraspiracle (usp) genes, which are both specific types of transcription factors known as
nuclear hormone receptors, form a heterodimer that when bound to ecdysone, activates
the expression of the early genes (Henrich and Brown 1995). ' In fact, multiple nuclear
hormone receptors in Drosophila have been shown using northern blot analysis to
respond to ecdysone pulses (Sullivan et al. 2002). Recent genomic studies in D.
melanogaster indicate that thousands of genes are regulated throughout development
(Arbeitrnan et al. 2002) and thousands of genes respond to changes in ecdysone
concentrations (Beckstead et al. 2005). As the genomic age advances, most (possibly all)
of the identities of the loci that are expressed within chromosomal puffs will be revealed,
as well as the genes that interact with them, enabling an improved understanding of how
flies develop.

The response of a developing fly to ecdysone pulses, however, is not uniform
among stages. Juvenile hormone (JH) modulates the nature of the response to an
ecdysone pulse, through a currently unknown mechanism (Henrich and Brown 1995,
Flatt et al. 2005). Like PTTH, JH is released by the corpora allota in response to
temperature, photoperiod, and hormonal signals that are relayed from the nervous system
(Kalthoff2001, Flatt et al. 2005). When J H is in high concentrations, an ecdysone pulse
initiates the next larval molt. However, when an ecdysone pulse occurs in the absence of

JH, metamorphosis is initiated.

Though development is predictable, environmental factors are known to affect the
development of flies and this plasticity can occur via several mechanisms. Temperature
affects blow fly development times, with lower temperatures causing slower development
(e.g. Anderson 2000, Grassberger and Reiter 2001). Blow fly larvae also diapause (an
environmentally induced delay of development) at low temperatures and under short
photoperiods (Tachibana and Goto 2004, Tachibana et al. 2005), though different
populations can exhibit different propensities for this behavior (Greenberg and Kunich
2002). In Drosophila, insufficient diet has also been shown to cause delays in
development as well as small body size. Nutritional plasticity is thought to be influenced
by the insulin receptor pathway (Shingleton et al. 2005). Temperature sensitive insulin
receptor (Inr) mutations can affect development time if their products are inactivated
before the acquisition of a critical size (when a larva has accumulated enough energy to
pupate). If the larva has fed enough that it has sufficient energy to complete
metamorphosis, however, the inactivation of Inr protein will cause a decrease in body
size. Similar delays in blow fly development have been observed when L. sericata and
Calliphora vicina (Diptera: Calliphoridae) were exposed to different diets (Kaneshrajah
and Turner 2004, Clark et al. 2006, Tarone and Foran 2006), indicating the potential for a
similar response in these species.

Unfortunately, little is known about the molecular control of blow fly
development. Though similarities in developmental biology among cyclorrhaphan flies
can be inferred from research in D. melanogaster and other dipterans (reviewed by
Henrich and Brown 1995), the specific molecular controls of development are not always

the same across all species. For instance, differences in ultraspiracle expression between

Chironomus tentans and other insects have been observed, with the C. tentans gene
located within a late puff, indicating a potentially novel regulatory mechanism in the
ecdysone response of this fly (Henrich and Brown 1995). There is also a difference
between L. sericata and Sarcophaga bullata for hsp90 expression during diapause, with
S. bullata down-regulating the gene during diapause while expression of the L. sericata
gene is unchanged (Tachibana et al. 2005). However, many genes have comparable
functions and regulation in Cyclorrhaphan flies. There are similarities between the
Calliphoridae and Drosophilidae in bicoid protein functions (McGregor 2005), which
determines the anterior region of embryos, though Calliphorid bicoid fails to fully rescue
Drosophila mutants due to differences in regulatory sequences among species. Two
other genes, Wingless (wg) (Mellenthin et al. 2006) and slalom (sll) (Ali et al. 2005), have
highly conserved gene expression patterns in L. sericata and D. melanogaster. Cell
signaling, as directed by wg, occurs in the same tissues and in the same patterns on wing
imaginal discs and sll is expressed in the salivary glands of larvae in both species.
Observations of comparable and dissimilar gene expression patterns in Cyclorrhaphan
flies highlight a need for specific information regarding the molecular genetics of blow
fly development to identify when molecular results can be generalized across all flies and

when they are specific to a lineage.

Evolution and Ecology of Blow Flies
To fully understand blow fly development, it is necessary to interpret the results
of developmental research in the context of blow fly evolution and ecology. The

phylogenetic relationships among the Diptera (Figure 1.2) were recently and extensively

reviewed by Yeates and Wiegmann (2005). The earliest known Diptera are estimated to
have diverged from other insects 233 Mya based on fossil evidence, and 248—283 Mya
based on molecular evidence. The most primitive flies are the Nematocera, which
includes mosquitoes and crane flies. They typically have 4—8 larval instars and pupae
with visible appendages. They are generally aquatic and depend on water/moisture for
reproduction and/or survival. The next infraorder of flies is the Brachycera, which are
estimated to have diverged approximately 208 Mya (fossil) to 194—241 Mya (molecular).
They are a monophyletic group as compared to the Nematocera, with many traits, such as
male genital rotation and the number of ganglia in the nervous system, that are
intermediate to the Cyclorrhapha and Nematocera.

The Cyclorrhapha are another well-supported monophyletic infraorder. They
have reduced larval head structures and three larval instars that are both presumed to be
adaptations that are related to pupation within a puparium. They diverged from other
flies approximately 130 Mya (fossil) to 122—172 Mya (molecular). Within this group are
the Schizophora, which eclose from the puparium with a ptilinum. The Schizophora
include the Drosophilidae and Calliphoridae and are estimated to have diverged 80 Mya
(fossil) to 71—1 13 Mya (molecular). However, the Drosophilidae belong to the
Acalypterata, whereas the Calliphoridae belong to the Calypterata. The major distinction
between the two clades is the presence of a calypter (a specialized wing structure attached
to the basal portion of the wing) in the Calypterata. The Drosophila and Musca (the
latter also belonging to the Calypterata) genera are estimated to have diverged
approximately 70 Mya (fossil) to 29—76 Mya (molecular). Based on the similarities in

highly derived dipteran characteristics, molecular and developmental data from the

Drosophilidae and other cyclorrhaphan flies are likely to be directly relevant to blow fly
biology, and may be used as a guide when data do not exist for the Calliphoridae.

Many of the Calliphoridae have successfully occupied a niche as the primary
successional species on carrion. Blow flies are capable of locating carrion within hours
of death (Catts and Haskell 1990), allowing them to regularly find and dominate a rich
but ephemeral food source. The sporadic occurrence of carrion results in a potential
tradeoff between body size and age at metamorphosis. Such a tradeoff has been studied
in spadefoot toads (Morey and Reznick 2000, Day and Rowe 2002, Gomez-Mestre and
Bucholz 2006), as well as in mites (Plaistow 2004) and butterflies (Kingsolver 2000,
Kingsolver et al. 2001) and it seems that blow flies are susceptible to similar ecological
forces as these species. For example, spadefoot toads are under intense selection to
complete development before their vernal ponds dry up, mirroring the need for blow flies
to utilize carrion before it disappears. However, selection has also been shown to favor
larger adult toads that require longer feeding periods, resulting in a tradeoff between the
two phenotypes. Additionally, the tradeoff between development time and body size is
associated with the evolution of plasticity in both traits that has also been observed in
blow flies. The Calliphoridae exhibit a range of minimum development times and body
sizes (Kamal 1958, Anderson 2000) and limited nutrition can both delay development
and lead to smaller body sizes (Kaneshrajah and Turner 2004, Clark et al. 2006). Given
the similarities between the toad and fly systems, information on blow fly development

may also contribute to the study of the evolution of age and size at metamorphosis.

10

Nematocera

 

 

 

 

  
   
 
   
  
  

 

 

 

 

 

 

 

Tlpuloldea ’ in,“
Tabanomorpha Stratiomyomorpha // .
Xylophaghomorpha
Brachycera Tabanomorpha / \
Nemectrlnoidea
Muscomorpha Aslloldea I"
Empldoldea P I
Eremoneura Asa-“la (part) ‘_
Cyclorrhapha Phoroldea flea
Hippobosoordea . ' ,,...:'
Musooidea z"
Schizophora ‘_
Oestroidea fl
Acalyptrata ‘

Figure 1.2. A phylogeny of the Diptera. Taken from Yeates and Wiegmann (2005).
The Nematocera are the most primitive flies and include mosquitoes and crane flies.
Cyclorrhapha are the “higher flies”, which have three larval instars and form a puparium.
The Schizophora, which include the Calliphoridae and Drosophilidae, have a specialized
structure, a ptilinum, which is used to break the lid off of the puparium during eclosion.
The Drosophilidae belong to the Acalyptrata, as they lack calypters on their wings. The
Calliphoridae belong to the superfamily Oestroidea, which is part of the Calyptrata.
Evolutionarily, the Drosophilidae and Calliphoridae are closely related families within
the Diptera, which means that they are expected to share many molecular, developmental,
and morphological similarities including three larval instars, the formation of a puparium,
and the presence of a ptilinum.

An ecological factor that has been shown to affect the biology of body size and
development time of many organisms is latitude. This is best demonstrated by
Bergmann’s rule, which states that endothermic organisms are larger in cooler climates

(Ridley 1996). However, the same concept applies to insects in many cases. Latitudinal

clines for several traits and genotypes are well documented in the Drosophila (reviewed

11

by Powell 1997). Dobzhansky demonstrated that chromosomal inversions in populations
of D. pseudoobscura and D. persimilis were geographically distributed along a latitudinal
grade. Similarly, the alcohol dehydrogenase locus has been shown to vary in such a
manner, as has body size in populations of D. melanogaster on multiple continents.
Latitudinal variation occurs in other dipteran species as well. For example, Rhagoletis
pomonella (Diptera: Tephritidae) exhibits latitudinal differences in development time
(Feder et al. 2003 ), highlighting the potential for latitudinal effects on blow fly
development.

Given the potential effects of latitude and other ecological factors on blow fly
development, adaptation to the local environment must be considered when evaluating
developmental traits for specific populations. Though differences in body size and
development time within the Calliphoridae have been briefly mentioned in the literature
(Greenberg 1991, Grassberger and Reiter 2001), the potential for genetic differences for
these traits, due to local adaptations, have not, until recently, been discussed as potential
lines of future research on blow fly development (Tarone and Foran 2006, Goff 2007).
Differences among recorded development times for L. sericata (Kamal 195 8, Greenberg
1991, Anderson 2000, Grassberger and Reiter 2001) could be explained by genetic
differences among strains, though the interpretation of variation in development time is
clouded by potential plasticity caused by variable rearing conditions among the studies.
The observed variation in development times indicates a need for quantitative genetic
comparisons among populations of blow flies to differentiate between genetic and

environmental effects on body size and development time traits in the Calliphoridae.

l2

Forensic Entomology

The main purpose of the dissertation research presented in the following chapters
was to improve the precision and accuracy of PMI estimates made with entomological
evidence, through the application of molecular and quantitative genetic methods.
Forensic entomology is a powerful tool that can aid in estimating a minimum PMI during
death investigations (Catts and Haskell 1990; Greenberg and Kunich 2002) and has long
enabled investigators to use tables that outline the minimum development times of each
immature stage (e.g. Kamal 195 8) to estimate the ages of flies associated with remains.
Blow fly development is dependent on temperature (e. g. Anderson 2000, Grassberger and
Reiter 2001), so if investigators know the developmental stage of the oldest flies
collected as evidence and have historical weather data for the scene, they can determine
the window of time necessary for a species to reach that stage. The time since insect
colonization of remains is generally assumed to be the minimum PMI.

The standard method for predicting a PMI from insect evidence has remained
relatively unchanged for decades (Catts and Haskell 1990). The static nature of the
approach is due, in part, to the general success of the method, which has been upheld
numerous times in American and international courts (Greenberg and Kunich 2002).
However, its continued acceptance has resulted in the persistence of a number of caveats
associated with PM] predictions based on fly evidence. One of these is that, since each
developmental stage gets progressively longer through fly development, a PMI prediction
obtained from later stages will necessarily incorporate a much larger window of time.
For example, Kamal (1958) found that, for the species L. sericata growing at 267°C, the

second larval instar lasted 9-26 hours while the longest stage, pupation, was 5—11 days.

13

At lower temperatures pupation can be even longer (Anderson 2000; Grassberger and
Reiter 2001), and age/PMI estimates must be correspondingly broad.

One method for generating a more precise age estimate within developmental
stages is to include body size in the PMI prediction process. As blow fly larvae feed,
they increase in size in a generally linear fashion, with relatively little variance (Wells
and Kurahashi 1994, Grassberger and Reiter 2001; Greenberg and Kunich 2002). Thus,
during the feeding stages linear regression can be used to refine age estimates. However,
the approach highlights the second caveat: larvae shrink when feeding ceases during the
third instar (Wells and Kurahashi 1994, Grassberger and Reiter 2001; Greenberg and
Kunich 2002) and exhibit a larger variance in body size than previous stages.
Additionally, pupae do not change in size as they age. These facts mean that the last two
(and longest) developmental stages provide relatively imprecise (though generally
accurate) PMI estimates, resulting from their long durations, variance in body size, or
unchanging body size.

The last caveat associated with the status quo approach for predicting blow fly
age is that it can sometimes be difficult to distinguish between feeding and postfeeding
third instar larvae. The determination of a third instar state is made by observing the
visibility of tissue in the crop (Figure 1.1) and the cessation of feeding, which can be
difficult in some cases (reviewed by Anderson 2000). If investigators do not note the
feeding behavior of larvae when collected from a body, the only physical evidence a
forensic entomologist will have to work with is the visibility of tissue in the crop.
Several factors, including starvation of larvae and the leakage of crop contents into

storage solution over time can make distinguishing between the feeding and postfeeding

l4

stages of the third instar difficult or impossible if crop contents alone are relied upon
(Anderson 2000). The inability to identify the developmental state of a third instar larva
will result in a larger window associated with an age estimate, as both stages must be
included in the PMI estimate.

The shortcomings of predicting blow fly age based on developmental stage and
body size may be addressed by including new information in the PMI prediction process.
With the arrival of the genomic age biologists have developed new tools that enable them
to assay gene expression levels at relatively low cost. From these, a detailed
understanding of gene regulation has emerged (reviewed by Kalthoff 2001). As
eukaryotes (including blow flies) develop, a variety of proteins are required at various
times, thus the cellular transcriptional machinery initiates the expression of RNA from
many genes in a temporally variable pattern. Given the highly specific temporal and
spatial requirements of gene expression during development, the expression levels of
these genes may be useful predictors of age as they are up- or down-regulated in a
reliable fashion, and thus have the potential to aid forensic entomologists in more
accurately estimating blow fly age.

The developmental similarities among Cyclorrhaphan flies mean that genes
known to vary throughout the development of D. melanogaster are excellent a priori
candidates for study in the context of forensic entomology. Two recent genomic studies
in Drosophila (Arbeitrnan et al. 2002 and Beckstead et al. 2006) demonstrated that
thousands of genes have predictable and temporally variable gene expression. Of these,
there are myriad gene expression patterns that can be used to indicate different points in

development. For example, in D. melanogaster, the gene Amalgam is expressed at its

15

highest levels during early pupation, while CG17814 is expressed at its highest levels
during late pupation (Arbeitrnan et al. 2002). Knowing the expression levels of both
genes could help distinguish among early, middle, and late pupal development.

Though gene regulation in D. melanogaster demonstrates a theoretical capacity to
predict blow fly age, gene expression profiles must be produced in a forensically useful
blow fly species to enable precise PMI predictions. The species studied in this research
was L. sericata; chosen because it is a common fly encountered in forensic entomology
that is globally distributed (Greenberg 1991). The Lucilia genus has also been included
in multiple molecular studies, mostly due to the economic effect of L. cuprina
infestations of Australian sheep (East and Eisemann 1993). This meant that gene
sequence information could be obtained from the public domain, or easily sequenced in
L. sericata by designing PCR primers from L. cuprina sequence, thereby limiting the
effort spent on acquiring gene sequences. By focusing on Lucilia genes, which were
known to vary throughout Lucilia and/or Drosophila development, it was possible to

assess the development of L. sericata in terms of nine gene expression profiles.

The Use of Developmental Plasticity and Gene Expression for Blow Fly Pest Control
The expression profiles produced in this research also have agricultural and
human health implications. Four of the genes studied (chitin synthase, acetylcholine
esterase, ecr, and usp) encode proteins that are targets of insecticides (Ware and
Whitacre 2004) and alleles of another (resistance to organophosphate-I , an alpha
carboxylesterase) can affect the toxicity of organophosphates (Newcomb et al. 2005).

Any variation in the expression of these genes during development or among strains may

16

be useful in identifying specific insecticides to use for targeted control of a specific
developmental stage or strain. Additionally, certain temperature treatments will likely
result in differential susceptibility to insecticides if temperature is shown to change
expression profiles of target genes. Knowledge of temperature variation in insecticide
target expression can identify an optimal season to apply specific insecticides and help
insect control personnel maintain facilities (such as barns) at an optimal temperature to
maximize the effectiveness of an insecticide treatment. Armed with data for variation in
insecticide-targeted gene expression, it should be possible to develop insect control
programs that maximize the effectiveness of insecticides, thus limiting the amount
needed to control a pest species.

Knowledge of factors affecting variation in body size may also be useful in
limiting the effects of myiasis events on sheep and cattle. L. cuprina extracts have been
used to elicit an immune response to myiasis that decreases the amount of biomass lost to
parasitization by limiting larval size (East and Eisemann 1993), thereby ameliorating
economic damages. Other factors that affect body size of agriculturally important blow
flies could be used in place of, or in conjunction with, fly extracts to further limit the
amount of damage they caused. Additionally, such an approach could be incorporated
into an integrated pest management scheme that combines the manipulation of

developmental plasticity with insecticide applications to better control pest species.

Research

The dissertation is divided into multiple chapters, each detailing experiments

designed to better understand L. sericata development through the application of

17

molecular and quantitative genetic methods. The first set of experiments revealed
conditions that affected plasticity in development stemming from variation in protocols
used for rearing flies. Four earlier publications contain data on the laboratory growth of
L. sericata that outline various growth rates for the species (Kamal 1958; Greenberg
1991; Anderson 2000; Grassberger and Reiter 2001). The authors used different rearing
procedures and populations of flies, making observed developmental variation among the
studies impossible to dissect. Likewise, no author compared grth under laboratory
conditions to growth on carrion. Hence a series of experiments was undertaken to
determine how the different rearing methods described in the literature affected
development, and which combination of rearing conditions best mimicked grth on
cadavers.

The second set of experiments was devoted to understanding grth differences
among strains and between replicates raised at different temperatures. The approach also
enabled the development of a statistical application, generalized additive models
(GAMs), to make predictions with non-linear length and weight data, as well as gene
expression profiles in future work. Various GAMs, which use likelihood statistics to
incorporate multiple non-linear variables into a prediction, were constructed and
compared as to their abilities to predict blow fly age (Hastie and Tibshirani 1990, Wood
2006). The models were first tested against each other using stage, length, and weight
data from 2559 immature flies. In addition to non-linear body size data, three regional
strains, originating from California, Michigan, and West Virginia, were included in the

study to determine if regional strains develop similarly. The flies were also raised at two

18

different temperatures, to understand potential effects of temperature on variation in PMI
predictions.

The third set of experiments dealt directly with gene expression during egg
development, through the analysis of a modest set of three genes. Eggs represent the
shortest developmental stage in flies (~10 hours at 32°C), so they were used in a
feasibility study to determine if gene expression levels would be useful in identifying
more specific periods of development within a stage. If estimates of age can be made
more precise in eggs, it is likely that estimates of longer stages will be more precise as
well. Much is known about Dipteran embryogenesis (e.g. McGregor 2005), which
indicates that it should be possible to differentiate between young and old eggs, based on
the expression levels of various patterning genes. Three genes were assessed during L.
sericata egg development to demonstrate the principle that gene expression can be used
to estimate blow fly ages.

Since the egg expression profiles yielded promising results, the forth and most
extensive line of research was conducted. A subset of 958 individuals from the 2559
staged, measured, and weighed larvae and pupae in the GAM experiment were profiled
for the expression of 9 developmentally regulated genes (three other genes were removed
as they were uninformative). All strains were assessed at 335°C, and gene expression in
the CA and MI strains was also assessed at 20°C. GAMs were constructed using gene
expression levels/size/stage, and compared to GAMs using size/stage to predict age. In
the next chapter, a blind study is detailed in which the ages of larvae and pupae were
predicted using GAMs developed from the gene expression data. The successful

validation of the methodology was a critical part of the research, because the results of

19

any statistical modeling must be confirmed using independent data (Scheiner and
Gurevitch 2001).

During the aforementioned GAM research it was noted that some individuals in
all replicates failed to form a puparium, even as adults were eclosing from that replicate.
Forensically, the collection of non-maturing (Peter Pan) individuals at a crime scene
could be very misleading, resulting in inaccurate PMI estimates. However, the ability to
recognize such individuals might be useful for identifying flies that should not be
included in a PMI analysis, as they would skew results, and may also indicate that pupae
were missed in sampling. Individuals in a state of arrested development are also an
important focus of research as they could be diapausing (Tachibana and Goto 2004),
estivating (Liu et al. 2006), or developmental mutants. Molecular information identifying
any gene that may be involved in delaying development is useful for studying plasticity
and may also be used to control pest species, as described above. Given the potential
benefits of studying this state, ten (or as many as available) Peter Pan individuals were
collected from each replicate. Of ~120 flies, 55 were profiled to determine if they
exhibited differential expression patterns as compared to normal postfeeding third instars.

The last project undertaken was a comparison of newly collected strains of L.
sericata, to confirm preliminary observations from the statistical modeling research that
showed temperature and strain effects on development time and body size. From the
California, Michigan, and West Virginia flies, 720 pupal length and weight
measurements, and 60 minimum development times were assessed to determine how
biological strain and temperature affected these traits. Genetic effects on development

time or body size may be indicators of differential selection among the strains, possibly

20

driven by environmental factors in the ecoregions that each strain originated from (Figure

1.3).

 

Figure 1.3. Ecoregions of the United States. Map from work by Hargrove and Hoffinan
(2004). The white dots indicate approximate origins of the CA, MI, and WV flies
studied. Regions of similar color have comparable elevations, soil, and climates. Both
the MI and WV strains were located in ecosystems that were mapped in shades of blue,
while the CA strain came from a yellow region.

Each section in this dissertation revealed information regarding one or more
aspects of developmental variation in the blow fly L. sericata. Any improved
understanding of development time (the main phenotype of interest in forensic
entomology) and body size (a phenotype of secondary interest) will be useful in
achieving more accurate and precise PMI predictions with entomological evidence.
Assessing the environmental and genetic effects on body size and development time can
also help to determine the conditions that should be followed in standard operating

procedures for rearing flies in the laboratory. Likewise, genetic and ecological effects on

development can be used to tailor a PMI estimate to the regional strain involved in an

21

investigation. Most importantly, gene expression data provide a new form of
information, independent from body size and developmental stage, which can improve
the precision of age estimates for later developmental stages. The use of GAMs to
achieve this goal allows for improved descriptions of error that will satisfy the Daubert
standard of scientific evidence, ensuring the place of forensic entomology in the
courtroom. Through the application of quantitative and molecular genetic approaches, a
more precise and accurate method of PMI prediction should emerge, while producing

data of value to agricultural, medical, and evolution research.

22

CHAPTER 2: PLASTICITY IN BLOW FLY GROWTH

Introduction

Forensic entomologists rely on published data of blow fly development to
estimate the time since initial colonization of remains, thus extrapolating a postmortem
interval (PMI) (Catts and Haskell 1990). PMI estimates based on entomological
evidence have been widely and successfully presented in legal proceedings, however the
laboratory study of blow fly development, on which these estimates are founded, has
never been standardized. Because of this, entomologists may utilize different blow fly
developmental data sets, which can lead to variable PMI predictions. Further, a lack of
scientific standardization has the potential to call into question the overall accuracy of
entomological evidence (see Saks and Koehler 2005).

Prominent examples of differing laboratory rearing methods and resultant data
sets can be found for the widely distributed green blow fly, L. sericata (Diptera:
Calliphoridae) (Mei gen) (Kamal 195 8, Greenberg 1991, Anderson 2000, Grassberger and
Reiter 2001). These data sets all present a developmental time scale from egg to adult.

In his work, Karnal (1 95 8) recorded only the duration of each developmental stage, while
Grassberger and Reiter (2001) and Greenberg (1991) also measured the length of
maggots until pupation, and Anderson (2000) measured crop length throughout
development. Each of these studies utilized different fly-rearing techniques, varying in
the quality and type of food, the quality of pupation substrate, and the destructiveness of
sampling. Likewise, the authors measured fly development at different temperatures, and

reported development data in assorted ways (minimum, average minimum, mode, and

23

maximum growth). The resulting picture of L. sericata development is clouded, with
relatively small differences in minimum development time among all studies, while
Anderson (2000) characterized a notably longer minimum development time at
temperatures similar to the others. Unfortunately, direct comparison of these studies is
impossible, as experimental conditions and genetic background of the flies varied among
them. Further, even though the data sets were generated with a goal of relating larval
development to PMI estimates on corpses, no attempt was made to tie laboratory-
established growth rate data to ecologically relevant larval development on carrion.

Development time is a quantitative trait that is expected to vary due to both
genetic and environmental factors (Mackay 2001; Conner and Hart] 2004).
Understanding genetic and environmental effects on quantitative traits is best
accomplished by altering one variable while keeping all others constant, and a limited
number of such experiments have been conducted in a forensic entomological context.
For instance, Kaneshrajah and Turner (2004) demonstrated that C. vicina reared under
otherwise constant conditions showed variable growth when raised on different organs,
and Wells and Kurahashi (1994) indicated that differences in rearing protocols were the
likely source of discrepancies regarding development times of Chrysomya megacephala
(Diptera: Calliphoridae). Likewise, high-density rearing conditions that increase maggot
mass temperatures were shown to shorten development times of C. megacephala
(Goodbrod and Goff 1990). Recently, L. sericata was found to exhibit variable growth
patterns depending on the species and tissue type on which cohorts were raised (Clark et
al. 2006). Certainly it appears that rearing conditions can have a major impact on the

developmental timing of calliphorids.

24

Just as environmental factors influence calliphorid development, intra-specific
differences have the potential to produce variation in fly developmental times. The field
of ecological genetics is replete with cases demonstrating the effects of genetic
background on quantitative traits (reviewed by Mackay 2001; Conner and Hart] 2004).
Developmental variability has been documented in many fly species, including strains of
Drosophila (Diptera: Drosophilidae) (Johnson and Schaffer 1973, Oudman et al. 1991 ,
Hoffmann and Harshman 1999, Parsch et al. 2000), Rhagoletis pomonella (Diptera:
Tephritidae) (Feder et al. 2003), and Scathophaga stercoraria (Diptera: Scathophagidae)
(Blanckenhom 2002). Since each L. sericata study referenced above was conducted on
different populations, it is impossible to separate the effects of environment and genetics
on fly development. Potentially, any (perhaps all) differences among L. sericata studies
could be explained by genetic variation among strains, however this would only be
demonstrated if each strain was raised using the same experimental protocol. Unless
standard rearing conditions are adopted, such comparisons are impossible.

The potential influence of the environment and genetics on quantitative traits, and
in particular development time, led to the hypotheses tested herein that L. sericata growth
is plastic with respect to rearing conditions, and that fly development on carrion will best
be predicted by a specific combination of laboratory conditions that affect this plasticity.
Temperature and humidity are already known to affect development time (Greenberg
1991, Anderson 2000, Grassberger and Reiter 2001) and mortality (Wall et al. 2001) in
this species, so these conditions were held constant to investigate the effects of other
rearing variables. Likewise, the flies in these experiments originated from the same

source population, allowing genetic differences to be largely ruled out as a source of

25

developmental variation. By changing the exposure of a single strain of L. sericata to
specific environmental conditions, several questions related to the hypotheses were
addressed. In particular: 1) Do laboratory rearing conditions affect the development time
of L. sericata? 2) Are any developmental differences caused by laboratory rearing
conditions large enough to explain the variation observed among published grth data
on this species? And 3) Does grth generated under laboratory conditions accurately

reflect larval development of L. sericata on a carcass?

Materials and Methods
Fly Collection and Rearing.

L. sericata adults were collected from the Michigan State University campus in
East Lansing, Michigan throughout the spring, summer, and fall of 2004, and were used
to establish a general population cage of approximately 200 flies. Species identification
was done using multiple keys, two independent identifications, and by comparing the
DNA sequence of a 798 base pair region of the mitochondrial cytochrome oxidase I gene
to published sequences on the NCBI website using the BLAST link. Forward primers for
DNA amplification were GATCAGTAGTAATTACAGCT, and
TAATATTGCTCATGGAGGAG, while reverse primers were
TTGACTTTTTAATATCTTAG, and CCTAAGAAATGTTGAGGGAAG. Polymerase
chain reactions were run for 35 cycles by denaturing at 95°C for 30 seconds, annealing
primers at 50°C for 30 seconds, and extending amplicons for one minute at 72°C.

Sequences were generated on a CEO 8000 capillary electrophoresis system, using a CEO

26

DTCS Quick Start Kit and the manufacturer’s suggested protocols (Beckman Coulter,
Fullerton, CA).

Experimental rearings were canied out between January and March of 2005. To
minimize the loss of genetic variation during this period, the population was expanded to
three cages of more than 100 individuals, from which 20—50 migrants were transferred as
pupae to the other cages each generation. Generations were allowed to overlap until the
cage required cleaning, which was done monthly while the next generation was in the
juvenile form.

Cages of adult flies were provided water and honey ad libitum. Beef liver was
supplied as a protein source one day prior to oviposition. On days that eggs were
collected, fresh liver was placed into a cage in the late morning to mid aftemoon. Cages
were checked every 15—30 min until oviposition was observed. Approximately 250—
1000 eggs (1—4 egg masses) were removed one hour after the first observation of
oviposition. The egg masses were immediately transferred to fresh liver and placed into
a l-liter glass canning jar (Alltrista, Muncie, IN), with a breathable cloth screwed on as a
lid. Jars were placed into a temperature-controlled incubator at 25°C (+/-0.5°C) with a
12:12 hour light and dark cycle. A beaker filled with water was kept in the incubator,
which provided a relative humidity of 25% (+/- 4%).

Several treatments were examined to assess the influence of rearing variables on
the development time of specific immature stages, and on total immature development
time (Table 2.1). These considered the freshness of food, moisture of food, type of
pupation substrate used, orientation of the substrate with respect to food, transfer of

larvae to fresh pupation substrate, and destructiveness of sampling. The influence of

27

meat freshness was tested by providing cohorts with 40g of liver every day (fresh meat
daily or FMD) or 120g of liver every third day (no fresh meat daily or NFMD). Paper
towel treatments received fresh meat daily, which was placed on a moist paper towel
(FMDPT). The influence of pupation substrate was examined by providing either clean
sand (Fainnount, Wedron, IL) or vermiculite (Therm-O-Rock West, Chandler, AZ) to
jars containing postfeeding third instars. The influence of food orientation with respect to
pupation substrate was tested by either placing meat on top of the substrate at the egg
stage, or placing the substrate on top of meat when larvae reached the postfeeding third
instar stage. Fresh pupation substrate was tested by removing 125 postfeeding third
instars from individual cohorts and placing them into ajar with 500ml of fresh pupation
substrate. The transfer treatments were taken from cohorts with far more that 300
individuals in the jar, meaning larval density was much greater in untransferred than
transferred treatments. Destructive sampling was assessed by permanently removing or
not removing 12 individuals from a cohort each day.

Experimental cohorts were checked approximately every twenty-four hours,
except jars with eggs, which were checked every half hour until they hatched, and pupae,
which were observed throughout the day until eclosion occurred. Length measurements
were taken throughout larval development, incorporating the 12 most mature larvae in all
treatment groups (either the largest maggots or postfeeding maggots lacking blood in
their crops). Ruler-measured lengths of the maximum body extension (to the nearest 0.5
mm) were determined using a stereomicroscope for first instars (due to their small size)
or by eye for all other stages. Advances in developmental stage were recorded to the

closest 15 minutes, however given that most animals were observed once per day,

28

development time variation of less than one day was indistinguishable from sampling
time variation. All experiments were conducted in the same temperature controlled
incubator, with jars rotated within the incubator daily.

Development of larvae on mammal carcasses was performed using three Sprague-
Dawley rats from breeding colonies at MSU, sacrificed by C02 asphyxiation within two
days of egg placement on the body. The rats weighed approximately 500g and were in
excess of the feeding needs of individual cohorts (larvae utilized approximately half of
the carrion before the postfeeding stage). An egg mass collected in the manner detailed
above was placed along the mouth of the rat. Rat carcasses were set in an open plastic
bag, which was placed into a styrofoam container with an opening cut from the lid. A
screen was fitted between the container and the lid to prevent escape of postfeeding
larvae. Animals were reared at 25°C (i05°C) and 25% (i4%) relative humidity, with
maggot length and the duration of developmental stages recorded as above. Larvae from
rat treatments were transferred to sand substrates to pupate.

Statistical Analyses.

Owing to unbalanced data (Table 2.1), MANOVA could not be used, thus
analyses of variance were examined using Type III AN OVAs (Scheiner and Gurevitch
2001). This approach removes the variance from variables other than the one of interest,
and compares the variance remaining to the dependent variable. ANOVA and regression
statistics were performed with the R statistical package (R Development Core Team

2004) at a < 0.05 significance.

29

Development times in hours and accumulated degree-days (ADD), including
standard deviations, were calculated for every significant treatment type and for rat
cohorts. ADD was calculated using a base temperature of 10°C.

Graphs of larval grth were produced using the R statistical package. Curves
were plotted by non-linear quantile regression using smoothing parameters that yielded
curves comparable to published data from Greenberg (1991), Wells and Kurahashi
(1994), and Grassberger and Reiter (2001). Treatments in the comparisons include FMD
cohorts that were transferred to fresh pupation substrate, FMDPT cohorts that were
transferred to fresh pupation substrate, and NFMD cohorts that were not transferred to
new pupation substrate. The plots included average and 95% confidence intervals, from
the day flies hatched until the first day pupae appeared, which were then compared to
averages of larval growth on rats. Data from Grassberger and Reiter (2001) were also
compared to larval development on rats, as that study included grth at 25°C. For these
analyses a locally weighted sum of squares (lowess) curve was plotted through the

estimates using R.

Results
Species Identification.

Morphological identification of flies indicated that all were L. sericata. To
confirm identification, a 798 base pair mitochondrial cytochrome oxidase 1 sequence
(NCBI accession number DQ062660) was obtained from a collected adult fly. A BLAST
search showed it was identical to a cytochrome oxidase 1 sequence from a L. sericata

population in Ontario, Canada (accession number L14947). The closest 13 NCBI gene

30

sequences were from L. sericata, with a maximum difference of 4 base pairs (<1%),
confirming the species identification.
Developmental Plasticity.

The pre-pupation period for this fly population (reared at 25°C) ranged from 145—
2645 hours (6—11 days), while the duration of egg to adult was 329—5055 hours (14—21
days), with all data given in Appendix 2.1. Throughout the experiment replicate
treatments followed synchronized growth trajectories during the feeding stages, with a
small number of individuals lagging. In contrast, postfeeding larvae within a treatment
advanced to pupation gradually over a week. Eclosion took place over a week also.

Development times for stages and treatments are given in the Appendix and are
summarized in Table 2.2 (using both hours and ADD). Linear models showed that
development among treatments did not exhibit statistical differences in the shortest
stages—the egg or the first two instars (a single exception is detailed below)—nor did
these stages significantly influence total development time (data not shown). In contrast,
the feeding portion of the third instar (F=1852, df=1, P=0.00013, R2=0.35), the
postfeeding stage of the third instar (F=27.67, df=l , P<0.0001, R2=0.44), and pupation
(F=5359, df=1, P<0.0001, R”=0.62) significantly affected overall development times.

Substrate type and its placement had no significant effect on development during
any stage. Other treatments examined (Table 2.1) significantly impacted development
time (Figure 2.1 and Table 2.2), while the stage at which that impact occurred differed.
FMD accelerated development compared to treatments that received supplements every
third day during the feeding portion of the third instar (F=12.19, df=1, P=0.0015),

although it was also a significant variable in the duration of the second instar (F=8.336,

31

df=1, P=0.0072). Accordingly, the two treatment types that developed in 14—16 days
were FMD. FMDPT also resulted in faster growth during the feeding portion of the
lifecycle compared to treatments without paper towels (F=206.8, df=1, P<0.0001). Moist
paper towels were not necessary for the most rapid overall development, given that the
fastest recorded time from egg to eclosion was from a FMD transferred treatment [329
hours (cohort 14 in Appendix 2.1)], however they promoted consistently faster
development (Figure 2.1). Once feeding ceased, the moisture of food did not contribute
to developmental variation (postfeeding third instar F =0.8439, df=1, P=0.37 for FMD and
F=1.677, df=1, P=0.21 for F MDPT), however transferring larvae to fresh substrate
significantly shortened the amount of time spent as postfeeding third instar larvae

(F =1 7.59, df=1, P=0.00022). The results indicate that handling larvae during the study
did not impede development.

Destructive sampling did not influence larval stages, but significantly increased
the pupal stage (F=49.13, df=1, P<0.0001). Finally, variables were assessed together to
determine their relative influence on total immature development. Each had significant
effects on total development time (FMD: F=4.644, df=1, P=0.039; FMDPT: F=8.019,
df=1, P=0.0079; Transfer to fresh substrate: F=4.454, df=1, P=0.043; Destructive

sampling: F=26.l4, df=1, P<0.0001).

32

1a. Moat 1b. Paper Towel

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0 —q-— --—r-——
8 ‘ o : § "
o
8 _ 8 _
V v
E - . g
3 ' o
I I
8 a 8 a
v v
o
s- .8. , i 1
: l l
I r T r
FMD NFMD No Yes
Treatment Treatment
1c. Transfer 1d. Destructive Sampling
8 —t 8 —l
V V
E E z i
i o s :
v e
I r r I
No Yes No Yes
Treatment Treatment

Figure 2.1. Developmental variation among Lucilia sericata cohorts by treatment
type. Boxplots of total development time (hours) for each of the 37 liver-fed cohorts.
The line within the box represents the median development hours, the box represents the
development times between the 25th and 75mpercentiles, and the ‘whiskers’ (outer-most
lines) represent the 5th and 95th percentiles. 1a: fresh meat daily or no fresh meat daily
(FMD vs. NFMD); 1b: paper towel (moist paper towel placed under meat); lc: transfer:
transfer of larvae to fresh substrate for pupation; 1d: destructive (removal of 12
individuals each day). Note: treatments were in combination with other treatment types
(Table 1) that had significant effects on development time. For instance, the two outliers
in the FMD boxplot (1a) are those that were also destructively sampled.

33

Development on Carrion.

The pre-pupal growth of larvae on rats was compared to the statistically
significant experimental treatments, as well as grth observed by Grassberger and
Reiter (2001) (Figures 2.2 and 2.3). The results displayed in Figure 2.2 show that the
shape and rate of larval growth curves for FMDPT treatments most closely matched the
three cohorts reared on rats.

Figure 2.3 displays the growth of larvae during the first three days of
development, when growth rate is relatively constant. A linear regression demonstrated
different rates of growth among treatments, which were 0.20 mm/hr, 0.10 mm/hr, 0.12
mm/hr, 0.21 mm/hr, and 0.23 mm/hr, for Rat, NFMD, FMD, F MDPT, and Grassberger
and Reiter (2001) respectively, with R2 values of 0.92, 0.77, 0.90, 0.95, and 0.99. The
regression model showed that length varied significantly with age (F=7099, df=1,
P<0.0001), while the effect of treatment types on length was also statistically significant
(F=281.8, df=4, P<0.0001), as was the interaction between age and treatment type
(F=155.0, df=4, P<0.0001).

Figure 2.4 compares the development of the flies reared on rats to development of
liver-fed treatments in this study. Cohorts on rats developed in a manner that was most
similar to the observed maximal development of liver-reared flies (i.e., FMDPT and some
FMD treatments), with development times between 333 and 337 hours (about 14 days).
Further, growth on rat carcasses was much less variable than the grth of liver

treatments.

34

21. Growth on NFMD 2b. Growth on FMD

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

:4 3—1
91- 9:-
s
O O
A F_ ~ - A F-l
E \ E
13. ~ .2.
i ”' i it ..
.1 .1
c- co~
'- Q—i
04-! N-
l l l l l T 7 T I T
50 100 150 200 250 50 100 150 200 250
Hours Hours
2c.GrowthonFllDPT 2d.GrowflronRataCompandtoGR
:— :-
’0
91- 93-1
A 81 A 2.
E E
E. E.
t? ”‘ i: ”‘
.J .J
0- 0'1
'- ‘—
N" N—
T l l l I l l
200 250 50 100 150 200 250
Hours Hours

Figure 2.2. Growth curves of Lucilia sericata on liver versus growth on rat carrion.
Non-linear quantile regression curves created from the lengths of maggots in daily
collections of each treatment type. 2a: meat added every 3rd day, no moist paper towel,
larvae were not transferred to fresh substrate to pupate; 2b: fresh meat daily, no moist
paper towels, larvae transferred fresh substrate to pupate; 2c: fresh meat daily, moist
paper towel used, larvae transferred to fresh substrate to pupate; 2d: the locally weighted
sum of squares curve of data estimated from Grassberger and Reiter (2001) plotted
against larval grth on rat canion. Numbers of cohorts plotted for each treatment were
3, 4, 6, and 6 for Rat, NF MD, FMD, and FMDPT respectively. The solid line on each
curve is the 50th percentile plot from cohorts raised on rats. Treatments are shown as
dashed lines, with the thicker dashed line representing the 50th percentile and the thinner
lines representing the 97.5th and 2.5th percentiles (95% confidence intervals). Confidence
intervals for the rat cohorts are present in 2d.

35

Comparison of Growth Rates

 

Length (mm)

 

 

 

 

Figure 2.3. Linear growth of Lucilia sericata on liver versus growth on rats.
Regression lines of the same treatments displayed in Figure 2.2, for the first three days of
growth—the linear phase of development. Line types used to indicate treatments are the
same as in Figure 2.2.

 

 

 

 

 

 

 

 

 

Liver vs. Rat
§ a
l'.’
8
I
8 _.
e
S _
n
£=
r 1
Liver Rat
Food Source

Figure 2.4. Development times of Lucilia sericata cohorts raised on liver versus rat
carrion. Comparison of total development hours produced by the 37 liver-fed cohorts in
this study to the development of the three rat-fed cohorts. Development time on rats was
much less variable than growth on liver, with a development time most similar to the
fastest growing liver-fed cohorts. Boxplot design is as in Figure 2.1.

36

Discussion
Environmental Components of Variation in the Development of L. sericata.

The green blow fly is a widely distributed species of great forensic importance.
Several authors have examined different fly populations reared under various
environmental conditions, and perhaps not surprisingly, the developmental times differ
from one another, with Kama] (1958), Greenberg (1991), and Grassberger and Reiter
(2001) estimating faster minimum development times than Anderson (2000). This
variability could result from genetic differences among populations, but could equally
result from dissimilarity in the conditions under which the animals were reared. Further,
none of the authors compared the laboratory growth of flies to that on actual carcasses.
In the current study, designed to estimate variation in developmental rates resulting from
environmental differences, a single population of L. sericata was grown under laboratory
conditions that mimicked those used in the earlier studies, and these treatment were
compared to larval development on carrion.

Given the minimum development times of the treatments detailed here, the fastest
fell within the standard errors for L. sericata reared at 22°C by Greenberg (1991) and
Grassberger and Reiter (2001), and is close to the mode reported by Kamal (1958), which
is a common forensic entomology resource. Likewise, the slowest minimum
development time for flies in this study was longer than the developmental minimum at
233° C found by Anderson (2000). This indicates that environmental variation alone can

potentially explain all differences in developmental rates detailed in earlier studies.

37

Results of these experiments demonstrate that variation in food moisture and
pupation substrate have a significant influence on the growth of L. sericata; variation in
rearing conditions generated a developmental difference of up to 7.4 days. Most notably,
treatments designed to maintain meat moisture during feeding shortened the development
of larvae. F MDPT treatments significantly shortened the feeding portion of third instar
larvae, and produced a much smaller developmental range (Figures 2.1 and 2.2). These
results accentuate the importance of considering food moisture when rearing fly larvae.
Grassberger and Reiter (2001) provided larvae with fresh liver daily, resulting in a similar
growth rate at 25°C. Other studies have included moist sawdust, paper towels, or wood
chips underneath meat (Kamal 195 8, Goodbrod and Goff 1990, Anderson 2000), which
would be expected to hold moisture. Interestingly, moist paper towels changed the
lifehistory table of FMD treatments toward the Greenberg (1991) estimate of third instar
duration, which is approximately one day shorter than that of Grassberger and Reiter
(2001). Unfortunately, Greenberg’s (1991) report was vague about how flies were raised
so it is unclear what other factors could be involved, but food moisture may play a role in
the differences in third instar development time observations between these authors.

Transferring postfeeding larvae to a fresh substrate for pupation significantly
shortened the time spent at this larval stage. The postfeeding portion of blow fly larval
development is generally variable (Wells and Kurahashi 1994) and L. sericata is
exceptional among blow flies for wandering far from its food to pupate (Anderson 2000).
This may mean that L. sericata searches for a specific set of environmental cues for
pupation, making the postfeeding stage susceptible to disturbance. The conditions that

produced the fastest growth in this study yielded a postfeeding stage duration of two to

38

three days. Kamal (1958) provided sawdust with food, and observed a mode postfeeding
duration of 90 hours at 26.7°C, with a minimum of 48 hours and a maximum of 192
hours. His mode observation is similar to untransferred treatments in this study, which
lasted a day longer than transferred cohorts. Greenberg (1991) reported an average
postfeeding time of 108 hours at 22°C while Grassberger and Reiter (2001) reported 94
hours at 20°C and 87 hours at 25°C (the temperature at which this research was
conducted). With little information on rearing conditions described by Greenberg (1991),
the shorter times reported in the latter study are hard to explain, but Grassberger and
Reiter (2001) reared their flies with dry sawdust in jars, which may have resulted in the
shorter average duration, given that the treatment seems similar to the transfer treatments
in the research presented here.

There is little information in the literature that helps explain the developmental
variability between transferred and untransferred postfeeding larvae found in the current
study. Three plausible explanations for this phenomenon are density of individuals in
each cohort, moisture differences between old and fresh substrates, and difference in odor
between the treatments. Larval density seems unlikely to have had much influence on
development time. Several treatments that were transferred to sand had larvae that had
congregated on the substrate surface, and these densely packed cohorts still pupated in a
timely manner. On the other hand, a lack of moisture and odor are both plausible agents
behind the accelerated onset of pupation in transferred larvae. The sensitivity of larvae to
moisture during feeding (outlined above) indicates that moisture is a potential cue for the
cessation of feeding, with maggots actively searching out wet areas (tissues) while

feeding, and reversing this behavior when heading towards pupation. Likewise, blow

39

flies are attracted to odors associated with decay (Catts and Haskell 1990, Chaudhury et
al. 2002, Hall et al. 2003), thus it might be advantageous to be attracted to putrefying
odors during feeding, followed by a pre-pupation move away from such odors.

Destructive sampling was found to be unimportant in larval development, yet was
the only significant variable affecting the duration of pupation. The delay in pupation
most likely resulted from the elimination of the earliest individuals to form a puparium,
which were destructively sampled (removed) by necessity. Given these findings, studies
of pupal development rates that require destructive sampling should consider its effects.
Other Potential Sources of Variation.

The data presented demonstrate the effects of differing rearing treatments on this
population of L. sericata. It should be noted however, while most variation in growth
existed among treatments, within-treatment variation was also observed. A portion of
this could be explained by unmeasured environmental factors, as only a small number of
rearing modifications were tested. Certainly, factors not considered in this study are
likely to impact developmental differences.

Likewise, though environmental conditions were found to be highly significant in
the development of L. sericata, genetic variation among fly populations used in different
studies could potentially be just as important in understanding developmental variability.
It is necessary to remember that each publication mentioned above outlined the
development of flies that originated from a different ecogeographical region. There is
precedence for population effects on the development of blow flies and several related
species (Johnson and Schaffer 1973, Greenberg 1991, Oudman et al. 1991, Hoffmann and

Harshman 1999, Parsch et al. 2000, Blanckenhorn 2002, Ames and Turner 2003, Feder et

40

 

al. 2003). Genetic makeup is likely to affect other populations of blow flies, although
these have been largely untested. Genetic differences, including potential interactions
between genotype and environment, may be important sources of developmental
variation when comparing populations of L. sericata.

Optimal Rearing Condition Using Liver and Growth on Rat Carrion.

One might expect that blow flies have evolved to develop fastest under natural
conditions of carrion decomposition. If this is the case, the fastest growth rate obtained
in laboratory rearings would be expected to mimic the growth. of flies living on carrion at
the same temperature. In the current study, L. sericata development on rat carcasses was
most similar to flies reared under high moisture conditions (Figure 2.4). This finding
helps address concerns raised by Kaneshrajah and Turner (2004) and Clark et al. (2006)
who observed a significant effect of tissue type on the growth of C. vicina and L.
sericata, respectively. Kaneshrajah and Turner (2004) were critical of rearing flies on
liver, as it seemed to delay development. This delay was similar to slower developing
treatments observed on desiccated liver in the current study, suggesting that larval rearing
should take place on non-desiccated substrates to best mimic growth on a corpse.
Applications to Forensic Entomology.

L. sericata development is plastic, at a level that alone could explain differences
in the species’ published developmental times. This finding highlights two important
factors that need to be considered when estimating a PMI based on blow fly
development. First, the discrepancies among development data sets can potentially be
explained, in toto, by differences in laboratory rearing protocols used to develop such

timetables. Accordingly, establishment of a common set of rearing conditions, which

41

best relate to growth on carrion, is critical if direct comparisons are to be made among
datasets, and if these data sets are to be used in legal proceedings. Second, because
forensic entomologists use a quantitative trait (development rate) and decomposition
ecology to make PMI estimates, researchers conducting studies on development time
must aim to address the effects of both genetics and environment on their findings. By
doing so, the forensic community can achieve a greater understanding of how important
each of these factors is to forensic entomology.

A final consideration regarding entomological evidence involves its legal use in
general. In the wake of judicial decisions that place a far greater emphasis on systematic
analyses, known error rates, and statistical probabilities (see Daubert v. Merrell Dow
Pharmaceuticals, 509 US 579 (1993), and KumhoTire Co. v. Carmichael, 526 US 137
(1999)), forensic scientists are under increasing pressure to conduct research, present
legal analyses, and draw conclusions in a methodical and scientifically replicable way,
while relying less on generalized knowledge and personal experience. The field of
forensic entomology, although based on sound scientific principles, can currently be
included among an assemblage of forensic disciplines that may be called into question
with regards to repeatability and standardized techniques (see Saks and Koehler, 2005).
Efforts to establish calliphorid laboratory rearing protocols that best portray fly
development on cadavers, and to standardize those techniques for future research, are
central to meeting the demands of Daubert and Kumho. Such endeavors are necessary if

forensic entomological evidence is to be routinely accepted in courts of law.

42

.338 .692“ 368 e. no

coon—Q QSE 3382 ”330% 59$ 60m: 88625 5:83 ”owmbmnsm cam £80 an mo Seton 05 we 8 03593 a mo Q8 no 883
we? too.“ moccb oumbmnsm .Bwbmnsm gunman amoa mo 48 com 3 wobommcwb 0.53 mEmE BE“ wcmuoofimom mag ”Lehman; .25..
wazmfiwm some 3 tonoo 06 80¢ @96an 2.32sz S ”Awfianamv o>zo§moa ADSEZV you .8 A92”: 33V 308 swoon ”Hafiz

 

Q ”8% mm ”8:532:8/ v we.» 2 ”mo? m ”8.» v ”QSEZ
mm H02 3 6.3m mm HoZ em Hoz mm Hoz mm HOSE
338. 89$ oombmnsm 895 03535 command. o>uosbmom ~82

ate—.8 tour—9r: rm 2: .8.“ 89¢ «non—«a9; —.~ ozah

43

.633 8me E08 a no

econ—m QSE 3382 ”330,—. Logan dew: 88693 8:83 “oombmnsm saw been. .8 mo 80309 05 a 8 88693 a mo m8 so 383
83 coon 22:5 magnum 68:33 8:33 smut mo A8 com 9 watchman: 203 magma e55 wcmeeotmom mmfi command. .08:
wEEEwm some a tosoo 05 88m @9688 £96363 3 ”chzafimmv o>uo=bmom AQSEZV Ho: 8 525 35c :88 €on ”832

 

Q ”8.» mm ”oxfiowfio> v ”mo> ma ”8% m ”8.» v HAux/EA
mm H02 3 65m mm HoZ em Hoz mm HoZ mm HOSE
330,—. 59mm oagmnsm 3ch magnum commas? 0392:an $02

32:8 3.3»: R 2.. he see .558; 3 92a...

44

 

333R: awaflé: amazes: aafimfis 224$: @0336: £53.36:
«5.2.432 Nasal: Erna?» woman? anaemem :fiwmem Edna: 03.253082
3.83.3: 5.33:: 9938.5 5.38.3 835.3: 63.33: 3:32:
33.32. 8.33.2.0. $3.8 ”$33.6 $63.3 mmmaneew 3.5.: 3:855
8:39.: 20.83:: @333: Stains: genes: $.53: 3.033:
3.84.8». 323.3: N333” ”332.3. sigma 8.33m Siam: 523 oz
A3232: $93.3: 833.5: Gmfinm: 354:6: €336: 3332:
233.8% 35: 2.2.35.3 2.23.5. £6.52 mmmfimem $8.8 £22...
£33 9.: 93nd: : $332.3 Rani. _ m: 2.5%? : an. :23: 5. 39mm:
8.3.353 2.0333 5:34am 2.33.8 $2.38 2.3.58 Edam: 338.23 oz
$25.23 Going: @343: $38.: 2 _.mfimm.e: engage: €1.84:
3.0323 3.23m: 2.239% 4.33.8 2.9.2.8. 3.3.3.3 «33% 338.25
Quads: $39.2; : E: 38. S: gimme: 23%;va Awnonwhm: 2 m. _ #3:
$333. 8.»?38 35323 23.3% 3.24:3 59.3.2 EH3: 9&2
A8838: Gus“: _ : sawing $.33: gimme: 633.0: €365:
8.34%; 03:: 3.2392 mnmfimoé. $38.8 2.3.08 9.3.5.0. 9:
:38. «95 wcmcootmom alSmE Em 3%.: EN alga: .L mum Enema.

63.9.28 Sauad we 3:253...

Maw—«o.— anfinwmm .5.“ «aka—Yvonne: 682.558.“ :5“ PE...— 5 2.53.5: FEE—Sm um 88: inane—0.5: 03.8.5. .N.N 03.;

45

CHAPTER 3: THE USE OF GENERALIZED ADDITIVE MODELS
IN ANALYZING FORENSIC ENTOMOLOGICAL DATA

Introduction

Daubert, et al. v. Merrell Dow Pharmaceuticals (509 US 579 (1993)) was a pivotal
ruling for forensic scientists, in which the US Supreme Court declared that the Federal
Rules of Evidence (particularly Rule 702), and not Frye (Frye v. United States (293 F.
1013, 1014, DC. Cir. (1923)), were the standard for scientific evidence and expert
testimony. In doing so, the High Court placed the burden of assessing the validity—and
thus admissibility—bf scientific evidence on the trial judge, based on five main criteria:
Has the technique in question been tested; Do standard operating procedures (SOPs) exist
for the technique; Has the technique been subjected to peer review and publication in the
appropriate literature; Is the technique widely accepted by the relevant scientific
community; and finally, What is the known or potential error rate of the technique?
DNA-based evidence has set the ‘gold standard’ for meeting Daubert requirements,
largely satisfying all of them. In contrast, many of the forensic sciences and resultant
expert testimony are based on practitioners’ training and experience, often with little
consideration for SOPs, method testing, potential error rates, or publication, even when
the technique is generally accepted. As an example, the National Institute of Justice
recently posted a solicitation for the study of fingerprints/fiiction ridges, though certainly
this method of identification is extremely well-established. Other areas of forensic
science fare far worse (Saks and Koehler 2005).

Forensic entomology falls between these extremes. The predictable growth of

carrion-feeding flies has long been used to estimate the time a body has been exposed to

46

insects, and thus to estimate a postmortem interval (PMI). Using larval size and
developmental stage to approximate age is well supported by research and observations
in developmental biology, and this forensic technique is widely described in the scientific
literature (e.g., Greenberg 1991, Grassberger and Reiter 2001). Likewise, countless legal
rulings have assured its admissibility, just as countless juries have been guided by
entomological testimony. However, scientists have reported different growth rates for
immature flies (Kamal 1958, Greenberg 1991, Wells and Kurahashi 1994, Anderson
2000, Grassberger and Reiter 2001) and court qualified experts have come to incongruent
conclusions about a PMI based on the same entomological evidence, depending on which
growth data were utilized (e.g., California v. Westerfield, CD 165805 (2002)). This
problem stems, at least in part, from a general failure to develop SOPs, and also from not
fillly considering the amount of variation present in larval growth (or more precisely, to
account for error rates inherent in estimates of larval age), two of the major tenets of
Daubert. The difficulty in estimating error is exacerbated by the fact that blow flies grow
in a non-linear fashion and have variable size distributions at different ages, unequally
affecting age estimates of developmental stages (Wells and Lamotte 1995).

The research presented here was designed to investigate the variability that occurs
in larval and pupal growth of blow flies in order to discern which of a suite of variables
have the largest influence on estimating age, and to explore the possibility of placing
confidence intervals around juvenile age estimates. Using three regional strains of the
blow fly Lucilia sericata (Diptera: Calliphoridae) (Meigen), collected in California,
Michigan, and West Virginia, a data set containing linear (developmental stage, strain,

rearing temperature) and non-linear (length and weight) measures was established.

47

 

Generalized additive models (GAMs) were developed taking these variables into account,
examining the level to which each influenced/predicted the percent of immature fly
development (Hastie and Tibshirani 1990, Wood 2006). Similar GAMs have already
been used to assess the effects of cadmium on the growth of L. sericata cohorts (Moe et
al. 2005), and were assessed here for their potential use as tools in predicting blow fly
development percent. The utility of a model was then tested on an independent data set
(larvae reared on rat carcasses), focusing on developmental stage and length. GAM
predictions of larval development percent were plotted against true age to assess the error

of the predictions and to define confidence intervals for these estimates.

Materials and Methods
Species Identification.

Wild L. sericata were collected in California (CA), Michigan (MI), and West
Virginia (WV), from the UC Davis campus in June of 2005, the Michigan State
University campus starting in May 2005 (which were provisioned with new flies
occasionally throughout the summer), and from the West Virginia University campus in
August of 2005. Adult individuals from each strain were identified by key (Hall 1948
and Gorham 1987), with independent confirmations, and through mitochondrial
cytochrome oxidase 1 gene sequencing (Tarone and F oran 2006).

Growth Experiments.

Cohorts of flies were raised in a round robin design, in which CA and M1 were

reared in one block, followed by CA and WV, and WV and MI, between 9/1/05 and

10/24/05. Flies ranged from two to five generations removed from their natural

48

population. Cohorts were initiated by placing fresh liver into the cages of adult flies,
which was checked regularly for eggs. When oviposition occurred the time was recorded
and meat and eggs were removed 1 h later. Cohorts were placed in either 20i0.5°C or
33.5:t1.8°C incubators under a 12:12 h light cycle at 255% relative humidity. Incubator
temperature fluctuation was noted using a HOBO data logger (Onset Computer, Boume,
MA). Eggs were transferred to fresh liver, which was placed on a moist paper towel in 1
L jars, covered with a breathable fabric lid, based on rearing conditions previously found
to best mimic those on carrion (Tarone and F oran 2006). Cohorts were given fresh liver
daily until postfeeding third instars were observed, at which point 250 individuals
(335°C treatments) and 375 individuals (20°C treatments) were transferred in batches of
125 to 1 L jars containing 500 mL of fresh sand as a pupation substrate.

Length and weight of 25 59 larvae/pupae were recorded, starting approximately 24
h after eggs were laid. Length was measured with a ruler based on the furthest extension
of a larva to the nearest ‘/2 mm. Wet weight of live individuals was measured on a Cahn
27 Automatic Electrobalance (Cahn Instruments, Cerritos, CA) to the closest l/ 100 mg.
Developmental stage was assessed by observing feeding larvae microscopically, by
visible crop length and migrating behavior for postfeeding larvae, and puparium
formation for pupae. Ten larvae were removed from a cohort and measured/weighed,
twice daily (in the morning and late afternoon). Ten pupae were collected once daily and
measured/weighed; 5 individuals were collected if less than 10 were available.

Earlier research showed that the destructive sampling of pupae delayed the
appearance of adults (Tarone and F oran 2006). To account for this, pupal age was

calibrated to the day of pupation. This means that pupal samples were assessed in groups

49

that pupated within 24 h of each other (i.e. 0—1 , 1——2, 2—3, etc. day old puparia) with the
minimum development time for pupation being the minimum development time for any
individual within a collective group of pupae.

Forensic entomologists generally assess fly grth progression using a measure
of relative age, allowing them to take into account the substantial influence of
temperature on development. Given that multiple variables had the potential to affect
immature fly growth rates in the current research, including understood (e. g.,
temperature) and questioned (e. g., fly strain) factors, a method that would allow growth
progression to be compared directly among all flies was required. Development percent,
or the relative (developmental) age of an individual, was used to assess the extent to
which a fly had progressed towards maturation (eclosion). This measure, often used for
relative developmental comparisons (e.g., Rogina and Helfand 1995, Rogina et al. 1997,
Anderson 2000), permitted individuals at all points in development to be compared,
which would be impossible if, for instance, temperature and fly strain varied in their
influence on growth. Development percent was calculated by determining the age in
hours of an individual, then dividing the age by the minimum total development time of
that experimental replicate. As an example, if an individual was sampled 100 h after
oviposition and the minimum development time for the replicate was 285 h, then the
individual was considered 35% developed.

The laboratory growth of larvae on rats have been described previously (Tarone
and Foran 2006) and differed from the measured cohorts primarily in food source and
temperature (25°C). Three cohorts of Michigan L. sericata larvae were reared on rat

carcasses and the developmental stage and length of twelve individuals were recorded

50

 

daily from each cohort through the first day that puparia were observed. These data were
used to predict age. The ethical guidelines of the Michigan State University Laboratory
Animal Resources unit were followed, adhering to IACUC requirements.

Statistical Analyses.

GAMs were developed using the mgcv library in the R statistical package (R core
development team 2004). The models use likelihood statistics to predict a value (e. g.,
age) based on various input data. GAMs relate non-linear data such as fly length and
weight to the predicted value (e. g., development percent) using smoothed, non-linear
mathematical fimctions (Hastie and Tibshirani 1990, Moe et al. 2005, Wood 2006). In
this manner, the relationship of two non-linear variables to each other can also be
included in GAMs (Wood 2006), so a length-by-weight term was also evaluated.
Distributions must be applied to the functions used to make predictions in a GAM, which
is done through a link function. Based on the results of residual plots produced for the
models, a gamma distribution (instead of a normal distribution) with a log link function
was most appropriate for the models evaluated. Diagnostic plots were compared among
models in order to confirm the validity of distributional assumptions in a model and to
compare the predicted versus true age. The first plot was a quantile-quantile graph of
modeled data versus data from samples. If the assumptions of a model are correct, this
line is straight. The next plot was a graph of residuals against predictions. The data
should be evenly distributed above and below zero, with no difference in residuals along
the linear predictor axis; unevenly dispersed residuals indicate that the assumed data
distribution in the model is inaccurate. The third plot was a comparison of the

distribution of residuals, which should appear as a bell curve (most error is small, with

51

rare instances of larger error). The final plot was a graph of true (response) versus
predicted (fitted) values for all data used to construct the model. For simplicity’s sake
this will be referred to as the Y = X line, or Y (predicted age) = X (true age). The most
precise models have all predicted age values clustered close to the Y = X line, with no
gaps in the line. A gap in predictions results in an aging inaccurary because an individual
of an age found in a gap will necessarily be predicted as either older or younger than it
actually is. More detailed information on GAM can be found in (Hastie and Tibshirani
1990, Moe et a]. 2005, Mansson et al. 2005, Wood 2006).

Models generated several statistics. For linear models the statistic used to explain
how closely data match a model is R2; as length and weight data are non-linear the
apposite statistic for GAMs is the percent deviance explained (Wood 2006). Degrees of
freedom or estimated degrees of freedom (a non-linear equivalent) were determined, as
was a P-value, which was based on the likelihood of a variable being predictive of age.
P-values in GAMs are considered estimates because likelihood statistics do not yield
actual P-values, but do provide values that are similar and can be used to estimate the
more familiar statistic. These estimates can vary by up to two times the actual P-value
(Wood 2006), thus terms were not considered significant unless P-values were less than
0.025. Additionally, multiple variables were included in some models, requiring a
Bonferroni correction that resulted in a significance threshold of P<0.0042. Given the
inherent inaccuracy of estimated P-values, they were only used to identify informative
terms or terms that were candidates for removal from a model owing to intermediate or

non-significant P-values. The inclusion or removal of a term, however, was ultimately

52

decided by the statistic used to compare models: the generalized cross validation (GCV)
score, which is an information criterion that is lower for better models (Wood 2006).

Six terms were used to develop models: fly developmental stage, length, weight,
length-by-weight, strain, and temperature. Stage, strain, and temperature were
considered linear variables, and length, weight, and the two plotted against each other
were non—linear. This resulted in 63 possible models, hence only a subset is presented
here. The first six models examined each variable by itself, while the remaining 12
combined variables to assess improvements gained (as measured by a decrease in GCV)
from including specific terms. Developmental stage was considered the primary variable,
as all forensic entomologists include this in PMI predictions. Body size is also often
incorporated into PMI estimates, thus length and weight were added to several models, as
well as being examined in combination. Next, the influences of strain and temperature
were tested through inclusion with the more familiar variables (stage, length, weight).
Similarly, since length-by-weight is a somewhat novel measure it was evaluated in
combination with the three standard variables, and then with all variables.

Finally, a GAM incorporating the standard variables used to age flies in forensic
entomological enquiries, developmental stage and length (Kamal 1958, Greenberg 1991,
Anderson 2000, Grassberger and Reiter 2001, Tarone and F oran 2006), was tested
against an independently derived data set. The model-based predictions of larval
development percent for three previously collected fly cohorts raised on rats were plotted
against their true development, comparing them to the predicted 95% confidence
intervals for the model (precision) and the Y = X line (accuracy). Confidence intervals

were superimposed over the predictions made for rat cohorts (using the quantreg library

53

in R) by plotting locally weighted sum of squares curves through the 97.5th and 2.5th

percentiles.

Results
Species Identification.

Flies collected from the three states were identified as L. sericata based on both
visual verification, visual confirmation by an independent entomologist, and cytochrome
oxidase 1 sequence data (accession numbers DQ868503, DQ868523, and DQ868524 for
CA, MI, and WV respectively). Sequences obtained from the CA strain, the M1 strain,
and the WV strain were 428 and 227 non-overlapping base pairs, 774 continuous base
pairs, and 776 continuous base pairs in length, respectively. BLAST results for the
sequences showed the closest match for all was to L. sericata, with 100 % similarity to at
least one other L. sericata sequence. The next closest species match was L. cuprina with
a 98% to 99% similarity (5—8 base pairs difference).

Immature Development.

Figure 3.1 depicts a plot of fly length against percent juvenile development. The
feeding portion of the lifecycle makes up the initial 25%, and shows a linear increase in
length. The postfeeding third instar, where body size decreases and variation in size
increases, is found from approximately 25—50%. The relatively unchanged second half
of the plots is the pupal stage. Weight results displayed the same pattern (data not
shown), and both demonstrated that the distribution of sizes in the feeding stages was
much smaller than it was in postfeeding third instar larvae and pupae. Minimum and

maximum development percents for each stage of development were: First instar = 5.5—

54

1 1.0%; Second instar = 7.4—15.4%; Feeding third instar = 12.6—26.0%; Postfeeding third
instar = 19.1—60.1%; and Pupa = 43.2—100% (Figure 3.2).

Size was influenced slightly, but significantly, by temperature and strain. CA
individuals tended to be larger than MI, which were larger than WV (Figure 3.3).
Differences in size among strains were not observed during feeding stages, but were
observed once feeding ceased (Figure 3.3) as each strain initiated the postfeeding third
instar at different points in development, resulting in variation in average pupal sizes.
Also, growth at 20°C yielded larger individuals on average than did growth at 335°C,
presumably due to a change in the relative rate of development for feeding larvae (Figure
3.3). Size differences caused by both strain and temperature were repeatable, though
average differences were well within the variation observed for size traits (e. g., Figure
3.1), resulting in an overlap of body sizes among all strains and both temperature

treatments.

55

Length Throughout Development

 

15

i
08

O O
38

1O
4

Length (mm)

(man-m. «D
00 G) mount). 0 O
000

 

 

 

Percent

Figure 3.1. The lengths (mm) of 2559 immature L. sericata throughout immature
development (percent of development values are on a 0—1 scale). Note the tight
distribution of sizes during the earlier, linear growth phase compared to the more variable
postfeeding third instar and pupal stages.

Distribution of Development Percent by Stage

 

—'——

1.0

 

0.8

 

 

 

 

Percent
0.6

 

0.4
1

 

 

 

 

 

 

 

g . E —‘——
I 1 l I I
13! 2nd 3rd 3rdPF Pups

Figure 3.2. A plot of the distribution of development percents for individuals at each
developmental stage. As development progressed, the proportion of the lifecycle spent
in a stage increased. 3rd indicates the feeding portion of the third instar; 3rdPF indicates
the postfeeding stage of the third instar.

56

Figure 3.3. The lengths and weights of individuals throughout development from the
6 cohorts. Growth is compared by strain and by temperature. Solid lines represent the
average for all strains or both temperatures. a) Length (mm) plots for each strain. The
largest strain, denoted by the line with short bars and spaces, was CA, and the smallest
strain, designated by the line with short bars separated by dots, was WV. The M1 strain
was close to the average size and is represented by the spaced line with long bars and
short spaces. Less size variation existed during the feeding portion of the lifecycle (when
size was increasing) than in the postfeeding and pupal stages. b) Length plots comparing
growth at 20°C versus 33.5°C. Growth at 20°C is represented by the spaced line with
short bars separated by dots and 33.5°C is represented by the line with short bars and long
spaces. The higher temperature resulted in a growth curve that had a steeper slope during
the linear growth phase of development; individuals from these treatments peaked in
body size proportionally faster than cooler treatments, which resulted in smaller body
sizes as pupae. c) Weight (mg) plots for each strain. Comparisons among strains were as
in (a). d) Weight plots for the two temperature treatments, with similar results as in (b).

57

Length (mm)

W019” (mg)

15

10

20

Length Throughout Development by Streln

 

 

 

 

 

0.2 0.4 0.6 0.8

Percent

Welght Throughout Development by Streln

1.0

 

 

 

 

 

LengthThroughoutDevelopmentbyTelnpenture

 

L009"1 (mm)

 

 

 

 

Welght Throughout Development by Temperature

 

W059” (m9)

 

 

 

 

58

Assessing Statistical Models.

All models demonstrated acceptable levels of error in the diagnostic plots (Figure
3.4a—t), indicating that the use of a gamma distribution with a log link function was
appropriate. A comparison of all models examined (Table 3.1) displayed the utility of
GAMs to predict development percent when different variables were included. Stage
was the single most informative variable (GCV = 0.045), while length and weight
garnered less information (GCV = 0.126 and 0.144 respectively); all were statistically
significant (P < 0.0001; Table 3.2). The length-by-weight term (model 4) provided an
intermediate level of information in assessing development (GCV = 0.059). Temperature
and strain were not significant predictors of age by themselves (models 5 and 6; GCV =
0.358 for both) and only provided useful information (P < 0.0001 and a decreased GCV)
when combined with other variables (e.g., model 18). Predictions with length and weight
yielded similar results to model 4, approaching, but not improving upon, the explanatory
power of stage alone (model 12; GCV = 0.064). Any model that included stage and at
least one body size measure explained 90.8—92.6% of the deviance in the data and GCV
scores of 004—0034, with the model that included all variables garnering the highest
percent deviance explained and the lowest GCV.

All models were limited in predicting the ages of postfeeding third instars and
pupae, generating artificially narrow age ranges. Gaps between stages were most
dramatic in model 1 (developmental stage alone), wherein individuals were simply
predicted to be the average age of that stage, although true ages were continuous.
Inclusion of body size improved predictive precision in the early stages, but not for

postfeeding third instars and pupae. As an example, in model 10, which included

59

developmental stage and length, postfeeding third instars that were 19.1—60.1%
developed (Figure 3.2) were given a restricted age range of 30.7—40.3% (95% confidence
intervals in Figure 3.4g). The gap between feeding and postfeeding third instars closed
somewhat in more complex models; model 18, which utilized all available data, showed
no gap between these stages (Figure 3.4h), although the data still did not cluster along the
Y = X line at the level seen during feeding. The inaccuracy of predictions remained for
pupae in all models, where true pupal ages were between 43.2—100% of immature
development, but 95% of predictions for pupae using model 18 had fitted values between
61.9—81.2%. Interestingly, predicted ages throughout this range were made for pupae of

any true age; that is, there was no slope to the pupal data as there was for the other stages.

60

Figure 3.4. A comparison of diagnostic plots for model 10 (a, c, e, g) and model 18
(b, d, f, h). Model 10 predicted age using length and stage, while model 18 used all data
to make age predictions, and explained the most deviance in the data. Panels a) and b)
are quantile-quantile plots, assessing the validity of each model’s assumptions. The line
is relatively straight and increasing for both models, indicating that the distributional
assumption of the model did not violate the other assumptions of the model, however
model 18 generated a smoother line. Panels c) and (1) show the distribution of residuals
throughout the lifecycle; in both models residuals are of equivalent size for all ages,
showing that there is no bias in residuals based on age. Panels e) and t) are the
distributions of residuals, which are normally distributed thus the gamma distribution was
acceptable to use with these data. Panels g) and h) are plots of true (response) versus
predicted (fitted) values for data used to construct the models. Fitted values accurately
predicted true age through linear (feeding) development (the first 25% of the lifecycle),
after which the first gap in prediction appeared. This indicates that an overestimation of
ages for postfeeding larvae between c. 19.1 and 30% developed was produced using
model 10. Model 18 was less likely to overestimate age for individuals at this point in
the lifecycle, but predictions were still biased toward overestimating age in young
postfeeding third instars. Neither model represents a noticeable improvement over using
stage alone for aging pupae. Predicting pupal age with just developmental stage resulted
in an age estimate between 43.2 and 100% of immature development. The most
predictive model (model 18) plots fitted values between 61.9 and 81.2 % (pupae). At
either end of this predicted range, true values could be between 43.2 and 100%. As with
all size/stage models, error increased with age.

61

SampleQuantites

Frequency Residuals

Response

0.6

-0.2 0.2

-0.6

0.6

0.2

-0.6 -0.2

100

0

1.0

0.6

0.2

Normal Q-Q Plot

 

 

 

 

Frir
23

III
-3 -101

Theoretical Quantiies

Residuals vs. Fitted

 

         

 

 

 

 

 

 

 

 

 

 

 

 

8

O

O
-1 O o

8 a Q

I I T I

0.2 0.4 0.6 0.8 1.0
Fitted Values
Histogram ofreslduels
H.—

I I I I I I 1
-0.6 -0.2 0.2 0.6
Residuals
Response vs. Fitted Values

1
0 O
-' O
O O
l’
l I I I I
0.2 0.4 0.6 0.8 1.0
Fitted Values

SampleQuentiies

Frequency

h

Response

62

0.4

-0.4 0.0

0.4

-0.4 0.0

200400

0.6 1.0

0.2

 

 

 

 

NormalQ-QPlot
O
at
I I I I I I I
-3 -1 0 1 2 3

Theoretical Quantiies

Residuals vs. Fitted

 

 

 

 

 

0.2 0.4 0.6 0.8

1.0
Fitted Values

Histogram of residuals

7

 

 

 

 

I I r I III 1 1
'03 -0.2 0.2 0.6

Residuals

Response vs. Fitted Values

 

   

 

 

       

 

I I I l

0.2 0.4 0.6 0.8 1.0

Fitted Values

GAM Validation with Independent Data.

The utility of model 10 (developmental stage and length) was examined through
analysis of the previously collected and independently produced rat carcass data set.
Consistent with the finding above, error in larval age estimations increased with age
(pupae were not considered here as length does not change during the stage). A plot of
true versus predicted age (Figure 3.5) shows that age predictions generated for the rat
data, when compared to known ages, spanned the Y = X line and were generally
consistent with (inside) the 95% confidence interval provided by the diagnostic plot for
model 10. In the feeding stages (<26% of total development) the predictions were
approximately +/-5% (or less) of the true age. However, in postfeeding third instars, ages
were initially overestimated, then clustered close to the line, and eventually disbursed
well below Y = X, consistent with the expectation that postfeeding individuals could not
be precisely aged using length. The model also continued to predict a narrower range of
ages for postfeeding larvae (32.5%—40.1%) as compared to their true ages (22.9%—

50.2%).

63

Predicted vs True Percent

 

 

 

 

O s
GI 00
g a I 08
I
<9. .. '
o r
E a, r
i o--+ ,
o
i . '
.5 o' ‘ , - "' o'8' ' 8 °8
a . . as — - -$- - .
n”. P. _. ’ 0
0 I
, I
N _ ’ I
o , ' ,8.
I I
,_
0' "r a ’
I I I I I
0.1 0.2 0.3 0.4 0.5
TruePercent

Figure 3.5. A plot of predicted versus true ages of 252 individual flies raised on rats
as estimated with a GAM for stage and length (model 10). A plot of predicted versus
true development percents of 252 larvae raised on rats as estimated with a GAM for
developmental stage and length (model 10). Development is displayed to 50% because
length measurements ceased when pupation occurred. The solid line represents the Y
(predicted age) = X (true age) line. Dashed lines represent 95% confidence intervals for
the predictions based on the data in (a). Model 10 accurately (data span Y = X) and
precisely (ranging c. 5% above and below Y = X) predicted age for the feeding portion of
the lifecycle (<26% of development), when body size increased in a linear fashion. As
expected, precision decreased greatly once the postfeeding third instar was reached,
although overall error for the model was within predicted levels.

Discussion

The requirements of Daubert necessitate standardized methodologies and
knowledge of potential error, two criteria where several forensic sciences, including
entomology, may be found lacking. Previously we examined how variation in published

rearing protocols, which are not standardized among laboratories, affect grth rates of

64

juvenile blow flies (Tarone and Foran 2006). In the current research the ability to
conduct statistical analysis of blow fly growth and aging, including confidence intervals,
error rates, and model comparisons, was tested. From a practical standpoint, the
methodology allows for direct estimates of error that should satisfy both scientific and
Daubert considerations. For instance, if stage and length are used to estimate that a larva
is 15% developed, model 10 generates a 95% confidence interval of approximately 10—
22% (Figure 3.5). Using a published minimum development time for L. sericata of 288 h
at 267°C (Kamal 1958), an age estimate of 29—52 h is produced, with the requisite error
described. For postfeeding third instars, an estimate of 40% developed (115 h)
corresponds to a 95% confidence interval of approximately 22— 60%, or 63—173 h.
Though the precision necessarily decreases in the latter stages, the window of time placed
around that prediction is now mathematically defined. The methodology also has the
flexibility to incorporate other data that may be collected.

Through the modeling used in this study, several key points became apparent.
First, developmental stage was the single most predictive factor in the models assessed,
explaining 89.5% of the deviance in the data. Logically this makes sense, as stage is a
direct measurement of developmental progress. In contrast, the non-linear
measurements—weight and length—although still significant, proved far less effective in
predicting development, while strain and temperature (genetic and environmental factors)
were by themselves not significant predictors. The ability of stage, length, and weight to
estimate age was greatest during the earliest phases of development, but for different
reasons. Egg, first instar, and second instar are by far the shortest developmental stages

in flies (Figure 3.2), thus identification of any of these simply described development

65

more accurately than did the much longer third instar and pupation. Weight and length
on the other hand are related to feeding, and changed in a relatively linear fashion during
the early larval stages, including the first portion of the third instar (e.g., Figure 3.1), but
once feeding ceased their utility dropped dramatically due to the reversal in body size and
larger overall variance in length and weight. Likewise, pupal size was of little utility as it
is static throughout the stage. As a result, model 18, which used all available
information, predicted a restricted pupal development of 61 .9% to 81.2% (Figure 3.4h,
95% confidence intervals) and showed no specificity (that is, the youngest or oldest
individuals were equally likely to be placed anywhere within that range). This means a
pupal development prediction with the best GAM was essentially the same as using stage
alone. Given these effects on predictive ability, it is not surprising that adding weight
and/or length to stage resulted in minimally improved models.

Second, error increased as development progressed for all models, indicated by
the gaps in predictive ability and the widening confidence intervals for successive
developmental stages (Figure 3.4g,h), which were most pronounced in postfeeding third
instars and pupae. The increasing inaccuracy of age approximation as fly development
progresses has been noted in the literature (Wells and Kurahashi 1994, Wells and
Lamotte 1995), and forensic entomologists account for it in PMI estimates by giving
large age ranges to postfeeding flies, although these rarely include an objective estimate
of error. The latter study (Wells and Lamotte 1995) used linear models to estimate blow
fly age based on length data, and yielded an increase in error for older larvae. The
similar findings indicate that there is a limit to the precision in blow fly age predictions

that can be achieved when only developmental stage and body size are evaluated. Owing

66

to this, alternative developmental data independent of basic grth are likely required to
increase the accuracy of PMI predictions, and in the future, traits that change regularly
during fly development, such as hormone titers or gene expression, may be useful in
generating a more specific PMI.

Third, the limited influence of fly strain and rearing temperature on development
is an important consideration as it indicates the models have value regardless of where
flies are collected or at what temperature they develop. This is not to say that
temperature is unimportant when making PMI estimates—it is critical, and is always
considered when estimating PMI (usually as accumulated degree days). However,
temperature did not alter developmental patterns to any large extent, although lower
rearing temperatures did result in slightly larger individuals overall for all strains, a
finding we continue to investigate. Similarly, the strains of flies examined had different
average sizes during development (Figure 3.3). For both criteria the distributions of body
size throughout development overlapped, so these data modified age predictions
minimally. Also, there was little difference in size among strains during the feeding
stages of the lifecycle, where size best predicts age, thus size variability resulting from
strain adds no confounding information during those stages.

Fourth, any given forensic case may present the entomologist with different data
from which to estimate fly age. While developmental stage was the most useful datum
for the development estimates in this study, other data, such as weight and length,
increased their accuracy. Using a model that incorporates all available data can help
ensure that investigators make the best possible prediction with the information they

receive, while maintaining an understanding of the limitations inherent in that model. An

67

estimate of the relative reliability of a PMI prediction (based on the GCV and percent
deviance explained) provides an understanding of its value in interpreting evidence.
Overall, generalized additive models offer a useful means of incorporating information
from multiple linear and non-linear variables to predict blow fly age, variables that can be
accommodated even if they change from one case to the next.

Finally and most importantly, a comparison of modeled development predictions
to the independently derived rat data made it possible to assess error rates and produce
confidence intervals in these estimates. Individuals less than 26% developed (feeding
larvae) generated the most accurate predictions; when 12 individuals of the same age
were sampled from a cohort, the predictions clustered around the known age in all
instances (Figure 3.5). In contrast, postfeeding stages had a much larger error rate. It is
worth noting that even when model 10 yielded its best estimates, there was still about
10% total variance in predicted ages of larvae (note again, at a true age of 15%, the 95%
confidence interval for rat data predictions was between c. 10—22%, thus stage and size
were not ‘perfect’ in estimating development even in the youngest individuals. The
utility of the methodology presented here is that it establishes a defined way to produce
confidence intervals around entomologically based PMI predictions, regardless of fly age.
Until new and independent variables that change in a predictable manner during
development are incorporated into age estimates, this error will necessarily exist, and
increase with age, however through these models that error can objectively be
determined. Equipped with such knowledge the forensic entomologist can relay to the
court the level of error found in a PMI prediction. Through this feat, one of the major

requirements of Daubert is more fully addressed.

68

Table 3.1—The 18 generalized additive models for predicting development percent
assessed in this experiment.

MM Development Percent= Percent fl

1 Stage 89.5 0.045
2 s(Length) 63 .3 0.126
3 s(Weight) 65.7 0.144
4 s(Length,Weight) 86.8 0.059
5 Temperature 0.022 0.358
6 Strain 0.0041 0.358
7 Stage+Strain 89.5 0.044
8 Stage+Temperature 89.7 0.044
9 Stage+Strain+Temperature 89.7 0.044
10 Stage+s(Length) 91 .2 0.038
1 l Stage+s(Weight) 90.8 0.04

12 s(Length)+s(Weight) 85 .9 0.064
13 Stage+s(Length)+s(Weight) 91 .6 0.036
14 Stage+s(Len gth )+s(Wei ght)+s( Len gth,Wei ght) 92 0.03 5
1 5 Stage+s(Length)+s(Weight)+Temperature 91 .8 0.036
16 Stage+s(Length)+s(Weight)+Strain 91 .6 0.036
1 7 Stage+s(Len gth)+s( Wei ght)+Temperature+Strain 92 0.036
1 8 Stage+s(Length)+s(Weight)+s(Length,Weight)+Temperature+Strain 92.6 0.034

s(variable)—Indicates that a smoothing curve was used in the GAM for this variable.
s(Length,Weight)—Indicates that a smoothed contour surface of length plotted against
weight was used in the GAM.

Development percent-Indicates the variables used in each model to predict development
percent.

Percent-Indicates the percent deviance explained.

GCV—Generalized cross validation score; lower scores are better models for predicting
development percent.

69

Table 3.2—Summaries of the estimated significance of each variable in a model.

Model Parameter

VOIO'IPOON—A

CD

10

11

12

13

14

15

16

Stage
Length
Weight
Length,Weight
Temperature
Strain

Stage

Strain

Stage
Temperature
Stage

Strain
Temperature
Stage
Length
Stage
Weight
Length
Weight
Stage
Length
Weight
Stage
Length
Weight
Length ,Weight
Stage
Length
Weight
Temperature
Stage
Length
Weight
Strain

df or edf Chi-cg

4
8.734
8.026
26.11

1
2
4
2
4
1
4
2
1
4

8.318
4
4.738
6.345
8.277
4
7.33
3.902
4
2.381
4.765
26.99
4
7.356
3.953
1
4
7.272
4.236
2

70

13658
3275.5
2629.3
11420
0.68
0.13
13757
17.408
13669
3.26
13767
17.195
3.0566
4921.9
120.38
6209.5
143.62
4109.9
3668.9
912.31
30.746
58.757
630.48
2.8875
17.601
114.66
901.98
33.427
59.039
12.533
865.21
33.759
64.411
10.695

P-value

<0.0001
<0.0001
<0.0001
<0.0001
0.41
0.94
<0.0001
0.0002
<0.0001
0.071
<0.0001
0.0002
0.081
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
0.3
0.003
<0.0001
<0.0001
<0.0001
<0.0001
0.0004
<0.0001
<0.0001
<0.0001
0.0048

Table 3.2 Continued
Model
1 7

18

df—Degree of freedom.

Parameter

Stage
Length
Weight
Temperature
Strain

Stage
Length
Weight
Length,Weight
Temperature
Strain

df or edf

4
6.612
4.182

1

2

4
3.842
8.498
16.03

1

2

Chi-cg

871.25
34.945
71.738
15.726
13.646
606.21
9.5722
5.3936
115.11
32.872
34.101

edf—Estimated degree of freedom (non-parametric variables).
Chi-sq—Chi-squared score.
P-value—As estimated from likelihood scores for each parameter in a model.

Figure 1

71

P-value

<0.0001
<0.0001
<0.0001
<0.0001
0.0011
<0.0001
0.044
0.76
<0.0001
<0.0001
<0.0001

CHAPTER 4: GENE EXPRESSION IN BLOW FLY EGGS

Introduction

Insects found on human remains can be useful in estimating a postmortem interval
(PMI) during death investigations (Catts and Haskell 1990). Primary among these are the
blow flies (Diptera: Calliphoridae), whose state of development when collected fiom a
corpse can be compared to published tables of juvenile fly growth, in order to
approximate when the eggs were deposited. As development continues, the larvae pass
through three instars and then move away fi'om their food source in order to pupate. For
many necrophagous fly species, including the widely distributed blow fly Lucilia
sericata, growth rates are well defined (e. g., Kamal 195 8, Greenberg 1990, Anderson
2000, Grassberger and Reiter 2001). However, developmental stages necessarily exist
over a period of time, in some cases several days, making precise PMI estimates difiicult.
Given this, any information that could be added to fly development stage data has the
potential to generate a more precise PMI.

While outward characteristics such as body size or instar have generally been
used to estimate fly age, other traits that are developmentally regulated, including the
differential expression of genes, offer great potential as an independent source of data for
estimating blow fly age. Developmental biology research has uncovered numerous
instances of gene expression changes throughout maturation (see Kalthoff 2001, and
references therein). Flies have been particularly well studied in this regard (Henrich and
Brown 1995, Arbeitrnan et al. 2002, Skaer et al. 2002, Sullivan and Thummel 2003,

Luders et al. 2003, Ali et al. 2005, McGregor 2005), including the Calliphoridae.

72

Predictable changes in gene expression during development led to the hypothesis tested
here, that differential gene expression could be used to make more precise PMI
predictions, by effectively breaking a developmental stage into smaller developmental
units. Towards this goal, mRNA levels of three genes differentially expressed in
Drosophila melanogaster eggs (Arbeitrnan et al. 2002), bicoid (bcd), slalom (sll), and
chitin synthase (cs), were assayed in L. sericata. Eggs were chosen because there is no
quantitative means of assessing their degree of maturity, and if egg aging is attempted at
all, investigators must rely on a qualitative evaluation of embryos, making it difficult to
objectively divide the stage into developmental subgroups. bcd is required early in egg
development, defining the anterior end of the egg during the formation of the anterior-
posterior axis in Cyclorrhaphan flies (McGregor 2005), including the Drosophilidae and
Calliphoridae. sll affects dorsal-ventral patterning (Luders et al. 2003), and is also highly
expressed in the salivary glands of D. melanogaster and L. sericata larvae (Ali et a1.
2005). cs was profiled as chitin formation is required for the proper assembly of the
larval cuticle (Tellam et al. 2000). Transcript abundances were assessed to directly test
the hypothesis that developmental stages of forensically important flies can be better
defined by combining expression information from specific genes, resulting in more

precise age estimates, as well as a more precise prediction of PMI.

73

Materials and Methods
Species Identification and Egg Collection.

L. sericata was collected in East Lansing, Michigan, and was identified visually
and genetically as previously described (Tarone and Foran 2006). A fly cage at room
temperature was presented with beef liver and examined every 15 minutes until females
were observed laying eggs, which was allowed to continue for one hour. Egg masses
(comprised of approximately 250 eggs) were placed on a moist paper towel in a petri dish
at 32°C, and whole masses were collected hourly until hatching of the remaining eggs
was observed. Sampled masses were immediately fixed in RNA Later (Applied
Biosystems, Foster City, CA) and stored at -80°C. Two replicates were collected for each
hourly age period. Prior to RNA extraction, egg masses were thawed and sliced with a
razor blade into fifths, resulting in the analysis of 10 groups of eggs for each one-hour
collection span. The first eggs hatched between 9 and 10 hours, thus the 8—9 hour time
span was the oldest analyzed.

Gene Sequencing and Primer Design.

Expression levels of bed, sll, and cs were compared to the steady-state expression
of two housekeeping genes (rp49 and ,6 tubulin 5 6D). L. sericata gene sequences were
available for bed, sll, and rp49 on the National Center for Biotechnology Information
website (www.ncbi.nlm.nih.gov), thus quantitative PCR primers were designed directly
from them using Applied Biosystems Primer Express software. [3 tubulin 56D and cs
sequences were obtained using primers for the D. melanogaster and L. cuprina genes
respectively (Table 1), taken from NCBI. PCR consisted of 35 cycles of denaturing at

95°C for 30 seconds, annealing primers at 55°C for 30 seconds, and extending amplicons

74

at 72°C. Extension times were 4 min for es and 2 min for [3 tubulin 5 6D. Sequences were
generated on a CEQ 8000 capillary electrophoresis system using a CEQ DTCS Quick
Start Kit and the same primers, following the manufacturer’s protocols (Beckman
Coulter, Fullerton, CA). PCR products were analyzed via agarose gel electrophoresis; a
single peak in dissociation curves of quantitative PCR (see below) confirmed the
electrophoretic evaluation.
cDNA Synthesis and Quantitative PCR.

Ninety RNA samples were isolated from egg masses in a 96-well format using an
ABI PRISM 6100 Nucleic Acid PrepStation and the manufacturer’s solutions and
protocols (Applied Biosystems). Eggs were placed in 300p.L of lysis solution without
use of a pre-filtration plate. RNA was eluted fi'om plates with 100uL elution solution and
incubated at 37°C for 1 hour with 70 units of DNase-I (RNase-Free, Roche, Basel,
Switzerland) and Ambion DNaseI buffer (Applied Biosystems). The enzyme was
inactivated at 75°C for 10 minutes and the RNA was precipitated using 110uL of
isopropanol followed by centrifiIgation at maximum speed for 1/2 hr at 4°C. Two 70%
ethanol washes followed, using the same centrifuge settings. RNA samples were allowed
to air dry for 15 minutes, at which point 32 uL Ambion RNase-Free water (Applied
Biosystems) was added to the pellet prior to freezing at -80°C.

cDNA was synthesized using a TaqMan Reverse Transcriptase kit (Applied
Biosystems) primed by oligo(dT) 16-mers according to the manufacturer’s instructions,
including 30uL of RNA in a final volume of l20uL. Gene expression levels were
assessed by quantitative PCR on an Applied Biosystems 7900HT using SYBRgreen PCR

mastermix (Applied Biosystems) in lSuL reactions on a 384-well plate. Each reaction

75

received 1.5 uL cDNA, 7.5 uL SYBRgreen PCR mastermix, and l uL each of forward
and reverse primers. The Applied Biosystems recommended thermal cycling parameters
were used with the exception that PCR cycles were increased to 50 and a dissociation
curve was produced for every reaction. Results were considered valid if a single peak
was present in the dissociation curve, indicative of a single amplicon being produced.
Optimized primer concentrations, based on trial runs designed to ascertain concentration
combinations that provided the largest signal to noise ratio in dissociation curves, are
found in Table 4.1.

Reactions without reverse transcriptase acted as controls to confirm that
amplification in quantitative PCR was not due to residual DNA. A known (positive)
cDNA sample was analyzed in triplicate during every run, allowing for comparisons
among 96-well plates, resulting in ninety cDNA samples (10 per time point, in duplicate),
six negative controls (PCR mix with no DNA), and six positive controls being assessed
for each locus.

Statistical analyses and the construction of plots were performed in the R
statistical program (R core development team 2004). Linear regression models were
analyzed via type III AN OVA. Standardized gene expression through time was plotted
for samples that yielded detectable levels of a transcript. The use of gene expression to
assess age was examined via generalized additive models (Hastie and Tibshirani 1990,
Wood 2006), which produce a statistic, percent deviance explained (similar to R2),
assessing the extent to which a variable influences the data. Predictions (fitted values)
for the data were plotted against true ages (response), allong for evaluation of the

model’s ability to predict the egg ages.

76

Final CT values for all loci were generated using the average of duplicate PCRs.
CT values of rp49 and fl tubulin 5 6D were averaged and subtracted from those of the
developmentally regulated genes to obtain a standardized CT. Regression curves were
drawn through standardized plots. Binary gene expression values (1 = present, 0 = not
present) were also assessed to determine if the presence or absence of gene expression
corresponded to a particular age. A locally weighted sum of squares curve was drawn
through the resultant plot. Generalized additive models then allowed prediction of egg
age with CT scores and binary values. One model used binary cs expression and CT
information from the other loci to predict age (N = 55). The other estimated age with CT
data for all loci (N = 33). Sample sizes were smaller than the total as only egg masses

that provided data for all loci were included in analyses.

Results

L. sericata sequences for ,6 tubulin 5 6D and cs are listed under the National
Center for Biotechnology Information accession numbers EF056211 and EF056212
respectively. cs best matched its L. cuprina homolog (98% with no gaps) and ,6 tubulin
5 6D exhibited the closest similarity to the fly Glossina morsitans morsitans tubulin beta-
1 gene (86% identity with no gaps); no Lucilia sequence was available for the latter
comparison.

Eighty-four of the 90 samples yielded rp49 and ,6 tubulin 56D profiles, which
demonstrated consistent expression levels throughout egg development. There was a

significant positive relationship in expression of the two housekeeping genes (P<0.0001,

77

R2=0.63), confirming their utility as internal standards. Of the 84 samples, bcd, cs, and
sll had an undetectable transcript level in 23, 31, and 20 samples respectively.

The developmentally regulated genes demonstrated qualitative and quantitative
differences in expression throughout egg development. cs was the only gene that showed
a consistent binary (on/off) pattern, with egg masses less than two hours old never
producing the transcript, while those 6 hours and older always expressed the gene,
therefore cs expression state could be plotted during egg development (Figure 4.1). The
presence of the transcript was a statistically significant predictor of age (P<0.0001,
R2=0.59).

Each of the genes had a different quantitative expression pattern (Figures. 4.2—4;
note that the displayed CT values are inversely related to gene expression levels).
Though only expressed after hour 2, cs transcripts significantly increased during egg
development (P=0.0004, R2=0.21). Conversely, bed and sll transcripts were at their
highest levels and lowest variation at 0—2 h, and significantly decreased as development
proceeded (P=0.0003, R2=0.19 and P=0.0023, R2=0.13 respectively).

Finally, generalized additive models were used to predict egg ages based on the
gene expression data. The first model used the binary expression data for cs and CT
scores for bcd and sll to predict egg age. It explained 72.1% of the deviance in the data
and accurately enabled the identification of egg masses as either 0—4 or 2—9 h old (not
shown). Next, CT scores for all three genes were used to predict age, which explained
76.7% of the deviance in the data. When predicted versus true ages were plotted (Figure
4.5), estimated ages followed the True = Predicted line, with 30 of 33 predictions within

2 h of the true age.

78

Binary chitin synthase Expression

 

1.0

A - - A -
- O v v v -

 

0.8
J

0.6

cs Expression State
0.4

0.2

 

 

0.0

 

Hours

Figure 4.1. Binary gene expression profile for cs at 32°C in L. sericata eggs, from 0—1
through 8—9 hours of development. 0 indicates no detectable expression of the gene
and 1 indicates detectable expression of cs. Expression was not detected from 0—2 hours.
Between 2 and 6 hours some eggs clusters expressed cs and some did not. Afier 6 hours,
all egg clusters expressed cs.

79

Standardized chitin synthase Expression

 

15
1

Standardized cs Threshold Cycle

 

 

 

 

Hours

Figure 4.2. Standardized expression of cs in L. sericata eggs, from 0—1 through 8—9
hours of development. The CT was standardized against the average of rp49 and ,6
tubulin 5 6D CT values and plotted through time. The regression of cs CT over time was
also included. High expression levels are indicated by low CT values. cs was not
expressed from 0—2 hours and then its expression increased.

80

Standardized bicoid Expression

 

O
OO

Standardized bcd Threshold Cycle

 

 

 

 

Hours

Figure 4.3. Standardized transcript abundance of bed in L. sericata eggs, from 0—1
through 8—9 hours of development. Standardization was as in Figure 4.2. bcd gene
expression was highest from 0—2 hours, then transcripts decreased in abundance.

81

Standardized slalom Expression

 

15
1

Standardized sll Threshold Cycle

 

 

 

 

é
o
g. . , .
3 8
I i T I I
0 2 4 6 8
Hours

Figure 4.4. Standardized transcript abundance of sll in L. sericata eggs, from 0—1
through 8—9 hours of development. Standardization was as in Figure 4.2. sll
expression was highest from 0—2 hours, then tended to be lower as eggs developed,
though variance in expression was high.

82

Response vs. Fitted Values

 

q 0 0CD
-* CIDCDO

Response
2 3 4 5 6 7 8
I
o

 

 

 

Fitted Values

Figure 4.5. Predicted (fitted) versus true (response) ages for a generalized additive
model that made age predictions for the 33 egg masses that expressed bed, sll, and
cs. Estimated ages were within 2 h of the true age in 30 of the 33 cases.
Discussion

The goal of this research was to examine the feasibility of using gene expression
to more precisely age immature flies of forensic interest, consequently generating more
accurate estimates of PMI. The loci examined demonstrated statistically significant,
though noisy, trends in expression levels throughout egg development. Additionally, egg
masses less than 2 h old did not express cs, while egg masses older than 6 h always
expressed the gene. Following on efforts to predict adult mosquito age using gene

expression and multiple regression (Cook et al. 2006), generalized additive models were

used to predict egg ages. When CT scores were available for all loci, 91% of predictions

83

were within 2 h of the true age, while the binary cs data combined with bed and sll CT
scores separated the egg masses into two distinct groups.

A key factor in aging flies using gene expression is, of course, examining loci
likely to vary in expression levels during the developmental period being examined. In
eggs, genes that are important for developmental patterning (e. g., dorsal/ventral) are
crucial for successful growth of the individual, thus their expression is under strict
biological control. During very early fly embryogenesis high levels of maternally
derived bed and sll products are necessary to properly establish biological axes (Kalthoff
2001, Arbeitman 2002, Luders et al. 2003, McGregor 2005). In the current study,
transcript levels of both genes dropped steadily through embryonic development,
although neither became undetectable. In D. melanogaster, bed is detectable only during
early embryonic development (Arbeitman et a1. 2002), thus its expression throughout
embryogenesis in Lucilia, albeit at decreasing levels, is somewhat puzzling
Developmental heterogeneity among eggs may have resulted in this phenomenon, while
it is also possible that bed serves some unknown secondary function in blow flies, or that
the transcript is stable but untranslated in older eggs. In contrast, the existence of sll
transcripts at the end of the egg stage can be accounted for, likely resulting from
endogenous sll expression commencing in the developing salivary glands (Luders et al.
2003). Finally, cs, which is required only for production of the larval cuticle (Tellam et
al. 2000), followed a different and predicted transcriptional course, wherein the earliest
portion of egg development was defined by an absence of transcripts, the middle portion
of development, as larval cuticle begins to form, was represented by low to intermediate

levels of cs expression, and the highest levels were found late in egg development. Most

84

importantly, all three loci showed significant trends in egg transcript expression over
time, and taken together increased the precision of egg age estimates.

Given that gene expression has the potential to more accurately age flies of
forensic interest, other factors need to be considered, including both the feasibility, and
legal acceptance, of the methods. The molecular techniques employed in this study have
been widely utilized in developmental molecular biology, and as important, could readily
be implemented in most laboratories equipped for forensic DNA investigation. They also
provide information that can be used to generate predictable error rates/confidence
intervals, meeting one of the major tenets of Daubert, an important consideration of any
new forensic protocol. The methods are amenable to microarrays (e. g. Arbeitman et al.
2002) and robotics, potentially producing simplified and high throughput blow fly aging
analyses. Finally, DNA-based methodologies overall have been widely accepted in
courts of law, thus new but related methods should have less difficulty overcoming
Daubert challenges.

The data presented here demonstrate that even the briefest phase of fly
development, the egg stage that lasts only several hours, can be divided into smaller
periods using gene expression data. Naturally, other stages of fly development,
particularly those that last longer and therefore are more forensically challenging (e. g.,
the third instar and pupation) can be examined using these methods as well. Addition of
more developmentally regulated genes into the analysis should further increase the
precision of age estimates by providing more age-informative data. The final outcome of
this is a more precise age given to developing blow flies, resulting in more precise

estimates of PMI.

85

ooo _
ooo o
ooo _
ooo o
oov
oov
oov
$.00
oov
$60
oov
oov
oov
oov

 

.5382 Mom :28 3 wow: 3253 we 928 5555050 :25 05
mowocoo 535.5850 6035.6 203 £255 05 EEK Sea 8558 05 88:5 L385: 5583‘

$232
$28:
58.132
332.22
Rs: :2

H<HUOUUOU<U<<<HOH<OO
<OOHOUUU<H<HUUOHU<<O

UOHHOH<UPHOUH<O<UU<U
<U<CUH<UOHU<EUU<O<OU
E<OUO<OHOUU<U<O<OO

Sow— _m< OEO<<OUHU<<OO<<EOH<<U<

vnmomo><
Ehomo><
N _ Nowomm
Q Snowm—
_ Gomomm
_ .momon—m
©3394
omwhom?‘

2: 5:22.350 Lena—I... Z a. 353134.-

H<<UOUUOEOHOO<EOU
<HO<<EUP<<U<UUOOU<<UUH
<0<HOOOHOHH<UHOHUOOH<O
<UU<H<HUU<<O<OOU<OUUO
U<O<<OHO<<<<<OH<UOO<UU<
U<<UHOUUOOE<<<HOUUH
<<OE<UUHHU<UOOHU<HU<<U

355800805850
good

mfiocosoom

wfiocozoom

wcmocozoom

wcmocozoem
mum 35:55
5.. 35:55
m8 2:555
«on. 35:55
mus 3:555
an: a55.55
mos ”25:55
Mom 9:55:50
mom 3:83.55
«Us 2:555

leolzeulam

mm 5
E 5

E 533
E 5.3.;
a a}?
a $33
m =5

a :5

e 5a

e 5:

a 5:. .

e 5:.
m 8%
a some
3.5...—

 

éa... a... a 5...: seats .3. e35.

86

CHAPTER 5: PREDICTING BLOW FLY AGE WITH LARVAL AND
PUPAL GENE EXPRESSION

Introduction

Blow flies (Diptera: Calliphoridae) are important parasites, parasitoids, and
primary successional species on carrion. Given their parasitic and necrophagous life
histories, they affect human activities regularly, as they can spread disease (Faulde et al.
1995) and are parasites of sheep (East and Eisemann 1993), cattle (Crystal and Ramirez
1975), and humans (Sherman 2000), which can cause economic damage and human
health hazards. Also, the carrion-feeding behavior of many blow flies makes them
valuable evidence in death investigations. Forensic entomology is a forensic science
discipline that can aid investigators in the determination of a postmortem interval (PMI)
(Catts and Haskell 1990, Greenberg and Kunich 2002). The estimate of a PMI is possible
because necrophagous flies, especially blow flies, are capable of colonizing remains and
progress reliably through development. Blow flies can lay eggs on wounds or orifices
within hours of death. The eggs hatch into larvae, which feed on carrion and progress
through three instars (periods of larval development that are separated by molts of the
cuticle). Eventually, third instar larvae leave the food source to form a puparium,
wherein metamorphosis occurs, leading to the eventual eclosion of the adult. Each
developmental stage has a defined, temperature dependent, duration that can be used to
estimate the age of an insect collected as evidence. Reported datasets outlining the
minimum development times of each stage (Kamal 1958, Greenberg 1990, Wells and
Kurahashi 1994, Anderson 2000, Byrd and Allen 2001, Grassberger and Reiter 2001,

Grassberger and Reiter 2002) can be used to backtrack from the developmental stage of

87

an insect associated with a decedent to the time that insects colonized the body. The
period of colonization is generally assumed to be the minimum PMI.

The fundamental methods for estimating blow fly age have remained unchanged
for decades (Greenberg and Kunich 2002). An investigator collects or receives insect
evidence that was associated with a body. The species and developmental stage are
identified. Since development is temperature dependent, historical temperature data,
usually from weather stations (Catts and Haskell 1990), can be used to estimate the
temperature at the death scene. Using a development table that outlines the durations of
each stage for the identified species, in conjunction with temperature data, it is possible to
determine the minimum and maximum time necessary for that species to achieve the
stage that was collected as evidence, resulting in a calculation of the time of colonization
of the remains (Catts and Haskell 1990). However, there are several problems associated
with the method. As blow fly development progresses, stage durations increase resulting
in accurate, but imprecise, estimates of fly age. The imprecision is due to the
progressively larger windows of time that must necessarily be placed around an age
estimate. For example, according to Grassberger and Reiter (2001), the first instar of
Lucilia sericata (Diptera: Calliphoridae) (Meigen) lasts a minimum of 24 hours at 20°C,
while the pupal stage is at least 209 hours. The windows of time around an age estimates
are longest during the postfeeding third instar and pupal stages. The postfeeding third
instar has a minimum duration that ranges from 82 hours at 34°C to 200 hours at 17°C,
while the pupal stage can last between a minimum of 120 hours at 34°C to 442 hours at

17°C (Grassberger and Reiter 2001).

88

Body size measurements, such as length and weight, can improve the precision of
age estimates, but only help to refine predictions for first, second, and early third instars,
when the animals are actively growing. Body size does little to refine predictions of
postfeeding third instars, as larvae cease feeding and actually begin to shrink during this
period (Greenberg 1990, Anderson 2000, Grassberger and Reiter 2001). They also
exhibit increased variance in body size, which adds to the difficulty of using size to
estimate age (Wells and Kurahashi 1993, Wells and Lamotte 1995, Chapter 3).
Measuring length and weight does not help in pupal age estimates as pupae do not change
in size during development. The failure of body size to refine age estimates for the last
two stages of the immature lifecycle, means that predictions of age during those times
remain imprecise. In addition, difficulties in distinguishing between feeding and
postfeeding third instar larvae (reviewed by Anderson 2000) can compound the problem
as predictions of undetermined third instars must include estimates for the entire stage,
instead of just the feeding or postfeeding portion.

The difficulties associated with the status quo approach to estimating a PMI based
on blow fly evidence must also be considered in the context of the legal requirements of
forensic sciences. In the United States, the Daubert ruling on scientific evidence
(Daubert v. Merrell Dow Pharmaceuticals, 509 US 579 (1993)) has raised the
expectations for all forensic sciences, as it requires scientific methods to be tested and
published, using standard operating procedures. The procedures applied must be
commonly accepted in the field of study. Likewise, Daubert requires an understanding of
error associated with the methods employed. Forensic entomology is based on solid

principles in developmental biology and is supported by a body of literature that is

89

accepted by the scientific community; however, different growth rates for immature blow
flies have been reported, in part due to a lack of standardized rearing protocols
(Greenberg 1991, Grassberger and Reiter 2001, Kamal 1958, Wells and Kurahashi 1994,
Anderson 2000). Further, forensic entomologists have estimated different ages for the
same insect evidence, depending on which growth data they use (California v.
Westerfield, CD165805). Similarly, there is a dearth of research on estimating error rates
of PMI estimates derived from blow fly development tables (Wells and Kurahashi 1993,
Wells and Lamotte 1995). Clearly, by addressing the lack of standard operating
procedures and statistical understanding in the field, the acceptability of forensic
entomology in US courts will be bolstered.

A new approach to predicting blow fly age may be warranted to improve
shortcomings of the status quo method of PMI prediction. Any data that can be used as a
reliable indicator of specific developmental time periods, especially within the
postfeeding third instar and pupal stages when growth is non-linear, will be useful in
making more precise age estimates. Gene expression levels are a potential source of new
information that could help improve the precision of PMI predictions. As organisms
develop they require the expression of proteins in various tissues at specific times in
development (Kalthoff 2001). The temporal requirements of cells for proteins means that
the genes encoding them will be up— and down-regulated in accordance with the
developmental needs of an organism, and may be useful in predicting the age of an
individual. The theoretical application of gene expression analyses to blow fly age
prediction is well supported by genomic research in the model organism Drosophila

melanogaster (Diptera: Drosophilidae). Time-series microarray analyses in Drosophila,

90

which have assessed the expression levels of thousands of genes at once, indicate that
there are myriad temporally predictable gene expression patterns during fly development
(Arbeitman et al. 2002, Beckstead et al. 2005). By choosing the right suite of genes that
are up- or down-regulated at different times in development, it may be possible to
identify very specific portions of development. For example, during pupation the genes
Amalgam and CG17184 are expressed at their highest levels during early and late
pupation, respectively (Arbeitman et al. 2002). With an evaluation of the expression
levels of just these two genes, it should be possible to distinguish between early, middle,
and late pupal development. Given the molecular, genetic, and physiological similarities
between the Drosophilidae and the Calliphoridae (McAlpine 1989, McGregor 2005, Ali
et al. 2005, Yeates and Weigmann 2005, Mellenthin et al. 2006), the Drosophila model
can help to target blow fly genes with an a priori expectation of informative regulation.
Ultimately, predictable temporal variation in gene expression must be
demonstrated in blow flies to make estimates of blow fly age, and thus PMI, feasible.
The blow fly species studied herein, L. sericata, was chosen because it is globally
distributed and forensically informative (Kamal 1958, Greenberg 1990, Anderson 2000,
Grassberger and Reiter 2001) that has been studied at a molecular level in several
instances (e. g. Mellenthin et al. 2006). Gene sequences have also been produced for a
sister species, L. cuprina, due to the economic effect of myiasis on Australian sheep OEast
and Eisemann 1993). This means that gene sequence information could be obtained from
the public domain, or easily sequenced in L. sericata by designing PCR primers from L.

cuprina sequence.

91

The research detailed herein was designed to address the hypothesis that estimates
of L. sericata age could be made more precise by including gene expression data in the
prediction process. To determine the usefiilness of gene expression, profiles of 12 genes
were produced from a time-series collection of immature L. sericata that was comprised
of 958 individual larvae and pupae. Since gene expression levels are continuously
variable (e. g. Gibson and Weir 2005) they can be considered quantitative traits, which are
subject to the influence of genetic and environmental factors (Mackay 2001, Conner and
Hartl 2004). If gene expression is to be used to predict blow fly age, then influences on
variation in expression must be accounted for. Accordingly, the potential for genetic
effects was assessed by evaluating gene expression among three regional strains of L.
sericata (originating from Davis, CA; East Lansing, MI; and Morgantown, WV).
Likewise, the profiles produced at two rearing temperatures (20°C and 33.5°C) were
assessed, to determine if temperature affects gene expression and the age estimates
resulting from them. The data produced enabled the evaluation of gene expression as a
means of predicting blow fly age, revealing valuable information regarding the use of

transcript level abundances to estimate a PMI.

Materials and Methods
Species Identification, Fly Rearing, and Collections.

L. sericata rearing methods and species confirmation are detailed in Chapter 3.
Each collected individual was immediately frozen in RNAlater (Applied Biosystems), at

—80°C afier length, weight, and developmental stage were recorded.

92

Sequencing of L. sericata Loci. Several loci required sequencing before quantitative
PCR primers could be designed. These included 13 Tubulin 56 D, chitin synthase (both
detailed in Chapter 4), acetylcholine esterase, ecdysone receptor, ultraspiracle, white,
scalloped wings, and rhodopsin 3. Sequences from the closest dipteran relative available
at (www.ncbi.nlm.nih.gov) were used to design PCR primers for the locus, targeting at
least a 300 bp segment. Sequencing primers and the sequences used to design them are
listed in Table 5.1. Sequencing methods were as in Chapter 4, with slight modifications
of annealing temperature, extension times, and the number of cycles depending on the
gene in question. Sequences were subsequently compared to known sequences, via
BLAST comparison (N CBI), to confirm that the appropriate sequence had been obtained
(Table 5.2).
RNA Extraction.

RNA was isolated from a subset of the 2559 individuals described in the Chapter
3. Five individuals per time point were examined from the Michigan and California
replicates raised at 20°C or 33.5°C, and from the West Virginia replicates raised at
33.5°C. The RNA isolation method from Chapter 4, conducted in a 96-well format on an
ABI PRISM 6100, was modified for use with larvae and pupae. Modifications stemmed
from the fact that adding too much lysed tissue to the wells prevented solutions fi'om
being drawn through the filters. Flies were ground in 300 uL of lysis solution by hand
with a sterile pestle. To prevent filter clogs, lysates from larvae greater than 10 mg in
size and fiom pupae were diluted in RNA lysis solution (Applied Biosystems); 20 uL of
larval lysates or 40 uL of pupal lysates, were placed into 300 uL of additional RNA lysis

solution. The dilutions were determined in preliminary experiments to establish the

93

largest volume of preliminary lysate that did not clog the filters of the 96-well RNA
isolation plates. The diluted 300 uL lysates of individual larvae and pupae of any size
were then drawn through a 96-well filter plate (Applied Biosystems), which removed
large particles, larval cuticle, and pieces of puparium from the solution before it was
added to the RNA isolation plate. All other steps were followed using the manufacturer
suggested protocol, with a final RNA elution volume of 100 uL. After RNA purification
a DNase I reaction removed any remaining DNA contamination as described in Chapter
4.
Reverse Transcription and Quantitative PCR.

cDNAs and controls were developed as in Chapter 4, except that a High Capacity
cDNA Archive Kit (Applied Biosystems) was utilized , according to the manufacturer’s
instructions. Once the no reverse transcriptase controls were shown to be negative (DNA
free) using the rp49 primers, quantitative PCR was performed with primers for all genes
(loci, qPCR primers, and primer concentrations are in Table 5.1). All qPCR products
yielded one product of the appropriate size when checked by gel electrophoresis, and
dissociation curves were consistent with a single product. Any reactions that had deviant
dissociation curves were eliminated from the study.

Quantitative PCR was performed using the same cycling parameters as in Chapter
4, however quantitative PCR was set up and analyzed in a 384-well format using a
Biomek 2000 Automated Workstation (Beckman Coulter), not by hand. qPCRs were
performed with 10 uL reactions, which consisted of 2 uL of cDNA. Power SYBR Green

PCR Master Mix was used in the reactions.

94

Samples of cDNA from an individual were divided into two wells and the average
CT was used as the score for that gene and individual. Scores were standardized against
an internal standard value derived from the average of the housekeeping gene CTs (rp4 9
and ,6 tubulin 5 6D) by subtracting the housekeeping score from the gene CT. In addition,
reactions were standardized against each other by setting the CT for the positive control
reactions to the same value on every 384-well plate in the experiment (positive control
reactions were run with rp4 9 primers and a standard cDNA sample).

Statistical Analyses.

Statistics and graphs were produced using the R statistical package (R core
development team). Several plots were used to understand the data. Gene expression
was compared for each developmental stage using boxplots of standardized gene
expression CT scores for each locus. Descriptions of boxplots can be found in Chapter 3.
Standard CT values for each locus were plotted against minimum development percents
(as detailed in Chapter 3), and locally weighted sum of squares curves were drawn
through the non-linear data, allowing comparisons of average expression to temperature
treatments and average expression among strains. For some graphs (see Results), a gene
for which no transcript could be detected was assigned a CT of 50 (the maximum cycle
tested) if the housekeeping genes gave a positive result. This kept the graphs from being
skewed due to missing data. Values were then plotted against minimum development
percents. Corrected expression plots were placed next to standard plots for a locus (no
transcript results were removed). Left-right panel comparisons demonstrated an increase
in the standardized CT value in the right panels at points where gene expression was

absent.

95

GAMs (Hastie and Tibshirani 1990; Wood 2006) were developed to assess the
effects of gene expression (quantitative and binary data), body size, developmental stage,
temperature, and strain on predicting minimum development percent. Each variable was
assessed on its own, or together with only the linear data (the stage, temperature, and
strain variables were only included if the variable was not a significant predictor of age
on its own), to determine its usefulness in predicting blow fly age. The variables were
used in larger models to determine the best GAMs for predicting blow fly age (Table
5.3). The models construct smoothing curves for non-linear data, which are cumulatively
used to predict the age of a fly. However, the distribution of data around that smoothed
curve must also be identified. The distribution is applied to the smoothed curve via a link
function, thus each model was assigned a distribution and link function based on the
criteria discussed in Chapter 3. GAMs produce generalized cross validation (GCV) and
percent deviance explained (PDE) statistics, which are also explained in Chapter 3.

The sigrrificances of specific terms in the models were determined. Genes were
considered significant predictors of age if the estimated P-value associated with the
individual gene by itself, or in combination with developmental stage and/or temperature,
was below 0.025—0.006 (depending on the Bonferroni correction required). Once
individual genes were established as significant predictors of development percent, they
were used in combination with other terms in larger GAMs. Comparisons of the GCV
and PDE scores from the models helped to determine if gene expression was useful in
predicting blow fly age. Diagnostic plots of error in the models were also constructed

and are explained in Chapter 3.

96

Results
Gene Sequences.

Fly strains were identified as L. sericata (Chapter 3). The sequenced genes
underwent a BLAST search on the NCBI website (www.ncbi.nlm.nih.gov) (Table 5.2).
All exhibited a close match to the same gene in another fly species. Two genes also
demonstrated 99—100% matches to L. sericata sequences that were published after this
project commenced.

Quantitative PCR and Statistical Modeling.

The housekeeping genes demonstrated a strong positive correlation with each
other (r = 0.84), indicating that they were good surrogates for estimating RNA
concentrations (Figure 5.1). A line with a slope of 1 and a Y-intercept of —1 .2 (the
average CT difference between the two genes) ran directly through the plot of rp49 CT
versus the tubulin CT (Figure 5.1). The difference derived fi'om the use of the average of
both housekeeping genes resulted in standardization scores that were well within the
range of standardized CT values for the other genes analyzed; indicating that most of the
variance in CT scores was not due to variation in housekeeping gene transcript levels

(Figure 5.2 as compared to Figures 5.3—5.41.).

97

Positive Correlation Among Housekeeping Genes

 

Rp49 CT

 

 

 

 

20 25 30 35

Beta Tubulin CT

Figure 5.1. The positive correlation between housekeeping genes. The CT scores

followed a line with a slope of 1 (after accounting for the average CT difference of 1.2)
with a correlation of 0.84.

98

Variance Due to Housekeeping CT Ditierenoe

 

 

 

 

 

g
E
3 ° '
l-
v ' -
I
? - o
I I I I
0.2 0.4 0.6 0.8 1.0
Percent

Figure 5.2. Variance in gene expression due to the standardization against
housekeeping genes. Since the average of both housekeeping genes was used for
standardization, one half of the difference between rp49 and the beta tubulin gene was
plotted throughout development. The variance in this plot was relatively constant and
indicates that an approximate difference in expression of up to 4 fold (2 CT units) could
be explained by variation in housekeeping gene expression.

Of the 2559 flies detailed in Chapter 3, 1025 were used to produce gene
expression profiles. Once ~100 samples had been profiled, preliminary plots of the data
indicated that wg, SCI, and rh3 were unlikely to provide useful information (Figures 5.3—
5) and were dropped from further study. 95 8 of the samples yielded partial or full
profiles of the remaining 9 genes. These were comprised of 48 first instar larvae, 79
second instar larvae, 13S third instar larvae, 334 postfeeding third instar larvae, and 362
pupae. There were 260 and 272 individuals profiled from the CA and MI strains
(respectively) raised in the 20°C treatments. Likewise 149, 121, and 156 individuals

from the CA, MI, and WV strains (respectively) were profiled fi'om the 33.5°C

treatments. Full gene expression profiles were obtained for 501 samples.

99

Plots were constructed to show the expression levels of each gene by
developmental stage and by development percent. GAMs were used to assess the
usefulness of individual gene expression profiles (quantitative and binary) for predicting
blow fly age. Table 5.3 lists all GAMs investigated, with models 2—14 assessing the
relative usefirlness of individual genes. Table 4 lists the significance and degrees of
freedom for terms in the models. The results of GAMs and plots for individual genes are

listed below (Tables 5.3-4; Figures 5.3—41).

scalloped wings Expression

 

 

 

 

o
o
o
In
N " o
‘8 o o
I. go 0 o o o
‘8 D 0
e ‘3. ° ° ° ° 8 °
5 0 e 0 o
O o
C 00 0° 3
S In ° 0 0 ° 9
- o
‘D " 00° 0 o
8
o o
O o o o e
o 00
‘0- —i o 0
0° O
I I I I I
0.2 0.4 0.6 0.8 1.0

Development Percent

Figure 5.3. Standardized gene expression of scalloped wings throughout the
immature development of L. sericata. No apparent pattern of gene expression was
observed throughout development.

100

 

 

 

 

o o
‘0 — o o
N 8 0° 0 0
80° ,
8 ° 8
‘ <9
i- 8 0 8
o o o
E o o
‘2
g 9 7 8 o
a o 0 ° 9 8
o (b 8 o O
o g 0 g o) 8
o
9— 0% 9 0° 8 ° Q) o
030 o 0
o
o
I I I I I
0.2 0.4 0.6 0.8 1.0

Development Percent

Figure 5.4. Standardized gene expression of Wingless throughout the immature
development of L. sericata. No informative pattern of gene expression was observed
throughout development.

101

rhodopsin 3 Expression

 

 

 

 

o o
In __ o
N 0 o 8 o o
o o
8 oo o
I-— o o
0 a « <8° o
B o 0 Q’ o
'5 Do 0 O 0
g 00 0 go 0 o o o o
E O o O O
In _ o O
rt) '- 0g 0 (9 0 ° 8
8’ 0 co 8 ° 9
0:8 Q 0 0 ° 8
‘ w 8 o O
0% O
S “ o
o o
O
o
I I I I I
0.2 0.4 0.6 0.8 1.0

Development Percent

Figure 5.5. Standardized gene expression of rhodopsin 3 throughout the immature
development of L. sericata. No apparent pattern of gene expression was observed
throughout development.

Expression of Informative Genes.

The most informative genes for forensic purposes are those that have stark
changes in regulation during development, particularly the latter portion of development,
where aging flies is most difficult. Within the third instar: cs, ecr, hsp60, hsp90, rop-I ,
w, and usp were up- or down—regulated between feeding and postfeeding third instars.
The most informative overall were cs and ecr. cs was strongly down-regulated during the
postfeeding third instar, and it was often not expressed at all (or expressed at an
undetectable level). Conversely, ecr was highly up-regulated during the postfeeding
stage. The reliable expression differences of cs, ecr, and, to a lesser extent, the other
genes noted above are usefiil indicators of the postfeeding condition and can be used as
molecular markers for that stage. The overall behavior of each gene was as follows.

The first gene evaluated was cs (Figures 5.6—9; note throughout that higher CT

scores indicate lower gene expression levels). Of the 958 individuals assessed, cs was

102

not detected in 102. The gene was expressed at consistent levels, except during the
postfeeding third instar, when it was expressed at lower levels (Figure 5.4) or not at all.
cs tended to be expressed at higher levels at lower temperatures (Figure 5.5), with the M1
strain expressing less cs at lower temperatures than the CA strain (Figure 5.6), while no
expression level differences were apparent among the three strains at the high
temperature treatment (Figure 5.7). The CA strain exhibited a stronger tendency to not
express cs during the postfeeding third instar at 33.5°C (Figure 5.7). GAMs assessing age
with CT scores for this gene were only significant predictors of development percent
when included with developmental stage (Table 5.4). The GAM that included the binary
cs term (Model 2) exhibited a 0.3% higher PDE than the GAM that assessed age in terms
of stage alone (Model 1), while the quantitative data (Model 6) exhibited an increase of

1.6% in PDE (Table 5.3).

 

 

 

 

 

chitin synthase Expression by Stage
:3 o
o
o
o
a 4 a
E
p.
o n
v- "i e
8 2 r—
8 3 8
r3 2 — £— r.‘ Jil— ' .4.
(O u I I I
'0 . I ' I
c A
5 m _ l::j Ed :j . E
o - o O o
o
I I I I I
1 st 20d 3rd 3rdPF Pupa

Developmental Stage

Figure 5.6. Standardized chitin synthase CT scores plotted by stage. The postfeeding
third instar stands out as expressing significantly less cs than the other stages.

103

csExpresslonetHighandLewTemper-atures csExpreseionetnghendLowTemperahlres

 

 

25
l
00

15
r

Slammed CT
10
1
Standardized CT
10 15
I L

   

 

 

 

 

 

 

O -‘ o J
o o
—T I I I I I I I I I
02 04 0.6 08 10 0.2 04 06 08 10
PercentDeveioped PercentDeveloped

Figure 5.7. Standardized chitin synthase CT scores plotted by development percent
for (a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 33.5°C. The shift in the red line between (a) and (b) indicates a cluster
of individuals that did not express the gene at the time point where gene expression is at
its lowest levels for this locus.

104

cs Expression Among Strains at High Temperatures cs Expression Among Strains at High Temperatures

 

 

25

Standardized CT
10
I
Standardized CT
10 15
r r

 

 

 

 

 

 

O "‘ o —l
I I I I I I T fl I I
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
Percent Developed Percent Developed

Figure 5.8. Standardized chitin synthase CT scores plotted by development percent
at 33.5°C for (a) all individuals that expressed the gene and (b) in all 958 individuals,
with NA expression values assigned a CT of 50. The red line indicates gene expression
in the CA strain. Blue indicates gene expression in the M1 strain. The green line
indicates gene expression in the WV strain. The shift in the red line between (a) and (b)
indicates a cluster of individuals that did not express the gene at that time point.

105

a. b.

 

 

«Expression BetweenStralnsatLowTemperatures cs Expression Between Strainsat LowTemperatures
“Ni 4
'0 -l
N
8 4

20
r

Standardized CT
10
I
Standardized CT
10 15
L 1

 

 

 

 

 

 

O ‘ o _
I r I I I I I I I I
02 04 06 0.8 10 0.2 04 06 08 10
Percent Developed Percent Developed

Figure 5.9. Standardized chitin synthase CT scores plotted by development percent
at 20°C for (a) all individuals that expressed the gene and (b) in all 958 individuals,
with NA expression values assigned a CT of 50. The red line indicates gene expression
in the CA strain. Blue indicates gene expression in the MI strain. Few individuals did
not express the gene at this temperature. The CA strain expressed more of cs than the M1
strain, though in the same pattern.

Plots of hsp60 expression are in Figures 5.10—13. hsp60 was not detected in 106
samples. When grouped by stage, hsp60 was expressed at its highest levels in the first
two instars, at intermediate levels during the feeding portion of the third instar, and at its
lowest levels during the postfeeding third instar and pupal stages (Figure 5.10). When
plotted against development percent, hsp60 expression decreased from hatching until
early pupation, and then increased in abundance until eclosion. Expression of hsp60 was
not affected by temperature (Figure 5.11). However, the CA strain expressed the gene at
higher levels (though in the same pattern) than the M1 strain at 20°C (Figure 5.13). While
the sample size was small, at 33.5°C there was no obvious difference in expression
among the strains, but the CA strain did express the gene at relatively high levels around

40 percent development (Figure 5.12). The gene was a significant predictor of

106

development percent in the GAM assessing its utility (Table 5.4). The GAM that

included the hsp60 term (Model 7) exhibited a PDE of 14.7% (Table 5.3).

heat shock protein 60 Expression by Stage

 

 

 

 

 

 

 

 

 

o
8
8
in _, ° 0 o
1'- O O
5 9 o
o 0
§ 0 —o—
2 9 s 4—
‘s : r__.__,= fi
20 q—
5 I + L_] I I
B '0 "‘ ‘ ' :
9 E a -
m ‘ T
I I 4 o
o -‘ o 8
o
0
j j l l l
1st 2nd 3rd 3rdPF Pupe

Developmental Stage

Figure 5.10. Standardized heat shock protein 60 CT scores plotted by stage. Gene
expression was highest during the first two stages and lowest during the last two stages.

107

hsp60 Expression at High and Low Temperatures

 

15
1

Standardized CT

 

 

 

 

Percent Developed

Figure 5.11. Standardized heat shock protein 60 CT scores plotted by development
percent for all individuals that expressed the gene. The black line is the lowess curve
for all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 33.5°C. The expression levels of this gene decrease from
hatching, until 60 percent development (early pupation), then increase until eclosion.

108

hsp60 Expression Among Strains at High Temperatures

 

Standardized CT

 

 

 

I I I I I
0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.12. Standardized heat shock protein 60 CT scores plotted by development
percent at 33.5°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
green line indicates gene expression in the WV strain. The strains expressed the gene in

the same general pattern, though the CA strain had higher expression around 40 percent
development.

109

hsp60 Expression Between Strains at Low Temperatures

 

Standardized CT

 

 

 

0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.13. Standardized heat shock protein 60 CT scores plotted by development
percent at 20°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the M1 strain. The
CA strain expressed more of this gene than the MI strain, at this temperature. Few
individuals did not express the gene at this temperature.

Plots of hsp90 expression are in Figures 5.14—17. hsp90 was not detected in 11
samples. When grouped by stage, hsp90 was expressed at stable levels except during the
feeding portion of the third instar, when it was expressed at lower levels than the other
stages (Figure 5.14). When plotted against development percent, hsp90 expression
demonstrated different expression patterns depending on the temperature at which
individuals were raised (Figure 5.15). During the first quarter of development
(approximately) there was a decrease in transcript abundance, which was much more
pronounced at 33.5°C than at 20°C. At the high temperature transcript abundance
increases until eclosion, but at the lower temperature expression was maintained at a
constant level. At 33.5°C, there was no obvious difference in expression among strains,

though the point of minimal hsp90 expression occurred at a later time in the CA strain

than it did in the other two (Figure 5.16). At 20°C both strains demonstrated a similar

110

pattern of hsp90 expression, with the CA strain expressing the gene at elevated levels.
The gene was a significant predictor of development percent when included in a GAM
assessing development percent with CT scores for hsp90 along with temperature and

developmental stage, which were also significant terms in that model (Table 5.4). The

GAM (Model 8) exhibited a PDE increase of 0.6% compared to Model 1 (Table 5.3).

neatshockprotelnSOExpressionbyStage

 

15
1
o

 

 

 

 

 

 

 

Standardized hsp90 CT
5
l

 

 

 

 

 

 

 

 

 

m _ o
l
o
1 l l l l
tat 2nd 3rd 3rdPF Pupe
Developmental Stage

Figure 5.14. Standardized heat shock protein 90 CT scores plotted by stage. Gene
expression was lowest during the postfeeding third instar.

111

hepOOExpressionatl-ilgnandLowTempeI-atures

 

15
1
0

Standard: ed CT
5
1

 

-5
l
o

 

 

 

Figure 5.15. Standardized heat shock protein 90 CT scores plotted by development
percent for all individuals that expressed the gene. The black line is the lowess curve
for all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 33.5°C. The expression levels of the gene decreased from
hatching, until approximately 25 percent development (when postfeeding larval
development is attained), then increased until eclosion when flies were raised at the high
temperature. At the low temperature, expression was higher than at high temperatures,
and expression was maintained at a steady level after 25 percent development.

112

hspoo Expression Among Strains at High Temperatures

 

in
‘_-i

10
1

Standardized CT
5
1

-5
1

 

 

 

I I I I I
0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.16. Standardized heat shock protein 90 CT scores plotted by development
percent at 33.5°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the M] strain. The
green line indicates gene expression in the WV strain. Expression decreased until the
postfeeding third instar (around 25 percent) then increased until eclosion. The strains
expressed the gene in the same general pattern, though the time of minimum expression
occurred later in the CA strain.

113

hspOO Expression Between Strains at Low Temperatures

 

52—1

 

 

 

 

 

p.
O
i
e “’ ‘
g N
S
a)
o _ /v
‘1’ 4
I r I I I
0.2 0.4 0.6 0.8 1.0
Percent Developed

Figure 5.17. Standardized heat shock protein 90 CT scores plotted by development
percent at 20°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
CA strain expressed more of this gene than the M1 strain, at this temperature.

Plots of ace expression are in Figures 5.18—21. ace was not detected in 322
samples. When grouped by stage, ace was expressed in decreasing levels through the
third instar, and then it increased in expression until eclosion (Figure 5.18). When
plotted against development percent, ace expression demonstrated two different
expression patterns depending on the temperature at which individuals were raised
(Figure 5.19). During the first quarter of development (approximately) there was a
decrease in transcript abundance, which was much more pronounced at 33.5°C than at
20°C. At 33.5°C transcript abundance increased until eclosion, but at the lower
temperature expression was maintained at a constant level. At 33.5°C there was no
obvious difference in expression among the strains. Though sample sizes per strain were
small, the point of minimal ace expression occurred at a later time in the CA strain than it

did in the other two strains (Figure 5.20). In the 20°C replicates, both strains

114

 

demonstrated a similar pattern of expression, with the CA strain expressing the gene at
elevated levels (Figure 21). In a GAM (Model 9), ace was a significant predictor of
development percent when temperature and developmental stage were included in the
model. The binary term was significant by itself (Table 5.4). The non-parametric GAM
(Model 9) exhibited a PDE increase of 1.7% compared to Model 1 (Table 5.3). The
GAM that included binary ace expression (Model 3) demonstrated a PDE of 3.6% (Table

5.3).

acetylcholine esterase Expression by Steps

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m!
“‘ e
o 8
o
8‘ a B e
I- o —.— I '7— o
O —r— : ' '
8 In- - I : i
0" I
t ed '
9 i ‘ l::l
rn . . I
.0- 2 ; _L_
__;__ o 4
o
o
0
o I I I I I
1st 2nd 3w 3rdPF Pupe
DevelopmentelStege

Figure 5.18. Standardized acetylcholine esterase CT scores plotted by stage. Gene
expression decreased through the third instar, then increased until eclosion.

115

 

 

 

 

 

 

 

 

 

 

ace Expression at High and Low Temperatures ace Expression at High and Low Temperatures
In a
N o 0 8 ‘
g _
'6 15 8 —
8 8
u a
e g 53 -
i . ._
.0 a
O
l I l l l
o 2 o 4 0.6 o a 1 0
Percent Developed Percent Deveioped

Figure 5.19. Standardized acetylcholine esterase CT scores plotted by development
percent for (a) all individuals that expressed the gene and (b) in all 958 individuals,
with NA expression values assigned a CT of 50. The black line is the lowess curve for
all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 33.5°C. The shift in the lines between (a) and (b) indicates a
cluster of individuals that do not express the gene at the time point where gene expression
is at its lowest levels for this locus. Many individuals did not express ace between
approximately 20 and 50 percent development (late third instar through early pupation).
This time period mirrors the period of least ace expression at the high temperature
treatment. Expression increased through development in the low temperature treatments.

116

 

 

 

 

 

 

 

 

ace Expression Among Strains at High Temperatures ace Expression Among Strains at High Temperatures
3 " 8
a J a 4
s m 5 8 '
" E 8
5
E In ._
§ .—
2 — a
(I) e _
.,, _
.1, _
o I I I I I o _ I I I I I
02 04 06 08 1.0 0.2 04 06 03 10
Percent Developed Percent Developed

Figure 5.20. Standardized acetylcholine esterase CT scores plotted by development
percent at 33.5°C for (a) all individuals that expressed the gene and (b) in all 958
individuals, with NA expression values assigned a CT of 50. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
green line indicates gene expression in the WV strain. The shift in these lines between
(a) and (b) indicates a cluster of individuals that do not express the gene at that time
pomt.

a . b .
ace Expression Between Strains at Low Temperatures ace Expression Between Strains at Low Temperatures

 

 

Shridudzod CT
0 15
I
Standardized CT
15
I

 

 

 

 

 

 

° I I I T I o T I I I I
0.2 0.4 0.5 0.8 1.0 0.2 0.4 0.6 0.8 1.0

Percent Developed Percent Developed

Figure 5.21. Standardized acetylcholine esterase CT scores plotted by development
percent at 20°C for (a) all individuals that expressed the gene and (b) in all 958
individuals, with NA expression values assigned a CT of 50. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
CA strain expressed more ace than the M1 strain (a). MI was also more likely to not
express ace.

117

Plots of ecr expression are in Figures 5.22—25. ecr was not detected in 17
samples. When grouped by stage, ecr was expressed in slightly increasing levels until the
postfeeding third instar, when expression was significantly greater than other stages, then
expression decreased through pupation until eclosion (Figure 5.22). When plotted
against development percent, ecr expression increased until achieving maximum
expression at approximately 35 percent, then expression decreased. This pattern was the
same for cohorts raised at both temperatures (Figure 5.23). However, at 33.5°C, the M1
strain expressed less ecr than the other strains (Figure 5.24) and at 20°C the M1 strain
expressed less ecr than the CA strain, except at the point of maximum expression, when
both strains converged to the same expression level for this gene (Figure 5.25). A GAM
assessing age with ecr CT scores was a significant predictor of development percent

(Table 5.4). This GAM (Model 10) exhibited a PDE of 6.6% (Table 5.3).

ecdysonereceptorExpresslonbyStage

 

 

 

 

 

 

 

 

 

p.
o —'_ q— -—r—-
gro- E I 8 —E—
i E31 + ‘
, . . E . E]
a °q —“ : ES _:_
0’ - - a
o --t— -O- -
o —-e-— o
'91 ° 0 o
o
i T l W T
tst 2nd 3rd 3rdPF Pupe
DevelopmentalStage

Figure 5.22. Standardized ecdysone receptor CT scores plotted by stage. Gene
expression increased through the postfeeding third instar, then decreased until eclosion.

118

ecrExpressionatnghandLewTemperatures

 

Standardzed CT
5
I

-5

 

 

 

 

l 1 i l T
02 04 0.6 08 10
PercentDeveloped

Figure 5.23. Standardized ecdysone receptor CT scores plotted by development
percent for all individuals that expressed the gene. The black line is the lowess curve
for all individuals, the blue line is the curve for expression at 20°C, and the red line is the
curve for expression at 33.5°C. The expression levels of this gene increase from
hatching, until approximately 35 percent development (during the postfeeding third
instar), then decrease until eclosion.

ecr Expression Among Strains at High Temperatures

 

53-4

10
I

Standardized CT
0 5
L I

 

 

 

0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.24. Standardized ecdysone receptor CT scores plotted by development
percent at 33.5°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
green line indicates gene expression in the WV strain. The strains expressed the gene in
the same general pattern, though the M1 strain expressed less ecr than the other strains.

119

ecr Expression Between Strains at Low Temperatures

 

15

‘1

10
I

 

Standardized CT
0
1

 

 

 

0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.25. Standardized ecdysone receptor CT scores plotted by development
percent at 20°C for all individuals that expressed the gene. The red line indicates
gene expression in the CA strain. Blue indicates gene expression in the MI strain. The
CA strain expressed more of this gene than the MI strain, at this temperature, though they
converge to the same maximum expression level dtu'ing the postfeeding third instar.

Plots of rop—I expression are in Figures 5.26—29. Of the 958 individuals
assessed, rop—I was not detected in 99 samples. When grouped by stage, rap—I was
expressed in increasing levels through the postfeeding third instar, then expression
decreased through pupation until eclosion (Figure 5.26). When plotted by development
percent, rap—1 increased in abundance until approximately 25 percent development
(when raised at 33.5°C), or until 35 percent development (when raised at 20°C) (Figure
5.27). The shift in these curves closely mirrored the shift in body size (seen in Figure
3.3), which did not occur with all genes. After achieving maximum expression, the
abundance of message decreased until eclosion. At high temperatures, there was no
obvious difference in expression of rop—I among the strains (Figure 5.28). However, at
the low temperature treatment, expression was consistently higher in the CA strain

compared to the M1 strain (Figure 5.29). In the GAM assessing age with rop—I CT

120

scores, the gene was a significant predictor of development percent when included with
stage (Table 5.4). Though both terms were significant predictors of development percent,
the GAM (Model 11) exhibited no PDE increase when compared to Model 1. However,
the GCV score was lower for Model 11 when compared to Model 1 (Table 5.3),
indicating that the inclusion of rap—I will help make better estimates of development

percent.

resistance to organophosphate-1 Expression by Stage

 

 

 

 

 

 

 

 

 

3 0
I‘ll __ o O 0 3
° 0 o
l- _"" —v—
o : . ° .9.
'|- o l O i
Q ‘- d
9 E E31 +8 o
'8 ' I r
.5 ' . . r
'2 ' I I l
8 : El '
C ___I._
e In - -
., s - E :
o _1_ E
o __.i_ 0
o J o o
O
l T l l V
131 2nd 3rd 3rdPF Pupa

Developmental Stage

Figure 5.26. Standardized resistance to organophosphate 1 CT scores plotted by
stage. Gene expression increased through the postfeeding third instar, then decreased
until eclosion.

121

rop-1 Expression at High and Low Temperatures

 

Standardized CT

 

 

 

 

Figure 5.27. Standardized resistance to organophosphate 1 CT scores plotted by
development percent for all individuals that expressed the gene. The black line is the
lowess curve for all individuals, the blue line is the curve for expression at 20°C, and the
red line is the curve for expression at 33.5°C. The expression levels of this gene increase
from hatching until approximately 25 percent development at high temperatures and until
approximately 35 percent development at low temperatures, then gene transcript
abundance decrease until eclosion.

122

rep-1 Expression Among Strains at High Temperatures

 

15
I

10
I

Standardized CT

 

 

 

I I I I I
0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.28. Standardized resistance to organophosphate 1 CT scores plotted by
development percent at 33.5°C for all individuals that expressed the gene. The red
line indicates gene expression in the CA strain. Blue indicates gene expression in the M1
strain. The green line indicates gene expression in the WV strain. The strains expressed
the gene in the same general pattern.

123

 

rep-1 Expression Between Strains at Low Temperatures

 

15
I

10
i

Standardized CT

 

 

 

Percent Developed

Figure 5.29. Standardized resistance to organophosphate 1 CT scores plotted by
development percent at 20°C for all individuals that expressed the gene. The red line
indicates gene expression in the CA strain. Blue indicates gene expression in the MI
strain. The CA strain expressed more of the gene than the M1 strain, at this temperature,
though both followed the same pattern.

Plots of w expression are in Figures 5.30—33. w was not detected in 233 samples.
When grouped by stage, w was expressed in increasing amounts from the first instar
through the postfeeding third instar, then expression decreased in pupae (Figure 5.30).
When plotted by development percent, w increased in abundance until approximately 25
percent development (when raised at 33.5°C), or until 35 percent development (when
raised at 20°C) (Figure 5.31), though this pattern was subtler than in the rop—I plots.
Individuals from the low temperature treatments were also more likely to express w than
individuals from high temperature treatments (Figure 5.31). After achieving maximum

expression, the abundance of the transcript decreased until eclosion. At high

124

 

temperatures, there was no obvious difference in expression of w among the strains
(Figure 5.32), but the CA strain was the least likely to express the gene at that
temperature and the M1 strain was the most likely to express it. However, at the low
temperature treatment, expression was consistently higher in the CA strain compared to
the MI strain, with the M1 strain likely to not express the gene during pupation (Figure
5.33). In GAMs assessing both the binary and non-parametric expression of w, the gene
was a significant predictor of development percent (Table 5.4). Binary expression
(Model 4) explained 0.43% of the deviance in the data. When w CT scores were used to

predict development percent (Model 12), a PDE of 3.55% was attained (Table 5.3).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

whiteExpressionbyStage
19.1 °
8 O
3 ' 8
8‘ g 8 0 g
E —— —— . §
.1._ .
2" ' i +
2— [’41 i
1.- “— T‘ _L _5_ ‘8—
O
O
O
o- O
I

tst 2nd 3rd 3rdPF Pups

Developmental Stage

Figure 5.30. Standardized white CT scores plotted by stage. Gene expression
increased through the postfeeding third instar, then decreased until eclosion.

125

 

 

 

   

 

 

 

 

 

 

w Expression at High and Low Temperatures w Expression at High and Low Temperatures
In _ 0
N
a _
g _
8 _
I- I-
E 52 _ g m
5 E - 3 2 —
In -1 Ir: _
0 O
O
O 4 0 O -1 00
l l l l l 7* T l l
02 04 06 08 10 0.2 0.4 06 08 10
Percent Developed Percent Developed

Figure 5.31. Standardized white CT scores plotted by development percent for (a) all
individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 33.5°C. The shift in the lines between (a) and (b) indicates a cluster of
individuals that did not express the gene at the time point where gene expression is at its
lowest levels for this locus. The expression levels of the gene increased from hatching
until approximately 25 percent development at high temperatures and until approximately
35 percent development at low temperatures, then gene transcript abundance decrease
until eclosion. High temperature treatments were less likely to express the gene.

126

w Expression Among Strains at High Temperatures w Expression Among Strains at High Temperatures

 

 

0..
N

20
I

Standardrzed CT
10 15
I I
Standardrzed CT
1 5
L

 

 

 

 

 

 

0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0

Percent Developed Percent Developed

Figure 5.32. Standardized white CT scores plotted by development percent at 33.5°C
for (a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the
CA strain. Blue indicates gene expression in the M1 strain. The green line indicates gene
expression in the WV strain. The shifi in the lines between (a) and (b) indicates that CA
was most likely to not express w and MI was most likely to express the gene.

127

a. b.

 

 

wExpreesionBetweenShalnsatLowTemperatures wExpressionBetweenStralnsatLowTemperatures
81
g 4
g _

15
1

StandardizedCT
10
l
StandardizedCT
10 15
l l

 

 

 

 

 

 

O “i o 4
I I I I T I 1 T n—
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
Percent Developed Percent Developed

Figure 5.33. Standardized white CT scores plotted by development percent at 20°C
for (a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the
CA strain. Blue indicates gene expression in the M1 strain. The CA strain expressed
more w than the M1 strain (a). MI was also more likely to not express w during pupation
(the last half of development).

Plots of usp expression are in Figures 5.34—3 7. asp was not detected in 109
samples. When grouped by stage, usp was expressed in increasing amounts from the first
instar through the postfeeding third instar, and then expression decreased in pupae
(Figure 5.34). When plotted by development percent, asp increased in concentration
from hatching until the postfeeding third instar, then expression decreased through
pupation (Figure 5.35), with no obvious difference in expression among temperature
treatments. The three strains exhibited little difference in expression at 33.5°C (Figure
5.36). However, at 20°C MI and CA expressed asp in a similar pattern, with CA
consistently expressing more of the gene than the M1 strain. A GAM assessing

expression of usp was a significant predictor of development percent when stage was

128

included in the model (Table 5.4). When asp CT scores were used to predict

development percent (Model 13), a PDE increase of 0.6% was attained (Table 5.3).

 

 

 

 

 

ultraspiracieExpressionbyStage
o
g-
5 22— g
o o
g '7‘ _,_ 0 J.
E 2- [:q l ' -8- I
_a_ o o
o-I
m o
l I I I r T
tst 2nd 3rd 3rdPF Pupe
DevelopnnntalStege

Figure 5.34. Standardized ultraspiracle CT scores plotted by stage. Gene expression
increased through the postfeeding third instar, then decreased until eclosion.

129

uap Expression at High and Low Temperatures

 

Standardized CT

 

 

 

 

0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.35. Standardized ultraspiracle CT scores plotted by development percent
for all individuals that expressed the gene. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 33.5°C. The expression levels of the gene increased from hatching until
approximately 35 percent development, then gene transcript abundance decreased until
eclosion.

usp Expression Among Stains at High Temperatures

 

15
1

 

Standardized CT
10
I

 

 

-5

 

I I I I I
0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.36. Standardized ultraspiracle CT scores plotted by development percent at
33.5°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the M1 strain. The green
line indicates gene expression in the WV strain. The strains expressed the gene in the
same general pattern.

130

usp Expression Between Strains at Low Temperatures

 

20
I

15
I

Standardized CT
10
1

 

 

 

0.2 0.4 0.6 0.8 1.0

Percent Developed

Figure 5.37. Standardized ultraspiracle CT scores plotted by development percent at
20°C for all individuals that expressed the gene. The red line indicates gene
expression in the CA strain. Blue indicates gene expression in the MI strain. The CA
strain expressed more of the gene than the MI strain, though both followed a similar
pattern.

Plots of sll expression are in Figures 5.38—41. Of the 958 individuals sampled,
281 did not express the gene. When plotted by stage, expression increased through the
feeding portion of development, then decreased until eclosion (Figure 5.38). When
plotted by development percent, expression increased until approximately 25 percent
development, and then decreased until approximately 50 percent development, when
expression levels increased again until eclosion (Figure 5.39). There was little difference
in expression patterns between temperatures, but individuals from high temperature
treatments were much less likely to express the gene (Figure 5.39). When expression at
the high temperature was assessed, it was clear that the lack of expression occurred in
both the CA and WV strains (Figure 5.40). The expression of sll at low temperatures was

similar among the CA and MI strains, but MI was much less likely to express the gene

131

(Figure 5.41 ). When GAMs utilized sll expression to predict development percent, both
the binary and quantitative expression levels of the gene were significant predictors of
age (Table 5.4). The GAM that utilized standardized sll CT scores (Model 14) exhibited
a PDE of 9.17%. The GAM using the binary expression data (Model 5) required the
assessment of stage, strain, and temperature with sll to achieve an increase of 0.3% in

PDE compared to Model 1 (Table 5.3).

 

slalom Expression by Stage
m -1
N o
O O
0
g -
O O
O
o O
o o 0
O

15

——‘—-

Standardized sll CT
1 0
1

ESéEEE?

 

 

 

in —I Q o
O o O
o —r
o
I I I f I
tst 2nd 3rd 3rdPF Pupa

Developmental Stage

Figure 5.38. Standardized slalom CT scores plotted by stage. Gene expression
increased through the third instar, then decreased until eclosion.

132

 

 

   

 

 

 

 

 

 

sll Expression at High and Low Temperatures all Expression at High and Low Temperatures
R
O
O O
8 _ O
l- ‘5.’ — 1-
o o
g r
t 9 ' i
8 a
m _
O - O __
O O
l I l l l l T I T
02 04 06 08 10 0.2 04 0.6 08 10
Percent Developed Percent Developed

Figure 5.39. Standardized slalom CT scores plotted by development percent for (a)
all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The black line is the lowess curve for all
individuals, the blue line is the curve for expression at 20°C, and the red line is the curve
for expression at 33.5°C. The shift in the lines between (a) and (b) indicates a cluster of
individuals that did not express the gene at the time point where gene expression was at
its lowest levels. The expression levels of the gene increased from hatching until
approximately 25 percent development, then gene transcript abundance decreased until
approximately 50 percent, and then increase again until eclosion. High temperature
treatments were less likely to express the gene.

133

sll Expression Among Strains at High Temperatures sll Expression Among Strains at High Temperatures

 

 

20
I
25
41

Standardized CT
10
1
Standardized CT
10 15
l I

 

 

 

 

 

 

Percent Developed Percent Developed

Figure 5.40. Standardized slalom CT scores plotted by development percent at
33.5°C for (a) all individuals that expressed the gene and (b) in all 958 individuals,
with NA expression values assigned a CT of 50. The red line indicates gene expression
in the CA strain. Blue indicates gene expression in the MI strain. The green line
indicates gene expression in the WV strain. The shift in the lines between (a) and (b)
indicates that CA was most likely to not express sll and MI was most likely to express the
gene.

a . b .
sll Expression Between Strains at Low Temperatures sll Expression Between Strains at Low Temperatures

 

 

10..
N

8'

Standardrzed CT
10
l
Standsrdrzed CT
10 15
l l

 

 

 

 

 

 

Percent Developed Percent Developed

Figure 5.41. Standardized slalom CT scores plotted by development percent at 20°C
for (a) all individuals that expressed the gene and (b) in all 958 individuals, with NA
expression values assigned a CT of 50. The red line indicates gene expression in the
CA strain. Blue indicates gene expression in the M1 strain. The strains expressed the
gene in a similar pattern (a). MI was more likely to not express sll.

134

The expression profiles of all developmentally regulated genes are shown in
Figure 5.42. All GAMs that included expression data from multiple genes (Models 16—
23) were found to have lower GCV scores than the control models. Likewise, all but one
model (Model 19 which used a gaussian distribution with gene expression data) exhibited
an increase in PDE compared to the control models (92.1%— 95.7% PDE compared to
88.2%— 91.8%; Table 3). Model 19 had a PDE of 91 .3%, which was 0.5% lower than
Model 15 and 3.1% better than the PDE for Model 1. (See information on specific stages
below.)

When diagnostic plots of the models were compared the improvement generated
by incorporating gene expression was substantial. Figures 5.43—50 depict diagnostic
plots, which assess error for specific models tested in this experiment. The control
models (Figures 5.43 and 5.44) show non-gene expression data only, while the rest
display control data and some form of gene expression data used to predict development
percent. The lower right quadrant of the plots depicts the predicted (Fitted) versus true
(Response) development percents of all individuals used to make the model. In model 15
(which used non-gene data, Figure 5.44), error increased throughout development and a
gap existed between larval and pupal predictions. All gene expression models decreased

error and closed the gap in predictions that were present in Figures 5.43 and 5.44.

135

Comparison of Gene Expression Profiles

 

 

 

m—
,_
l3
0
L-
2'
men
a:
C
o
(D
E
E
“00‘
C
.9
(D
o—

 

 

 

 

Percent Developed

Figure 5.42. The combined standardized expression (CT score) patterns for all nine
genes throughout L. sericata immature development. The lowess curves for each gene
represent expression for all individuals that expressed the gene. The combined
expression patterns of the genes were used to predict blow fly age with GAMs.

136

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Normal Q-Q Plot
co
0 fi'é'
‘0 _
.2
s s -
3
O -l
g o -i
E I
N a
(I)
g 4;)”
i i i l l T i l
-3 -1 0 1 2 3
Theoretical Quantiles
Histogram of residuals
o F'-
‘9 ' "r-
r-
E 8 - ~—
0 ‘-
3 r—fl
S
u. 8 - __
O _ j l—i

 

I I I r I I I
-0.6 -0.2 0.2 0.6

Residuals

Residuals

Response

0.6

-0.2 0.2

-0.6

1.0

0.6

0.2

Residuals vs. Fitted

 

CDC
0

 

 

 

L 1 i L
cummfi

-l com-Ml:

loom-Ina)

T I I I I r
0.1 0.3 0.5 0.7

Fitted Values

Response vs. Fitted Values

 

 

 

Fitted Values

Figure 5.43. Diagnostic plot for Model 1, which used developmental stage to predict
L. sericata development percent (on a scale of 0—1). A gamma distribution was used
with this model. Descriptions for each panel type can be found in Figure 3.4. Predicted
(Fitted) values represent a range of true (Response) values. Error increased with age, and
gaps exist between predictions for each developmental stage.

137

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

<0. 0 <0.
(I: o o
2
E 2 N
g g g as
O .12
g N. 3 N
E ‘i’ ‘1 CI’
8 (D to
0 £ 0
' I I I I I I l
—3 —1 0 1 2 3
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0
‘2
Q)
E o a
a) O
a ‘— 8
g, a
u. 8 c:
O
F—T'_T—T_i—i_'1
-06 -0.2 02 06
Residuals Fitted Values

Figure 5.44. Diagnostic plot for Model 15, which used developmental stage, strain,
temperature, length, and weight to predict L. sericata development percent (on a
scale of 0—1). A gamma distribution was used with this model. Descriptions for each
panel type can be found in Figure 3.4. Predicted (Fitted) values represent a range of true
(Response) values. Error increased with age, and a gap exists between predictions for
postfeeding third instar and pupal ages.

138

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

Sample Quantiles
—0.4 0.0 0.4
m l I i I
\
c

Residuals

 

 

 

 

 

-3 -2 -1 0 1 2 3 0.2 0.4 0.6 0.8 1.0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0
E
>‘ G)
8 2
:5 8
3 8
i 8 a:
O
i—lfi—fi—l
-0.4 0.0 0.2 0.4 0.2 0.4 0.6 0.8 1.0
Residuals Fitted Values

Figure 5.45. Diagnostic plot for Model 18, which used developmental stage, strain,
temperature, length, weight, binary gene expression for four genes, and quantitative
gene expression for nine genes to predict L. sericata development percent (on a scale
of 0—1). A gamma distribution was used with this model. Descriptions for each panel
type can be found in Figure 3.4. Predicted (Fitted) values represent a range of true
(Response) values. The increase in error with age has diminished compared to other
models and the gap between predictions for postfeeding third instar and pupal ages has
shrunk.

139

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

O
(I)
g N.
° 7 2
a ‘ 3
2 3 - s
g _ a:
(B
(I) N
o‘ -o
i I l i l I l l
-3 -2 -1 o 1 2 3 0.0 0.2 0.4 0.6 0.8 1.0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0
S
>
a a E
(D
= 8
8 c, 8
E V r:
O
l l i l l l
-0.2 0.0 0.1 0.2 0.3 0.0 0.2 0.4 0.6 0.8 1.0
Residuals Fitted Values

Figure 5.46. Diagnostic plot for Model 19, which used the same parameters as Model
18 to predict L. sericata development percent (on a scale of 0—1), but a gaussian
distribution was used. Descriptions for each panel type can be found in Figure 3.4.
Predicted (Fitted) values represent a range of true (Response) values. The increase in
error with age has diminished compared to Model 15 and the gap between predictions for
postfeeding third instar and pupal ages has shrunk. However, compared to Model 18, this
model exhibits more error in predictions of younger individuals.

140

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N. _ ° N
(n O O
é’
g o' ‘ g o'
0 0. _ 2 0.
9 0 g; o
g n:
m _
(I) N N
? g T i 1 i l l ?
—3 -2 -1 o 1 2 3 0.0 0.2 0.4 0.6 0.8
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
8
‘- l—'___‘
O
é‘ a
8 — a
g o F s
u. "5 n:
o I 1.:
I—I—I—I—fi
-O.2 -0.1 0.0 0.1 0.2 0.0 0.2 0.4 0.6 0.8
Residuals Fitted Values

Figure 5.47. Diagnostic plot for Model 20, which used the same parameters as Model
18 to predict L. sericata development percent (on a scale of 0—1), but a gaussian
distribution was used with this model and all genes expression scores were anchored
against hsp60 expression. Descriptions for each panel type can be found in Figure 3.4.
Predicted (Fitted) values represent a range of true (Response) values. The increase in
error with age has diminished compared to Model 15 and the gap between predictions for
postfeeding third instar and pupal ages has been eliminated. However, compared to
Model 18, this model exhibits more error in predictions for younger individuals, though it
is an improvement over predictions made with Model 19.

141

The last two models were similar to Model 21, but strain, temperature, and non-
significant terms were removed. When compared to the models that preceded them, there
was little change in the diagnostic plots (Figure 548—49 and Figures 5.49—50). Likewise,
the statistical evaluations of the models revealed little change in the overall performance
of the models compared to the model that preceded them. Model 22 had PDE and GCV
scores of 94.6% and 0.0059, which represented no change in PDE and a GCV increase of
0.0003 compared to Model 21. Likewise, the removal of temperature (Model 23
compared to Model 22) resulted in PDE and GCV scores of 94.7% and 0.0059 for Model
23, which represented a 0.1% increase in PDE and no change in GCV compared to

Model 22.

142

 

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

In N. _ 00 N
9 O o
E ‘- 52 e-
2 — a
C'- n:
E
3 0 ~ 0
l l
o
l l I l l l
-3 -2 -1 0 1 2 3 0.0 0.2 0.4 0.6 0.8 1.0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
.8
> a:
3
a 3
“.3 o w
LL In a:
O
ififi—ifi
-0.2 -0.1 0.0 0.1 0.2 0.0 0.2 0.4 0.6 0.8 1.0
Residuals Fitted Values

Figure 5.48. Diagnostic plot for Model 21, which used the same parameters as Model
18 to predict L. sericata development percent (on a scale of 0—1), but a gaussian
distribution was used with this model and all genes expression scores were anchored
against length measurements. Descriptions for each panel type can be found in Figure
3.4. Predicted (Fitted) values represent a range of true (Response) values. The increase
in error with age has diminished compared to Model 15 and the gap between predictions
for postfeeding third instar and pupal ages has been eliminated. However, compared to
Model 18, this model exhibits more error in predictions for younger individuals, though it
is an improvement over predictions made with Model 19.

143

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Q 0
In 0 °
2 _
E 2
a— (U
8 o ‘ .13
w
it - a
to "I
m cl: 1
O
i i 1 l i i i
-3 -2 -1 o 1 2 3 0.0 0.2 0.4 0.6 0.8 1.0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0
:9
> O
s 2
3 <- 8
O" tn
9 an
LL. 8 n:
O
i i i i i l
—o.2 0.0 0.1 0.2 0.3 0.0 0.2 0.4 0.6 0.8 1.0
Residuals Fitted Values

Figure 5.49. Diagnostic plot for Model 22, which used the same parameters as Model
21 to predict L. sericata development percent (on a scale of 0—1), but parameters
that were non-significant predictors of age in Model 21 (length, weight, cs, w, strain)
were removed. Descriptions for each panel type can be found in Figure 3.4. Predicted
(Fitted) values represent a range of true (Response) values. The increase in error with
age has diminished compared to Model 15 and the gap between predictions for
postfeeding third instar and pupal ages has been eliminated. However, compared to
Model 18 this model exhibits more error in predictions for younger individuals, though it
is an improvement over predictions made with Model 19. Removing strain, cs, and w,
had little effect on the predictions made with the model compared to predictions made
with Model 21.

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

CO ('0
o 0 o 0
In 0 O
2
E 52
,_ m
8 o 13
U)
E: a
m "'.
"’ c.’
-3 -2 -1 0 1 2 3 0.0 0.2 0.4 0.6 0.8 1.0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0
:9
> en
ti 8 g
:3 ‘- O.
0' w
9 ll)
u. 8 n:
O
l l l l i i
-0.2 0.0 0.1 0.2 0.3 0.0 0.2 0.4 0.6 0.8 1.0
Residuals Fitted Values

Figure 5.50. Diagnostic plot for Model 23, which used the same parameters as Model
22 to predict L. sericata development percent (on a scale of 0—1), except that
temperature was removed from this model. Descriptions for each panel type can be
found in Figure 3.4. Predicted (Fitted) values represent a range of true (Response)
values. Removing temperature had little effect on the predictions made with this model
compared to predictions made with Model 22.

Finally, the differences between GAMs with or without gene expression data were
examined for the most difficult to age stages, the postfeeding third instar and pupation.

Here the influence of expression data was most vivid. As mentioned, there is very little

145

age information to be gained during the latter stages of development, as the animal has
ceased changing in size. This leaves large errors in age estimates using conventional
aging techniques (including stage/length/weight/temp/strain), with PDE values
plummeting to 36.2% for third instars (Figure 5.51) and 15.8% for pupae (Figure 5.52).
When gene expression data are included, these jump to 79.8% (Figure 5.53) and 78.2%

(Figure 5.54) respectively.

146

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

c
in N .. “l -
2 0 o
“a
g - 8 1
3
g o' 8 o o
g- I:
a - ..
N N
d -06” o' H O
' I f T T I I I ' I I 1 I T I
-3 -2 -1 0 1 2 3 0.20 0.30 0.40
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
— «D.
r o
8 ‘ we
>. o o
8 8 - _— ‘é’
a a 2’.-
8 8 a s
I. m
LL 8 .4 Cr o'
(‘l
o -— o
I I I I I 1
-0.2 0.0 0.1 0.2 0.3 0.20 0.30 0.40
Residuals Fitted Values

Figure 5.51. Diagnostic plot for postfeeding third instar incorporating
stage/length/weight/ temp/strain. Descriptions for each panel type can be found in
Figure 3.4. Note the extensive scatter in response vs. fitted values.

147

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o

co J m
In 0' C
.1; J
g 1— % P
a o .3 o'
2 ._ ' E; ._
g o' - a: 0'
(U i _ l
m m m

o' -O o'

I I I I I I I I l

-3 -2 -1 0 1 2 3
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values

D

ID
3 a)
o- %
2 d.)
u. 2 (I

I I I I I I I Fl
—0.3 —0.1 0.1 0.3
Residuals Fitted Values

Figure 5.52. Diagnostic plot for pupae incorporating stage/length/weight/
temp/strain. Descriptions for each panel type can be found in Figure 3.4. Note the
extensive scatter in response vs. fitted values.

148

Normal Q-Q Plot Residuals vs. Fitted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o
m .-
2
’5
8. - a 8.
O o .3 o
o d "75
'g 6‘?
w 2 8
‘0 c5 ' c; ..
l O l
I I I I I I I I I I
-2 -1 0 1 2 0.25 0.35 0.45
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
8 - ~—
3 ~ I» ‘3.
g o _ 2 o
a N '— 8
8 ‘ 8 8
II 52 .. CE 0'
o 1 .—-. r-‘l— 13 8
I I r I 0'
-0.10 0.00 0.05 0.25 0.35 0.45
Residuals Fitted Values

Figure 5.53. Diagnostic plot for postfeeding third instar incorporating
stage/length/weight/ temp/strain and expression data. Descriptions for each panel
type can be found in Figure 3.4. Note the tightening in response vs. fitted values over

Figure 5.51.

149

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Normal Q-Q Plot Residuals vs. Fitted
o o
N N
._ .- ‘ .- . , °o2°
g v- U) ‘- § :7".
' "‘ _' ' "‘ O a? 0 3.
g Q g C O O 3i:;'0::“ if."
(I) — ."‘:, ' ' ‘3
g "l g It“; ‘\.‘ ; 9.. g
E T'- ‘-. - '.r_...|:'u.a O o
s <;-> - s - °°a>
P— ‘20 00 O
I I T I I I I I I I I I
-3 -2 -1 0 1 2 3 0 4 0.6 0 8 1 0
Theoretical Quantiles Fitted Values
Histogram of residuals Response vs. Fitted Values
0 _ _—
co
g o .. _ §
0) V
a _ a
s .. a
U. N " CK
‘ I
o .-
I I I I I
-0.2 -0.1 0.0 0.1 0.2
Residuals Fitted Values

Figure 5.54. Diagnostic plot for postfeeding third instar incorporating

stage/length/weight/ temp/strain and expression data. Descriptions for each panel
type can be found in Figure 3.4. Note the tightening in response vs. fitted values over
Figure 5.52.

Discussion

Two important conclusions, regarding PMI predictions made with insect
evidence, can be drawn from the research presented in this section. The first, and most
important, is that the hypothesis that gene expression data can improve predictions of

blow fly age was mathematically and graphically demonstrated (Tables 5.3, 5.4, and

150

Figures 5.43—54). The second is that the inclusion of strain and temperature into a GAM
used to predict development percents of immature L. sericata had little effect on the
outcome of predictions. Accordingly, it is reasonable to expect that the technique can be
applied to different populations, raised at variable temperatures, with little overall effect
on the results of a prediction.

The effective use of genetic data was most impressive during the third instar and
pupation. This was anticipated, as it is these stages that are the longest, and are where
only broad estimates of fly age can currently be made. The first two instars are typically
about a day in length (temperature dependent), thus knowledge of developmental stage
alone is an excellent predictor of age, and is where most of the precision in knowing
stage originates. Size is helpful during the feeding portion of the third instar, however
once the larvae cease growing, precision tumbles. This is not fully reflected in the
general GAMs, as they look across development, not at specific parts of development.
Once comparisons within these stages were done however, the utility of gene expression
data became obvious. Relatively little was learned from the standard traits in postfeeding
third instar larvae, with only about 36% of the deviance in the data being explained. This
stems from the body size change (shrinkage) that occurs once feeding ceases and the
increase in variation that occurs with the shrinkage. During pupation, where size changes
little or not at all, the PDE dropped to 16%. Addition of gene expression data made both
of these values jump (80% and 78% respectively). It is where standard aging techniques
are at their worst that genetic data provided the greatest improvement in predicting blow

fly age.

151

 

In addition to increasing the precision of blow fly age predictions, the molecular
methodology is favorable to the standard approach for other reasons. First, the use of
GAMs and their related statistics allows for a detailed understanding and description of
error, an important part of the Daubert requirements for scientific evidence (see Chapter 3
for more on this point). Second, almost any forensic laboratory that is qualified to work
with DNA can conduct gene expression analyses, assuming they are using quantitative
PCR (many laboratories are). Finally, with robotics and microarrays (Arbeitman et al.
2002, Beckstead et al. 2005) the technique can be automated and miniaturized. All of
these qualities make the analysis of gene expression data a powerful fiiture tool for
forensic entomological PMI predictions.

There are several non-forensic benefits that have resulted from this research as
well. The information on expression levels of the insecticide-related genes cs, ace, ecr,
usp, and rap-1 may aid in the control of L. sericata and other fly species. The first four
genes are several of the main targets of insecticides (Ware and Whitacre 2004), as they
all involve integral aspects of fly biology, that cannot be lost without causing death: cs is
critical to the formation of an insect cuticle (Tellam et al. 2000), ecr and usp form a
heterodimer molecule that is necessary for the proper progression of development
(Henrich and Brown 1995), ace is critical to the function of nerves that interact with the
neurotransmitter acetylcholine (Chen et al. 2001), and mutants for rop-I (an alpha
carboxylesterase) can enable resistance to organophosphates (Newcomb et al. 2005),
which are insecticides that target acetylcholine-binding nerves. The high-resolution data
of the expression levels of all of these genes, throughout the immature lifecycle, reveals

an interesting picture that will be useful in designing insect control programs.

152

Each insecticide-interacting gene varies in concentration throughout development.
Two genes (cs and ace) are even likely to not be expressed at all during the third instar.
Periods of high or low expression of a gene may indicate points in development where
different chemicals may be more or less effective in killing these insects. It will be
possible, using this and similar research, to identify stages of development that are likely
to be more or less resistant to a specific insecticide. In the case of ace, the third instar
was a period of low expression for this gene, thus organophosphates may be differentially
effective as a poison for that stage than for the other immature stages. Such information
will enable a more educated decision as to the type of insecticide to use for pest control,
based on the life stage that will be targeted. Additionally, 20°C increased ace expression,
which, again, may alter the effectiveness of organophosphates. If it is found that certain
temperatures are more conducive to insect control, operations that use insecticides will
know the temperatures to maintain their facilities to ensure maximal effectiveness of
insecticide treatments. Temperature effects on insecticide resistance may also be used to
guide the seasonal use of these chemicals by applying them when they are most likely to
kill pests.

The gene expression profiles also provide a hi gh-resolution picture of transcript
level abundance throughout development, which is information that is useful to
evolutionary biologists. One current goal of evolutionary and medical biologists is to
determine the nature of evolution in gene regulation and phenotypes, through
comparative genomics (Collins 1998, Yeates and Wiegmann 2005). Evolutionary
biologists aim to gain insight into speciation and adaptation (i.e. Skaer et al. 2002,

McGregor 2005, Yeates and Wiegmann 2005), while the medical field will need to

153

understand how to transfer treatments for diseases in mammalian model organisms to
humans or predict when treatments will not transfer across species (Varki and Altheide
2005, Khaitovich et al. 2006). Both fields will benefit from emerging comparisons of
gene regulation among related species.

Dipteran biology is uniquely poised to be a leading contributor to the emerging
field of comparative genomics. The order contains D. melanogaster, which is an
important model organism with a sequenced genome. Currently, there are eleven other
Drosophila genomes that are complete (www.flybase.org), as well as projects to
sequence a number of other fly genomes and to sequence 50 genomes in D.
melanogaster. In addition, the pest status of a number of other Dipterans, including
mosquitoes, the Tse-tse fly (Diptera: Muscidae), and the Mediterranean fruit fly Ceratitis
capitata (Diptera: Tephritidae) have resulted in two complete mosquito genomes
(Anopheles gambiae and Aedes aegypti) and several ongoing genome sequencing
projects. Given the propensity of the Diptera for developing into pests, along with the
plummeting costs of sequencing, the list of published fly genomes is likely to grow. All
of this genomic sequence allows for powerful comparisons of any gene of interest among
species. As an example, spliceosomes were recently compared among 11 insect
genomes, revealing important similarities and evolutionary differences among their
snRN A genes (Mount et a1. 2007). Of the genomes compared, 8 were Dipteran (6 were
Drosophilids).

Currently, most research in comparative genomics involves sequence
comparisons, though the tools exist now to investigate how gene expression affects

phenotype evolution as well. Already, such a comparison within the family Calliphoridae

154

has demonstrated that acrostichal bristle types are determined by heterochrony in the
expression of proneural genes during the growth of imaginal discs (Skaer et al. 2002). As
technology and genomic sequences become available, more of these sorts of comparisons
will occur.

The data reported here address several questions that will be critical for
understanding gene expression evolution: How much variation in expression is there?
What are the distributions of gene expression levels? How does genotype affect an
expression profile? How does the environment affect an expression profile?

The first two questions are important for several reasons. The nature of gene
expression variation will be important for the choice of statistical tools used to analyze
and compare profiles. Additionally, selection can only occur on variable traits, so
information on gene expression variation (if it affects a phenotype) is an indicator of
variation that selection can act upon. Unfortunately, information on gene expression
variation throughout development, which can be used to answer these questions, is rare.
Typically, comparisons of gene expression are limited by cost so temporal profiles tend
to have very few replicates at any one time-point. As an example, Arbeitman et al.
(2002) performed two replicate analyses at any one time-point. Likewise, studies that
include sufficient replication to characterize distributions of variation in gene expression
tend to focus on one or a few time points (Powell 1997, Montooth et al. 2003, and Tarone
et al. 2005).

This dissertation produced temporal profiles of gene expression, with sufficient
replication throughout the immature lifecycle to develop a detailed picture of how

transcript abundances of the 9 genes in the research vary throughout maturation. Based

155

on Figures 5.6—5.41, several trends were apparent. First, variation in the expression
levels of all of these genes, at any point in development, was at least one order of
magnitude (one CT unit is approximately a two fold change in expression). Second,
variation tended to be uniform throughout development for any given locus (with several
exceptions). Third, different loci exhibited different amounts of variation in expression
levels. Fourth, there was typically a non-normal (gamma) distribution for expression
levels, with most loci demonstrating a tail along the low concentration side of the average
CT score.

The information on gene expression variation produced in this dissertation has
several important repercussions. First, models of gene expression will likely need to
account for ample variation in gene expression (ten fold or more) and should assume a
gamma distribution before a normal distribution. Second, for any locus the distribution
and variation in expression of that gene will be specific to that locus. Also, variation at
one time point can reasonably be assumed to be uniform throughout development, unless
there is specific data to suggest otherwise. Last, there is sufficient variation in gene
expression for selection to act upon. Such knowledge will be important as researchers
develop models to explain the effects of gene expression variation within a pathway or
network.

Understanding the nature of genotype and environment effects on gene expression
will also be important in future comparative genomic endeavors. In this research, there
were consistent differences in transcript abundances for several loci that varied by
genotype and temperature, resulting in the subtle but significant effects of strain and

temperature on GAM predictions of age (Table 5.3). Variation ranged from very little in

156

the case of hsp60 (which only varied by strain at low temperatures) to very large in the
case of ace (which was affected by temperature and strain). One key difference between
temperature and strain effects was that changes in expression among strains tended to
result in a shift of the same pattern along the standardized CT axis (i.e. rap-1 in Figure
5.29), while temperature differences could result in vertical shifts in expression (i.e. cs in
Figure 5.7), completely different patterns in expression (i.e. hsp90 in Figure 5.15), or a
shifi in expression along the time axis (i.e. rap-1 in Figure 5.27).

Knowledge of effects on expression variation will aid future research in several
ways. First, the importance of environmental effects was highlighted, as they seemed to
exert more influence on some loci than genetic differences among strains. Moreover, the
types of changes that the two temperatures produced were varied (shifts along both axes
and new expression patterns), while the effects of genetic differences were to maintain
the overall pattern of expression but shift the average concentration of transcripts
produced by a locus, by less than one order of magnitude. Such knowledge will be
important to future comparative genomic research as it will enable investigators to
design, analyze, and interpret experiments with these effects in mind.

The results from this chapter have contributed to the bodies of knowledge in
several different fields of biology. Basic information on the variation in gene expression,
throughout development, (for the 9 loci investigated) will be useful in designing,
analyzing, and interpreting comparative genomic data. For the five genes that interact
with insecticides, this information will be valuable in designing insect control programs
for L. sericata and similar blow flies that will maximize the effectiveness of the

insecticides used to control the insect. Most importantly, the main hypothesis of this

157

research, that reliable gene expression throughout development can help produce more

precise predictions of blow fly age, was demonstrated.

158

Quantitative PCR and sequencing primers used in these experiments. Tub=j3 T ubulin 56
D. Rp49=rp49. ChS=chitin synthase. Hsp60=heat shock protein 60. Hsp90=heat shock
protein 90. AcE=acetylcholine esterase. EcR=ecdysone receptor. Usp=ultraspiracle.
Rop=resistance to organophosphate 1. W=white. Sll=slalom. Wg=wingless.
Scl=scalloped wings. Rh3=rhodopsin 3. Template refers to accession numbers of the
sequences used to construct the primers. NA in the template column indicates sequences
that were obtained by this research, that are not yet submitted to NCBI. nM refers to the
end concentration, in nM, of primers in a PCR reaction.

159

Table 5.1. Primers used for sequencing and qPCR for all loci in this experiment.

Primer
Tub R1
Tub F1
unb R
unb F
qu49 F
qu49 R
ChS F2
ChS R2
th8 F
thS R
qHsp60 F
qHsp60 R
qHsp90 F
qHsp90 R
AcE F4
AcE R4
chE F
chE R
EcR F4
EcR R3
chR F
chR R
Usp F1
Usp R1
qup F
qup R
qRop F
qRop R
W F1

W R1
qW F
qW R
qSll F
qSll R
qu F
qu R
Sci F4
Scl R3
chI F
chl R
Rh3 F2
Rh3 R3
th3 F
th3 R

Function

Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR
Sequencing
Sequencing
Quantitative PCR
Quantitative PCR

Sguence
CACCAGATCGTTCATGTTGC

CGAGACCTACTGCATCGACA
ACCAGGCATGAAAAAGTGAAGAC
TCCGTAAATTGGCCGTCAAC
ACAATGTI'AAGGAACTCGAAGT'HTG
GGAGACACCGTGAGCGATTI'
GAACTGCCTATACCCGTGGA
GGATGTAAACACGCCGCTAT
GCCGACGGAGAACCTATACCA
GATGGCTGTCATTGTGGGTACA
CATCATTCCCGCCCTTGA
ATCTTCGGCAATAATGACCAAAG
AAGATCATTTGGCTGTCAAGCA
AGAAGGGCACGGAATTCAAGT
TATATGGGCTCCAGCAAAGG
ATGGTACCCGATTGCATCAT
CACCGGTTATGCCAGGTITI'
TGATCCCAAAGGCCAACATT
TTTCACCCTCGAGCAGTCTI'
CTTTCTTI'TCGCGTCGTTTC
GCATGCGGCCGGAAT
GCGTCG'ITTCATI'GCACACT
CGCAGGAGATAAAGCCAGAC
TGGTGTCGACGTGCATATT
CGAGCAAAAAGCCGAATCAC
TGCCTACGCGCAAAAAGG
GCCCCACTGTI'GAGCCATA
CCCGAGGATGTITGGGTAAGA
ACCGATCCTCCGCTCTTAAT
TGATATCCAAGAACGCCACA
ACAACAGCCAAGACTTGGACATAG
GCGCCCAGTGTCCTACCA
TCCAACGGCCACAATCTI'AAGTA
CGTTI'AGGTGTI'GCCGCAAT
TGTCTGGTI'CCTGTACGGTGAA
TI'ATCGCCAATAACACGGAAATT
CGCCATI'GTGAACGTGATAC
GCGAAAGCCAAAACTAC GAG
CGGAAGCGGCAGATTITT
TTCTCCGGGATTGGTGACA
CGGCAAATCCTI'ATCGAAAT
ACAAAACGTCCCCAACTTTC
ACTACGAATGCTI'ITATTGCC'ITATG
GCTI'TGCCAGTGAGTCATTTTACC

160

Template

NM_166357
NM_166357

EF05621 1
EF05621 1
A81 18976
AB1 18976
AF 22 1 067
AF221067
EF056212
EF 056212
AB1 18971
AB1 18971
AB1 18970
AB1 18970
U88631
U88631
NA
NA
U75355
U75355
NA
NA
AYOO7213
AYOO7213
NA
NA
AY691501
AY691501
U38899
U38899
NA
NA
AY926574
AY926574
AY926575
AY926575
U58977
U58977
NA
NA
AJ87841 1
AJ87841 1
NA
NA

In!
1000
1000
400
400
400
400
1000
1000

66167
400

66167
400
400
400
1000
1000

66167
400
1000
1000
400
400
1000
1000
400
400
400
400
1000
1000
400
400

66137
400
400

66167
1000
1000
400
400
1000
1000

66167
400

Table 5.2. Gene sequencing results.

Gene Sages
B tubulin 56 D Glossina morsitans morsitans
chitin synthase Lucilia cuprina

acetylcholine esterase Lucilia cuprina
ecdysone receptor Lucilia cuprina
Lucilia sericata

ultraspiracle Lucilia cuprina
Lucilia sericata
white Lucilia cuprina
scalloped wings Lucilia cuprina
rhodopsin 3 Calliphora vicina

Size

635
713
369
350
102
683
508
861
884
31 1

Percent

86
98
99
98
100
95
99
95
96
85

Species indicates the dipteran species that best matched a sequence. Size indicates the
length in base pairs of match results for the sequence BLAST on the NCBI website.

Percent indicates the percent similarity of that BLAST comparison.

161

.338 85 :_ 3335 296365 .«o .5585: 05 .6: Z .386 a .3 35.298 8530: 28:0: 65 .6: man :32: 85 :8 88m :o_82_m> 3.80 35.83% 05
.6: >00 4:38: 82:93—30: “2:2: 9 wow: 83 836:9, PE 05 .«o 38:88 086:3 39:8 35:25 a 35 8806s Aofimtgdimtmbm donbmg v.53
5E0? .5 5w:o_ :8 82:0 wEfiooEm 83865 Cam :0 Adm Benz—«SN 83 05% 85 :8 033 33808,: a 6:38:35 Augusta .3335 0.63 Sm c5 .3 .8: 63
333:; has: EREEwG .58 =a .88.: @3833 .o:% 85 :8 Sn: 5:3 05 «.8365 388:5 .32 8 £835 .39: 05 8 :ozsfibmmn 2.: 3.3% “:5
:28:& x5. 2: :5 Agfimsmw 8 «85.53 3on some 9 “323% 3:32:56 05 6:09.95 Boob: 823230: 8255.: 868: 2 :8: £232: 30:? of.

36 New $8.: 2 no mo 0: zbzcouccmfimamm. $2635: + Amocom.Emcmdmigéromflw "Eocene mm
33 0.3 $8.: \s no me o: :bzcmvccmfimzmm. _m>>wo_c_n + Amocmm.cumcmnvm+§.nvm+QEoH+mmflm "Enema mm
Sm 6.3 $86 .Aefimgvcnmmsmg 63825 + amass.negates;Cfsamfivmaexcts95.686 "2:86.”. a
an :8 Bad Eemgvcmsmamg 62,833 + 6286865993fear:waefléaewtmsm "68.8 ow
an m. 5 $8.0 2335366362 $38.56 + amcoemézdtsafsWasatsgwtmg "3:828 8
Sn foo, vmod :mocmEEmgmsmoEB + 3659992..vm+$>vm+3vm+an._.+c_mbm+mmEm "Enema 3
Sm Sm vmod :mocmEEmE 365916.13...vm+A>>vm+3vm+aEwh+cfibw+mmSw "Enema 2
Ba :8 Rod 28:66.th .638?555.5v?Scat:Vwasfléaemimg "3:86; 9
98 w. 5 986 .6866st 2962,.59.3szzgmifeacmdm 3&3 + 525.. 89m 1 59$ 2

to ta 3.0 2836562 saw u .528 E
as: new 5.3 28.65562 8651836 u .598 2
mum on...” 5.0 28.66662 9% u 5628 we
mam «.3 63.0 @3368. 8665686 n 28.3 :
Sm 8.0 moo .82va63 :6me "3:861 2
one Wow ovod .AmocwEEmB 6939:8318me u Emcee; m
:5 was $0.0 28.56662 magmaeflémqmsw 1 28.8 m
«8 S: :3 28.55662 88656 u 286“. .\.
8m 9% 63.0 28.6652 8396on 1 68.8 0
new now 25.0 28365562 6632656699:ch u 586a m
was 2.0 one 28.65662 2:5 u 286“. v
mom on and :mocmeemg 825 1:866 m
mam 6.3 95.0 28366562 865.685 n 28.2. N
was «.8 9.0.0 26266662 83w u 286“. s
a M8. Mud 55: 833569 6.266% 362

 

$5522.38 m3. 5 339%: £25.: «>523: comm—29:00 .m.m 03:.

162

Table 5.4. Estimated significance and degrees of freedom for variables in all GAMs.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Model Variable Linear chi-s dfl P-value
1 Stfle Y 8816.2 4 <0.0001
2 Stage Y 9031 .3 4 <0.0001

binary chitin synthase Y 21.087 1 <0.0001
3 binary acetylcholine esterase Y 46.911 1 <0.0001
4 binary white Y 5.209 1 0.023
5 Stage Y 9025.6 4 <0.0001
Strain Y 4.2436 2 0.12
Temp Y 22.828 1 <0.0001
binary slalom Y 8.0417 1 0.0047
6 Stage Y 8797 4 <0.0001
s(chitin synthase) N 23.093 3.99 0.0001
7 s(heat shoclgmtein 60) N 156.36 6.96 <0.0001
8 Stage Y 7926.3 4 <0.0001
Temp Y 28.806 1 <0.0001
s(heat shock protein 90) N 25.187 6.52 0.0005
9 Stage Y 6502.8 4 <0.0001
Temp Y 12.568 1 0.0004
s(acetylcholine esterase) N 30.128 7.17 0.0001
10 s(ecdysone receptor) N 82.1 79 5.46 <0.0001
11 Stage Y 7309.2 4 <0.0001
s(resistance to organophcyhate 1) N 62.44 5.47 <0.0001
12 s(white) N 28.493 6.15 0.0001
13 Stage Y 8167.2 4 <0.0001
sLuItraspimcle) N 27.006 3.58 <0.0001
14 s(slalom) N 83.34 5.85 <0.0001
15 Stage Y 483.29 4 <0.0001
Strain Y 23.608 2 <0.0001
Temp Y 94.234 1 <0.0001
s(Length) N 39.327 7.85 0.1 1
s(Weight) N 13.253 8.02 <0.0001
s(Length,Weight) N 68.628 1 5.2 <0.0001
16 Stage Y 460.93 4 <0.0001
Strain Y 16.914 2 <0.0001
Temp Y 115.13 1 <0.0001
s(Length) N 21 .342 7.13 0.0039
s(Weight) N 15.804 4.91 0.0073
s(Length,Weight) N 80.472 21 .9 <0.0001
binary chitin synthase Y 32.382 1 <0.0001
binary acetylcholine esterase Y 0.1385 1 0.71
binary white Y 0.7359 1 0.39
binary slalom Y 0.0736 1 0.79

 

163

 

Table 5.4. Continued.

 

 

 

 

Model V able Linear chi-e dfl P-val

17 Stage Y 232.56 4 <0.0001
Strain Y 3.9962 2 0.14
Temp Y 62.513 1 <0.0001
s(Length) N 30.504 5.61 <0.0001
s(Weight) N 8.6437 1 .27 0.0054
s(Length,Weight) N 37.219 1 1 .8 0.0003
s(chitin synthase) N 8.7356 2.98 0.034
s(heat shock protein 60) N 52.67 5.75 <0.0001
s(heat shock protein 90) N 16.536 1 .92 0.0003
s(acetylcholine esterase) N 80.381 7 <0.0001
s(ecdysone receptor) N 21.546 1 <0.0001
s(resistance to organophosphate 1) N 44.075 5.07 <0.0001
s(white) N 2.9499 1.63 0.17
s(ultraspiracle) N 23.967 5.25 <0.0001
s(slalom) N 1 .5654 1 0.21

18 Stage Y 232.56 4 <0.0001
Strain Y 3.9962 2 0.17
Temp Y 62.513 1 <0.0001
s(Length) N 30.504 5.61 <0.0001
s(Weight) N 8.6437 1.27 0.0054
s(Length,Weight) N 37.219 11.8 0.0003
s(chitin synthase) N 8.7356 2.98 0.034
s(heat shock protein 60) N 52.67 5.75 <0.0001
s(heat shock protein 90) N 16.536 1.92 0.0003
s(acetylcholine esterase) N 80.381 7 <0.0001
s(ecdysone receptor) N 21.546 1 <0.0001
s(resistance to organophosphate 1) N 44.075 5.07 <0.0001
s(white) N 2 .9499 1 .63 0.17
s(ultraspiracle) N 23.967 5.25 0.0004
s(slalom) N 1 .5654 1 0.21
binary chitin synthase Y 101.58 1 <0.0001
binary acetylcholine esterase Y 101.58 1 <0.0001
binary white Y 101.58 1 <0.0001
binag slalom Y 101.58 1 <0.0001

 

164

 

Table 5.4. Continued.

 

 

 

 

Var'i bl Line r chi dfl P-val
19 Stage Y 107.61 4 <0.0001
Strain Y 5.7489 2 0.057
Temp Y 22.098 1 <0.0001
s(Length) N 0.3813 1 0.54
s(Weight) N 4.9028 1 0.027
s(Length,Weight) N 51 .222 14.5 <0.0001
s(chitin synthase) N 4.8273 1 0.029
s(heat shock protein 60) N 73.642 4.34 <0.0001
s(heat shock protein 90) N 24.305 3.64 <0.0001
s(acetylcholine esterase) N 89.688 6.41 <0.0001
s(ecdysone receptor) N 18.27 1 <0.0001
s(resistance to organophosphate 1) N 51.348 6.19 <0.0001
s(white) N 0.6523 1 0.42
s(ultraspiracle) N 35.758 3.86 <0.0001
s(slalom) N 2.1667 2.15 0.37
binary chitin synthase Y 23.006 1 <0.0001
binary acetylcholine esterase Y 23.006 1 <0.0001
binary white Y 23.006 1 <0.0001
binary/3mm Y 23.006 1 <0.0001
20 Stage Y 120.09 4 <0.0001
Strain Y 11.099 2 0.0042
Temp Y 10.942 1 0.001
s(Length) N 6.7474 3.38 0.11
s(Weight) N 4.8594 1 .2 0.038
s(Length,Weight) N 24.463 9.38 0.0054
s(heat shock protein 60, chitin synthase) N 41.961 17.8 0.0015
s(heat shock protein 60, heat shock protein 90) N 39.926 7.75 <0.0001
s(heat shock protein 60, acetylcholine esterase) N 63.749 7.54 <0.0001
s(heat shock protein 60, ecdysone receptor) N 56.883 15 <0.0001
s(heat shock protein 60, resistance to organophosphate 1) N 58.023 15.4 <0.0001
s(heat shock protein 60, white) N 0.5825 1 0.46
s(heat shock protein 60, ultraspiracle) N 50.482 9.58 <0.0001
s(heat shock protein 60, slalom) N 0.2966 1.09 0.63
binary chitin synthase Y 18.171 1 <0.0001
binary acetylcholine esterase Y 18.171 1 <0.0001
binary white Y 18.171 1 <0.0001
binary slalom Y 18.171 1 <0.0001

 

165

 

 

 

 

Table 5.4. Continued.
Model Variable Linear chi-cg dfledf P-value

21 Stage Y 167.71 4 <0.0001
Strain Y 20.991 2 <0.0001
Temp Y 16.04 1 <0.0001
s(Length) N 0.7976 1 0.37
s(Weight) N 1 1.685 2.99 0.0091
s(Length,Weight) N 0.5645 0.37 0.21
s(Length, chitin synthase) N 2.7732 1 0.097
s(Length, heat shock protein 60) N 130.35 9.69 <0.0001
s(Length, heat shock protein 90) N 64.9 10.1 <0.0001
s(Length, acetylcholine esterase) N 103.54 14.2 <0.0001
s(Length, ecdysone receptor) N 38.517 1 <0.0001
s(Length, resistance to organophosphate 1) N 56.472 12 <0.0001
s(Length, white) N 1.1918 1 0.28
s(Length, ultraspiracle) N 107.65 14.9 <0.0001
s(Length, slalom) N 30.064 7.77 0.0002
binary chitin synthase Y 17.721 1 <0.0001
binary acetylcholine esterase Y 17.721 1 <0.0001
binary white Y 17.721 1 <0.0001
binary slalom Y 17.721 1 <0.0001

22 Stage Y 207.64 4 <0.0001
Temp Y 12.111 1 0.0006
s(Length,Weight) N 15.495 5.48 0.013
s(Length, heat shock protein 60) N 143.36 1 1 <0.0001
s(Length, heat shock protein 90) N 45.018 8.54 <0.0001
s(Length, acetylcholine esterase) N 96.334 13.1 <0.0001
s(Length, ecdysone receptor) N 67.053 13.6 <0.0001
s(Length, resistance to organophosphate 1) N 81.328 10.5 <0.0001
s(Length, ultraspiracle) N 1 1 1.35 15.1 <0.0001
s(Length, slalom) N 20.687 6.9 0.0046
binary chitin synthase Y 5.1045 1 0.024
binary acetylcholine esterase Y 17 1 <0.0001
binary white Y 0.0717 1 0.79
binary slalom Y 17 1 <0.0001

 

 

 

166

Table 5.4. Continued.

 

 

 

Model Variable Linear chi-e dfl P- lue
23 Stage Y 193.9 4 <0.0001

s(Length,Weight) N 18.197 5.39 0.0042
s(Length, heat shock protein 60) N 140.01 13.3 <0.0001
s(Length, heat shock protein 90) N 38.366 7.27 <0.0001
s(Length, acetylcholine esterase) N 104.41 1 5.4 <0.0001
s(Length, ecdysone receptor) N 74.27 14.8 <0.0001
s(Length, resistance to organophosphate 1) N 70.912 10.3 <0.0001
s(Length, ultraspiracle) N 142.34 16.4 <0.0001
s(Length, slalom) N 18.154 7.02 <0.0001
binary chitin synthase Y 4.0422 1 0.045
binary acetylcholine esterase Y 10.675 1 0.0012
binary white Y 0.0231 1 0.88
binarlealom Y 10.675 1 0.0012

 

The significance and degrees of freedom for all variables in all GAMs assessed in this
project. For genetic data, the simplest model is shown in which a gene was statistically

significant. Model 21 contains all available data; after that model, non-significant

variables were removed. All terms were significant predictors of age in at least one
model. Linear indicates whether (Y) or not (N) a term is linear. Df=degrees of freedom.
edf=estimated degrees of freedom. P-value=estimated P-value. See the Generalized

Additive Models section for greater details on models.

167

 

CHAPTER 6: VALIDATION OF GENE EXPRESSION BASED
PREDICTIONS OF BLOW FLY AGE WITH BLIND PREDICTIONS

Introduction

The results presented in Chapter 5 indicated that predictions of blow fly
development percent made with gene expression data are more precise than current
forensic entomology approaches. However, the predicted performance of a model does
not necessarily ensure that predictions will follow the expected pattern. To prove that a
statistical model is a valuable predictor of development percent it is necessary to validate
it (Scheiner and Gurevitch 2001), in this case by estimating the ages of unknown
immature blow flies in a blind study. If the predicted ages of individuals plot along a
True = Predicted line, then a model can be considered validated for use. In this chapter,
the predicted ages of L. sericata collected on rats were used to validate the use of gene

expression for PM] estimates.

Methods
Strain Collection and Identification.

To avoid the effects of inbreeding resulting from over-winter rearing, a new
collection of L. sericata was caught out of doors on the Michigan State University
campus in May of 2006. Individuals were identified visually and by C01 sequence, as
described in Chapter 3.

Collection of Unknowns.
Egg masses were placed in the mouths of three C02 asphyxiated rat carcasses as

described in Developmental Plasticity. Two cohorts were raised in incubators, one at

168

20°C and one at 33.5°C. Rearing conditions were the same as in Chapter 2. The third
cohort was raised in a sealed terrarium under outdoor ambient conditions. The terrarium
was covered with a tightly fitting foam lid, which had a hole cut in the middle to allow air
in. The hole was sealed by a screen, which kept other flies out of the terrarium.

Several individual larvae or pupae were collected daily from each of the three
cohorts. An independent researcher recorded the ages of the individuals, which were
stored in RNAlater (Applied Biosystems) at -—80°C. Pupae were collected as described in
Chapter 3 for the stable temperature treatments; in the terrarium no attempt was made to
separate pupae by age. Cohorts were checked daily until adults were observed, which
was recorded as the minimum development time for the cohort; all development percents
were calculated based on minimum development time.

For the ambient temperature experiment, temperature variance was noted using a
HOBO data logger (Onset Computer, Boume, MA). Temperature data were used to
calculate accumulated degree hours (ADH) by subtracting a base temperature of 10°C
fi'om the recorded temperature at any hour and adding the hourly values together. This
resulted in an accumulated degree hour (ADH) calculation. The total ADH until eclosion
was used for the minimum development time and the ADH at the time of collection was
divided by that number to calculate minimum development percent for the ambient
temperature cohort.

Converting RN Alater Size Measurements to Live Size Measurements.

Since the body size measurements in the previous section were based on live L.

sericata, and the unknown individuals were presented for analysis stored in RNAlater, it

was necessary to develop a protocol for converting size measurements of individuals

169

 

 

stored in solution to an estimate of live size. Unanalyzed individuals (N=400) from the
MI collections in Chapter 3 were remeasured and reweighed after approximately one year
of storage in RNAlater at —80°C. The new measurements and weights were plotted
against the recorded live data. Regression equations for the comparisons were used to
convert the sizes of the unknown individuals to usable data for GAMs.

Molecular Biology and Statistics.

Gene expression CT scores were obtained as described in the previous section,
which were used to predict the ages of blind study individuals. Several predictions were
made for each individual, beginning with a GAM that included length, weight, and
developmental stage to predict development percent. Likewise, GAMs were produced
utilizing body size measurements, developmental stage, and all available gene expression
information. Either the gamma or gaussian distributions were applied to the models,
depending on which better satisfied the assumptions of error (discussed in Chapter 3).
Models were also developed with gene CT scores, in which each gene expression term
was anchored [i.e. the term s(variable, gene) was used] against length measurements
[s(length,gene)], as well as hsp60 CT scores [s(hsp60,gene)] when hsp60 was detected
(see Table 5.3).

Two to four predictions were generated for each individual. Development
percents were predicted for all individuals based on body size and developmental stage
data. An additive model was also developed for each individual, using all available data.
If an individual was raised at a known temperature, then temperature was included in the I
model. Strain was not included, as few investigators will have that information. If a

diagnostic plot indicated that anchoring a prediction with length was not likely to

170

improve upon other predictions made with body size and stage (based on the description
of model types below), then a length anchored prediction was not used.

Predicted values of development percent were plotted against true development
percent, and were visualized with the Y = X (True = Predicted) line i 10%. A GAM was
classified as a Type 1 model if its diagnostic plot was similar to Model 15 from the
previous section, or as a Type 2 model if the diagnostic plot indicated no gap in
predictions and a decrease in the error throughout development. Predictions from Type 1
and Type 2 models were plotted and compared to predictions made with body size and

stage information only.

Results

The new strain of Michigan L. sericata was identified by visual identification and
through C01 sequence. A BLAST search on the NCBI website revealed that a 753 base
pair fragment of the C01 gene was most similar to another L. sericata sequence, with
100% similarity. The closest match to non-L. sericata sequence was a 99% match to that
gene in L. cuprina, with a 7 base pair mismatch.

Ambient temperatures spiked each day, then decreased to an evening low (Figure
6.1). Despite the fluctuations, the ADH based development percents were very similar to
the purely temporal development percents (Figure 6.2). Because ADH is generally used

in forensic entomology they were used for the analyses detailed below.

171

Temperatures During the Outdoors Experiment

 

01
O
1
1

e
1 e
1
1
e
1
1
1

 

 

(9AA
01001
e

1

l

1

1

1

r

(0
O
‘—
l

r

1

1

 

 

 

Temperature In Celsius
N
01
O
O

 

 

—L—LN

0010
1
1
1
1

8/7/2006 SIS/20% 8111/2m6 8(1312006 811 512015 8/1 712006 SHED/20% BI21l2CX£ 8123/2113 8/25/20%
0:00 0 (X) 0.00 0,00 0.00 0 00 0:00 0:00 0 00 000

Time and Date

Figure 6.1. Temperature plots for the ambient temperature experiment.
Temperatures spiked in the afiemoon as sunlight was directly on the terrarium.

 

Percent Home vs. Percent ADH" 0)

- 0.9964): + 1.2448
I 0.993

Percent in ADI-K10)

 

0 20 40 60 80 100 120
Percent in Hours

 

 

 

Figure 6.2. The regression of ADH (base 10) percents versus percent development in
hours. The values were strongly correlated, with an R-squared and slope almost 1.
However, the 1.2448 shift up the Y axis indicates that ADH is a better measure of
developmental progress.

Ninety individuals were used to obtain cDNA samples for the blind study. Four

of these were non-pupators (see the next section) and 86 were normally developing larvae

172

and pupae. Body size measurements were calculated based on the equations in Tables
6.1—2, which were specific to each stage. Seven flies were identified as first instars, 4
were second instars, 10 were feeding third instars, 23 were postfeeding third instars
(based on crop content), and 42 were pupae. The 86 individuals were divided among
temperature treatments with 34 raised at 20°C, 29 raised at 33.5°C, and 23 raised at
ambient temperature. Of the postfeeding third instar identifications, 5 were
retrospectively found to be feeding third instars, but predictions were made using the
identified stage, as an investigator would not have been able to recognize a misidentified
stage. F orty-nine individuals generated data that could be used in Type 2 unanchored
models. Fifty-six individuals received hsp60 anchored predictions. The ages of 71 flies
were predicted based on models that used gene expression data anchored by length.

The plot for predicted versus true age for a GAM using developmental stage and
body size to predict development percent (Figure 6.3) was used as the benchmark against
which all other predictions were compared. The plot also included predictions for the
four non-pupators (in the gray bar), which will be discussed in more detail in the next
section. For the purposes of validation, the non-pupators were not considered further. If
the hypothesized increase in precision occurred as predicted, then the results of plots
from the models that used gene expression to predict blow fly age should demonstrate an
improvement over the results in Figure 6.3. Additionally, the error associated with these
models was considered. Based on the diagnostic plots of gene expression GAMs (from
the previous section), approximately 90% of the data should be within 10% of the true
age. With realistic collection conditions, individuals that are younger than the minimum

age are expected (only a subset of flies on a body can be the oldest), thus

173

underpredictions of age are anticipated. However, the minimum development time of a
blow fly is the important information used by forensic entomologists, thus very little
overprediction of blow fly age should be tolerated in a model that will be used to predict
this datum. When body size and stage data were used to predict blow fly age (Figure
6.3), the predictions of larval age was within 10% of true development except for 6
individuals. However, many pupal ages were predicted as greater than 10% older or 10%
younger than their true ages. Likewise, few individuals were predicted to be between 40
and 60 percent developed, and no individuals were predicted to be greater than 75%

developed.

 

Predicted Develounent Percent

 

 

 

 

O 20 40 60 80 100

Two Development Percent

Figure 6.3. Predicted versus true development percents for all 90 individuals in the
blind study. Predictions were made with GAMs that did not use gene expression
information. The four points in the gray bar represent the four larvae that failed to
pupate, and were eliminated from analyses in this section.

All gene expression models (Figures 6.4—10) improved upon the GAM

predictions made with body size/stage data (Figure 6.3). When unanchored additive gene

174

 

expression predictions were compared to their true ages (Figure 6.4), 100% of the larvae
were predicted as within 10% of their true ages or younger than their true age. Similarly,
85.7% of pupae were predicted as less than 10% older than their true development
percents. Most importantly, pupal age predictions were more precise than the size/stage
predictions of age, as seen by the reduced size of the gap between larval and pupal age
predictions. Additionally, development percents were accurately predicted when true
percents were greater than 80%. When only the Type 2 additive model predictions
(which included models predicting development percent with complete, or almost
complete, gene expression profiles) were plotted against their true development percents
(Figure 6.5) 93.9% of individuals were predicted as within or less than 10% of their true
ages. In the three cases that larvae were predicted as older than 10% of true age, the
predictions were only marginally greater than 10% deviant from the true development
percent, which was not the case for predictions of individuals from the preceding models
(Figures 6.3—4). Length anchored predictions followed a similar pattern, overall and for
Type 2 models (Figure 6.6—7). Over-predicted development percents (>10%) with Type
2 models of this type only occurred twice (again, with less deviation from the true age
than the size/stage based model). However, models that used hsp60-anchored predictions
produced too many imprecise predictions to be considered useful and were not

considered further.

175

AllAddltlvePredlctloneVereueTmePercente

 

Predicted Development Percent

 

 

 

 

Tme Development Percent

Figure 6.4. Predicted versus true development percents for the 86 normally
developing individuals in the blind study. Predictions were made with GAMs that
used unanchored gene expression terms.

Type 2 Additive Predictione Venue True Percente

 

Predicted Development Percent

 

 

 

 

r r l r r
20 40 60 80 100

Two Development Percent

Figure 6.5. Predicted versus true development percents for 49 normally developing
individuals in the blind study. Predictions were made with GAMs that used
unanchored gene expression terms and were Type 2 models.

176

All Length Anchored Predictions Versus True Percent:

 

Predicted Development Percent

 

 

 

 

20 40 60 80 1 00

Two Development Percent

Figure 6.6. Predicted versus true development percents for 71 normally developing
individuals in the blind study. Predictions were made with GAMs that used length
anchored gene expression terms.

Type 2 Length Anchored Predictions Versus True Percents

 

Predicted Development Percent

 

 

 

 

True Development Percent

Figure 6.7. Predicted versus true development percents. Predictions were made with
GAMs that used length anchored gene expression terms and were Type 2 models.

177

The performances of Type 2 additive models were also evaluated by temperature,
but it was not possible to determine if certain temperatures were associated with more
accurate predictions than others; since 54.4% of profiles were predicted with unanchored
additive Type 2 models and each temperature was only a third of the remaining 49
individuals there was little data to compare. The 20°C predictions seemed the most
accurate, but more research (with larger sample sizes) will be required to establish
whether or not temperature (as well as other environmental factors) affect predictions
made with such data.

All models continued to demonstrate a bias toward predicting individuals to be
younger than their true ages. Adding gene expression information did not eliminate
underestimates of age; however, the frequency of underestimates was less than those

made with the models in Figure 6.1.

Discussion

The results of the blind study demonstrated the utility of using gene expression
for predicting blow fly age, and the use of GAMs to predict blow fly age by incorporating
gene expression data was validated. When complete or nearly complete gene expression
profiles (Type 2 models) were used to predict blow fly age, the predictions were much
more precise. Both gene based GAMs and GAMs that used length anchored gene
expression terms provided more precise predictions of age, generating far more blow fly
age estimates within 10% of actual age. In doing so, a number of problems with the

current PMI prediction process have been addressed. Currently, the range of

178

 

development percents for pupal samples is ranges from ~40—100% (see Chapter 3) if
standard entomological measurements are employed. The use of gene expression clearly
decreases this range to within ~10% of the true development percent. Likewise, the
predictions produced with gene expression GAMs yielded a more seamless transition of
age estimates from postfeeding third instar to pupa. As well, estimates of ages greater
than 10% above true age, which are the least desirable, remarkably decreased with these
models as compared to the pupal predictions without gene expression data.

Error in predicting individuals to be younger than they were was more common.
As this phenomenon occurred independent of the use of gene expression data it is not a
signature shortcoming of using RNA profiles to predict age. Indeed, the aging bias
occurred without the use of gene expression data. As previously mentioned in Chapter 2,
L. sericata development time varies, with only some individuals developing at the
maximum rate. Thus some individuals could truly be less mature than the most
developed of their cohort. In natural conditions, underestimates of age will be likely; as
females do not stop laying after the first female has oviposited on a corpse. The most
important data, however, from a blow fly age estimate is the predicted minimum
development time/PMI. Given that there is a maximum development rate, as long as a
model is not likely to predict individual ages that are older than their true age, then the
minimum development time will be precisely estimated, as was the case in the blind
study. This problem can be largely overcome by sampling multiple individuals. The
most developed can be used as an indicator of the minimum amount of time that flies
could have colonized remains, thus avoiding or ameliorating a problem with

underpredictions of age. Given the ability of GAMs to accurately predict blow fly

179

development percents using gene expression/body size/stage data, and with more
precision than GAMs that employed traditionally used forensic entomology measures
(size/stage), the application of these models in conjunction with transcript level
information is an attractive option for predicting development percents of immature L.

sericata for PMI estimates.

180

 

Table 6.1. Regressions of live length against preserved length for each stage.

 

 

 

 

 

 

 

 

 

 

Stage Regression Equation P-value R2

1st Instar Length=l .062*RNA Later <0.0001 0.76
Length+0.3

2nd Instar Length=1.548*RNA Later <0.0001 0.77
Length-0.48

3'a Instar Length=1.095*RNA Later <0.0001 0.89

(feeding) Length+2.2

3'a Instar Length=0.7371*RNA Later <0.0001 0.59

stfeeding) Length+5.2

Pupa Length=0.9059*RN A Later <0.0001 0.88

Length+0.65

 

Table 6.2. Regressions of live weight against preserved weight for each stage.

 

 

 

 

 

 

 

 

 

 

Stage Regression Equation P-value R2

1st Instar Weight=1.078*RNA Later <0.0001 0.87
Weight+0.07

2“d Instar WeighFl .O77*RNA Later <0.0001 0.98
Weight-0.04

3rd Instar Weight=l .009*RNA Later <0.0001 0.99

(feeding) Weight+0. 1 2

3rd Instar Weight=l .047*RNA Later <0.0001 0.97

(postfeeding) Wei ght-0.94

Pupa Weight=0.9541*RNA Later <0.0001 0.98
Weight+0.23

 

181

 

 

CHAPTER 7: GENE EXPRESSION ANALYSES OF NON-
PUPATIN G LARVAE

Introduction

Forensic entomology uses the predictable development times of blow flies
(Diptera: Calliphoridae) to estimate a postmortem interval (PMI) by backtracking from
the developmental stage of the most mature insects collected on a body to the time that
the insects colonized the remains (Catts and Haskell 1990, Greenberg and Kunich 2002),
a time that can approximate the PMI. Blow fly development is known to be plastic as
different food sources and rearing treatments can delay development (Kaneshrajah and
Turner 2004, Clark et al. 2006, Tarone and Foran 2006) and any factors that contribute to
plasticity can result in inaccurate PMI estimates.

In previous research on L. sericata (Diptera: Calliphoridae) (Meigen)
development (Chapters 3 and 6), a curious form of developmental plasticity was observed
in which postfeeding third instar larvae, as determined by behavior, the visibility of tissue
in the crop, and gene expression levels (both cs and ecr), were present at a time when
other members of their cohort had reached adulthood. The pupal stage between the
postfeeding third instar and the imago is the longest in L. sericata immature development
and can last for more than a week (Grassberger and Reiter 2001). Clearly, the collection
and analysis of these abnormally old larvae, dubbed herein as “Peter Pan” larvae, could
have a drastic effect on the accuracy of a PMI estimate if pupae were not collected at the
death scene. A forensic entomologist is unlikely to miss pupal evidence, however; one
may not be present at the time of collection, which may result in an untrained individual

overlooking pupal samples. If pupae are overlooked, “Peter Pan” larvae may be included

182

 

in a larval sample from a death scene. Any means of identifying “Peter Pan” larvae will
be usefiil in avoiding their analysis for PMI predictions and may also be an indication
that pupae were missed in sampling. During the previous observation of the “Peter Pan”
phenomenon (Chapter 3), 10 (or as many as were available) such individuals were
collected, and placed in an RNA preserving fixative. Fifty five were profiled for the
mRNA expression levels of 9 genes (Chapter 5), revealing usefirl information for
identifying these larvae. The profiles of informative genes were then compared to the
profiles of “Peter Pan” larvae collected in a blind study, to confirm that they can be used

to successfully predict the condition.

Methods

Ten “Peter Pan” larvae were collected from each replicate (Chapter 3), except in
the case of samples from the CA strain, which is discussed below. The mortality of
larvae in each cohort and the production of “Peter Pan” individuals were generally
observed for each strain, though no formal measures were made. Samples were fixed in
250 uL RNAlater (Applied Biosystems) and stored at —80°C. RNA was isolated and
converted to cDNA using previously described methods (Chapter 5). Quantitative PCR
analyses of these larvae were then conducted, using primers specific for 9 genes as
previously reported (Chapter 5). Quantitative PCR generates threshold cycle (CT)
values, which can be used to determine the relative concentrations of different transcripts
in an individual, that were standardized against housekeeping genes and a common
cDNA sample, allowing for a comparison of CT scores of “Peter Pan” larvae to normally

developing postfeeding third instars (Chapter 5). Quantitative PCR is based on the

183

exponential increase of target DNA in a reaction, resulting in low CT scores for genes
that are expressed at high levels. Analyzed individuals represented the three strains (CA,
MI, WV) and two rearing temperatures (20°C and 33.5°C) studied in the previous
experiment (Chapter 3). CT values of non-pupating larvae from 55 individuals were
plotted by stage, to determine if any genes exhibited notable differences between normal
postfeeding third instars and “Peter Pans”. For genes that exhibited differential
expression between normal and non-pupating postfeeding third instars, expression was
also compared to the profiles of “Peter Pan” larvae in a blind study (Chapter 6) to
determine if it was possible to predict the state in unknown flies.

All boxplots of expression profiles were made in R (R core development team)
with the blind study profiles shown in the right panel of the plots. Median values are
represented by a horizontal line, the middle fiftieth percentile of CT scores is represented
by the box, and the ninety-ninth percentile is contained within the whiskers of a plot.
Expression profiles for a gene were considered different if the fiftieth percentile box for

“Peter Pan” flies was beyond the median for that gene in postfeeding third instars.

Results

There were qualitative differences among the three strains studied (CA, MI, WV)
for both the percent of 250 postfeeding third instars that were “Peter Pan” individuals, as
well as mortality at 33.5°C. The CA strain was not likely to produce larvae that did not
pupate and, as a result, 10 “Peter Pan” larvae could not be sampled from all of the CA
replicates. Also, the WV strain was observed to experience much more mortality at the

high temperature treatment than the other two strains.

184

 

Nine gene expression profiles were produced for the “Peter Pan” larvae, allowing
comparisons to the transcript level profiles of normally developing L. sericata in previous
research (Chapter 5). Five of these genes were expressed within the range observed in
postfeeding third instars, whereas four genes (hsp90, rop-l, sll, usp) were identified as
having expression patterns that might be used to differentiate “Peter Pan” larvae from
normal postfeeding third instars. The hsp90 gene was up-regulated in larvae that had
arrested development (Figure 7.1), while the other three loci were down-regulated as
compared to postfeeding third instars (Figures 7.2—7.4). Expression plots (by stage) for
these genes were shown with their counterpart plots in the blind study (Figures 7.1—4).
Of the four genes that exhibited differential expression patterns in the 55 known non-
pupators, three of them (hsp90, sll, usp) expressed similar, abnormal patterns in blind
study “Peter Pans”, and could have been used to identify them as individuals that should

not be incorporated into a PMI estimate.

185

 

hepOi) Expression Compared to Non-Pupetors

 

15
L
O

10
1
~11»
O

 

 

 

 

 

 

 

 

 

 

 

 

 

Standardized hepOOExpreesion
5
1
1
1

 

 

 

Figure 7.1. Standardized gene CT scores for hsp90, plotted by stage, for all
individuals from the previous section, for the 55 non-pupator individuals from this

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Pupe

study, and the individuals in the blind study. The left panel represents expression

from a previous section (Chapter 5), with results from the 55 non-pupators (NP) placed
next to postfeeding third instars (3rdPF). The right panel represents gene expression for

the individuals in the blind study (Chapter 6), including the 4 non-pupators. 1St = first
instar. 2"d = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar.
NP = non-pupator or “Peter Pan” larvae. On average, hsp90 was expressed at higher
levels in non-pupators than in postfeeding third instars.

186

 

 

rep-1 Expression Compared to Non-Pupetors Blind Study rep-1 Expreeelon

 

 

15
l

O

O

O
0
E33:—

-‘# O 000 O

10
l
"100 o

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

StandardizedropExpreesion

 

 

 

 

 

 

 

 

 

 

 

 

._ 4 EB ' 1
8 E _:_ _E_ wr- - E E
o _'._ E : . _1_
. —.:,— : : =
o - O N - -_ _i._ 0
I I I I I I T I I I
181 2nd 3rd 3rdPF NP Pupe 3rd W NP Pupa
DevelopmentalStage DevelopmentalStege

 

Figure 7.2. Standardized gene CT scores for rap-1, plotted by stage for all
individuals from the previous section, for the 55 non-pupator individuals from this
study, and the individuals in the blind study. The left panel represents expression
from a previous section (Chapter 5), with results from the 55 non-pupators (NP) placed
next to postfeeding third instars (3rdPF). The right panel represents gene expression for
the individuals in the blind study (Chapter 6), including the 4 non-pupators. 1St = first
instar. 2"d = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar.
NP = non-pupator. On average, rop-I was expressed at lower levels in non-pupators than
in postfeeding third instars, though this pattern was not apparent in the blind study.

187

 

sllExpreeelonComperedtoNon-Pupetore
a '- o
O O
O
8 e o o o
e
:2 - °

StandardlzodsliExptmm
mm“,
111

 

 

 

o
T F T fl T 1
1st 2nd 3rd 3rdPF NP Pupe
Developmental Stage

BlindsmdyellExpreeelon

 

 

14
J

 

12
l

 

 

 

 

 

10
l

 

 

 

 

 

 

 

 

 

StanderdzedsiiExpreeslon

 

 

 

 

 

 

Developmental Stage

Figure 7.3. Standardized gene CT scores for sll, plotted by stage, for all individuals
from the previous section, for the 55 non-pupator individuals from this study, and
the individuals in the blind study. The left panel represents expression from a previous
section (Chapter 5), with results from the 55 non-pupators (NP) placed next to
postfeeding third instars (3rdPF). The right panel represents gene expression for the
individuals in the blind study (Chapter 6), including the 4 non-pupators. 1St = first instar.
2'“I = second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar. NP =
non-pupator. On average, sll was expressed at lower levels in non-pupators than in

postfeeding third instars.

188

 

uspExpresslonComperedtoNon-Pupetors BlindStudyusp Expression

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0 °_ 0
8-4
co—i .
. m.“ o —'_ i
T— :— _g_ e _._ “H —“ [___l t
£21 . .2. i i . ;
‘8 l:l F'_J l 8 : _.r_
g , gr.jfi uF—AF: 5v— .4...
2 “H 4— : _.-_ : r L g
_.i._ '4- O
a o . a M
O—t
in O 0‘
' T l V I I I I I I 1
1st 2nd 3rd 3W NP Pups 3rd 3rdPF NP Pupa
DevelopmentalStage DevoioprnentalStago

 

Figure 7.4. Standardized gene CT scores for usp, plotted by stage, for all
individuals from the previous section, for the 55 non-pupator individuals from this
study, and the individuals in the blind study. The left panel represents expression
from a previous section, with results from the 55 non-pupators (NP) placed next to
postfeeding third instars (3rdPF). The right panel represents gene expression for the
individuals in the blind study, including the 4 non-pupators. 1St = first instar. 2"d =
second instar. 3rd = feeding third instar. 3rdPF = postfeeding third instar. NP = non-
pupator. On average, usp was expressed at lower levels in non-pupators than in
postfeeding third instars.

Discussion

Potential Causes of the “Peter Pan” State.

The “Peter Pan” condition could result from a number of factors. Winter-induced
diapause, a cessation of development brought on by cool temperatures and short day
lengths, is known to occur in L. sericata and can vary widely among populations of this
species (Greenberg and Kunich 2002). The light cycle used (12: 1 2) in this study (see
Chapter 3) and the low temperature treatment (20°C) are both capable of inducing this
state (Tachibana and Goto 2004), indicating that “Peter Pan” larvae may be diapausing.
Additionally, the CA strain produced less non-pupating larvae than the other strains,

revealing differences in the genetic predisposition for producing larvae with arrested

189

development. Another difference among the strains also points to diapause, as the MI
and WV strains were collected from regions that experience snow and prolonged periods
of cold (thus necessitating an arrest of development to overwinter) and were more likely
to produce “Peter Pan” larvae, suggesting that diapause is the cause of the “Peter Pan”
state.

Alternatively, a summer diapause condition (known as estivation) might also be
activated in these flies (e. g. Liu et al. 2006). The authors observed a delay in the
development of the cotton bowl worm Helicoverpa armigera (Lepidoptera: Noctuidae)
that was caused by high temperatures. Estivation seems unlikely in this instance, as 20°C
is not warm enough to induce heat stress. Likewise, L. sericata are capable of developing
at higher temperatures than 33.5°C (Grassberger and Reiter 2001). It should be noted
though, that the WV strain experienced more mortality at the higher temperature
treatment than the other two strains, indicating that heat stress may be possible for some
strains at 33.5°C.

Last, the “Peter Pan” condition could be the result of mutations in genes that
regulate development. As an example, D. melanogaster insulin receptor (Inr) mutations
can delay larval development (Shingleton et al. 2005). It is possible that different alleles
of genes that regulate development exist in natural populations in high enough
frequencies to result in the observation of severely delayed development of larvae in
some strains. There are myriad examples of stage-specific lethal mutants in D.
melanogaster (www.flybase.org), including mutations of dre4 (Sliter and Gilbert 1992),
which is involved in the response to ecdysone, and E74A (Bialecki et al. 2002), which

regulates the proper progression of immature developmental stages. Clearly, many genes

190

are required for the proper development of flies as they undergo molts and
metamorphosis and genes that have mutant alleles with a lethal phenotype resulting in an
inability of third instars to pupate could potentially explain the phenomenon of non-
pupating larvae.

Interpreting Gene Expression Changes in “Peter Pan” Larvae.

Some information on the “Peter Pan” condition can be gleaned from the gene
expression changes that occur in these individuals, shedding light on the mechanism of
the developmental delay. A potential molecular marker of diapause is hsp90, which was
also profiled in this study, as it is up-regulated in larvae that are breaking diapause
(Tachibana et al. 2005). hsp90 expression in non-pupating individuals is commensurate
with the breaking of diapause, since it was up-regulated in “Peter Pan” larvae. The
“broken diapause” hypothesis is supported by the possibility that the initiation and
breaking of diapause prolongs C. vicina development at low temperatures, as observed by
Ames and Turner (2003). They noted an extremely long delay in development at very
low temperatures, which was disproportionately large compared to the delay that is
expected by decreasing temperature. The authors hypothesized that C. vicina begins and
then breaks diapause when raised in environments that are close to the point at which the
condition would be initiated, thereby increasing the delay in development. Such a delay
could also be possible in L. sericata that were raised in environments that activate
diapause, though the effects would be less pronounced in this experiment as the flies
were not raised under “borderline” temperature conditions.

Alternatively, high hsp90 levels could be symptomatic of a mutant condition that

results in hsp90 up-regulation. The simultaneous down-regulation of usp supports this

191

possibility, as usp and ecr interact to regulate other genes throughout fly development
(Henrich and Brown 1995). These loci covaried with hsp90 in 33.5°C temperature
treatments (Chapter 5), raising the possibility that the gene is regulated by the ecdysone
response, which is affected by juvenile hormone (J H) concentrations. High JH levels are
known to prevent pupation (Kalthoff 2001) and usp is also a putative target of JH binding
(Flatt et al. 2005). An interaction between usp and J H could explain the down-regulation
of usp in flies that do not pupate if “Peter Pan” larvae were mutants for the J H response,
possibly through faulty usp regulation.

The last gene that was down-regulated both in known and unknown samples was
sll. It is expressed in the salivary glands of L. sericata larvae (Ali et al. 2005) that
degrade during the postfeeding third instar. The “Peter Pan” state is an extremely
prolonged postfeeding third instar, thus the down-regulation of sll in these flies is likely
to be a result of extreme degradation of the salivary glands that express the gene,
meaning that it is a good marker for the condition, but unlikely to reveal useful
information as to why “Peter Pan” flies fail to pupate.

Given the reliable genetic regulatory changes associated with L. sericata larvae in
arrested development, the cause of the “Peter Pan” state is probably due to one of two
reasons: diapause or mutation. It seems most likely that the conditions of the experiment
contributed the temporary initiation of diapause in L. sericata larvae. The light cycle
(12: 12) and one of the temperature treatments (20°C) are known to induce the state in on
population of L. sericata (Tachibana and Goto 2004) and hsp90 was up-regulated in
larvae that experienced arrested development, in a manner that was consistent with the

breaking of diapause (Tachibana et al. 2005). Alternatively, the inability to pupate may

192

be due to mutations in genes that regulate development, as supported by the down-
regulation of usp in larvae that do not develop properly. If this is the case, there must be
a selective reason for the maintenance of lethal alleles at high frequency in all three
populations that were studied (CA, MI, WV). A selective advantage can occur if
heterozygotes provide a beneficial phenotype as compared to homozygotes. Also, it is
possible that there are lethal mutations that are linked to loci under strong positive
selection, allowing them to “hitchhike” along with the gene under selection. Both
scenarios are documented in population/quantitative genetic literature (reviewed in
Conner and Hartl 2004) and could explain how lethal mutations may persist in three
populations of blow flies, producing larvae that refilse to grow up.

Forensic Analyses of Non-pupating Larvae.

The analysis of “Peter Pan” larvae indicates that three of the nine loci examined,
hsp90, sll, and usp, could be used to eliminate individuals in a state of arrested
development from their use in age estimates, thus avoiding the increase in error
associated with their inclusion. Based on the results, low sll and usp abundance, along
with high hsp90 concentrations, helped identify a “Peter Pan” larva. Recognition of
expression patterns that are indicative of “Peter Pan” larvae will also enable investigators
to identify collections that missed pupae, as “Peter Pans” would be collected at times
when pupae and adults from their cohort should also be present. The analysis of
expression levels for sll, usp, and hsp90 may be instrumental in determining the
developmental status of a postfeeding third instar larva and whether or not it should be

used to predict a PMI.

193

CHAPTER 8: POPULATION AND TEMPERATURE EFFECTS ON
BODY SIZE AND DEVELOPMENT TIME

Introduction

Understanding how ecological conditions drive the evolution of body size and
development time is a fundamental topic of research in evolutionary biology (Oudman et
al. 1991, Morey and Reznick 2000, Kingsolver 2001, Day and Rowe 2002, Hallas et al.
2002, Feder et al. 2003, Conner and Hart] 2004, Plaistow 2004, Gomez-Mestre and
Buchholz 2006). One area of such research focuses on the fact that, when faced with
limited resources, many organisms that go through a metamorphosis experience a
tradeoff between developing quickly (ensuring survival to adulthood) and producing
larger, and more successful adults by delaying maturation. Spadefoot toads, which are
desert-dwelling and develop in vernal ponds, have been extensively studied in this
capacity, with different species delaying development in low resource conditions, while
others sacrifice body size in low nutrient conditions (Morey and Reznick 2000).
Spadefoot toads have much faster development times than ancestral toad species and
evolutionary relationships among them explain some of the differences in size and larval
growth periods (Gomez-Mestre and Buchholz 2006). Though the tradeoff between body
size and development time is documented, attempts to model it have been only partially
successful, as some species will produce the largest individuals when growth is at its
fastest, which led to hypotheses that developmental thresholds (minimum requirements
for life history events to occur) must be involved in the physiological processes that
determine body size and development time, with selection for certain thresholds resulting

in different relationships between size and age at maturity (Day and Rowe 2001). The

194

effects of thresholds in development have been partially upheld by research in the mite
species Scancassania berlesei (Michael), which noted the predicted shift in reaction
norms for development that regulated by thresholds, but also observed effects of maternal
nutrition and sex on the traits (Plaistow et al. 2004). Additionally, work on Pieris rapae
has demonstrated that thermal performance curves are a result of selection pressures on
development that can also affect the evolution of body size (Kingsolver 2000, Kingsolver
2001)

The ephemeral nature of blow fly resources (carrion and biological waste), which
occur sporadically in the environment and disappear quickly, and the global distributions
of some blow fly species could lead to selection for different body sizes and development
times in different populations. One key feature of organisms that experience body
size/development time tradeoffs is the evolution of plasticity (environmentally induced
changes in phenotypes; Day and Rowe 2002), which has been demonstrated in the blow
fly Lucilia sericata (Diptera: Calliphoridae) (Meigen) (Clark et al. 2006, Tarone and
Foran 2006). Size and age at maturity in L. sericata are responsive to changes in
laboratory rearing conditions such as moisture and type of food respectively. The
evolution of plasticity and the potential for a tradeoff between body size and development
time in this species mean that L. sericata is another possible model for studying the
evolution of age and size at maturity.

Studying blow fly development also has applied uses. Blow flies cause myiasis
(parasitization) in humans (Faulde 2001), sheep (East and Eisemann 1993), and cattle
(Ramirez and Crystal 1975). Also, the predictable development of blow flies in optimal

growth conditions can be used as a biological clock to help determine a postmortem

195

interval (PMI) in death investigations (Catts and Haskell 1990, Greenberg and Kunich
2002). Both myiasis and PMI predictions are intimately associated with the immature
development of blow flies, thus any contribution to the understanding of development
also contributes to the ability of researchers to improve PMI predictions and limit the
damage caused by myiasis.

The research presented here stemmed from previous observations of
developmental variation among populations of the forensically informative blow fly L.
sericata. Different authors have reported varying minimum development times for this
species, with some indicating faster minimum development times for this species (Kamal
195 8, Greenberg 1991, Anderson 2000, Grassberger and Reiter 2001), as well as
differences in the development times of Russian and US populations (but, with little
detail explaining the support for the observation; Greenberg 1991). Some of the variation
among studies may be due to differences in laboratory rearing conditions known to affect
development time of L. sericata (Clark et al. 2006, Tarone and F oran 2006), though other
factors may also be involved. In previous research, the effects of temperature and
biological strain were considered throughout development (Chapter 3, Chapter 5), and
consistent differences were observed between cohorts of L. sericata raised at different
temperatures and among three biological strains originating from California, Michigan,
and West Virginia. Low temperatures prompted larger third instar and pupal sizes
(length and weight), and the California strain was consistently the largest, while the West
Virginia strain was consistently smallest. Differences among the strains appeared to be
due to variation in the timing of the cessation of feeding during the third instar (Figure

3.3). The observed effects of temperature and strain on the age and size at L. sericata

196

 

maturity suggest that both the environment and genotype are important considerations
when predicting minimum development times of this species. As with body size, gene
expression was also susceptible to temperature and strain effects (Chapter 5), further
supporting the idea that different populations of L. sericata exhibit variable
developmental patterns. Indeed, variation in body size and development time (the core
phenotypes of interest to forensic entomologists) has been observed in multiple
Cyclorrhaphan species (reviewed in Tarone and Foran 2006).

The explanation for temperature and strain effects on body size, development
time, and gene expression levels, can be found in quantitative genetic theory (reviewed in
Conner and Hartl 2004). Any continuously variable phenotype, such as length or
minimum development time, is subject to the influence of genetics, the environment, and
the interaction between the two, which can be detected through reaction norm plots and
ANOVA (Scheiner and Gurevitch 2001, Conner and Hartl 2004). Because forensic
entomologists use quantitative traits (development time, length, weight) to determine
blow fly age predictions (Kamal 1958, Greenberg 1991, Anderson 2000, Grassberger and
Reiter 2001), the phenotypes can be studied with a quantitative genetic approach to
determine if and how the environment and genotype affect them. Knowledge of such
factors can increase the accuracy of forensic entomological predictions of blow fly age by
identifying genetic and environmental conditions that affect the maturation of the
Calliphoridae.

Genetic variation for body size and development time must also be considered in
terms of new approaches designed to more precisely predict development percents of

blow flies based on gene expression data (Chapter 5). If minimum development times

197

are the same across populations, development percent predictions do not need to be
calibrated to biological strain. However, if minimum development times are quite
different across populations, predictions will be indicative of different ages for different
strains, requiring a calibration appropriate to the strain of interest. A calibration would
pose a problem to the forensic science community, as it would require a database of
development times for any region where forensic entomology is employed or post priori
studies to determine the specific minimum development time of a population of interest.
Though the differences in body sizes and development times were consistent after
hundreds of measurements for all strains/temperatures (within replicates), the number of
replicates per temperature per strain (2) was small, and variability could theoretically be
due to chance. To definitively demonstrate that the growth of individuals from each of
the strains is affected by genotype and temperature, it was necessary to analyze a larger
sample size.

The research presented here was designed to more rigorously test the hypothesis
that genotype and the environment (temperature) affect the development time and body
size of L. sericata. New flies were collected from Davis, CA; East Lansing, MI; and
Morgantown, WV. Measurements of body size in 720 individuals and minimum
development times from 60 replicates were analyzed to determine that biological strain
and temperature affect these traits. Knowledge of such effects will be instrumental in
decreasing error in PMI estimates made with blow fly data and may contribute to a better
understanding of the evolutionary of tradeoff between body size and development time

by providing data on this phenomenon in a new species.

198

Materials and Methods

New collections of L. sericata were made in the summer of 2006 from the same
locations as the flies studied in Chapter 3 (which occurred in 2005), and they were
identified as L. sericata, through visual identification and cytochrome oxidase 1 sequence
(Tarone and Foran 2006). Ten cohorts of each strain were induced to lay eggs as
previously described (Chapter 3). A cohort was assigned to a temporal block, which was
composed of all cohorts laid on that day. Cohorts were split between 20°C and 33.5°C
temperature treatments, and were raised as described in Chapter 3, except that pupae
were left in the pupation substrate, not destructively sampled. For a replicate, twelve
individual pupal lengths (to the nearest 0.5 mm) and weights (to the nearest 0.01 mg)
were measured, as previously described (Chapter 3), on the 3rd day of pupation.
Likewise, the minimum development time (to the nearest 0.5 hours) for a replicate was
recorded.

This scheme resulted in 720 measurements of individual size: 240 for a strain,
with 120 measurements in a temperature treatment. Similarly, 20 minimum development
times per strain were recorded (10 per temperature treatment). Since a block was split
among 20°C and 33.5°C treatments, which were composed of multiple replicates of a
strain, the data were analyzed in R (R core development team) using a split-plot AN OVA
(Scheiner and Gurevitch 2001), with strain nested within temperature, which was nested
within temporal block. Mean, median, minimum, and maximum values for traits were
reported for the 2006 study. Pupal data from Chapter 3 were also analyzed to determine
if results were similar between experiments, but these data were not considered as

thoroughly, as they were based on two replicates per strain.

199

Boxplots (described in Chapters 2 and 7) and reaction norm plots were
constructed in R to compare weight (a more precise measure of size than length, which
was estimated to the nearest 0.5 mm) and development time variation among strains and
between temperatures. For all reaction norm plots, the Davis, California strain (CA) is
represented by red, the East Lansing, Michigan strain (MI) is represented by blue, and the

Morgantown, West Virginia strain (WV) is represented by green.

l
Results {

All three strains were identified as L. sericata. Sequence from all strains resulted
in BLAST hits to other published L. sericata cytochrome oxidase 1 sequences. A 361 bp
sequence from the CA strain was 100% similar to known L. sericata sequence (accession
number DQ868523). A 762 bp sequence fi'om the M1 strain was 100% similar to known
L. sericata sequence (accession number L14947). A 494 bp sequence fiom the WV
strain was 100% similar to known L. sericata sequence (accession number DQ868523).
In all cases, the most similar sequences in the BLAST searches were L. sericata
sequences, which were distinguishable from L. cuprina sequences by at least 1%
sequence identity.

Minimum development times for the 2005 results were within the ranges of
development times for the 2006 experiment. At 33.5°C, the mean minimum development
time was 257.9 hours (minimum: 238 hours, maximum: 283.5 hours) for the 2005
experiment and 288.7 hours (median: 283.4 hours, minimum: 237 hours, maximum: 357
hours) for the 2006 experiment. When strains in the 2006 experiment were compared

(Figure 8.1), CA developed slower (mean: 308 hours, median: 312.3 hours, minimum:

200

264 hours, maximum: 357.5 hours) than MI (mean: 276.1 hours, median: 278.8 hours,
minimum: 237 hours, maximum: 311 hours) or WV (mean: 282.1 hours, median: 274.3
hours, minimum: 260.5 hours, maximum: 333 hours) in the high temperature treatment.
At 20°C, the mean minimum development time for the 2005 experiment was 503.7 hours
(minimum: 468.5 hours, maximum: 561 hours) and 466.1 hours (median: 455.5 hours,
minimum: 428 hours, maximum: 572 hours, Figure 8.2) for the 2006 experiment. In the
low temperature treatments, the CA strain (mean: 458.9 hours, median: 455.5 hours,
minimum: 431.5 hours, maximum: 497.5hours) developed, on average, faster than MI
(mean: 463.9 hours, median: 454.5 hours, minimum: 430 hours, maximum: 548 hours)
and WV (mean: 475.5 hours, median: 455.5 hours, minimum: 428 hours, maximum: 572
hours), as the others had some recorded minimum development times above 500 hours
(Figure 8.2—3). Similarly, all rank order comparisons of mean development times among
the strains were maintained between 2005 and 2006 with CA developing fastest at 20°C
and slowest at 33.5°C, while Ml grew fastest at 33.5°C and at an intermediate pace when

raised at 20°C (Figures 8.1—3).

201

Averages of Mlnlmum Development Times at High Temperature

 

360
r

 

 

 

 

 

280
1

 

 

 

 

 

Minimum Development Time (h)
300
r

 

 

 

 

 

 

260
1

240
J

 

 

 

 

I I 1
CA Ml WV

Strain

Figure 8.1. Boxplots of minimum development time comparisons among strains at
33.5°C in 2006. The CA strain developed more slowly compared to the other strains.

202

Averages of Minimum Development Tlmes at Low Temperature

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0
1‘37
0
o—
3%
E
r-fia
E
E
o
is“ ——~—
30 -
Ew—r
3
g“
E
20
(O—
V
o—
3
I r 1
CA Ml WV
Strain

Figure 8.2. Boxplots of minimum development times comparisons among strains at
20°C in 2006. The strains exhibited similar median and fastest development times, but
CA had a faster mean development time as compared to the other strains, which had
some minimum development times that were longer than 540 h.

203

 

Development Times of Different Strains

 

550
1

500

450

Minimum Development (11)
400

 

 

 

 

 

8
0')
O
O
t")
O
in
N
I l l I I I I I
20 22 24 26 28 30 32 34
Temperature (C)
b .
Development Times of Different Strains
O
m —1
in

3

EO\

d)

5%"

O.

.9

Q)

a§—

O

E

.gfifi

2
o-ut
8

250

 

 

 

r I I r l r
20 22 24 26 28 30 32 34

Temperature (C)

Figure 8.3. A comparison of average minimum development times for each strain in 2005 (a) and in
2006 (b). CA is red, MI is blue, WV is green. The M1 mean minimum development time was consistently
faster than WV, though their development times were comparable. The CA mean was fastest at 20°C and
slowest at 33.5°C in both years. Development was faster at 33.5°C.

204

Body size comparisons between 2005 and 2006 were largely similar, though the
2006 experiment yielded larger animals than the 2005 experiment. When raised at
33.5°C, the mean length was 6.9 mm (minimum: 5.5 mm, maximum: 8 mm) in 2005 and
7.2 mm (median: 7 mm, minimum: 6 mm, maximum: 9 mm) in 2006, while mean
weights were 23.12 mg (minimum: 10.66 mg, maximum: 49.14 mg) for 2005 and 27.29
mg (median: 38.3 mg, minimum: 19.18 mg, maximum: 63.03 mg) for 2006. Average
weights at the high temperature were largest for the CA strain (mean: 28.63 mg, median:
29.04 mg, minimum: 13.01 mg, maximum: 42.28 mg), intermediate for M1 (mean: 28.16
mg, median: 27.6 mg, minimum: 14.1 mg, maximum: 40.11 mg), and smallest for the
WV strain (mean: 25.08 mg, median: 24.81 mg, minimum: 16.22 mg, maximum: 37.86
mg) in the 2006 data (Figure 8.4). When L. sericata was raised at 20°C mean length was
7.4 mm (minimum: 5.5 mm, maximum: 10 mm) for 2005 and 7.9 mm (median: 8 mm,
minimum: 6.5 mm, maximum: 9.5 mm) for 2006 and average weights were 29.05 mg in
2005 (minimum: 10.66, maximum: 49.14) and 39.36 mg (median: 38.3 mg, minimum:
19.18 mg, maximum: 63.03 mg) in 2006. The mean weights of low temperature
treatments in 2006 revealed CA to be the largest (mean: 44.8 mg, median: 44.77 mg,
minimum: 30.27 mg, maximum: 63.03 mg), M1 to be intermediate in size (mean: 38.33
mg, median: 37.33 mg, minimum: 26.86 mg, maximum: 49.86 mg), and WV to be the
smallest strain (mean: 34.96 mg, median: 34.66 mg, minimum: 19.18 mg, maximum:
48.25 mg; Figures 8.5). Strain effects on body size followed this trend as well; when

both measures for body size were compared among strains, the CA strain was

205

consistently the largest and the WV strain was consistently the smallest in 2006 (though

the WV average length was longer than MI in 2005) (Figures 8.6—7).

Weight Averages at High Temperature

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

—r—
o l
V_‘ r ‘
I
‘ §
m I
on" L o
__,__
A I
U)
as“
E
.9’
‘1’ t0
3 N“
I
' l
o ' '
Nd : :
' r
' l
: __
e— .
I c—‘—
_A—
I I I
CA Ml WV
Strain

Figure 8.4. Boxplots of average weights compared among the strains at 33.5°C in
2006. On average, CA was larger than MI, which was larger than WV.

206

Weight Averages at Low Temperature

 

60
1

 

 

 

 

 

 

 

 

Weight (mg)
40
1

 

 

 

 

 

 

 

30
1

 

20
1

 

 

 

CA Ml WV

Strain

Figure 8.5. Boxplots of average weights compared among strains at 20°C in 2006.
On average, CA was larger than MI, which was larger than WV.

207

Pupal Lengths of Different Strains

 

Length (mm)
8
//

 

 

 

 

 

(D _-
r 1 1 r r r 1
20 22 24 26 28 30 32 34
Temperature (C)
b .
Pupal Lengths of Different Strains
‘0. _
0')

Length (mm)
7.0 7.5
//

 

 

6.0

 

1 1 1 1 1 1 1
20 22 24 26 28 30 32 34

Temperature (C)

Figure 8.6. A comparison of average lengths for each strain in 2005 (a) and in 2006
(b). CA is red, MI is blue, WV is green. The CA mean length was the largest in both
years. WV was the smallest in 2006, but MI was smallest at 33.5°C in 2005. Lengths
were greater for 20°C treatments.

208

Pupal Weights of Different Strains

 

40
1

Weight (mg)
30
1

 

20

 

 

 

r r r I 1 1
20 22 24 26 28 30 32 34

Temperature (C)

Pupal Weights of Different Strains

 

Weight (mg)
40 50 60
1 1 L

30
1

 

 

I r 1 r 1 1 1
20 22 24 26 28 30 32 34

 

Temperature (C)

Figure 8.7. A comparison of average weights for each strain in 2005 (a) and in 2006
(b). CA is red, MI is blue, WV is green. The CA mean weight was the largest and WV
was the smallest. Weights were greater for 20°C treatments.

209

In both years, ANOVA demonstrated the significance of the observed differences
among strain and between temperatures on body size and minimum development time. A
comparison of pupal sizes and minimum development times from 2005, though based on
only two replicates per temperature per strain, resulted in a statistically significant effect
of strain on length (P=0.034) and a borderline significant effect of temperature on length
(P=0.085). Results were similar when weight was assessed, with a significant effect of
strain on weight (P=0.03) and a borderline significant effect of temperature (P=0.08).
However, when minimum development time was assessed, there were significant strain
(P=0.001) and temperature (P=0.03) effects on the trait, as well as a significant
interaction between temperature and strain (P=0.0071).

The comparisons of the 2006 strains showed statistically significant effects of
strain and temperature on all three traits assessed. Length was significantly affected by
strain (P<0.0001) and temperature (P<0.0001). Likewise, weight was affected by strain
(P=0.0079) and temperature (P=0.0001). Minimum development times, though they
were significantly affected by strain (P=0.049) and temperature (P<0.0001), were not

significantly affected by the interaction between the two variables (P=0.2).

Discussion
Temperature Effects on Body Size.

The observation of high temperatures resulting in smaller body size is consistent
with a pattern that has been recorded for approximately 80% of insect species (Sibly and
Atkinson 1994). These observations present a paradox, since faster growth promotes

larger body sizes when development is accelerated by improved nutrition (Berrigan and

210

Charnov 1994). Many attempts have been made to model development to understand
why insect sizes decrease in higher temperatures, seeking to explain body size as a
function of factors such as larval mortality, limited lifespans, gonad development,
increased molecular metabolism, and spatial/temporal heterogeneity in the environment,
(Berrigan and Charnov 1994, Sibly and Atkinson 1994, Perrin 1995, Atkinson and Sibly
1996). Two factors, explored in these analyses could potentially explain the phenomenon
in L. sericata. First, development is dependent on biochemical metabolism and the
mechanical acquisition of food. If temperature increased the rate of feeding at a slower
rate than the catalysis of food, flies may be smaller due to a proportionally larger increase
in metabolism. This theory has been supported by observations of a slower increase in
fish feeding as compared to metabolic activity in warmer conditions (Perrin 1995), and
modeling research on growth (Perrin 1995, Atkinson and Sibly 1996). In L. sericata,
there is a proportional increase in the rate of growth for high temperature treatments as a
function of development percent (Figure 3.3), indicating that a disproportionately large
increase in molecular metabolism may affect blow fly size. However, higher mortality at
high temperatures has also been proposed as a mechanism for promoting the paradoxical
growth observations associated with higher temperatures (Berrigan and Charnov 1994,
Sibly and Atkinson 1994). Essentially, if larvae are more likely to die in warmer
conditions than in cooler temperatures, due to factors such as heat stress or anoxia, then
there may be selection to complete development quickly at the expense of body size. The
observation of higher mortality at 33.5°C, especially in the smallest strain (WV), supports
this theory. Given these results, it is likely that one or both of the mechanisms is driving

the evolution of small body sizes at high temperatures.

211

 

 

 

The Evolution of Age and Size at Metamorphosis.

The research presented here demonstrates several interesting features that help to
understand the evolution of L. sericata age and size at metamorphosis. First, body size
was affected by both genetic strain and temperature. The CA strain was consistently the
largest, while WV was smallest. Likewise, lower temperatures resulted in larger pupae.
Second, there was no straightforward relationship between body size and development
time. At 20°C, the smallest strain developed for the longest period of time and the largest
matured, on average, in the least amount of time. Fast development times of larger
strains indicate the presence of selection on developmental thresholds as proposed by
Day and Rowe (2001). However, at 33.5°C, MI developed fastest, while CA took the
longest to mature. The change in rank order of minimum development times, which
occurred in the same pattern for both years, suggests that the nature of selection on the
threshold, or thresholds, influencing development time/body size in L. sericata is
different among strains. The selection for fast development times in one temperature and
not the other is evidence that the ecosystems in which the studied strains live may be
pushing them toward different development times. Variable selection may also be
pushing the strains toward alternative optimal temperatures for body size, as seen in P.
rapae (Kingsolver 2000, Kingsolver 2001). Differential selective pressure is not hard to
imagine given the dissimilarities between the ecoregions that each strain originated from
(Figure 1.3).

Many ecological factors can potentially influence fly development, but one,
latitude, is known to affect body size and development time in D. melanogaster (Oudman

et al. 1991, Hallas et al. 2002) and Rhagoletis (F eder et al. 2003) respectively. The

212

effects of latitude differences on size and age at metamorphosis in fly species indicate
that it may be possible to determine a series of guiding principles for predicting minimum
development time/body size based on the ecological factors (including latitude).
However, such effects are likely to be specific to blow flies. For example, body size is
known to increase with latitude in Drosophila (e. g. Hallas et a1. 2002), but the largest
(CA) and smallest (WV) strains originated from similar latitudes, indicating a different
mode of selection for body size in L. sericata. Accordingly, comparisons of Lucilia
populations to Drosophila populations will likely lead to a fuller understanding of how
the environment and genotype influence Dipteran phenotypes.

Year-to-Year Variation.

One interesting feature of the research presented here was the general similarity
between the 2005 and 2006 studies. CA was consistently the largest and, with the
exception of length comparisons in 2005, WV was consistently the smallest strain.
Likewise, the rank order of mean development times was the same in both years.

However, despite these comparative continuities, there were differences in
average minimum development times and sizes among strains collected in different years.
There are several potential explanations for this observation. The easiest explanation is
that the small sample size (2 replicates per temperature per strain) in the 2005 experiment
was randomly biased as compared to the 2006 experiment, which had 10 replicates per
temperature per strain. Another possibility is that the destructive sampling in the 2005
study contributed to slower pupal development. This idea is supported by the observation
that destructive sampling delays the appearance of adults (Tarone and Foran 2006).

However, the fact that the 33.5°C treatments were more similar to each other than the

213

20°C treatments and that the 2005 study took sampling into account (Chapter 3), indicates
that destructive sampling did not substantially contribute to the slower 20°C development
observed in the 2005 study. Alternatively, the differences could be due to year-to-year
genetic variation within each population. This is supported by the observation of short-
terrn and long-term changes in chromosomal inversion frequencies in a population of
Drosophila subobscura (Rodriguez-Trelles et al. 1996), indicating the possibility that the
differences observed herein were the result of a temporal shift in local development times
and body sizes. However, the three strains were collected in different US states, but
exhibited similar changes of approximately 50 hours in mean development times at 20°C
(Figure 8.3) between 2005 and 2006. If temporal variation explains the differences
among strains, it seems unlikely that three different strains experienced similar shifts in
development times unless a major climatic effect influenced each.

Forensic Entomology.

The research presented here demonstrated several important features of L.
sericata developmental variation, which are useful for understanding blow fly body size
and maturation times in the context of forensic entomology. First, significant effects of
both temperature and genetic strain on the traits were observed, which have clear
ramifications for their use in the forensic sciences. The subtle differences among the
three strains and two temperatures for body size were previously shown to have little
effect on predictions of minimum development percents (Chapters 3 and 5), but the best
estimates of age, when using body size measures, may result from populations with
known average body sizes. More importantly, it is clear that the predictions of minimum

development percent will be more accurate if they are calibrated to the strain of interest.

214

One can never know the development times of all populations in every year, but it may
be possible to understand the rules governing the evolution of this trait. In determining
those rules it will be possible to decrease errors in age predictions resulting from
populations that differ from laboratory-based tables of development times.

The variation in minimum development times among strains may be seen as
detracting from the significance of the improved precision in age predictions garnered
through gene expression analyses (Chapter 5, Chapter 6), but these are, in reality, two
different problems with the PMI prediction process. The first problem in predicting a
PMI with blow fly evidence is an issue of precision rooted in developmental biology and
stems from the need to understand how mature a blow fly happens to be. For example, if
a blow fly develops in 500 hours, an investigator would prefer that an estimate of 70%
through development be accompanied by a window of 325—3 75 hours as opposed to 275—
425 hours. If accurate, a shorter window of time around a PMI estimate allows the
investigator to rule out more potential suspects and potentially focus on the correct
suspect in a death investigation. The precision of PMI estimates with blow fly evidence
has been improved with recent work that utilizes gene expression data to predict age
(Chapter 5, Chapter 6) and this problem in estimating blow fly age is likely to decrease as
molecular and statistical tools develop.

However, the second problem with predicting a PMI with blow fly evidence is an
issue of accuracy that is rooted in population/quantitative genetics and stems from the
fact that phenotypes can vary among blow fly strains, which can lead to an inaccurate
PMI estimate if the development of a specific population is not in accordance with the

data used to make an age prediction. As an example, the ability to precisely predict a

215

development percent of 70% does not mean that the predicted development percent is the
same for two different populations. At a given temperature, if one population develops in
500 hours, while another develops in 470 hours, an estimated development percent of
70% would correspond to an age of 350 hours and 329 hours (respectively), potentially
resulting in an inaccurate estimate of age. The occurrence of strain effects on minimum
development time in this research indicates a need for population-specific data to avoid
inaccuracy in PMI estimates.

Population-specific predictions of age may be aided by evolutionary ecology
theory, which seeks to explain phenotypic variation in terms of ecological effects on
those traits (reviewed in Conner and Hartl 2004). By studying a series of populations, the
guiding principles behind the evolution of development times in L. sericata (and other
blow flies) can be determined, allowing for more accurate PMI predictions. Such rules
can be used to calibrate estimates with published data to the population studied. One
possible means of developing a strain-specific PMI would be to combine predicted
development times with published data, which would be used as a prior in Bayesian
analyses (Scheiner and Gurevitch 2001), thereby developing a posterior blow fly age
estimate tailored to the population of interest. By determining the factors that affect blow
fly development times, more accurate PMI estimates can be determined with

entomological data.

216

CHAPTER 9: DISCUSSION

Improved Predictions of Blow Fly Age
Introduction.

The goal of the research detailed herein was to improve PMI predictions that are
based on blow fly evidence. There are several challenges facing traditional forensic
entomological PMI predictions that can be overcome through the application of
knowledge gained from the results of this dissertation. First, both the accuracy and
precision of blow fly age estimates can be increased by employing molecular and
quantitative genetic methods. Any research that contributes to a firller understanding of
factors that affect development times in L. sericata also reveals information that helps to
make more accurate age predictions, as an entomologist will be better equipped to
estimate the minimum development time of a blow fly in a variety of conditions and for
different populations. Similarly, research that can help to decrease the window of time
provided with a PMI prediction will increase precision. Second, two parts of the Daubert
requirements for scientific evidence, standard operating procedures and a statistical
understanding of methods, have not been fully considered in forensic entomology. Both
of these requirements were addressed by the research conducted herein. The research
evaluating environmental and genetic effects on variation in blow fly development
demonstrated a need for standard operating procedures and will be useful in developing
them, as the results presented can be used to design standard procedures for raising blow
flies in the laboratory. Likewise, the use of GAMs enabled a statistical understanding of

age estimates made with blow fly data. Finally, the methods employed in the dissertation

217

provide logistical improvements over other forensic entomology approaches that make
their application appealing. The widespread use of nucleic acid technologies in forensic
science laboratories means that most can employ the molecular techniques used for aging
blow flies and since they have the potential to increase laboratory productivity, they will
be viewed by investigators as an attractive option to pursue.

Improving Accuracy and Precision in Forensic Entomology.

The results of the research presented here show that it is possible to achieve more
accurate and precise PMI estimates with entomological data. The use of gene expression
levels and GAMs to predict blow fly age led to a clear improvement in GCV and PDE
statistics, particularly for postfeeding third instars and pupae. The statistics provide
information that can be used to compare models predicting a response of interest (in this
case, minimum development percent of L. sericata) and show that blow fly age estimates
that include gene expression analyses are more precise than those that do not. The use of
GAMs also enables a graphical assessment of the increase in precision garnered by gene
expression data, as seen by a comparison of the fitted vs. response plots (Chapter 5). The
accuracy of PMI predictions was also aided by molecular analyses. In assessing gene
expression differences among feeding, postfeeding, and “Peter Pan” third instar larvae,
the ability to distinguish amongst them was enhanced, thereby decreasing errors in stage
identification that would result in inaccurate predictions.

The quantitative genetic assessment of blow fly development will also lead to
increases in the accuracy of PMI predictions. The research investigating rearing
conditions that affect development time (Chapter 2) and the effects of genotype and the

environment on development time and body size (Chapters 3 and 8) produced results that

218

will aid in correctly identifying the underlying factors that contribute to ambiguities
between fly growth seen in practice and growth as it appears in the literature. Any
information. on factors (both genetic and environmental) that influence the two
phenotypes of greatest interest to a forensic entomologist, body size and development
time, will help investigators produce better theories on the evolution of blow fly
development that can be used to achieve more accurate PMI estimates. Likewise, the
ability to recognize environmental conditions that influence development may help
investigators in determining the most relevant developmental data for a case.
Addressing the Daubert Standards in Forensic Entomology.

The results of the research presented here also helped address two of the
requirements of the Daubert ruling, standard operating procedures and an understanding
of error. The reasons for, and results of, the research in Chapter 2 demonstrate the need
for standard operating procedures in rearing protocols used to produce development
tables for forensically informative blow flies. Clearly, there is plasticity in fly
development, as it was possible to achieve a large range of replicable development rates
depending on the rearing treatments used to raise L. sericata. The comparison of
development of flies on rats to flies raised on liver indicates that only a small subset of
liver-based rearing protocols, in particular ones that keep the food source moist, produce
growth that is comparable to growth under forensically relevant conditions (growth on a
body). This demonstrates that the a standard operating procedure for the rearing of flies
in the laboratory should produce growth rates that mimic the development of those flies

under the most forensically relevant conditions—on carrion. Likewise, as more

219

quantitative genetic influences on blow fly development are revealed, they can be
incorporated into the protocols for predicting a PMI with entomological data.

The application of molecular biology techniques has also made the analysis of
blow fly data more amenable to standard operating procedures. The molecular biology '
community has detailed guidelines for the methods used herein. When DNA
identifications of individuals were a new tool in the forensic sciences, a National
Research Council report was compiled by a group of experts in the fields relevant to the
new technique and its application. The report outlined a series of reccomendations that
resulted in a set of standards for forensic DNA identifications (NRC 11). Consequently,
DNA analysis is considered the “gold standard” in forensic science (Saks and Koehler
2004). The meticulous requirements of the community may also lead to a similar benefit
for forensic entomology. Accordingly, the wealth of information on, and recommended
uses of, gene expression technologies, including quantitative PCR (e.g. Mocellin et a1.
2003, Bustin et al. 2005, and Puskas et al. 2006) and microarrays (e. g. Puskas et al.
2006), can be used to develop a set of guidelines in forensic entomology. In this manner,
input from the scientific community, which contributed to the optimal use of DNA
evidence, can be used to guide the development of molecular protocols in forensic
entomological predictions of blow fly age.

The statistical requirements of Daubert have also been more fully addressed in
this dissertation. Using GAMs, it was possible to graphically demonstrate expected error
through fitted vs. response plots as well as statistically demonstrate expected error with
the PDE score. The use of GCV scores can also help investigators determine the most

likely model to use/trust for a PMI estimate. Finally, the validation of GAM predictions

220

of blow fly age with traditional forensic entomological data and gene expression data has
ensured that the technique may, one day, be used in court as its ability to reliably predict
age has been characterized in blind studies. These statistical results are particularly
important, since research of this type is rare in forensic entomology.

Logistical Improvements Associated with Molecular Predictions of Blow Fly Age.

Gene expression analyses provide some logistical improvements that make PMI
predictions based on such data appealing. First, the molecular methods studied herein
can be conducted by more scientists, as there are a greater number of trained molecular
biologists than trained forensic entomologists. Crime laboratories generally seek the
advice of the few outside entomologists in the country, while the molecular methods used
in the research presented here can be easily applied in most molecular biology
laboratories and should expand the accessibility of forensic entomology to more cases by
enabling any laboratories equipped to do DNA research to analyze blow fly samples.
Though it is unlikely that such analyses will be employed within crime laboratories, as
they do not encounter insect evidence frequently enough to justify in-house analyses, the
number of potential laboratories that can provide investigators with data for predicting
blow fly age will increase if molecular aging techniques are used.

Second, molecular biology research is amenable both to the use of robotics and
miniaturization on microarrays, which allow thousands of genes to be analyzed
simultaneously. These features should enable more rapid casework and help to decrease
the costs of applying gene expression analyses for estimating a PMI. During the research
presented here, the production of a gene expression profile cost approximately $15 per

individual larva or pupa. With roughly 50% of samples providing full expression

221

profiles, the analysis of 100 individuals would cost approximately $1500 and yield 50 full
profiles, which is sufficient to provide a statistically valid sample size. Given the large
expenses involved a homicide case, a $1500 price tag is not prohibitive and with
miniaturization the cost will likely decrease while increasing the number of loci assessed
(Dahl et al. 2007).

Finally, the use of GAMs and gene expression data can help investigators deal
with the types of evidence they receive. Forensic scientists must often work with non-
ideal evidence. Sample sizes can be small, requiring a limited analysis so evidence can
be reevaluated at a later date. Any technique that accommodates this need will be
preferred by investigators. Also, some entomological samples may be damaged in
storage, with resultant weight, length, or stage determinations impossible to assess.
Through the use of gene expression analyses, samples do not need to be completely used,
allowing for future analyses. Likewise, if certain data (e.g. weight measurements) are
unavailable, a GAM can be developed that allows for a statistical understanding of the
model specific to the available data. This will enable investigators and triers of fact to
evaluate how much weight can be assigned to a PMI prediction based on entomological
evidence.

Future Research in Forensic Entomology.

A solid foundation has been established for future studies in forensic entomology.
One of the most obvious directions to proceed is to profile gene expression levels in the
many other forensically informative fly species (e.g. Hall 1948, Kamal 1958, Catts and
Haskell 1990, Anderson 2000, Greenberg and Kunich 2002). Comparisons of gene

regulation among cyclorrhaphan flies indicate that gene expression is likely to be similar

222

among species (6. g. determination of the embryonic anterior axis by bcd; McGregor
2005), but continuity in gene expression profiles across species is not guaranteed (e.g.
differential timing in the expression of proneural genes between the Calliphora and,
Phormia genera; Skaer 2002). Each species will require profiling, just as development
tables are needed for each (Catts and Haskell 1990). As gene expression research in blow
flies develops, a series of loci should be settled on for all species. The identification of
individuals with DNA evidence is based on standard loci (Chakraborty et al. 1999),
which enables a detailed understanding of the error resulting from these well-studied l
short tandem repeats. This example should be followed in forensic entomology to
achieve the same reliability and standardization that has been reached with DNA
identifications. Accordingly, the best loci for estimating blow fly age should vary
throughout the development of most or all of the common forensically informative blow
fly species and the number of genes necessary for assessing blow fly age should be
determined by the precision that can be gained by adding loci to an age estimate.
Additionally, the 9 genes studied herein are not necessarily the best for predicting
a PMI. The genes studied in this research were chosen from a short list of available
Lucilia sequences, based on a priori information indicating that their expression levels
would change during development, thus enabling improved PMI estimates. However,
there are thousands of genes in cyclorrhaphan genomes (Yeates and Wiegmann 2005)
and, as demonstrated by genomic work in Drosophila (Arbeitman et al. 2002, Beckstead
et al. 2005), thousands of unique gene expression profiles. Clearly, many more loci could
be assessed to determine an optimal set for predicting fly age. The most important reason

for adding genes to the PMI prediction process can be seen in the persistence of error in

223

pupal estimates (Chapter 5), wherein the PDE scores for gene expression based estimates
was 78%, which, while a vast improvement over the 16% score associated with estimates
that did not use gene expression, can be improved by the addition of more loci that are
regulated during pupation. Candidate genes may be found by searching Drosophila and
other dipteran gene expression data in the literature. Genes that exhibit changes in
regulation at any point during development are all a priori candidates for research. The
Amalgam and CG17814 loci studied by Arbeitman et al. (2002) are two examples of
expression changes in genes that may increase precision in pupal age estimates. As more
genes are included, arrays of gene sequences (such as the ones used in Arbeitman et al.
2002) could be used to determine age with hundreds or thousands of gene expression
levels, which will enable more precise estimates of age.

There were also temperature effects on gene expression that indicate transcript
level analysis may be a useful means of identifying blow fly development in different
environments. Doing so could potentially enable investigators in determining molecular
markers for conditions (like temperature differences) that contribute to plasticity in blow
fly development, which will help them avoid errors in entomologically based PMI
estimates. The ace gene demonstrated this point best, as it was expressed at a high level
at low temperatures, while high temperatures caused down-regulation of the gene during
the third instar. Consequently, ace expression levels (and hsp90, which behaved
similarly) should help confirm or refute that evidentiary flies developed at the
temperature assumed by investigators. Any gene expression changes that are indicators
of environmental effects that could influence developmental plasticity (e. g. starvation,

disease, diapause) have the potential to be useful in identifying when flies were exposed

224

to those conditions, allowing for more accurate PMI estimates. The research on “Peter
Pan” larvae shows that arrested development in third instars can be identified by
expression differences in the hsp90, sll, and usp genes. Identification of misleading
individuals will help to prevent errors associated with their inclusion in a PMI estimate.
In fact, all of the states of the third instar (feeding, postfeeding, non-pupator) are
distinguishable by assessing gene expression differences, which is useful for improving
PMI estimates with third instar larvae as it can be difficult to distinguish among them
with physical evidence alone (Anderson 2000, Chapter 5, Chapter 7). However, to avoid
circular logic in the PMI prediction framework, any gene expression level employed in
the identification of a developmental stage should not be included in subsequent
development percent predictions. Stage exerts the largest influence on predictions
(Chapter 3), so a gene used to assign stage would have a disproportionately large
influence on a PMI estimate if it were also included in the within-stage prediction.
Accordingly, an expression profile used to recognize third instar states should be
independent of the set of genes predicting age, necessitating research to determine the
best loci for third instar identifications versus loci most useful for development percent
predictions.

Further work to improve the statistical understanding of PMI estimates made with
gene expression data is also necessary. There are other statistical options for gene
expression analyses besides GAMs that are also capable of predicting age with
multivariate, non-linear data, which may be valid techniques or even exhibit superior
performances in predicting blow fly age. Cluster analyses (Eisen et al. 1998) and

classification and regression trees (Brieman et al. 1984) are common statistical tools that

225

allow for a hierarchical organization of complex data that can indicate very specific
subsets within a data set by splitting it into iteratively smaller subsets. The two
Drosophila studies on genome-wide expression that were used as theoretical support for
this experiment relied on cluster analyses (Arbeitman et al. 2002, Beckstead et al. 2005),
indicating that these statistics are useful for assessing gene expression data. Neural
networks can also make predictions with complex data (V enables and Ripley 2002,
Ripley et al. 2004). These statistics calculate probability in a manner that is similar to
biological neural functions, by passing input data through nodes, which then approximate
output data. The ability of both methods to make predictions with complex data sets
indicates that neural networks and cluster analyses may be improvements over the GAM
statistic used herein. However, descriptions of neural networks are likely to be overly
complex for a jury to understand and the statistical support (such as model comparisons)
for cluster analyses and neural nets are not as straightforward as they are in GAMs as
they rely on thousands of randomizations of the data. It should also be noted that both
approaches are capable of making predictions that are too specific to the data set used to
design them and must be extensively trained against new data to avoid that problem
(Brieman et al. 1984, Ripley 2004). Additionally, Bayesian analyses may be useful for
developing population-specific predictions of minimum development times (Scheiner and
Gurevitch 2001). These methods merge previously recorded data, or “priors”, with data
that are specific to the population of interest to make posterior predictions. In the context
of forensic entomology, posterior predictions may be superior estimators of PMI, as
published data for development times and gene expression levels could be combined with

a small data set from a population of interest (such as the strain-specific development

226

times in Chapter 8), providing more accurate estimates of age. All of these statistical
methods remain unexplored by the forensic entomology community and warrant further
consideration.

Finally, the genetic and environmental variation in body size and development
time among L. sericata strains should be more extensively characterized. Though the
effects of temperature and strain on these traits were statistically significant, there were
only three population comparisons made. This low level of replication makes it difficult
to interpret the roles that local adaptation due to ecological differences played in the
evolution of these traits among the strains studied. The next step in understanding the
major underlying factors that influence body size and development time evolution in L.
sericata and other blow flies is to study more populations and potential ecological
influences that could be driving the observed differences among traits. Any ecological
components that differ among the environments where a blow fly species is found could
be the subjects of future studies, though efforts should be directed at identifying factors
that explain most variation in development traits. The goal in determining ecological
factors that affect blow fly development is to articulate a unifying theory for how the
traits evolve among populations that could be used to predict development times/body
sizes of unknown populations. An improved theoretical understanding of these traits
could enable the use of correction factors and/or Bayesian analyses to achieve more

accurate and population-specific predictions of age.

227

Controlling Blow Fly Populations with Gene Expression and Plasticity Data

The results of this dissertation point to several means of improving the control of
blow fly pests. The gene expression analyses conducted herein (Chapter 5) identified
expression level differences among stages of L. sericata development in four loci that
encode insecticide targets (cs, ace, ecr, usp), as well as variation in the expression of a
gene that can decrease the toxicity of organophosphates (rop-I). Such information can be
used to determine stages that are likely to be more susceptible to specific insecticides,
which may allow for the most efficient use of chemicals to kill specific blow fly
developmental stages.

Additionally, the research provided ample information regarding environmental
effects on development time (Chapters 2, 7, 8), body size (Chapters 3, 8), and gene
expression levels (Chapter 5) throughout the development of L. sericata. Low
temperatures were shown to increase ace expression in larvae (Chapter 5), indicating that
temperature differences could potentially affect organophosphate resistance. Likewise,
body size and/or development times varied in response to food moisture (Chapter 2),
fresh pupation substrate (Chapter 2), and temperature (Chapters 3, 8). Finally, arrested
development in L. sericata postfeeding third instars was reliably associated with
differences in the expression levels of hsp90, sll, and usp (Chapter 7), which may provide
clues as to how to environmentally or chemically induce a delay in development. Any
means of interrupting blow fly development or decreasing the size of blow fly pests
(thereby preventing or decreasing the damage caused during myiasis) may be useful in

developing management plans to ameliorate the damages caused by blow fly pests.

228

Conclusions

Several conclusions can be drawn from the research presented here. First, gene
expression analyses improve the accuracy and precision of PMI estimates, especially in
postfeeding third instars and pupae, which currently provide the least precise age
estimates with traditional approaches. Transcript abundances were useful in
distinguishing amongst feeding, postfeeding, and “Peter Pan” third instar larvae, helping
to avoid errors derived from misidentified third instar states. More importantly, the
developmental profiles produced enabled significantly superior statistical descriptions of
development, allowing for a mathematically defined understanding of error associated
with age estimates.

Second, blow fly size, development time, and gene expression levels were shown
through ANOVA and GAM to be affected by both genotype and the enviromnent.
Rearing conditions influenced body sizes and development times. The CA, MI, and WV
strains deve10ped at different rates and had different average sizes. Finally, gene
expression levels significantly varied according to genetic strain and/or temperature
treatments.

Third, studying blow fly developmental variation can contribute to the
understanding of the evolution of age and size at metamorphosis. Such knowledge will
help to provide a framework for developing standard operating procedures for the rearing
of flies in the laboratory, addressing a key requirement of the Daubert standard for
scientific evidence. Additionally, information on strain-specific development can be used

to achieve more accurate PMI estimates by tailoring them to local populations.

229

Fourth, GAMs are useful for predicting blow fly age. They were used herein for
estimating age with gene expression data, as well as with more traditional developmental
stage and body size measures. Likewise, with this statistical approach, it was possible to
demonstrate that, though significant, the effects of strain and temperature on L. sericata
body size may be ignored with minimal effects on the precision of development percent
predictions. The ability to introduce and remove variables from these models, and
understand the effects on resultant predictions, will also be of value to investigators both
in allowing them to work with all available data and enabling them to more fully address
the statistical requirements of the Daubert standards for scientific evidence.

Last, the study of L. sericata development can improve scientific understanding in
multiple fields of biological research. The data presented have provided preliminary
information on potential influences of the environment on the evolution of L. sericata
body size and development time traits. The molecular data provided a picture of the
effects of strain and temperature associated with gene expression profiles that may aid
future research in comparative genomics and information on the distributions associated
with profiles of developmentally regulated genes. In addition, data on plasticity and the
expression of specific genes can contribute to future research on the control of insect
pests. Finally, and most importantly, the accuracy and precision of PMI estimates made
with blow fly evidence was significantly improved, as was the ability to address the

Daubert standards for scientific evidence.

230

APPENDIX

231

 

92 oz 0588; oz 02 we, 9% 2
mm oz a=§§§ 02 we» oz as: _ a
mm oz amaesfi> oz oz 02 as; 2

3m oz 2:865» 02 02 8> as; a
8 oz a==2EO> 02 oz a; @352 w
2 oz s=§§o> 02 02 oz @352 a
em 02 228E5> oz war 02 92m 0
8 oz e=3§o> we» oz oz Q2» m
em 02 058:8; 02 oz oz 9% 4
om 02 3:38; 02 02 a; as; M

3m oz a=§§o> oz oz we, 9&2 N

2m 02 s=§§o> 02 oz 02 9&2 _

wwm $38. Home; Batmnsm $ch owmbmnsm command. 0385.539 “no: fiasco

63.6.23 emsfiau he 3.5.—8 =n Se...“ mount, 3:222: 6.3 9.933: 9.5-«En: 95 he 552:6 was 35:53:. Ad mmcaeaa<

232

 

mm 3% BSSzoS83> 07S 02 oz SSE cm
em 02 Sufism oZ 3% oZ SSE m N
Sam 02 Siam 02 oz 02 SSE vm
mm 02 98m 3% 07S 02 SSE mm
o S 02 Siam 3% oz 02 SSE mm
em 3% SSS—em OZ oz 02 SSE S N
am 3% 98m 02 3% oZ SSE om
m .w S 3% 98m 02 3% oZ SSE a S
m .w S 3% Socwm OZ oz 02 SSE a S
m S oZ Secem oZ 3% oZ SSE n S
m S 02 28m 07S 02 oZ SSE S
S N 02 BSSsoSgo> 3% oz 02 SSE m S
m .w S 02 eSSSsoS83> oZ 3% oZ SSE VS
Wm S 02 BSSsoSELo> 02 oz 02 SSE m S
mum S338. 393 mumbmnsm 395 oabmnzm 3%ch 3>S§E3S S32 togou

233

 

m .w S 02 Siam oZ 3% 02 EMS ow
on 02 28m 02 3% oZ SSS mm

m .om oZ Squaw 07S 3% 02 3m w m
m Sum 3% BSSSQSEBSV 02 oz 02 SSE n m
m .vm 3% BSSSUSEB> 02 07S 02 SSE om
m .3 3% oSSS30S§o> oZ 3% 07S SSE m m
m .vm 3% BSSsoSSEo> oZ 3% oZ SSE Sum
m . S m 3% Siam oz 02 07S SSE m m
mm 02 Siam oz 02 OZ SSE mm
mm. S N 02 8:322:3/ oz 07S 02 SSE S m
m . S m 3% Siam oZ 3% oZ SSE om
mm oZ SSS—mm oZ 3% oZ SSE mm

m n. S m 02 oSSS30S§o> oZ 3% o7S SSE mm
mm 3% oSSSsoSSEo> oZ 3% o7S SSE hm
wmm S330 {S1 33E uSmbmnsm 395 oSmbmnsm Hammad; o>SSosb3S SmoSS tonoo

234

 

deS 3mm: vmm 353 De m5: mnsm
2.: NSv NoS Sam Sm mndm mmdm
No.2 omv an wSS Sm mndm 3.3
3.2 mmwov owS m. S mm mmmm Dem
093 m.mom mwm on on em mém
m0: mmomv m. SoS 3.02 W; W; cm
3.2 m~.mmm mmdf we we mm mm
3.: ace 0an mmdo ow mmmm mdm
56S mméwm flung mo me mm mm
mndm 03v vwm Ova mmfiv mnSN om
anS 0mg vmm 3.3 we mmdm new
35S nmv ONE m.ooS 0% Se cm
36S Sfioh $503 Sfioh «mam ucSSeoobmonS @533“: .855 Em .835 SEN 355 ES

235

 

 

ww.mS mmm mNNoS m5: m.mm mam em
wwéS 5mm mag we 3.: Sum mmsm
3.3 5mm DwoS 3 mm?» Sum mmfim
wstS 5mm flumoS £139 3 mm mm
S53 Sow wwS 3.3 m5? mném mnfim
00.2 mmm WNOS mmdn mSm mm 3.8
8.2 mmm 9N2 wads mSN mm mndm
Sw.mS mSmm mwS we mm m.mm mdm
Sw.mS mSmm voS ow mm Wmm mdm
wo.mS mmdhm SS m5: we m5mm mndm
no.2 ohm moS Wmo we mmmm mndm
8.2 mum ooS ma mm WNN Wmm
SWMS amm moS mime mNNm Wmm mm
SSS msmm ovS mmdm mmdm mam mm
m>mS SSoS. 3303 S809 «SE aSSeooetmonS 3533“: 835 Em 335 SEN .865 ES

236

 

 

ww.mS mmm m.wwS mdv mm m.NN mm
QMS m.mmm 8S no em Wmm Sm
3.: 5mm Wm: 33 em mind mm
2.3 m.wmm WSS was Sum mm mm
3.: Wmmm woS Wmm m.mm mm mm
EMS Smm owS mdv Sum mm mm
3.2 Smm mnwwS mmdm Wmm mm mm
56S mmcnmm WNSN mdh mndm mndm mmém
0S vwm mméoS mmdo mndm mm.wm WMN
3.: 0mm 903 3:8 mmdm Strum Sum
So.mS mmdcm m.wwS Won mnfim mndN mmém
2.3 cam 3.th m5: mndm mm.wm mam
3.2 memm mm.moS Se. mmdm mN.wN vm
ww.mS mmm mmNoS m5: mam mdm Sum
m>8S 88S. 330: S881 835 ucSSeooSSmoAS @533”: .835 Sam .855 SEN 83.5 ES

237

 

.S.N 335. 5 3 38 $355 .8813 88508 35:8 S58H 8S 8:0: 5mm 8 mmm :85 E8 SumIS mtonoov 38:8 S5
-35 SS 55: Wmom 8 mmm :85 3mg 3:5 5383335 EDESESS .885 58 5:0: 5 S3582 5 553 .355 53252335 S98
35 E83 95: 383 S333 35 8V Enos 5 58.598 38 3:5 SS< .3033me 50:8 5.3 85 335.5, .3 3285288 35 38

S33: cmS< 686.23 3.5.835 So 5853 SS 385 53282335 SSS 52555 35 E8 33% :03 .So 355 5353335 5:555 of.

238

LITERATURE CITED

Ali RA, Mellenthin K, Fahmy K, Da Rocha S, and Baumgartner S. 2005. Structural
conservation of the salivary gland-specific slalom gene in the blowfly Lucilia sericata.
Dev Genes Evol. 215:537-43.

Ames C and Turner B. 2003. Low temperature episodes in development of blow flies:
implications for postmortem interval estimation. Med Vet Entomol. 172178—86.

Anderson GS. 2000. Minimum and maximum development rates of some forensically
important Calliphoridae (Diptera). J Forensic Sci. 45:824—32.

Arbeitman MN, Furlong EEM, Imam F, Johnson E, Null BH, Baker BS, Krasnow
MA, Scott MP, Davis RW, and White KP. 2002. Gene expression during the lifecycle
of Drosophila melanogaster. Science. 297:2270—5.

Atkinson D and Sibly RM. 1996. On the solutions to a major life-history puzzle. Oikos.
77:359—65.

Beckstead RB, Lam G, and Thummel CS. 2005. The genomic response to 20-
hydroxyecdysone at the onset of Drosophila metamorphosis. Genome Biology. R99.

Berrigan D and Charnov EL. 1994. Reaction norms for age and size at maturity in
response to temperature: a puzzle for life historians. Oikos. 70:474—8.

Bialecki M, Shilton A, Fichtenberg C, Segraves WA, and Thummel CS. 2002. Loss
of the ecdysteroid-inducible E75A orphan nuclear receptor uncouples molting from
metamorphosis in Drosophila. Dev Cell. 32209—20.

Blanckenhorn W. 2002. The consistency of quantitative genetic estimates in field and
laboratory in the yellow dung fly. Genetica. 1 14:171—82.

Brieman L, Friedman J, Stone CJ, and Olshen RA. 1984. Classification and
regression trees. CRC Press LLC: Boca Raton, FL.

239

Bustin SA, Benes V, Nolan T, and Pfaffl MW. 2005. Quantitative real-time RT-PCR--a
perspective. J Mol Endocrinol. 34:597—601.

Byrd JH and Allen JC. 2001. The development of the black blow fly, Phormia regina
(Meigen). Forensic Sci Int. 120:79—88.

Catts EP and Haskell NH. 1990. Entomology and Death: A Procedural Guide. Joyce’s
Print Shop, Inc.: Clemson, SC.

Chakraborty R, Stivers DN, Su B, Zhong Y, and Budowle B. 1999. The utility of
short tandem repeat loci beyond human identification: implications for development of
new DNA typing systems. Electrophoresis. 20:1682—96.

Chaudhury MF, Welch JB, and Alvarez AL. 2002. Responses of fertile and sterile
screwworrn (Diptera: Calliphoridae) flies to bovine blood inoculated with bacteria
originating from screwworm-infested animal wounds. J Med Entomol. 39:130—4.

Chen Z, Newcomb R, Forbes E, McKenzie J, and Batterham P. 2001. The
acetylcholinesterase gene and organophosphorus resistance in the Australian sheep
blowfly, Lucilia cuprina. Insect Biochem Mol Biol. 31:805—16.

Clark K, Evans L and Wall R. 2006. Growth rates of the blowfly, Lucilia sericata, on
different body tissues. For Sci Int. 1562145—9.

Collins FS, Patrinos A, Jordan E, Chakravarti A, and Gesteland R. 1998. New Goals
for the US. Human Genome Project: 1998-2003. Science. 5389:682—9.

Conner JK and Hartl DL. 2004. A primer of ecological genetics. Sinauer Associates,
Inc.: Sunderland, Massachusetts.

Cook PE, Hugo LE, Iturbe-Ormaetxe I, Williams CR, Chenoweth SF, Ritchie SA,
Ryan PA, Kay BH, Blows MW, and O'Neill SL. 2006. The use of transcriptional
profiles to predict adult mosquito age under field conditions. PNAS. 103:18060-5.

Crystal MM and Ramirez R. 1975. Screwworm flies for sterile-male release: laboratory
tests of the quality of candidate strains. J Med Ent. 12:418-422.

240

Dahl A, Sultan M, Jung A, Schwartz R, Lange M, Steinwand M, Livak KJ, Lehrach
H, and Nyarsik L. 2007. Quantitative PCR based expression analysis on a nanoliter
scale using polymer nano-well chips. Biomed Microdevices. [Epub ahead of print].

Day T and Rowe L. 2002. Developmental Thresholds and the Evolution of Reaction
Norms for Age and Size at Life-History Transitions. Am Nat. 159:338—50.

East [J and Eisemann CH. 1993. Vaccination against Lucilia cuprina: the causative
agent of sheep blowfly strike. lmmunol Cell Biol. 71 :453—62.

Eisen MB, Spellman PT, Brown PO, and Botstein D. 1998. Cluster analysis and
display of genome-wide expression patterns. PNAS. 95: 14863—8.

F aulde M, Sobe D, Burghardt H, and Wermter R. 2001. Hospital infestation by the
cluster fly, Pollenia rudis sensu stricto Fabricius 1794 (Diptera: Calliphoridae), and its
possible role in transmission of bacterial pathogens in Germany. Int J Hyg Environ

Health. 203: 201-204.

Feder JL, Berlocher SH, Roethele JB, Dambroski H, Smith JJ, Perry WL,
Gavrilovic V, Filchak KE, Rull J, and Aluja M. 2003. Allopatric genetic origins for
sympatric host-plant shifts and race formation in Rhagoletis. PNAS. 100210314—9.

Flatt T, Tu MP, and Tatar M. 2005. Hormonal pleiotropy and the juvenile hormone
regulation of Drosophila development and life history. Bioessays. 27:999—1010.

Gibson G and Weir B. 2005. The quantitative genetics of transcription. Trends Genet.
21 :616—23.

Goff ML. 2007. Insects and time since death: What do we really estimate? American
Academy of Forensic Sciences Annual Meeting. San Antonio, Texas.

Gomez-Mestre I and Buchholz DR. 2006. Developmental plasticity mirrors differences
among taxa in spadefoot toads linking plasticity and diversity. PNAS 103:19021—6.

241

Goodbrod JR and Goff ML. 1990. Effects of larval population density on rates of
development and interactions between two species of Chrysomya (Diptera:
Calliphoridae) in laboratory culture. J Med Entomol. 272339—43.

Gorham JR. 1987. Insect ad mite pests in food: an illustrated key. US Department of
Agriculture, Agriculture Handbook Number 655.

Grassberger M. and Reiter C. 2001. Effect of temperature on Lucilia sericata (Diptera:
Calliphoridae) development with special reference to the isomegalen- and isomorphen-
diagram. For Sci Int. 120:32—6.

Greenberg B. 1991. Flies as forensic indicators. J Med Entomol. 20:565—77.

Greenberg B and Kunich JC. 2002. Entomology and the Law Flies as Forensic
Indicators. Cambridge University Press: Cambridge, UK.

Hallas R, Schiffer M, and Hoffmann AA. 2002. Clinal variation in Drosophila serrata
for stress resistance and body size. Genet Res. 79: 141-8.

Hall DG. 1948. The blow flies of North America. Monmnental Printing Co.: Baltimore,
MD.

Hall MJ, Hutchinson RA, F arkas R, Adams ZJ, Wyatt NP. 2003. A comparison of
Lucitraps and sticky targets for sampling the blowfly Lucilia sericata.
Med Vet Entomol. 17:280—87.

Hargrove WW and Hofl'man FM. 2004. Potential of multivariate quantitative methods
for delineation and visualization of ecoregions. Environ Manage. 342839—60.

Hastie TJ and Tibshirani RJ. 1990. Generalized additive models. Chapman and
Hall/CRC: Boca Raton, FL

Henrich VC and Brown NE. 1995. Insect nuclear receptors: a developmental and
comparative perspective. Insect Biochem Molec Biol. 25:881—97.

242

Hoffmann AA and Harshman LG. 1999. Desiccation and starvation resistance in
Drosophila: patterns of variation at the species, population and intrapopulation levels.
Heredity. 83:63 7—43.

Johnson FM and Schaffer HE. 1973. Isozyme variability in species of the genus
Drosophila. VII. Genotype-environment relationships in populations of D. melanogaster
from the Eastern United States. Biochem Genet. 10:149—63.

Kalthoff K. Analysis of biological development. 2001. McGraw-Hill: New York,NY.

Kamal AS. 1958. Comparative study of thirteen species of sarcosaprophagous
Calliphoridae and Sarcophagidae (Diptera) I. Bionomics. Ann Entomol Soc Am. 51:261—
70.

Kaneshrajah G and B. Turner B. 2004. Calliphora vicina larvae grow at different rates
on different body tissues. Int J Legal Med. 118:242—4.

Khaitovich P, Enard W, Lachmann M, and Paabo S. 2006. Evolution of primate gene
expression. Nat Rev Genet. 7:693—702.

Kingsolver JG. 2000. Feeding, Growth, and the Thermal Environment of Cabbage
White Caterpillars, Pieris rapae L. Physiol Biochem Zool. 73:621-8.

Kingsolver JG, Gomulkiewicz R, and Carter PA. 2001. Variation, selection and
evolution of function-valued traits. Genetica. 112—113: 87—104.

Kulathinal RJ and Hartl DL. 2005. The latest buzz in comparative genomics. Genome
Biol. 6:201.

Liu Z, Gong P, Wu K, Sun J, and Li D. 2006. A true summer diapause induced by high
temperatures in the cotton bollworrn, Helicoverpa armigera (Lepidoptera: Noctuidae). J
Insect Physiol. 52: 1012—20.

243

Luders F, Segawa H, Stein D, Selva EM, Perrimon N, Turco SJ, and Hacker U.
2003. Slalom encodes an adenosine 3'-phosphate 5'-phosphosulfate transporter essential
for development in Drosophila. EMBO J. 22:363 5—44.

Ludwig MZ, Patel NH, and Kreitman M. 1998. Functional analysis of eve stripe 2
enhancer evolution in Drosophila: rules governing conservation and change.
Development. 125:949—5 8.

Mackay TFC. 2001. The genetic architecture of quantitative traits. Annu Rev Genet.
35:303—39.

Mansson RA, Frey JG, Essex JW, and Welsh AH. 2005. Prediction of properties from
simulations: a re—examination with modern statistical methods. J Chem Inf Model.
45:1791—803.

McAlpine JF. 1989. Manual of Nearctic Diptera. Canadian Government Publishing
Center.

McGregor AP. 2005. How to get ahead: the origin, function, and evolution of bicoid.
BioEssays. 27:904—13.

Mellenthin K, Fahmy K, Ali RA, Hunding A, Da Rocha S, and Baumgartner S.
2006. Wingless signaling in a large insect, the blowfly Lucilia sericata: a beautiful
example of evolutionary developmental biology. Dev Dyn. F eb;235(2):347—60.

Mocellin S, Rossi CR, Pilati P, Nitti D, and Marincola FM. 2003. Quantitative real-
time PCR: a powerful ally in cancer research. Trends Mol Med. 9:189—95.

Moe SJ, Kristoffersen AB, Smith RH, and Stenseth NC. 2005. From patterns to
processes and back: analyzing density—dependent responses to an abiotic stressor by
statistical and mechanistic modeling. Proc R Soc B. 272:2133—42.

Montooth KL, Marden JH, and Clark AG. 2003. Mapping determinants of variation in
energy metabolism, respiration and flight in Drosophila. Genetics. 165:623—35.

Morey S and Reznick D. 2000. A comparative analysis of plasticity in larval

244

development in three species of spadefoot toads. Ecology. 81: 1736—49.

Mount SM, Gotea V, Lin CF, Hernandez K, and Makalowski W. 2007. Spliceosomal
small nuclear RNA genes in 11 insect genomes. RNA. 1325—14.

National Center for Biotechnology website. www.ncbi.nlm.nih.gov

Newcomh RD, Gleeson DM, Yong CG, Russell RJ and Oakeshott JG. 2005. Multiple
mutations and gene duplications conferring organophosphorus insecticide resistance have
been selected at the Rop—l locus of the sheep blowfly, Lucilia cuprina. J Mol Evol.
60:207—20.

Oudman L, Van Delden W, Kamping A, and Bijlsma R. 1991. Polymorphism at the
Adh and alpha Gpdh loci in Drosophila melanogaster: effects of rearing temperature on
developmental rate, body weight, and some biochemical parameters. Heredity. 67:103-
15.

Parsch J, Russell JA, Beerman I, Hartl DL and Stephan W. 2000. Deletion of a
conserved regulatory element in the Drosophila Adh gene leads to increased alcohol
dehydrogenase activity but also delays development. Genetics. 156:219—27.

Perrin N. 1995. About Berrigan and Charnov’s life-history puzzle. Oikos. 73:137—9.

Plaistow SJ, Lapsley CT, Beckerman AP and Benton TC. 2004. Age and size at
maturity: sex, environmental variability and developmental thresholds. Proc Biol Sci.
271: 919—24.

Powell JR. 1997. Progress and prospects in evolutionary biology. Oxford University
Press: New York, NY.

Puskas LG, Menesi D, Feher LZ, and Kitajka K. 2006. High-throughput functional
genomic methods to analyze the effects of dietary lipids. Curr Pharm Biotechnol. 7:525—
9.

245

R Development Core Team. 2004. R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-00-
3, URL http://www.R-project.org.

Ripley RM, Harris AL, and Tarassenko L. 2004. Non-linear survival analysis using
neural networks. Stat Med. 23:825—42.

Rogina B and Helfand SL. 1995. Regulation of gene expression is linked to life span in
adult Drosophila. Genetics. 141 :1043—8.

Rogina B, Benzer S, and Helfand SL. 1997. Drosophila drop-dead mutations accelerate
the time course of age-related markers. PNAS. 94:6303—6.

Saks MJ and Koehler JJ. 2005. The coming paradigm shift in forensic identification
science. Science. 309:892-5.

Scheiner SM and Gurevitch J. 2001. Design and analysis of ecological experiments. 2"d
Ed. Oxford University Press. New York, NY.

Sherman RA and My-Tien Tran JM. 1995. A simple, sterile food source for rearing
the larvae of Lucilia sericata (Diptera: Calliphoridae). Med Vet Entomol. 9:393-8.

Sherman RA. 2000. Wound myiasis in urban and suburban United States. Arch
Intern Med. 160: 2004—14.

Shingleton AW. 2005. Body-size regulation: combining genetics and physiology. Curr
Biol. 15:R825—7.

Shingleton AW, Das J, Vinicius L, and Stern DL. 2005. The temporal requirements for
insulin signaling during development in Drosophila. PLoS Biol. 3:e289.

Sibly RM and Atkinson D. 1994. How rearing temperature affects optimal adult size in
ectotherms. Func Ecol. 8:486—93.

246

Skaer N, Pistillo D, and Simpson P. 2002. Transcriptional heterochrony of scute and

changes in bristle pattern between two closely related species of blowfly. Dev Bio.
252:31—45.

Sliter TJ and Gilbert L1. 1992. Developmental arrest and ecdysteroid deficiency
resulting from mutations at the dre4 locus of Drosophila. Genetics. 130:555—68.

Sullivan AA and Thummel CS. 2003. Temporal profiles of nuclear receptor gene
expression reveal coordinate transcriptional responses during Drosophila development.
Mol Endocrinol. 1 7:2 1 25—3 7.

Tachibana S, Numata H, and Goto SG. 2005. Gene expression of heat-shock proteins
(Hsp23, Hsp70 and Hsp90) during and after larval diapause in the blow fly Lucilia
sericata. J Insect Physiol. 51 :641—7.

Tachibana S and Numata H. 2004. Effects of temperature and photoperiod on the
termination of larval diapause in Lucilia sericata (Diptera: Calliphoridae). Zoolog Sci.
21:197—202.

Tarone AM and Foran DR. 2006. Components of plasticity in the development of a
Michigan population of Lucilia sericata. J Med Ent. 43:1023—33.

Tarone AM, Nasser YM, and Nuzhdin SV. 2005. Genetic variation for expression of
the sex determination pathway genes in Drosophila melanogaster. Genet Res. 86:31—40.

Tellam RL, Vuocolo T, Johnson SE, Jarmey J, and Pearson RD. 2000. Insect chitin
synthase cDNA sequence, gene organization and expression. Eur J Biochem. 267:6025—
43.

Varki A and Altheide TK. 2005. Comparing the human and chimpanzee genomes:
searching for needles in a haystack. Genome Res. 15:1746—58.

Venables WN and Ripley BD. 2002. Modern Applied Statistics with S (4th edn).
Springer: New York, NY.

247

 

Wall R, Pitts KM and Smith KE. 2001. Pre-adult mortality in the blowfly Lucilia
sericata. Med Vet Entomol. 15:328—334.

Ware WG and Whitacre DM. 2004. An Introduction to Insecticides (4th edition)
Extracted from The Pesticide Book, 6th ed. MeisterPro Information Resources:
Willoughby, Ohio.

Wells JD and Kurahashi H. 1994. Chrysomya megacephala (Fabricius) (Diptera:
Calliphoridae) development: rate, variation, and the implications for forensic
entomology. Jpn J Sanit Zool. 45:303—9.

Wells JD and Lamotte LR. 1995. Estimating maggot age from weight using inverse
prediction. J Forensic Sci. 402585—90.

Wood SN. 2006. Generalized additive models: an introduction in R. Chapman and
Hall/CRC Taylor and Francis Group: Boca Raton, FL.

Yeates DK and Wiegmann BM. 2005. The evolutionary biology of flies. Columbia
University Press: New York, NY.

248