3:51... 4§.§.¢1....:
9.... .23. lwrulx. .
.1 : v Ii! 35.1.:
’21:}.
I... r. .3
I p

3:33.
I! 5:».
i

1!.

an}.
an? ,

t .19.! . .
3:! . . q .
; . . ~,
r .1: 1 . .2 .31.. ..I..vauu
L: QyV. . , 5S...
4 1.. 3 ~
3

a... :fiuflxi
\rflfiafiwng u. n»... is»...

3:31... 4%.... .X
.. z 2

it!

Nix 3‘ Ilt£twbfi 35.5..
a... 142.....I) $9231.. 25:11.1{53-9:
. 1|.- ..»3 .\.5!I5.!t..§9 9".)

and. fa.
2va r: .2

( ~2‘

.. Kurt-.3...
:X {3321:3313 A

3.2.1.. is .

s. 31:32.52...

M. i3... .8“

. t . to.
42:..- 5...... In. .

 

 

x .
r... O \l!!.«

1. x31. 5.! .1]...
i sb..t‘;tln

 

 

 

 

$.12}:...z.ii.i.. . .. .2. . _ , 1 B
mggfifiés 9w§:$§ . . . .fiwpawfx . , . . . ; . “:1”.,.....+n.q

. "I? :. .lxlxr: . r .7... .35.: .t . 3.

 

3: . . .

 

 

 

goo7

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the
thesis entitled

THE USE OF ANTIMICROBIAL RESISTANCE PROFILES OF
FECAL ESCHERICHIA COLI TO IDENTIFY ORIGINS OF FECAL

MS.

 

CONTAMINATION OF SURFACE WATER

presented by

JANET CAROL BEAGLEY

has been accepted towards fulfillment
of the requirements for the

degree in Comparative Medicine and
Integrative Biology

 

W

 

0' Major Professor’s Signature

_Afc2tl 22, 2 00-4

Date

MSU is an Affirmative Action/Equal Opportunity Institution

c—o-o-n-o—n-

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/ClRC/DateDue.indd-p.1

 

THE USE OF ANTIMICROBIAL RESISTANCE PROFILES OF FECAL
ESCHERICHIA COLI TO IDENTIFY ORIGINS OF FECAL CONTAMINATION OF
SURFACE WATER
By

Janet Carol Beagley

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Comparative Medicine and Integrative Biology

2006

 

ABSTRACT

THE USE OF ANTIMICROBIAL RESISTANCE PROFILES OF FECAL
ESCHERICHIA COLI TO IDENTIFY ORIGINS OF FECAL CONTAMINATION OF
SURFACE WATER

By

Janet Carol Beagley

Fecal contamination of surface water and antimicrobial resistance both present
public health and environmental concerns. The purpose of this project was to determine
the prevalence of resistant E. coli in feces from a variety of species in the Red Cedar
watershed and to use discriminant analysis of antibiotic resistance profiles of E. coli to
identify sources of fecal contamination in the Red Cedar River (Michigan). We
hypothesized that there would be differences in antimicrobial resistance of fecal E. coli
from different species, and that discriminant analysis would be able to separate and
classify known-source isolates with greater accuracy than would be expected by chance.
E. coli was cultured from septic tank and sewage samples, livestock, pets, free ranging
wildlife, and river water samples, and one isolate per sample was subjected to
microdilution susceptibility testing for 16 antimicrobial agents. The overall percentage of
resistant isolates was greatest in livestock, followed by human (septic and sewage),
companion animal, river, and wildlife samples. Significant differences among species
groups were found for 10 out of 16 antimicrobials for resistant versus susceptible
categories and 14 out of 16 antimicrobials for minimum inhibitory concentrations (MIC).
The best four-group discriminant analysis model had an average rate of correct
classification (ARCC) of 46%. Using this model, 52.3% of river isolates were classified
as wildlife, 17.9% as pets, 16.4% as humans, and 13.5% as livestock, and positive
predictive values were 41% for classification as human, 71% for livestock, 37% for pet,
and 43% for wildlife. This study indicates that discriminant analysis of MIC values may

be a cost effective aid to identifying major sources of fecal contamination.

Acknowledgements

Funding for this project was provided by The Center for Comparative
Epidemiology (Michigan State University, College of Veterinary Medicine), the USDA,
and an NIH T-32 Training Grant. My major advisor, Dr. John B. Kaneene, provided
guidance and support throughout all stages of this project. Committee members, Dr.
Thomas R. Comer, Dr. Joseph Gardiner, and Dr. Scott Winterstein provided input during
the planning of this project, as well as carefiII reading and helpful comments during the
writing of this thesis. Katie May provided laboratory support, RoseAnn Miller helped
with statistical analysis, and Dr. Vilma Yuzbasiyan-Gurkan and Dr. John Baker provided
support and encouragement of my participation in the Comparative Medicine and
Integrative Biology graduate program. This project would not have been possible
without the help of numerous organizatiOns and individuals in collecting the diverse
samples necessary for this research. Septic tank samples were collected with the help of
Schunk Septic Service, Lashbrook’s Excavating and Septic Service, Howell Sanitary
Company, Brighton Septic, and Ball Septic Tank Company. Sewage samples were
collected with the help of the Lansing Wastewater Treatment Plant. Dog and cat samples
were collected with the help of Capital Area Humane Society, Ingharn County Animal
Shelter, The Humane Society of Livingston County, and many dog and cat owners.
Livestock samples were collected with the help of Dr. Ron Erskine, Dr. Dan Grooms,
MSU farms, and many local farmers. River coliforrn plates were provided by the

Michigan Department of Environmental Quality.

iii

Table of Contents

List of Tables .......................................................................................... vi
List of Figures ........................................................................................ vii
Introduction ............................................................................................ 1
Fecal Contamination of Surface Water ................................................... 1
Microbial Source Tracking ................................................................. 1
The Spread of Antimicrobial Resistant Bacteria ........................................ 2
The Red Cedar River (Michigan) ......................................................... 3
Study Rationale .............................................................................. 4
Overall Goal and Hypotheses .............................................................. 5
Objectives .................................................................................... 5
Overview ...................................................................................... 6
Chapter I: Literature Review: The Use of Antimicrobial Resistance Profiles and
Discriminant Analysis to Identify Sources of Fecal Contamination ........................... 8
Microbial Source Tracking to Identify Sources of Fecal Contamination ............ 8
Microbial Source Tracking Methods ...................................................... 9
Microbiological Methods .......................................................... 9
Chemical Methods ................................................................. 11
Genotypic Methods ................................................................ 11
Phenotypic Methods ............................................................... 12
Antibiotic Resistance Analysis for Microbial Source Tracking ...................... 12
Comparison of Antimicrobial Resistance Analysis with Molecular Methods... . . 14

Choice of Bacteria for Antimicrobial Resistance Microbial Source Tracking... ..14

Choice of Antimicrobials for Antimicrobial Resistance Microbial Source

Tracking ............................................................................ 16
Statistical Methods for Antimicrobial Resistance Microbial Source Tracking ..... 19
Multiple Antimicrobial Resistance Index ...................................... 19
Cluster Analysis ................................................................... 2O
Discriminant Analysis ............................................................ 21
Discriminant Function Analysis as a Statistical Method for Microbial Source
Tracking ............................................................................. 21
Discriminant Analysis: Estimation of Error ............................................ 22
Discriminant Analysis: Library Size and Representativeness ........................ 24
Discriminant Analysis: Choice of Grouping of Sources .............................. 26
Discriminant Analysis: Classification of Unknown Isolates .......................... 29
Conclusion .................................................................................. 31

Chapter II: Antimicrobial Resistance of Fecal Escherichia coli Isolates from Septic and
Sewage Sources, Companion Animals, Livestock, Free-Ranging Wildlife, and Surface
Water .................................................................................................. 33

Abstract ...................................................................................... 33

iv

Introduction ................................................................................. 34

Materials and Methods ..................................................................... 37
Study Site ........................................................................... 37
Sample Collection ................................................................. 37
Isolation of E. coli Bacteria ...................................................... 39
Antimicrobial Susceptibility Testing ........................................... 39
Statistical Analysis ................................................................ 40

Results ....................................................................................... 41
Isolation of E. coli ................................................................. 41
Antimicrobial Resistance ......................................................... 42
Species Comparisons Within Species Groups ................................. 44
Minimum Inhibitory Concentrations ........................................... 45

Discussion ................................................................................... 45

Conclusions ................................................................................. 52

Chapter III: The Use of Antimicrobial Resistance Profiles of Escherichia coli from
Livestock, Pet, Wildlife, and Human Sources to Identify Origins of Fecal

Contamination of the Red Cedar River (Michigan) ............................................. 68
Abstract ...................................................................................... 68
Introduction ................................................................................. 69
Materials and Methods ..................................................................... 72

Study Site ........................................................................... 72
Sample Collection and Bacterial Isolation ..................................... 72
Susceptibility Testing ............................................................. 72
Descriptive Statistics .............................................................. 73
Principal Component Analysis ................................................... 73
Discriminant Analysis ............................................................. 74
Results ....................................................................................... 75
Principal Component Analysis .................................................. 75
Discriminant Analysis ............................................................ 76

Model Selection .......................................................... 76

Model Evaluation ......................................................... 77
Resubstitution Versus Crossvalidation ................................ 78

Alterations of Species Groupings ...................................... 78
Classifications of River Isolates ........................................ 79

Discussion ................................................................................... 79
Conclusion .................................................................................. 84

Overall Summary, Conclusions, and Recommendations ....................................... 90
Summary ..................................................................................... 90
Conclusions ................................................................................. 91
Recommendations .......................................................................... 92

Literature Cited ...................................................................................... 93

List of Tables

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 3.1

Table 3.2

Table 3.3

Table 3.4

Class, Dilution Range, and Resistance and Intermediate Breakpoints of
Antimicrobials ...................................................................... 54

Common Multiple Antimicrobial (MAR) Resistance Profiles .............. 57

Percentages of Isolates Exhibiting Antimicrobial Resistance
(Antimicrobials Showing Significant Differences Among Groups)... . . . . .58

Percentages of Isolates Exhibiting Antimicrobial Resistance
(Antimicrobials Showing No Significant Differences Among Groups). . .59

Classification Table for K-Nearest-Neighbor, K=4, 16 Antimicrobials....87
Classification Table for K—Nearest Neighbor, K=4, 11 Antimicrobials....87

Comparison of Discriminant Analysis Models With Different Groupings
of Species ........................................................................... 87

Classification of River Isolates and Positive Predictive Values for
Classification as Human, Livestock, Pet, and Wildlife ...................... 88

vi

List of Figures

Figure 2.1
Figure 2.2
Figure 2.3

Figure 2.4
Figure 2.5
Figure 2.6
Figure 2.7
Figure 2.8
Figure 2.9
Figure 2.10
Figure 2.11
Figure 2.12
Figure 2.13
Figure 2.14
Figure 2.15
Figure 2.16
Figure 2.17
Figure 2.18
Figure 2.19

Figure 3.1

The Percentages of Isolates From Each Species Group Resistant or
Intermediate to One or More Antimicrobial Agent ........................... 55

The Percentages of Isolates From Each Species Group Resistant or
Intermediate to Two or More Antimicrobial Agents ......................... 55

The Percentages of Isolates From Each Species Group Resistant or

Intermediate to Five or More Antimicrobial Agents .......................... 56
MIC Values for Amikacin ....................................................... 60
MIC Values for Amoxicillan/Clavulanic Acid ................................ 6O
MIC Values for Ampicillin ...................................................... 61
MIC Values for Ceftiofur ......................................................... 61
MIC Values for Ceftriaxone ..................................................... 62
MIC Values for Cephalothin ..................................................... 62
MIC Values for Chloramphenicol .............................................. 63
MIC Values for Ciprofloxacin ................................................... 63
MIC Values for Cefoxitin ........................................................ 64
MIC Values for Gentarnicin ...................................................... 64
MIC Values for Kanamycin ...................................................... 65
MIC Values for Nalidixic Acid .................................................. 65
MIC Values for Streptomycin ................................................... 66
MIC Values for Sulfrnethoxamine .............................................. 66
MIC Values for Tetracycline .................................................... 67
MIC Values for Trimethoprim/Sulfinethoxazole ............................. 67

Principal Components Plot of Human (Septic and Sewage) Fecal
E. coli Isolates .................................................... , .................. 85

vii

Figure 3.2
Figure 3.3
Figure 3.4

Figure 3.5

Figure 3.6

Principal Components Plot of Livestock Fecal E. coli Isolates. . . . . . ..85
Principal Components Plot of Pet Fecal E. coli Isolates ...................... 86
Principal Components Plot of Wildlife Fecal E. coli Isolates. . . . ..86

Classification of E. coli Isolates From River Samples Collected in
Summer and Fall .................................................................. 88

Classification of E. coli Isolates From River Samples Collected When

Fecal Coliform Counts were Low (<6OO cfullOOml) and High (>1000
cfu/lOOml) .......................................................................... 89

viii

Introduction

Fecal Contamination of Surface Water:

Fecal contamination of water used for drinking, recreation, and agriculture poses
a serious threat to human, animal, and environmental health. Human and animal feces
may contain pathogens, including Shigella spp., Vibrio cholerae, hepatitis A virus,
Noroviruses, Escherichia coli OlS7:H7, Salmonella, Campylobacter, Cryptosporidium,
and Giardia. Numerous disease outbreaks have been attributed to these and other water-
borne organisms (Scott et al., 2002). Sources of fecal contamination in rivers and lakes
include sewage, failing septic systems, agricultural runoff, and wildlife. Health risks and
potential management strategies vary depending on the major sources of contamination.
Specifically identifying sources, however, has been difficult (Scott et al., 2002; Meays et

al,2004)

Microbial Source Tracking:

A number of genetic and microbiological methods have been used to characterize
fecal bacteria, such as Escherichia coli, in order to link them to a specific source. These
methods include bacteriophage identification, ribotyping, DNA sequence analysis with
PCR amplification, and antimicrobial resistance analysis (Meays et al., 2004). The goal
of each of these methods is to create a library of bacteria from known animal and human
sources and then use statistical methods, such as discriminant analysis, to classify
bacteria from an unknown source into one of the known-source categories. Although

molecular methods of characterization may have the potential for greater specificity, the

formation of large libraries is often cost-prohibitive. In contrast, antimicrobial resistance
is an inexpensive and relatively simple method of characterizing bacteria.

Antimicrobial resistance analysis is based on the principle that resistance to
certain antibiotics is selected for by antimicrobial exposure. Bacterial flora in the
intestines of humans and various types of animals are subjected to different types,
concentrations, and frequencies of antimicrobials. These differences should be reflected
in the resistance profiles of fecal bacteria from different species. In contrast to humans
and domesticated animals, wild animals have generally not been treated with antibiotics.
Although resistance is generally low in fecal bacteria from wildlife, it is possible for wild
animals to acquire resistant fecal bacteria via contact with antimicrobials or resistant
bacteria in the environment (Selvaratnam and Kunberger, 2004; Middleton and Ambrose,

2005)

The Spread of Antimicrobial Resistant Bacteria:

In addition to being a relatively inexpensive method of identifying sources of
fecal contamination, antimicrobial resistance analysis allows us to measure the impacts of
antibiotic use on bacteria from animals, humans, or the environment. Antibiotic
resistance is in itself a threat to human and animal health. Pathogenic bacteria are
increasingly becoming resistant to available antimicrobials, which compromises our
ability to treat infections successfully (Jones et al., 1997; Witte, 1998). Fecal
contamination may contribute to the spread of resistance by disseminating resistant fecal
bacteria into the environment. Monitoring resistance in intestinal bacteria from humans

and domesticated animals helps us to assess the direct effects of antimicrobial use, and

monitoring levels of resistance in wild animals and environmental samples provides

information on the broader, environmental impacts of antimicrobial use.

The Red Cedar River (Michigan):

The Red Cedar River arises in Cedar Lake and flows northwest approximately
73km, draining an 1186 km2 area in central Michigan. The river flows through wooded
and agriculture land, many small to medium sized towns, and the campus of Michigan
State University. In East Lansing, Michigan, it joins the Grand River, which ultimately
empties into Lake Michigan. The Red Cedar River and its tributaries and drains provide
mid-Michigan residents with numerous recreational opportunities and also serve as a
source of irrigation for crops.

Fecal coliform counts are conducted weekly in the spring through the fall on
samples from the Red Cedar River at the Michigan State University (MSU) campus,
other areas around East Lansing, and upstream at the town of Williamston, as part of a
collaborative water-quality monitoring program between Michigan State University and
the Michigan Department of Environmental Quality. Between 2000 and 2002 the
average percentage of river samples at the MSU campus that failed to meet the standards
for partial body contact recreation (1,000 cfu/ 100ml) ranged from 15 to 24% (Kline-
Robach and Witter, 2002). This resulted in frequent closure of the campus canoe livery.

In the past, both sewage outflow and agriculture have been indicated in
contamination of the Red Cedar River (Kline-Robach and Witter, 2002). Two large
cattle operations near Wolf Creek were presumed to be the cause of a spike of 39,667

fecal coliform cfu/ 100ml at the Williamston location in September of 2001. Near the

MSU campus, a sample from a major sewage discharge exceeded 4,000,000 cfu/ 100 ml
in August of 2002, a time when downstream sites were also elevated. Although both of
these cases involved very high levels of bacteria at specific locations that could be traced
to their sources, for the most part sources of bacteria isolated from the Red Cedar River

are unknown.

Study Rationale:

The development of this study was based on the need to identify sources of fecal
contamination in the Red Cedar River, the lack of general knowledge regarding the
impacts of fecal contamination on the spread of antimicrobial resistance, and the need for
additional information on the utility of antimicrobial resistance analysis and discriminant
analysis for microbial source tracking.

Our study is built on previous work done in our lab, which demonstrated that
discriminant analysis of antimicrobial resistance of E. coli may have potential as a means
of identifying sources of surface water contamination (Sayah et al., 2005). This study
found significant differences in resistance, using disk diffusion, between a variety of
species. Limited sample sizes restricted discriminant analysis to livestock and wildlife
source groups and resulted in an average accuracy of classification of 60.8% (Sayah,
2004)

I In order for a discriminant analysis tool to be accurate, the library of known
source isolates must include representatives from the full spectrum of possible sources.

We aimed to increase the representativeness of this library by incorporating human septic

tank samples, untreated sewage, dog and cat samples, and a greater number of free-
ranging wildlife samples.

In addition, we aimed to improve the ability of discriminant analysis to
differentiate and identify sources of fecal contamination by incorporating additional
antimicrobials and minimum inhibitory concentrations (MIC values) into our analysis.
One isolate from each sample was tested against a panel of antimicrobials that was
chosen based on its inclusion of agents that are normally effective against gram-negative
bacteria and are commonly used in human and animal medicine. Although previous
studies have used multiple concentrations of antimicrobials to characterize bacteria, to
our knowledge the use of MIC values as variables for discriminant analysis in microbial

source tracking studies has not been reported.

Overall Goal and Hypotheses:

The purpose of this study was to use discriminant analysis of antimicrobial
resistance profiles of fecal E. coli from known sources to determine the sources of fecal
contamination in the Red Cedar River. We hypothesized that there would be significant
differences in antimicrobial resistance of fecal E. coli from humans, livestock, wildlife,
and pets, and that discriminant analysis would be able to separate and classify known-

source isolates with greater accuracy than would be expected by chance classification.

Objectives:

The objectives of this study are to:

1. Review current literature for studies that involve the use of antimicrobial
resistance analysis as a means of identifying sources of fecal contamination

2. Test for differences in antibiotic resistance of fecal E. coli isolates from humans
(septic tanks and sewage), pets (dogs and cats), livestock (beef cattle, dairy cattle,
swine, and sheep), and free-ranging wildlife (ducks, geese, and deer) collected in
or near the Red Cedar watershed.

3. Determine the parameters, antimicrobials, and grouping of species that optimize
the ability of discriminant analysis to separate and correctly classify known-
source isolates.

4. Use this discriminant function to classify unknown source isolates cultured from
samples collected from the Red Cedar River at multiple locations between May

and December of 2004.

Overview:

This thesis is organized into three chapters. Chapter 1 is a literature review of the
use of antimicrobial resistance analysis for microbial source tracking. This chapter is
divided into sections based on the major issues and topics that have arisen in microbial
source tracking studies, followed by a conclusion that summarizes current knowledge and
provides recommendations for future research. Chapters 2 and 3 represent the
experimental components of this thesis, and each contains an abstract, introduction,
methods, results, discussion, and conclusion. Chapter 2 is a comparison of antimicrobial
resistance profiles of fecal E. coli from septic and sewage sources, pets, livestock, free-

ranging wildlife, and surface water. Chapter 3 describes the use of discriminant analysis

of antimicrobial resistance profiles from human, livestock, pet, and wildlife sources to

identify origins of fecal contamination in the Red Cedar River (Michigan).

Chapter I: Literature Review: The Use of Antimicrobial Resistance Profiles and

Discriminant Analysis to Identify Sources of Fecal Contamination.

Microbial Source Tracking to Identify Sources of Fecal Contamination

Fecal contamination of water is a widespread problem in the United States, as
well as in other countries throughout the world. Both urban development and intensive
agriculture have been implicated in increasing fecal bacterial contamination of surface
water (Wesikal et al., 1996; Mallin et al., 2000; Collins, 2004; Shehane et al., 2005).
Contamination may originate from point sources, such as industrial or sewage effluent, or
non-point sources, such as failing septic tanks or run-off from agricultural or other lands.
Feces fi'om human sources is often considered to carry a higher risk of microbes that are
pathogenic to humans, such as Shigella spp., Vibrio cholerae, hepatitis A virus, and
Noroviruses (Scott et al., 2002). However, animal feces may also contain bacteria,
viruses, and protozoa that are pathogenic to humans, including E. coli 01 57:H7,
Salmonella, Campylobacter, Cryptosporidium, and Giardia.

Normal intestinal bacteria, such as fecal coliforms and fecal streptococci, are used
to indicate fecal contamination of water used for drinking, recreation, or agriculture.
Associations between high numbers of indicator bacteria and land-use may suggest
potential sources of contamination. For example, a positive correlation between fecal
coliform levels and housing density may suggest human sources (Young and Thackston,
1999). However, fecal coliforms and streptococci are present in the intestines of all

warm-blooded and some cold-blooded species (Harwood et al., 1999), and thus

quantifying these bacteria is often insufficient to identify specific sources of
contamination.

When fecal indicator bacteria are present in surface water from mixed-use
watersheds, with urban, agricultural, and undeveloped land, the relative contributions of
various sources can be difficult to determine. Because identification of the sources of
fecal contamination is crucial for risk assessment and the development of management
plans, several methods of bacterial source tracking have been proposed to determine the

origins of fecal bacteria.

Microbial Source Tracking Methods

Microbial source tracking methods have been classified into four groups:
microbiological, genotypic, phenotypic, and chemical (Scott et al., 2002). Genotypic and
phenotypic methods generally rely on a database, or library, of bacterial isolates from
known sources. Bacteria from water samples or other unknown sources can then be
compared against this library of known sources and classified accordingly. In contrast,
microbiological and chemical methods look for characteristic organisms or chemicals in
the water that indicate either human or animal feces. Although these methods do not
require the construction of a known source database, they do require prior knowledge of

organisms or chemicals that are unique to human or specific animal sources.

Microbiological Methods:
An early microbiological source-tracking method involved determination of fecal

coliform/fecal streptococci (FC/F S) ratios in surface water. Human feces contain a

greater proportion of coliforms as compared with animal feces, and FC/F S ratios > 4
were considered to indicate a human source of fecal contamination (Scott et al., 2002;
Meays et al., 2004). However, variability in the survival rates of fecal streptococci and
coliforms indicates that this in an unreliable method, and F C/F S ratios are no longer
recommended for differentiating human and animal sources of fecal contamination
(APHA, 1998).

Current microbiological methods involve the detection of organisms that are
specific to human feces. For example, sorbitol-fermenting Bifidobacterium spp.,
anaerobic bacteria found primarily in human intestines, have been used as an indication
of human-derived fecal contamination (Rhodes and Kator, 1999; Bonjoch et al., 2005).
Human-specific viruses may also indicate human sources of fecal contamination,
including certain strains of Bacteroidesfragilis bacteriophages (J ofre etal., 1989) and
human enteric viruses (J iang et al., 2001; Lee and Kim, 2002; Fong et al. 2005). Other
species-specific enteric viruses have been used to indicate the presence of swine
(J imenez-Clavero et al., 2003) or bovine (Fong et al., 2005; Jimenez-Clavero et al.,
2005) feces. It has also been reported that E. coli isolates from human and animal feces
contain different serotypes of RNA bacteriophages, and attempts have been made to use
F+ RNA coliphages for tracking sources of fecal pollution (Hsu et al., 1995; Griffin et
al., 1999). Although the detection of organisms found only in human or certain animal
feces can be a very specific indication of sources of fecal contamination, the sensitivity of

these methods may be limited (Scott et al., 2002).

10

Chemical Methods:

Chemical methods of bacterial source tracking are based on detection of chemical
compounds, rather than microbiological organisms, that may be specific to human-
derived sources of contamination. Proposed compounds include caffeine (Burkhardt,
1999; Standley et al., 2000) and chemicals found in laundry detergents (Sinton et al.,
1998). Caffeine is present in the urine of humans who have ingested caffeine-containing
beverages or pharmaceuticals, and may be an indicator of human sewage. However,
dilution and chemical changes of compounds in the environment may decrease the
sensitivity of this method. Laundry detergent chemicals, such as fluorescent whitening
agents indicate general human or industrial sources, but do not necessarily indicate fecal

pollution (Sinton et al., 1998).

Genotypic Methods:

Genotypic methods use a variety of molecular techniques to create genetic
fingerprints, based on differences in DNA. In order to classify unknown isolates, a
database of fingerprints from known source isolates must first be created. Techniques
that have been used for bacterial source tracking include ribotyping (Carson et al., 2001;
Parveen etal., 1999), pulsed-field gel electrophoresis (Simmons et al., 1995), repetitive
DNA sequences (Dombeck et al., 2000), and ribosomal genetic markers in Bacteroides
spp. or Prevotella spp. (Bernhard and Field, 2000a, 2000b). Ribotyping involves
analysis of DNA fragments generated from restriction enzyme digestion of genes
encoding 16S rRNA. Fragments are separated by gel electrophoresis, probed, and

compared with a known-source library. Pulse-field gel electrophoresis is similar to

11

ribotyping, except instead of analyzing only DNA encoding rRNA, it uses the whole
DNA genome. Repetitive DNA sequences (Rep-PCR) involve PCR amplification of
DNA between adjacent repetitive extra-genie elements. Ribosomal genetic markers in
Bacteroides or Prevotella have been detected and characterized with length heterogeneity
PCR and terminal restriction length polymorphism (Bernard and Field, 200a, 2000b).
These host-specific molecular markers eliminate the need to first culture the organism,
but rely on the presence of these specific genes, which may be variable (Bernhard and

Field, 2000a, 2000b; Scott et al., 2002).

Phenotypic Methods:

Phenotypic methods involve classifying bacterial isolates based on some
phenotypic property. These methods are similar to genotypic methods in that isolates
from an unknown source are compared to a database of known source isolates. A variety
of phenotypic methods, including biochemical tests (Olsen et al., 1992), outer-membrane
protein profiles (Barenkamp et al., 1981), and multiple antimicrobial resistance analysis
(Parveen et al., 1997; Harwood et al., 2000; Graves et al., 2002, others), have been
proposed as ways of discriminating among various groups of bacteria. Of these,
antimicrobial resistance analysis has shown the most promise as a bacterial source

tracking method.

Antibiotic Resistance Analysis for Microbial Source Tracking

Antibiotic resistance analysis as a method of bacterial source tracking is based on

the underlying principle that bacteria flora present in the guts of various types of animals

12

are subjected to different types, concentrations, and frequencies of antibiotics. Selective
pressure within a specific group of animals selects for bacteria that possess specific
"fingerprints" of antibiotic resistance. (Scott et al. 2002).

Differences in antimicrobial use in humans and various animal species result in
differences in fecal bacterial resistance. Certain antimicrobial agents, such as
tetracyclines, are used extensively as feed additives for growth promotion in animal
agriculture (Prescott and Baggot, 1993). Correspondingly, a high percentage of bacteria
resistant to tetracycline have been cultured from the feces of tetracycline-treated swine
(Chee-Sanford et al., 2001; Sengelov et al., 2002). In contrast, novel and more expensive
antimicrobials, such as the newer fluoroquinolones, are oflen reserved for use in human
medicine. Studies indicate fluoroquinolone-resistant bacteria may be more common in
human feces, as compared with other species (Parveen et al., 1997; Webster et al., 2004).
Exposure of wild animals to antibiotics is minimal, and thus selection for resistant
bacteria should be lower than in humans and domesticated animals. In general, studies
have shown a low prevalence of resistant bacteria in wildlife (Routrnan et al., 1985; Guan
et al., 2002; Burnes, 2003).

Resistant bacteria have been cultured fi'om surface water, and studies have linked
these bacteria with sources based on resistance profiles. Higher percentages of resistant
and multiply resistant bacteria have been found in more developed watersheds, indicating
a correlation between human sources of contamination and increased resistance of fecal
indicator bacteria. Recent studies have used cluster analysis (Kelsey et al., 2003;

Webster et al., 2004) and discriminant analysis (Hagedom et al., 1999; Graves et al.,

13

2002; Whitlock et al., 2002; Carrol et al., 2005) of antimicrobial resistance profiles to

differentiate sources and classify unknown isolates.

Comparison of Antimicrobial Resistance Analysis with Molecular Methods

Antibiotic resistance analysis has been compared with molecular source tracking
methods. One study comparing antimicrobial resistance analysis, amplified fragment
length polymorphism (AF LP), and 16S rRNA gene sequencing on 319 E. coli isolates
from livestock, wildlife, and sewage, indicated that AF LP was most effective at
discriminating sources (Guan et al., 2002). Whereas genetic methods have been
considered to be more sensitive at identifying differences between isolates (Guan et al.,
2002; Scott et al., 2002), antibiotic resistance analysis has the advantage of being
relatively simple and inexpensive (Scott at al., 2002; Meays et al., 2004). This allows for
the creation of large known-source libraries that are representative of all potential sources
of fecal pollution. Additionally, the large genetic heterogeneity of E. coli makes
grouping of sources difficult (Lasalde et al., 2005), whereas using phenotypic traits limits
the number of variables considered. One suggestion is to use a combination of methods;
with antimicrobial resistance analysis one may quickly assess general sources of
contamination, whereas molecular methodology may indicate more precisely the specific

sources if needed (Crozier et al., 2002).

Choice of Bacteria for Antimicrobial Resistance Microbial Source Tracking

The ideal bacterium to use both to quantify and to trace fecal contamination is

non-pathogenic, easy to culture and enumerate, and an accurate reflection of the presence

14

of fecal material and associated pathogens. Fecal coliforms and streptococci have been
used as indicator bacteria, and are thus the most logical choices for antimicrobial
resistance source tracking studies. Microbial source tracking studies have successfully
used fecal coliforms (Harwood et al., 2000; Whitlock et al., 2002; Burnes, 2003; Kelsey
et al., 2003), E. coli (Parveen et al., 1997; Guan et al., 2002; Webster et al., 2004; Carroll
et al., 2005), fecal streptococci (Wiggins et al., 1996; Hagedom et al., 1999; Harwood et
al., 2000; Bower, 2001; Geary and Davies, 2003;) and Enterococcus spp. (Crozier et al.,
2002; Graves et al., 2002; Choi et al., 2003; Booth et al., 2003; Wiggins et al., 2003) to
trace origins of fecal contamination.

Streptococci, unlike fecal coliforms, can regularly be cultured from treated
sewage (Miescier et al., 1982), and have thus been suggested as the better choice for
monitoring sewage-contaminated water. Additionally, streptococci survive longer than
fecal coliforms, especially in saline water, and may thus provide a better indication of
fecal contamination in marine environments. Prolonged survival of streptococci has also
been used as an argument against using it as an indicator of fecal contamination, as
environmental reservoirs may exist (Desmarais et al., 2002). One thought is that the
longer survival of streptococci may make it a useful indicator of a relatively longer-term
history of contamination, whereas the less persistent coliforms may provide a better
indication of recent contamination (Harwood et al., 2000).

Of the coliforms, E. coli is the most widely used fecal indicator and is advocated
by the United States Environmental Protection Agency as being strongly correlated with
risk associated with viral, bacterial, and parasitic pathogens of enteric origin (U SEPA,

2001; Edberg et al., 2000). However, recent evidence documenting the persistence and

15

then the increase of a specific genotype of E. coli in northern Minnesota suggests that E.
coli bacteria can survive over the winter months in temperate regions and subsequently
grow during the summer months (Ishii et al., 2006). In general, however, the moderate
survivability, ease of culture, and popularity as a fecal indicator have made E. coli and
other fecal coliforms a popular choice for bacterial source tracking studies.

To compare the efficacy of enterococci versus coliform source tracking, Harwood
et al. (2000) conducted a study in which both fecal coliforms and enterococci were
cultured from known and unknown sources and tested against similar panels of
antimicrobial agents. These two separate databases classified known source isolates with
similar accuracy. For the most part, unknown sources were classified into the same
source categories, regardless of whether the coliform or enterococci library was used.
This study suggests that the specific species of bacterium used for bacterial source
tracking may not be crucial, as long as it is consistent within a given study and

antimicrobial agents are chosen accordingly.

Choice of Antimicrobials for Antimicrobial Resistance Microbial Source Tracking
Bacterial source tracking studies using antimicrobial resistance analysis have
aimed to include antimicrobial agents which are commonly used in human medicine,
animal medicine, and animal agriculture. Common antimicrobials used include
arnoxicillin, ampicillin, chlortetracycline, cephalothin, erythromycin, nalidixic acid,
neomycin, gentamicin, and tetracycline. Burnes et al. (2003) included six antimicrobials
and found that spectinomycin, streptomycin, kanamycin, and arnpicilin were the most

important measurement variables, whereas gentamicin and tetracycline were relatively

16

unimportant for classification. Carrol et al. (2005) found that amoxicillin and
erythromycin provided the best separation between human and non-human isolates,
wheras livestock sources had the best separation for cephalothin and chlortetracycline.

Factors that may influence which antibiotics best separate source groups include
the choice of bacteria, the sources being considered, and the geographical area in
question. The choice of antibiotics should also be tailored to the species of bacteria being
examined. One study comparing streptococci and E. coli used the same eight
antimicrobial agents for each, but also included a ninth agent, vancomycin, for
streptococci (Harwood et al., 2000). Vancomycin was not included for fecal coliforms
because vancomycin resistance is an intrinsic characteristic of fecal coliforms. Webster
et al. (2004) points out that the antimicrobials that best differentiate species or species
groups may also vary depending on the geographical area of the study, due to differences
in popularities of prescribed antibiotics in different regions.

The number of antibiotics used in microbial source tracking studies has varied
between four (Geary and Davies, 2003) and 14 (Guan et al., 2002), with the majority of
studies using between eight or nine agents (Parveen et al., 1997; Harwood et al., 2000;
Bower et al., 2001; Crozier et al., 2002; Graves et al., 2002; Whitlock et al., 2002; Booth
et al., 2003; Webster et al., 2004; Carrol et al., 2005;). In general, the accuracy of
classification using discriminant analysis increases with increasing numbers of
antimicrobial agents. Harwood et al. (2000) found that the average rate of correct
classification (ARCC) was maximized by including all eight or nine (for coliform and
streptococci libraries, respectively) antibiotics tested. However, Wiggins et al. (1996)

noted that while in general the ARCC improved with increasing numbers of antibiotics

17

used in the analysis, the highest ARCC was actually obtained using four, not all five, of
the antibiotics tested. Similarly, Hagedom et al., (1999) initially tested isolates against
seven concentrations of 13 antimicrobials, but found that the best separation by source
was with six antimicrobials from two to four concentrations each. In this case,
chlortetracycline, erythromycin, neomycin, oxytetracycline, streptomycin, and
tetracycline were included in the analysis, whereas halofuginone, salinomycin,
streptomycin, amoxicillin, ampicillin, Chloramphenicol, rifampin, and vancomycin were
excluded from the analysis (Hagedom, 1999). Thus, although additional variables
generally increase the distinction between sources, the inclusion of antibiotics for which
resistance does not differ significantly between sources may introduce additional noise
and reduce the accuracy of the discriminant analysis.

Although initial studies tested whether an isolate was resistant or susceptible to
one concentration of an antibiotic (Parveen et al., 1997; Guan et al., 2002), more recent
studies have incorporated multiple concentrations of the antimicrobial agents (Hagedom
et al., 1999; Burnes et al., 2003; Carroll et al., 2005). The inclusion of more than one
concentration of each antimicrobial agent introduces additional variables into the
discriminant function and may improve the ability of the analysis to separate source
groups and thus correctly classify unknown isolates.

Another method of increasing the number of variables to potentially better
differentiate sources was introduced by Carrol et al. (2005). In this study, a library of
717 known source isolates from human, domesticated animal, livestock, and wild sources
were used to classify surface water isolates. Four concentrations each of eight antibiotics

were used, and additional variability was introduced by rating the growth of bacteria from

18

one to four. These ratings corresponded to no growth, filamentous growth, restricted

growth, and full growth.

Statistical Methods for Antimicrobial Resistance Microbial Source Tracking
Multiple Antimicrobial Resistance Index:

Early microbial source tracking efforts used what is known as the multiple
antibiotic resistance index (MAR index) as a means of identifying the presence of high-
risk sources of fecal pollution (Kasper et al., 1990; Krumpennan, 1983). Krumpennan
(1983) cultured E. coli from a variety of sources, including sewage, human anal swabs,
poultry, swine, cattle, and wildlife and tested for resistance to 12 antimicrobials. MAR
indices were then calculated for each isolate by adding the total number of antimicrobials
to which an isolate was resistant, and dividing by the number of antimicrobials tested.
MAR indices were also calculated for each species. This study showed that the highest
risk sources, namely humans, poultry, and swine, had the highest MAR indices. In this
way, they suggested MAR indices could be calculated to indicate the likelihood of
contamination of food from a hi gh-risk source. Similarly, Kasper et al. (1990) calculated
MAR indices for E. coli isolates cultured from surface water collected from urban and
rural areas. Urban isolates consistently had higher MAR indices, suggesting that MAR E.
coli may be useful in distinguishing hi gh-risk sources of fecal contamination and
assessing water quality. A shortcoming of MAR indices is that they only compare the
amount of resistance, not necessarily the specific antibiotics to which isolates are

resistant.

19

Cluster Analysis:

In cluster analysis, samples are grouped into clusters based on distances between
variables. This technique differs from discriminant analysis in that clusters are formed
without consideration of prior groups. In the case of antimicrobial resistance analysis,
isolates that share identical or similar antibiotic resistance profiles are grouped together,
and inferences can be made about common source origins of these isolates. Several
studies have used cluster analysis of antimicrobial resistance of fecal bacteria to compare
point and non-point sources (Parveen et al., 1997) and to classify surface water isolates as
being from human or nonhuman sources (Kelsey et al., 2003; Webster et al., 2004).

Parveen et al (1997) used a combination of MAR indices and cluster analysis of
MAR profiles to compare the antibiotic resistance profiles of E. coli bacteria from point
sources (sewage treatment effluent) and non-point sources (estuarine surface water).
MAR indices were higher for point source isolates, and cluster analysis showed increased
diversity (more clusters) of isolates from point versus non-point sources. Additionally,
profiles of isolates taken directly from human feces were similar to profiles from point
sources and profiles of isolates taken directly from animal feces were similar to profiles
from non-point sources. This study demonstrates the possibility of using cluster analysis
to differentiate between point and non-point sources of fecal pollution.

Studies have also used cluster analysis to compare unknown-source surface water
isolates with isolates from known sewage sources ( Kelsey et al., 2003; Webster at al.,
2004). Kelsey (2003) showed that isolates from one out of twenty-three surface water
sites were similar to sewage isolates, but for the majority of the sites the surface water

samples were grouped separately from the sewage samples. Although they concluded that

20

the majority of fecal contamination in the inlet was from non-human sources, it should be
noted that the lack of similarity between sewage and surface water resistance patterns
could be due to a lack of representativeness of sewage isolates. In contrast, Webster et
al. (2004) found that E. coli from estuaries in South Carolina contained isolates with
patterns that matched resistance patterns fi'om wastewater treatment plant influent, and

they concluded that human sources of fecal contamination were present in these areas.

Discriminant Analysis:

Discriminant analysis is a popular statistical tool used in antimicrobial resistance
bacterial source tracking studies. This method has the ability to both separate groups and
classify unknown-source isolates. Discriminant analysis and recent studies using this

technique are described in detail below.

Discriminant Function Analysis as a Statistical Method
For Microbial Source Tracking
Discriminant fimction analysis is a multivariate statistical method that can be used
to differentiate known groups or classes on the basis of several classification variables.
The resulting function, or classification rule, can then be used to classify individual
samples into groups or classes based on these variables. In the case of antimicrobial
resistance analysis, groups have been defined as the known sources of fecal bacteria, and
resistance or susceptibility to various antibiotics have been used as classification

variables. Discriminant analysis is applied to these samples from known sources, and the

21

resulting classification rule can then be used to classify bacteria from an unknown source,
such as surface water samples.

Parametric and non-parametric variations of discriminant analysis have been
described (Lachenbruch, 1975; Hand, 1981; McLachlan, 2004). If the variables have
approximately multivariate normal within-class distributions, then parametric
discriminant analysis methods are applicable. Parametric methods rely on a measure of
generalized squared distances, and can be based on the pooled covariance matrix,
yielding a linear function , or on individual within group covariance matrices, yielding a
quadratic function. Non-parametric methods that make no assumptions about the
underlying distribution of the data have also been described, and these include kernel and
k-nearest neighbor methods. Non-parametric methods use pooled or individual
covariance matrices to calculate Mahalanobis distances, and posterior probability
estimates of group membership for each class are evaluated based on estimated group-

specific densities.

Discriminant Analysis: Estimation of Error
With any discriminant function analysis, a certain amount of error will be present
in the ability to correctly classify samples fiom an unknown source into one of the known
source groups. Various methods have been described for estimating this error in terms of
average rate of correct classification (ARCC) (Hand, 1984; McLachlan. 2004). The
traditional resubstitution method involves using the same known-source isolates to create
and test the classification rule. In essence, a classification rule derived from discriminant

analysis of all known-source samples is used to classify these same samples, and the

22

percent of correctly classified isolates from each group is calculated. In contrast, the
cross-validation method involves removing each isolate and classifying that isolate based
on the library of remaining isolates. Because the isolate being classified is not included
in the library from which the classification rule is based, the cross-validation method is
considered less biased than the resubstitution method. An alternative to the cross-
validation method involves setting aside a random subset of isolates from-known sources
to be used to test the classification rule. Since the test isolates are not included in the
discriminant function analysis, this method is also considered to be relatively unbiased.
After known source isolates have been classified by one of these methods, the ARCC is
computed by adding the number of correctly classified isolates in all source categories
and dividing by the total number of isolates.

The ARCC value, even as determined by the less biased cross validation or test
data-set methods, may be biased by the inclusion of more than one isolate per sample.
Isolates from the same sample may be similar enough in resistance patterns that when one
isolate is taken out and reclassified, it will be correctly classified more often because
isolates from that same sample are still included in that library. Wiggins (2003) showed
that the ARCC decreased dramatically when all isolates from a given sample were
excluded in a cross-validation method, as compared with the typical cross-validation
method of simply excluding one isolate. Many studies include multiple isolates from a
given sample (Graves et al., 2002; Bower et al., 2001; Harwood et al., 2000; Wiggins et
al., 1999) and the bias inherent in this method of sampling is not always addressed.

Percentages of misclassified isolates can be calculated by adding the percentages

in the appropriate rows of the classification table produced by discriminant analysis,

23

excluding the percentages of correct classification (Harwood et al., 2000). Harwood et
al. (2000), argue that a source of contamination can legitimately be implicated in
pollution when the percentage of unknown sample isolates classified into a particular
source category exceeds the misclassification percentage. Similarly, as a way to
determine the lower limit for considering a source to be a significant contributor to a
watershed, Whitlock et al. (2002) averaged the expected frequencies of misclassification
from all sources and then added four times the value of the standard deviation to the
average. They proposed that if a source is found at levels above this value, termed the
minimum detectable percentage (MDP), it can be reasonably assumed that this is not the
result of misclassification of other sources and therefore is present in the watershed. As
an example, Wiggins at al. (2003) noted that even with an ARCC of 57% and an MDP of
25%, a library has the ability to identify sources that are present at average levels above
25%. Because the goal of microbial source tracking is to identify the major sources of

pollution, this level of detection may be adequate in many cases.

Discriminant Analysis: Library Size and Representativeness
The size of the library may be critical to the success of library-based bacterial
source tracking methods in predicting sources of fecal contamination (Whitlock et al.,
2002; Wiggins, 2003; Harwood et al., 2000). Library sizes for antimicrobial resistance
bacterial source tracking studies have ranged from as few as 166 total known source
isolates (Geary and Davies, 2003) to over 4,000 (Harwood et al., 2000). A few hundred
isolates fi'om each identified source has been suggested as the number necessary to

provide adequate discrimination between source isolates (Hagedom et al., 1999; Wiggins

24

et al., 2003). However, Carroll et al. (2005) demonstrated that a smaller source library,
consisting of a total of 717 isolates from human, domestic animals, livestock, and
wildlife, was sufficient for distinguishing human versus nonhuman sources in a relatively
small geographical area.

To test whether a library was large enough to avoid random grouping based on
stochastic processes rather than true relationships, Whitlock et al. (2002) randomly
assigned isolates to source categories and hypothesized that the ARCC for the randomly
assigned data set should approximate the probability that an isolate would be assigned to
a category by chance. For their library of 2,398 isolates, only negligible random
grouping appeared to be occurring, as demonstrated by ARCC values of 27.9% and
28.9%, when the probability that any one isolate would be assigned to one of four source
categories by chance was 25%. To demonstrate further the effectiveness of their library,
isolates were collected fiom around a leaking septic system. As expected, the majority of
isolates from these samples were classified as human.

It many cases, smaller libraries yield higher ARCC values (Wiggins et al., 1999;
Harwood et al., 2000), but these values are likely biased due to the small library size.
Harwood et a1. (2000) found a significant negative correlation between the number of
sampling events and the percentage of correctly classified isolates, leading to the
suggestion that higher correct classification rates may be due to the relative homogeneity
of antimicrobial resistance profiles from individual animal populations. In a study
designed to compare libraries of different sizes, Wiggins (2003) showed that although

discriminant functions based on small libraries tended to have higher ARCC rates, they

25

were less able to correctly classify non-library isolates than functions based on larger
libraries.

In evaluating library size, the most important consideration is whether or not a
library is representative of the resistance profiles of known sources. Hagedom et al.
(1999) suggested that one way to determine if a library is representative is to regularly
add samples of known source isolates to an existing library. If the individual correct
classification and the ARCC do not change significantly with the addition of these new
samples, then it can be assumed that the library is representative (Hagedom et al., 1999).
This method can also be used to test if the representativeness of a library changes over

time.

Discriminant Analysis: Choice and Grouping of Sources

In order for a library-based bacterial source tracking study to be applicable, the
known source library must contain isolates from all possible sources. Common sources
include sewage, septic tanks, dogs, cats, different livestock species, and wildlife. These
sources, however, may vary depending on the size, geography, and land-use of the area
being studied. Furthermore, the way in which known-sources are grouped may depend
on the purpose of the study and the questions being asked.

Bacterial source tracking studies over wide geographical areas or those that focus
on mixed-use watersheds need to include a greater number of potential sources. These
sources may be grouped into broad categories, or they may be considered individually.
In general, grouping sources into broader categories will result in higher ARCC values,

as there are fewer possible categories for an isolate to be classified into. For example, in

26

one of the first studies to use discriminant analysis of antimicrobial resistance analysis for
microbial source tracking samples from beef cattle feces, dairy cattle feces, turkey feces,
chicken feces, sewage influent, and pristine streams were used to classify unknown
isolates from polluted streams (Wiggins et al., 1996). A total of 1435 isolates from 17
samples were tested, and discriminant analysis resulted in an ARCC of 74% when all six
sources were considered individually. When samples were grouped into four sources,
cattle, poultry, sewage, and wildlife (pristine stream samples), the ARCC increased to
84%. When samples were grouped into two categories, either human or animal, the
ARCC was 95%.

If the primary question is whether or not water is being contaminated with human
fecal material, samples may be classified as being from human or non-human (animal)
sources. Advantages include a higher ARCC and a potentially more accurate
classification of isolates into these two source categories. This technique may be used to
confirm either the presence or absence of human fecal material, and the associated health
risks. Although human fecal contamination of water has been associated with a greater
incidence of disease, there are risks associated with animal feces as well. The ideal study
would differentiate isolates from humans and specific animal species, but limited
resources or specific objectives may favor broad groupings, such as human versus animal
sources.

Studies that primarily differentiate human fiom non-human fecal contamination
are presented by Burnes (2003) and Carrol et al. (2005). Burnes (2003) used antibiotic
resistance and discriminant analysis to differentiate human fiom non-human sources of

fecal contamination in a mixed-use watershed. A library of 1125 fecal coliform isolates

27

from sewage influent, dogs, cats, cattle, horses, chickens, and wild animals was created.
The ARCC for human versus non-human isolates was 94%, and the resulting
discriminant function was used to classify 400 surface water isolates as originating from
either human or non-human sources. Similarly, Carrol et al. (2005) used a library of 717
known source isolates from human, domesticated animal, livestock, and wild sources to
classify surface water isolates. Although the error of classification for four source groups
was considered too high to be useful, the ARCC value was 93.8% for human versus
animal sources. In this case, the relatively small library size may have limited the correct
classification rates when samples were grouped such that there were more than two
source categories.

If the desired outcome is to determine specific source species, one approach may
be to create multiple sub-libraries. For example, Crozier et al. (2002) used antibiotic
resistance analysis to determine origins of fecal bacteria in an impaired segment of the
Roanake River, in Roanoke County, Virginia. A library of 1562 isolates was created
fi‘om eight known sources, including horse, human, raccoon, sheep, chicken, cow, white-
tailed deer, Canada goose, and muskrat. Rather than immediately classifying unknown
river isolates into one of these eight categories, they proposed first classifying isolates
using a classification rule that discriminates between human, livestock, and wildlife
isolates. Non-human isolates were then further classified using either a four-way
firnction that discriminated between chicken, cow, horse, and sheep or a three-way
function that discriminated between deer, goose, and raccoon. They propose that the
creation of sub-libraries may be a better approach because it controls the number of

possible sources and therefore potentially reduces misclassification. The ARCC values

28

for a three-way wildlife function and an eight way all source function were 92% and
73%, respectively.

When a study is conducted within a specific watershed or defined area, the
sources may be very specifically tailored to that area. For example, Choi et al. (2003)
tested enterococci from potential sources of fecal bacteria at Huntington Beach,
California, a popular surfing spot that has been closed at times due to high fecal bacterial
counts. Sources included bird feces, urban runoff, coastal marsh sediment, and sewage
influent. This is one of the few studies that defines known sources to a specific area not
as species, but as routes, even though these routes may contain fecal bacteria from a
mixture of species. In fact, low classification rates for sewage was believed to be due to
its composition of bacteria from a mixture of sources. Nevertheless, classification rates
for bird feces and urban runoff were over 80%, and the classification rule was considered
useful for classifying unknown seawater isolates. This approach highlights the need to
consider applications of bacterial source tracking when choosing sources, as in this case it

is likely to be the routes of contamination, rather than species, that are managed.

Discriminant Analysis: Classification of Unknown Isolates
The purpose of microbial source tracking is to identify sources so that
interventions can reduce fecal contamination. Perhaps the strongest evidence for the
benefits of antimicrobial resistance analysis as a means of bacterial source tracking
comes from the analysis of sources of fecal pollution in Page Brook watershed in
Virginia (Hagedom et al., 1999). Discriminant analysis indicated that the majority of

isolates fiom the river were of cattle origin, and human sources were not implicated.

29

Based on the result of this study, portions of the stream were fenced to reduce livestock
access. This resulted in a decrease in fecal coliform numbers by 94%, which is direct
evidence of the importance and applicability of microbial source tracking studies.

Other studies have classified unknown isolates into sources that are consistent
with land-use, further validating this method as a microbial source-tracking tool. For
example, Whitlock et al. (2002) used discriminant analysis of antimicrobial resistance
profiles of fecal coliforms to estimate the major sources of fecal pollution into Stevenson
Creek, in southwest Florida. Significant percentages of fecal coliforms were classified as
originating fiom humans, wild animals, or dogs, but not cattle. This fits with land-use
patterns, as Stevenson creek runs through primarily urban and rarely agricultural lands .
The site most affected by dog feces, according to the analysis, was adjacent to a city park
where dog feces had been observed (Whitlock et al., 2002).

Another example of consistency between land-use and classification of unknown
isolates is given by Graves et al. (2002). This study used a library of 1174 known source
Enterococcus isolates to estimate a10% human, 40% wildlife, and 50% livestock
contribution to fecal contamination of Spout Run stream in Virginia. Human sources
contributed to fecal contamination at and to a lesser extent downstream from a
community with known failing septic systems, but not upstream of this community.

Many bacterial source-tracking studies suggest that multiple sources are
contributing to the fecal contamination of surface water in a given area. Geary and
Davies (2003) used discriminant analysis of antibiotic resistance profiles to examine
sources of fecal contamination into an important shellfish-producing estuary in Australia,

and they found sources of fecal contamination to the oyster beds was a relatively equal

30

mix from septic tanks, chicken, beef, and dairy farms. Another example is given by Choi
et al. (2003), who showed that some days bird feces was the major contributor to fecal
enterococci at Huntington Beach, California, and salt marsh and sewage plume were the
major contributors on other days.

Studies have found differences in sources of fecal contamination with variations
in environmental conditions, such as rainfall (Whitlock et al., 2002; Burnes 2003; Carrol
et al., 2005; Shehane et al., 2005) and temperature (Booth et al., 2003). Whitlock et al.
(2002) and Carrol et al. (2005) found that classification of surface water isolates revealed
a higher percentage from human sources during low rainfall conditions, despite overall
higher fecal counts during high rainfall conditions. These studies suggested the presence
of chronic low levels of human pollution in these areas, which was diluted with animal
sources during periods of heavy rainfall. In contrast, Shehane et al. (2005) found that in
an urban Florida watershed fecal coliforms from wild sources dominated during a
drought, but the relative frequency of fecal coliforms from human sources increased after
rainfall increased to normal levels. Booth et al. (2003), found a greater percentage of
human sources during cooler months, and suggested that seasonal conditions such as a
high water table and less evaporation of effluent could contribute to more septic drain

field failures and an increase in human loading rates during these months.

Conclusion
Discriminant analysis of antimicrobial resistance profiles of fecal indicator
bacteria is a promising technique for identifying sources of fecal contamination. Recent

studies have used these techniques to differentiate sources and groups of sources, as well

31

as to classify isolates from unknown-sources. The accuracy of these techniques,
measured by a variety of methods, varies greatly between and within studies. Factors that
may affect the accuracy of a discriminant function and the ability of the resulting
classification rule to classify isolates include the choice of indicator bacteria, the choice
of antimicrobial agents, the size of the known-source database, potential sources of fecal
contamination within a given geographical area, and the way in which potential sources
are grouped for analysis. Nevertheless, associations between indicated sources and land
use patterns, as well as reduction of contamination when indicated sources are eliminated,
demonstrate the applicability of these methods. Future directions may include
combinations of antimicrobial resistance analysis and genetic source tracking methods.
Overall, discriminant analysis of antimicrobial resistance profiles may be a relatively
quick and inexpensive method of indicating major sources of fecal contamination over

small or large geographical areas.

32

Chapter II: Antimicrobial Resistance of Fecal Escherichia coli isolates from Septic
and Sewage Sources, Companion Animals, Livestock, Free-Ranging Wildlife, and

Surface Water.

Abstract

Antimicrobial resistance in both pathogenic and nonpathogenic bacteria isolated
fi'om humans and animals is a significant public health concern, and the health and
environmental effects of the spread of resistant bacteria via fecal contamination of
surface water are largely unknown. The purpose of this study was to measure resistance
in E. coli bacteria from a variety of sources in a central Michigan watershed to assess the
prevalence of resistant and multi-drug resistant fecal E. coli and to determine if E. coli
fi'om various sources can be differentiated based on their antimicrobial resistance
profiles. Sensitivity testing against 16 common antimicrobial agents was done on fecal
E. coli from human sources (sewage and septic tank samples) companion animals (dogs
and cats), livestock (beef cattle, dairy cattle, swine, and sheep), free ranging wildlife
(deer, ducks, and geese), and surface water (The Red Cedar River). Overall, isolates
resistant to one or more or two or more antimicrobials were more commonly from
livestock, but isolates resistant to five or more antimicrobials were more commonly from
human (septic or sewage) sources. Resistant and multi-drug resistant E. coli were
isolated fi'om wildlife feces, although resistance was less common as compared with
other species and river samples. Compared with other groups, more livestock isolates
were resistant to Chloramphenicol, gentamycin, kanamycin, streptomycin,

sulfrnethoxazole, and tetracycline, and more human and pet isolates were resistant to

33

amoxicillin/clavulanic acid and ampicillin. Human sources also had the greatest
percentage of isolates resistant to nalidixic acid. MIC values may be a more sensitive
measure of differences between fecal sources, as significant differences in MIC values
between one or more species group were present for 14 of 16 antimicrobials and
significant differences in resistance and susceptible categorical classifications were

present for only 10 of 16 antimicrobials.

Introduction

Corresponding to the increase in the use of antimicrobials in human medicine,
veterinary medicine, and animal production, there has been an increase in the prevalence
of resistant bacteria over the past several decades (Prescott and Baggott, 1993; Houndt
and Ochman, 2000; Hughes and Datta, 1983). The emerging problem of antimicrobial
resistance is significant because resistance compromises our ability to treat bacterial
infections successfully (Witte, 1998; Jones et al., 1997; Neu, 1992). Normal intestinal
flora, such as Escherichia coli, are exposed to a range of antimicrobial agents, which may
select for the survival of resistant bacteria. Bacteria can become resistant to
antimicrobials via gene mutation, vertical transfer of resistance-encoding genes, or gene
transfer via mobile elements, such as plasmids, transposons, and integrons (Prescott and
Baggot, 1993). Resistant E. coli bacteria are of concern because they may cause disease
(Schroeder et. al., 2002) and because resistance genes may be transferred from E. coli to
other pathogenic bacteria (Witte, 1998; Neu,1992). Little is known about the
environmental and health impacts of the dissemination of resistant bacteria into the

environment through manure, sewage, or other fecal contamination.

34

Differences in antimicrobial use in human medicine, veterinary medicine, and
animal production result in differences in selective pressures on intestinal bacterial flora.
Low levels of certain antimicrobial agents, such as tetracyclines, are used extensively as
feed additives for growth promotion in animal agriculture (Prescott and Baggot, 1993).
In contrast, novel and more expensive antibiotics are often reserved for used in human
medicine. Because of these differences in the types, frequencies, and concentrations of
antimicrobials used, one would expect fecal bacterial resistance to differ between these
groups. Investigators have used antimicrobial resistance profiles to differentiate between
combinations of human, livestock, companion animal, and wildlife fecal bacteria
(Burnes, 2003; Geary, 2003; Guan et al., 2002; Whitlock et al., 2002; Parveen et al.,
1997; Wiggens et al., 1996).

In general, studies have shown a low prevalence of resistant bacteria in wildlife
(Burnes, 2003; Guan et al., 2002; Routman et al., 1985). Exposure of wild animals to
antibiotics is minimal, and thus selection for resistant bacteria should be lower than in
humans and domesticated animals. However, it is possible for wildlife to come in contact
with feeds containing antibiotics (Livermore, 2001), antibiotic residues (Kumar et al.,
2005), or resistant bacteria in the environment (Selvaratnam and Kunberger, 2004;
Fallacara, 2001). Routman et al. (1985) demonstrated more resistance in E. coli from
yellow baboons that frequented a human refuse pit in comparison with more isolatw
populations. Another study found that 95% of E. coli isolates from migratory Canada
Geese were resistant to penicillan G, ampicillan, cephalothin, and sulfathiazole

(Middleton and Ambrose, 2005). The presence of resistant bacteria in wildlife is of

35

concern to human health, as wild animals may become a reservoir of resistant bacteria
that could then be spread to humans (Cole et al., 2005).

E. coli bacteria found in the environment are presumed to have originated from
animal sources. (Desmarais et al., 2002; USEPA, 2001; Edberg et al., 2000) Because the
original source of feces may vary, it is possible to find a range of resistance in E. coli
collected from environmental sources. Selvaratnam and Kunberger (2004) found an
increased frequency of resistant bacteria in a creek adjacent to a sewage treated field, in
comparison with upstream and downstream sites. Other recent studies have shown that
the antimicrobial resistance profiles of fecal indicator bacteria can be used to predict the
sources of origin of these bacteria (Guan et al., 2002; Graves et al., 2002; Whitlock et al.,
2002; Harwood et al., 2000; Parveen et al., 1997). It has been suggested that the
accuracy of these bacterial source tracking efforts is highly dependent on the existence of
a library of isolates that represents the diversity of resistance in all possible sources
(Johnson et al., 2004; Wiggins et al., 2003).

Endeavors to monitor resistance in fecal E. coli have been undertaken for a
variety of reasons, including bacterial source tracking (Carroll et al., 2005; Webster et al.,
2004; Burnes, 2003; Guan et al., 2002; Whitlock et al., 2002) and assessing the impacts
of antimicrobial use (Hoyle et al., 2005; Middleton and Ambrose, 2005; Adetosoye,
1980). However, relatively few studies compare resistance and MIC values of E. coli
fi‘om a broad range of fecal sources within a given area. The purpose of this study was
to compare the antimicrobial resistance of E. coli bacteria from humans (sewage and
septic samples), livestock (beef cattle, dairy cattle, swine, and sheep), pets (dogs and

cats), free-ranging wildlife (deer, ducks, and geese), and surface water, in a central

36

Michigan watershed in order to 1) assess the prevalence of resistant and multi-resistant
bacteria in different species and the environment, 2) determine if E. coli from various
sources can be differentiated based on their antimicrobial resistance profiles, and 3)
prepare a library of known source isolates for firture bacterial source tracking studies.
We hypothesized that there would be differences in antimicrobial resistance of fecal E.
coli from wildlife, humans, livestock, and companion animals. It was expected that E.
coli isolates from free-ranging deer, geese, and ducks would be resistant to fewer
antibiotics than E. coli isolates from humans and domestic animals. E. coli isolates from
the Red Cedar River should exhibit a variety of resistance profiles, depending on the

original fecal source of these bacteria.

Materials and Methods
Study Site:
All samples were collected within or near the Red Cedar watershed, a 1,186 km2
area in Central Michigan. The Red Cedar River arises in Cedar Lake and flows
northwest approximately 73 km, where it joins the Grand River in East Lansing,

Michigan.

Sample Collection:

Samples were obtained fi'om the feces of free-ranging deer, ducks, and geese from
May-September, 2004. Deer feces were found by searching wooded or agricultural lands
within or near the Red Cedar Watershed. 95 deer fecal samples were collected from 15

locations. In addition, fecal material was collected from 10 fresh road-kill deer, for a

37

total of 105 deer fecal samples. A total of 108 goose fecal samples were collected from
17 different locations. An additional 19 banked goose samples, collected in 2003 for a
previous study within the Red Cedar watershed, were used. A total of 97 wild mallard
duck fecal samples were collected from 15 different locations.

Dog and cat samples were collected between May-December, 2004. Students at
the Michigan State University College of Veterinary Medicine volunteered to collect
fresh droppings from their animals, as well as friends' and neighbors' animals. Feces
were also collected from three local animal shelters. In total, 100 dog fecal samples and
105 cat fecal samples were collected.

A total of 125 separate septic tank samples were collected from May-September,
2004, with the help of seven different local septic tank pumping companies. Material
was sampled either directly from the tank or from the truck soon after it was pumped
from the tank. Sewage samples were collected from the inflow station at a local sewage
treatment plant, before the addition of any chemicals. A total of 157 samples were
collected on 17 different days over one year (September 2004- October 2005).

Banked E. coli isolates from livestock fecal samples collected during 2003 for a
previous project within the Red Cedar watershed were used (Sayah et al., 2005). Banked
isolates were from 31 beef cattle, 53 dairy cattle, 33 sheep, and 31 swine fecal samples.
An additional 66 beef cattle, 53 dairy cattle, 72 swine, and 24 sheep samples were
collected between January of 2005 and May of 2006, either from fresh fecal piles or
directly from animals.

The Red Cedar River was sampled weekly at six locations, three on the MSU

campus, two other East Lansing locations, and one Williamston location, between May-

38

December, 2004. Enumeration of fecal coliforms was done at the Michigan Department
of Environment Quality (MDEQ), and these plates were then brought to our lab for
confirmation of E. coli. In total, coliform plates from 339 separate river samples were

obtained.

Isolation of E. coli Bacteria:

Fecal samples from wildlife, livestock, and domestic animals, septic and sewage
samples, and isolates from the MDEQ coliform plates were streaked on MacConkey Agar
and incubated at 37 degrees Celsius (°C) for 18 to 24 hours. Three separate lactose and
bile salt precipitate positive colonies were selected for further confirmation. Biochemical
confirmation of E. coli bacteria was done using triple sugar iron, urea, oxidase, citrate,
methyl red, Voges-Proskauer, and indole tests. If one or more of these tests were
inconsistent with typical E. coli, a second attempt was made by selecting a different
colony from the original MacConkey plate or re-streaking the sample onto a new
MacConkey plate. One confirmed E. coli isolate per fecal sample was selected for

antimicrobial susceptibility testing.

Antimicrobial Susceptibility Testing:

Antimicrobial susceptibility testing was done on one isolate from each sample
using a Sensititre AutoInoculator, AutoReader, and the CMV7CNCD antimicrobial panel
(TREK Diagnostic Systems, Inc., Cleveland, Ohio). The CMV7CNCD Sensititre panel is

a 96-well plate containing cefoxitin at 0.5-16ug/mL, arnikacin at 05-4 ug/mL,

Chloramphenicol at 2-32 ug/mL, tetracycline at 4-32 ug/mL, cefiriaxone at 025-64

39

ug/mL, amoxicillin/clawlanic acid at 1/0.5-32/ 16 ug/mL, ciprofloxacin at 0.015-4
ug/mL, gentamicin at 025-16 ug/mL, nalidixic acid at 0.5-32 ug/mL, ceftiofur at 0.12-8
ug/mL, sulfamethoxazole at 16-512 ug/mL, trimethoprirn/sulfamethoxazole at 0.12/2.38-
4/76 ug/mL, cephalothin at 2-32 ug/mL, kanamycin at 8-64 ug/mL, ampicillin at 1-32
ug/mL, and streptomycin at 32-64 ug/mL (Table 2.1). This panel was selected because it
contains antibiotics used in human and animal medicine that normally have activity

1 against E. coli bacteria. Each plate was inoculated with 50 uL of a broth suspension
(made by suspending single isolate E. coli colonies in demineralized water to a density
equivalent to a 0.5 McFarland standard, and then mixing 10 uL of this suspension in 10
mL of Sensititre Mueller-Hinton broth). Plates were incubated at 34-36 °C for 18-24
hours and then read by the AutoReader, which uses florescence to detect growth at each
dilution. Minimum inhibitory concentrations (MIC) were recorded, and categorical
designations of resistant (R), susceptible (S), and intermediate (I) were made based on
breakpoint values set by the Clinical and Laboratory Standards Institute (CLSI). No
CLSI breakpoint value has been established for arnikacin, so isolates that were resistant at

the highest concentration tested (4 rig/ml) were considered resistant.

Statistical Analysis:

All statistical tests were done using SAS software, version 9.1. E. coli sources
were classified as human (sewage and septic), livestock (beef cattle, dairy cattle, swine,
and sheep), pets (dogs and cats), wildlife (deer, ducks, and geese), or river. Percentages
of isolates from each species group and the river resistant or intermediate to one or mere,

two or more, and five or more antibiotics were graphed in order to compare the

40

prevalence of resistant and multi-drug resistant isolates among the groups (Figures 2.1-
2.3). Separate chi-square tests were done for each antibiotic to test for significant
differences in resistance among isolates fi'om human, livestock, pet, and wildlife sources.
For the purpose of these analyses, resistant and intermediate isolates were grouped
together. For antibiotics for which the chi-square test was not valid (expected values <5),
F isher's exact tests were performed. Differences among specific species within groups
were also tested using chi-square tests.

For the majority of antimicrobials, <1 % of isolates fell into the category of
intermediate resistance. However, 39.4% of isolates were interrnediately resistant to
cephalothin, and for this reason a separate chi-square analysis was done for cephalothin,
comparing numbers of resistant, intermediate, and sensitive isolates among species
groups.

The percentages of isolates from the four species groups and the river with each
MIC value were graphed separately for each of the 16 antimicrobials (Figures 2.3-2.19),
and a Kruskal-Wallis test was performed for each antibiotic to test for significant

differences in MIC values between the four species groups.

Results
Isolation of E. coli:
E. coli bacteria was isolated and confirmed from 103/105 deer, 119/127 goose,

93/97 duck, 95/ 100 deg, 95/105 cat, 111/125 septic, 109/157 sewage, and 330/339 river

samples. E. coli was isolated and confirmed from all of the newly collected livestock

41

samples and confirmed from 148 banked livestock isolates, resulting in confirmed

isolates from a total of 103 beef cattle, 107 dairy cattle, 98 swine, and 55 sheep samples.

Antimicrobial Resistance:

In all, 781 of the 1423 isolates tested (55%) exhibited resistance (defined as
'resistant' or 'intermediate', based on CLSI standards) to one or more antibiotics.
Livestock samples showed the greatest percentage of isolates resistant or intermediate to
one or more antibiotics (64%), followed by river (58%), human (52%), pets (50%), deer
(47%) and waterfowl (46%).

Out of the 781 isolates exhibiting resistance, 217 isolates (15% of total) were
resistant or intermediate to two or more antibiotics. Livestock isolates had the greatest
prevalence of resistance to two or more (31%) (Figure 2.2) antimicrobials, whereas
human isolates had the greatest prevalence of resistance to five or more antimicrobials
(5.0 %) (Figure 2.3). Wildlife had the lowest percentage of isolates resistant to multiple
antibiotics, with 6.5% resistant to two or more antibiotics and 1.0% resistant to five or
more antibiotics. Multiple resistant river isolates were also relatively rare, with 9.1 %
resistant to two or more antibiotics and 1.5% resistant to five or more antibiotics.
Common combinations of two, three, four, five, and six antimicrobials to which isolates
were resistant to are listed in Table 2.2.

There was at least one human isolate resistant to each of the 16 antibiotics tested,
15/ 16 for pets, 14/16 for livestock, 13/ 16 for river, 12/16 for wildlife. Percentages of
resistant isolates within each species and species group are presented in Table 2.3

(Antimicrobials for which significant differences between species groups were noted) and

42

Table 2.4 (Antimicrobials for which no significant differences between species groups
were noted).

Chi-square (X2) or fisher’s exact tests revealed significant (p<0.05) differences
in resistance between species groups (human, livestock, pets, and wildlife) for 10 out of
the 16 antimicrobials tested (Table 2.3). Resistance to Chloramphenicol (X2=22.33,
degrees of freedom (d.f.)=3, p<0.0001), gentamicin (Fisher’s exact, p=0.0257),
kanamycin (X2=25.17, d.f.=3, p<0.0001), streptomycin (X2=88.25, d.f.=3, p<0.0001),
sulfarnethoxine (X2=80.03, d.f.=3, p<0.0001), and tetracycline (X2=219.54, d.f.=3,
p=0.0001) were all highest in isolates from livestock Resistance to ampicillin
(X2=51.00, d.f.=3, p<0.0001) and amoxicillin/clavulanic acid (X2=5.49, =3, p=0.0003)
were most common in pet and human isolates, and nalidixic acid (Fisher’s exact,
p<0.0001) resistance was most common in human isolates. Interestingly, when separate
chi-square tests were done comparing isolates from human and pet sources, nalidixic acid
was the only antibiotic for which there was a significant difference (X2=14.45, d.f.=1 ,
p=0.0001) between these two groups. 8.6% of human isolates and only 0.5% of pet
isolates were resistant to nalidixic acid.

When both ‘intermediate’ and ‘resistant’ isolates were considered resistant,
cephalothin was by far the antibiotic to which resistance was most common, with 53%
river, 46% pets, 45% deer, 45% waterfowl, 44% human, and 36% livestock being
resistant (Table 2.2). However, when only ‘resistant’ isolates were considered resistant,
9.5% human, 6.8% deer, 6.7% river, 4.3% waterfowl, 3.2% pets, and 3.0 % livestock
isolates were resistant to cephalothin. Differences in resistance to cephalothin among the

four species groups were significant both when comparing susceptible vs. resistant

43

(X2=7.87, d.f.=3, p=0.0487) and susceptible, resistant, and intermediate (X2=20.24,
d.f.=6, p=0.0025) isolates.

No significant differences among the four species groups were found for
resistance to arnikacin, cefliofur, ceftriaxone, ciprofloxacin, cefoxitin, or

trimethoprim/sulfamethoxazole (Table 2.4).

Species Comparisons Within Species Groups:

A higher percentage of sewage compared with septage isolates were resistant to
ampicillin (x2: 4.85, d.f.=1, p=0.0277) and kanamycin (x2=5.21, d.f.=1, p=0.0225).
There were no significant differences (p<0.05) between septic and sewage samples for
any of the other antibiotics tested.

Between pet species, there were marginally more cat isolates resistant to
ampicillin (X2=4.74, d.f.=l , p=0.0295) and there were more dog isolates resistant to
cephalothin (X2=7.65, d.f.=1, p=0.0057). No significant (p<0.05) differences between
resistance in dog and cat isolates were present for other antibiotics.

When comparing the four livestock species, there significantly more swine
isolates to resistant to ampicillin (X2=35.57, d.f.=3, p=<0.0001), kanamycin (X2=10.00,
d.f.=3, p=0.0186), streptomycin (x2=29.2o, d.f.=3, p<0.0001), sulfrnethoxazole
(x2=40.22, d.f.=3, p<0.0001), and tetracycline (x2=77.54, d.f.=3, p<0.0001). Although
percentages of resistant isolates were similar in samples from beef and dairy cattle for
most antibiotics, more beef cattle isolates than dairy cattle isolates were resistant to
sulfinethoxazole (X2: 7.22, d.f.=1, p=0.0072) and tetracycline (X2=8.46, d.f.=l

p=0.0036).

44

No significant (p<0.05) differences in resistance were present between wild deer,

geese, and ducks.

Minimum Inhibitory Concentrations:

Figures 2.4-2.19 represent the percent of isolates from each species group that
have an MIC at each concentration of each the antibiotics tested. Significant differences
(Kruskal-Wallis, p<0.05) were found in MIC values among the four species groups
(human, livestock, pets, and wildlife) for arnikacin (X2=87.93, d.f.=3, p<0.0001),
amoxicillin/clawlanic acid (X2=15.76, d.f.=3, p=0.0013), ampicillin (X2=15.03, d.f.=3,
p=0.0018), ceftiofur (x2=21.55, d.f.=3, p<0.0001), cephalothin (X2=8.77, d.f.=3,
p=0.0325), chlorarnphenicol (X2=13.96, d.f.=3, p=0.003), ciprofloxacin (X2=13.21,
d.f.=3, p=0.004), cefoxitin (X2=8.96, d.f.=3, p=0.0299), gentamicin (x2=22, d.f.=3,
p<=0.0001), kanamycin (X2=20.96, d.f.=3, p=0.0001), streptomycin (X2=88.17, d.f.=3,
p<0.0001), sulfamethoxazole (x2=75.92, d.f.=3, p<0.0001), tetracycline (x2=219.39,
d.f.=3, p<0.0001), and trimethoprim/sulfamethoxazole (X2=55.46, d.f.= 3, p<0.0001).
No significant differences were found among species groups for nalidixic acid (X2=6.61,

d.f.=3, p=0.0854) and cefiriaxone (x2=1.0112, d.f.=3, p=0.7985).

Discussion
The percentages of E. coli resistant to certain antimicrobials noted in this study
are similar to findings from other studies. Guan et a1. (2002) also found a higher
percentage of tetracycline- and streptomycin-resistant E. coli in livestock feces and a

higher percentage of ampicillin-resistant E. coli from human sources, as compared with

45

other species. The low prevalences of isolates resistant to cef’tiofur and gentamicin seen
here are also similar to those noted by Guan et al. (2002). The 5.4% prevalence or
resistance to Chloramphenicol in livestock isolates was greater than the 1.2% noted by
Guan et al. (2002) but lower than the 24% prevalence of resistance to Chloramphenicol in
healthy finisher pigs noted by Boerlin et al. (2005). Siegel et al. (1975) found resistance
to ampicillin from human samples ranged from 10 to 37%, which is comparable to the
prevalence of ampicillin resistance noted in this study. Similarities in species—specific
resistance between studies are likely due to similarities in trends of antibiotics usage
across different regions. ‘
Although a large percentage of isolates were classified as not susceptible to
cephalothin, the majority of these were actually intermediate, rather than resistant based
on CLSI breakpoints. Cephalothin was also the only antibiotic for which resistant or
intermediate classifications were most common in river isolates. Cephalosporins are
derived from natural substances, and it is possible that the widespread resistance to
cephalothin may be due in part to intrinsic resistance of E. coli, rather than selection for
resistance due to anthropogenic antimicrobial use. As a first-generation cephalosporin,
cephalothin is considered to have variable activity against gram-negative bacteria, as
compared with second- and third-generation cephalosporans. Notably, resistant isolates
to cefoxitin, cefiiofur, and ceflriaxone were rare across all species. Although it is likely
that intrinsic resistance of E. coli to cephalothin may be a complicating factor, the
significant differences in cephalothin resistance between species suggests that
antimicrobial use is also contributing to the cephalothin resistance noted in this study.

Antimicrobial resistance to cephalothin may also be perpetuated by use of other beta-

46

lactam antibiotics, as some of the beta-lactamase resistant factors produced by gram-
negative bacteria have been shown to be active against both penicillans and
cephalosporins (Richmond et al., 1971)

A higher percentage of livestock isolates were resistant to 6 out of 16 of the
antibiotics tested, as compared with isolates from human, pet, and wildlife sources.
Within livestock, there were more swine isolates resistant to each antibiotic tested, with
the exception of arnikacin, cephalothin, cefoxitin, and gentamicin. An overwhehning
majority (71%) of swine isolates were resistant to tetracycline. Furthermore, whereas
human isolates showed a greater percentage of resistance to ampicillin than livestock
overall, swine isolates actually showed the greatest percentage of resistance to ampicillin
(25%). Recent studies have shown a similarly high prevalence of resistance in isolates
from swine feces (Sayah et al., 2005; Schroeder et al., 2002a, Schroeder et. al., 2002b).
This is likely due to the move towards large, intensive confinement rearing systems and
the subsequent treatment of large populations simultaneously through medicated feed.
The use of low-level antibiotics in feed or water for growth promotion is common in the
swine industry. Prescott and Baggot (1993) estimate that 45% of pigs in the United
States are fed combinations of chlortetracycline and penicillin until they are 35 kg of
body weight and 40% are fed one or more antibiotics through the finishing period.
Frequent exposure to these antibiotics is likely the reason for the high percentage of
resistant isolates from swine feces.

There were a greater percentage of resistant isolates from beef cattle as compared
with dairy cattle for 12 out of the 16 antimicrobials tested, although this difference was

only significant for sulfamethoxazole and tetracycline. A previous survey of

47

antimicrobial treatment on farms in the Red Cedar watershed noted that the use of
chlortetracycline, sulfamethoxazole, tilmicosin, and enrofloxacin is more common in beef
than dairy cattle (Sayah et al., 2005). A greater percentage of dairy cattle isolates were
resistant to 3 out of 4 cephalosporines tested (cetiofur, cefiriaxone, and cephalexin),
although these differences were not significant. This trend may be due to the greater use
of ceftiofur on dairy farms as compared with beef farms (Sayah et al., 2005).
Interestingly, there were isolates resistant to Chloramphenicol from all four livestock
species, even though chloramphenical has been banned from use in food animals in this
country since the 19803 (Prescott and Baggot, 1993). It has been suggested that
Chloramphenicol-resistance genes may be maintained in E. coli by virtue of linkage to
other genes encoding resistance to antimicrobials still in use in food animal medicine
(Bischoff et al., 2005). Resistance in bacteria from livestock is cause for concern
because fecal bacteria are being released into the environment in large quantities and
because there is potential for resistance factors or resistant bacteria to be passed to
humans through meat and dairy products.

Human sources, pet sources, or both, had the greatest percentage of isolates
resistant to 7 out of the 16 antibiotics tested, although these differences were only
significant for amoxiciliin/clavulanic acid, ampicillin, and nalidixic acid. Other than a
small percentage of resistance to nalidixic acid seen in river isolates, humans and pets
were the only source categories that exhibited resistance to either of the fluoroquinolones,
ciprofloxacin and nalidixic acid (Table 2.3). These agents are not commonly used in
livestock production, although enrofloxacin usage was reported in a small percentage of

beef cattle in the Red Cedar Watershed (Sayah et al., 2005).

48

There were no significant differences in resistance profiles between isolates from
human (sewage and septic tank) samples and isolates from pets (dogs and cats) for 15 out
of 16 antibiotics tested. This may be due to similarities in the types of antibiotics used in
human and companion animal medicine. Whereas a number of antibiotics have been
banned for use in food animals due to drug residue and resistance concerns, few
restrictions or guidelines exist for antibiotic use in companion animals (Prescott et al.,
2002). Additionally, the close proximity of humans and companion animals in the home
environment may result in exchange of bacteria and resistance factors between these
groups. It is also possible for companion animal feces to end up in septic tanks and sewer
systems, as owners may flush animal feces down the toilet. Thus, the noted similarity in
resistance profiles between E. coli fiom humans and companion animal may reflect a true
similarity in gut flora or it may be due to the contamination of septic and sewage material
with companion animal feces. Additional studies looking at differences in resistance in
isolates fi'om septic tanks of houses with pets and those without, as well as isolates
directly from the feces of pet-owners and non pet—owners would provide further insight
into the relationship between resistant fecal bacteria in humans and in companion
animals.

Among free ranging wildlife, duck isolates showed the most resistance, followed
by deer, then geese, although differences in resistance between wildlife species were not
significant. Due to the challenges of collecting fresh feces from wild ducks, almost all of
the sampling locations were areas where ducks congregate and are fed by humans. The
trend towards more resistant isolates from duck feces may be due to this human contact.

In addition, ducks feed and spend almost all their time in the river and thus may be the

49

most likely of the three wildlife species to acquire resistant bacteria from river fecal
contamination. Further studies comparing resistance profiles of bacteria from wild ducks
in remote versus more developed areas would give insight into whether human contact or
fecal contamination of surface water or both contribute to resistance in this species.

Resistant fecal E. coli from wildlife species were rare relative to other groups.
However, 46% of wildlife isolates were resistant or intermediate to one or more
antibiotics. It is likely that antibiotic resistance in bacteria from wildlife is related to the
increasing association of wild animals and human settlements. Gilliver et al. (1999)
found widespread resistance in Enterobacteriaceae from wild rodents in northwest
England. Conversely, in Finland Osterblad et al. (2001) found very little resistance in
fecal bacteria fiom wildlife that rarely had contact with humans or anthropogenic
antibiotics. Contrasting studies show increased (Rolland et al., 1985) and no difference
(Routman et al., 1985) in resistance of E. coli from yellow baboon populations more
closely associated with human settlements than more remote populations. Recently,
concerns have arisen that wildlife may become reservoirs of resistant bacteria (Middleton
and Ambrose, 2005; Cole et al., 2005). Conjugative exchange of antibiotic resistance
plasmids in E. coli from migratory waterfowl has been demonstrated (Tsubokurea et al.,
1995), and migratory species, such as geese and ducks, may have the potential to collect
and disseminate resistance factors over wide areas.

Although the overall resistance of river isolates was low when compared with
humans, livestock, and pets, resistance was more common in river samples than in
wildlife. Ash et al. (2002) noted that antibiotic resistant bacteria were widespread in

samples from US. rivers, and resistant bacteria in rivers have been linked to human,

50

livestock, and wildlife sources (Carroll et al., 2005; Burnes, 2003; Graves et al., 2002;
Whitlock et al., 2002; Hagedom et al., 1999). Relatively high percentages of resistance
to certain antibiotics suggest that some of the bacteria in the river are originating from
non-wildlife sources. For example, 7.9% of river isolates were resistant to tetracycline,
second only to livestock. This implies that either some of these E. coli originated from
livestock, or genes for tetracycline resistance somehow increase the ability of E. coli
bacteria to survive in the environment. Although there have been some studies
investigating whether resistance plasmids alter bacterial survival (Anderson et al., 1974;
Smith et al., 1974; Grabow et al., 1975 ), evidence is inconclusive and it is that this is the
cause of increased tetracycline resistance in river isolates in this case.

When MIC values were compared between species, significant differences were
found for 14 out of the 16 antibiotics tested, as compared with 10 out of 16 when just
looking at resistance versus susceptibility. This suggests that comparison of MIC values
may be a better way to differentiate between species based on antimicrobial resistance
profiles. One limiting factor in bacterial source tracking efforts has been the inability to
differentiate source groups adequately with discriminant analysis functions (Lasald et al.,
2005; Harwood et al., 2000; Wiggins et al., 1996). Wiggins et al. (1999) also suggested
that the ability of discriminant function analysis to separate and correctly classify known
source isolates may be increased by using a larger number of antibiotic concentrations.

Bacteria resistant to multiple antimicrobials are of special concern, due to the
underlying threat that certain pathogenic bacteria will become resistant to all currently
available antimicrobial agents. Multiple drug resistance may be due to a single

mechanism of resistance that is effective against several antibiotics, or it may be due to

51

multiple mechanisms of resistance. It is common for bacteria exhibiting resistance to one
member of a class of antibiotics to also be resistant to other members of that same class
(Prescott and Baggot, 1993). However, in our study, none of the five most common two-
drug resistant profiles, the four most common three-drug resistance profiles, the four
most common four-drug resistance profiles or the most common five-drug resistance
profile contained drugs within the same antimicrobial class (Table 2.3). The most
common multi-drug profiles all included resistance to tetracycline, which suggests that
plasmid mediated resistance to tetracycline is often linked to resistance to other
antimicrobials.

The greater the number of antimicrobials to which an isolate is resistant, the
greater the potential threat to human health. Thus, we are particularly concerned with
penta-resistance, or resistance to five or more antimicrobial agents. Whereas multiple
antimicrobial resistance was present in 15% of all isolates, less than 2.5% exhibited
penta-resistance. Interestingly, although livestock showed the greatest percentage of
resistance to two or more and three or more antimicrobials, penta-resistant isolates were
most commonly from human sources. This is significant because it suggests that while
there is more overall resistance in bacteria from livestock, the resistance profiles that are

of greatest concern to public health may be originating from humans.

Conclusions
In summary, there were more livestock isolated resistant to one or more and two
or more antibiotics, but the majority of isolates resistant to five or more antibiotics were

of septic or sewage origin. Resistance was rare is wildlife species as compared with other

52

groups, but should continue to be monitored due to the implications of the dissemination
of resistant bacteria by wildlife species. Resistance in river isolates was more common
than in wildlife, which suggests that resistant bacteria are being released into the
environment via non-wildlife fecal sources. Our study shows that analysis of MIC values
rather than simple ‘resistance’ versus ‘susceptible’ categorical classifications may be a
more sensitive measure of differences between sources of fecal bacteria. Differences in
MIC values of E. coli from different fecal sources may aid in low cost methods of

tracking sources of fecal contamination.

53

Table 2.1: Class, Dilution Range, and Resistance and Intermediate Breakpoints

 

 

of Antimicrobials.
Class and Antimicrobial Dilution Range Resistance Intermediate
(ug/ml) Breakpoint‘ Breakpoint‘
(us/ml) lug/m1)
Cephalosporins
Cephalothin 2-32 32 16
Cefoxitin 0.5-16 32b 16
Ceftriaxone 025-64 64 32
Ceftiofur 012-8 8 4
Aminoglycosides
Amikacin 0.5-4 >4°
Gentarnicin 025-16 16 8
Kanarnycin 8-64 64 32
Streptomycin 32-64 64 32d
Penicillins
Amoxycillin/Clavulanic Acid 1/0.5-32/ 16 32/16 16/8
Ampicillin 1-32 32 16
Sulfonamides and
Potentiated Sulfonamides
Sulfarnethoxazole 16-512 512 256
Trimethoprim/Sulfamethoxazole 0. 12/2.3 8-4/ 76 4 2
Quinolones and Flouroquinolones
Ciprofloxacin 0015-4 4 2
Nalidixic Acid 0.5-32 32
Phenicols
Chloramphenicol 2-32 32 16
Tetracycline 4-32 16 8

 

a. Breakpoints based on CLSI guidelines. b. If resistant to 16 ug/ml, then MIC must be 32
try/u). or greater, so isolates resistant to 16 rig/ml were considered resistant. c. There is no
CLSI guideline for arnikacin resistance, isolates resistant at 4 rig/ml were considered
resistant. d. If susceptible at 32 rig/ml then considered susceptible, actual MIC is lower than

32 rig/ml.

54

Figure 2.1: The Percentages of Isolates From Each Species Group Resistant or
Intermediate to One or More Antimicrobial Agent

 

0)

% of Isolates
O

N
O
1

 

5‘

888

_L
o o
L

 

 

HUMAN
n=220

LIVESTOCK

n=360

y .
/
//
/
/

PET
n=190

l Intermediate
l Resistant

7/ A

 

WILDLIFE
n=315

 

Figure 2.2: The Percentages of Isolates From Each Species Group Resistant or

Intermediate to Two or More Antimicrobial Agents.

 

% of Isolates

 

 

HUMAN
n=220

LNESTOCK
n=360

PET
n=190

WILDLIFE
n=315

l Intermediate
I Resistant

 

RNER
n=329

 

55

 

 

Figure 2.3: The Percentages of Isolates From Each Species Group Resistant or
Intermediate to Five or More Antimicrobial Agents.

 

I Intermediate
I Resistant

% of Isolates

 

 

HUMAN LN ESTOCK PET WILDLIFE RN ER
n=220 n=360 n=190 n=31 5 n=329

 

 

 

56

Table 2.2: Common Multiple Antimicrobial (MAR) Resistance Profiles.

 

Common MAR Profiles n % (ALL) % (MR)
TET SUL 85 6.1 40.7°
TET CEP 80 5.8 38.3 °
TET STR 76 5.5 36.4 a
AMP CEP 62 4.5 297°
STR SUL 60 4.3 28.7 a
TET 301. STR 55 4.0 44.4b
TET SUL CEP 36 2.6 29.0 b
TET SUL AMP 33 2.4 26.6 b
TET STR AMP 33 2.4 26.6 b
TET CEP AMP 32 2.3 25.8 b
CEP AMP AM/CL 32 2.3 25.8 b
TET SUL STR AMP 25 1.8 342°
TET SUL STR CEP 22 1.6 30.1 °
TET SUL AMP CEP 18 1.3 24.7 °
TET STR AMP CEP 16 1.2 21 .9°
TET SUL STR KAN 15 1.1 20.5°
TET AMP CEP AM/CL 15 1.1 20.5 °
TET SUL STR AMP CEP 12 0.9 35.3d
TET SUL STR AMP KAN 11 0.8 32.4d
TET SUL STR AMP TRI/SUL 10 0.7 29.4 d
TET SUL AMP CEP AM/CL 10 0.7 29.4 d
TET SUL STR AMP CEP AM/CL 7 0.5 583°

n refers to the number of isolates with that MAR profile, %(ALL) refers to the
percentage of all isolates with that MAR profile, %(MR) refers to the
percentage of all isolate with that MAR profile out of all isolates resistant to a:
> one antibiotic, b: >2 antibiotics, c: > 3 antibiotics, d: > 4 antibiotics, e: > 5
antibiotics. AM/CL = amoxicillin/ clavulanic acid, AMP = ampicillin, CEP =
cephalothin, KAN = kanamycin, STR = streptomycin, SUL =
sulfamethoxazole, TET = tetracycline, TR/SU = trimethoprim
sulfamethoxazole.

57

Table 2.3: Percentages of Isolates Exhibiting Antimicrobial Resistance

(Antimicrobials Showing Significant Differences Among Groups).
AMI
CL AMP CEP CLO GEN KAN NAL STR SUL TET

 

Septic
n=111 4.5 11.7 42.3 0.9 0.9 0.0 6.3 1.8 5.4 4.5
Sewage
n=109 6.4 22.9 45.9 0.9 0.9 4.6 11.0 1.8 4.6 6.4
Human

n=220 5.5 1 7.3 44.1 0.9 0.9 2.3 8.6 1.8 5.0 5.5

Beef
n=103 1.0 4.1 32.0 5.2 3.1 2.1 0.0 11.3 17.5 30.9
Dairy
n=107 0.9 1.9 43.0 1.9 0.9 3.7 0.0 6.5 5.6 14.0
Swine
n=98 2.7 24.6 28.2 8.2 3.6 11.8 0.0 32.7 39.1 70.9
Sheep
n=55 0.0 10.9 45.5 7.3 0.0 7.3 0.0 20.0 14.6 41.8
Livestock
n=363 1.4 10.6 36.0 5.4 2.2 6.2 0.0 17.6 20.1 39.6

009
n=95 2.1 13.7 55.8 0.0 1.1 1.1 1.1 2.1 5.3 4.2
Cat
n=95 8.4 26.3 35.8 1.1 0.0 1.1 0.0 5.3 5.3 6.3
Pets
=190 5.3 20.0 45.8 0.5 0.5 1.1 0.5 3.7 5.3 5.3
Duck
n=93 1.1 4.3 51.6 1.1 1.1 1.1 0.0 2.2 3.2 5.4
Goose
n=119 0.0 0.0 39.5 0.8 0.0 0.0 0.0 0.0 0.0 1.7
Deer

n=103 1.0 1.9 44.7 1.0 0.0 0.0 0.0 1.9 2.9 3.9
Wildlife
n=315 0.6 1.0 44.8 1.0 0.0 0.3 O 1.3 1.9 3.5

River
n=329 1.8 4.9 52.6 0.6 0.6 0.9 1.5 3.0 2.7 7.9

ALL
n=1423 2.5 9.6 45 2.0 0.9 2.4 1 .8 6.2 7.5 1 3.5

Abbreviations for antimicrobial agents: AM/CL= amoxicillin/clavulanic Acid,
AMP= ampicillin, CEP= cephalothin, CLO= Chloramphenicol, GEN= gentamicin,
KAN= kanamycin, NAL= nalidixic acid, STR= streptomycin, SUL=
sulfamethimazole, TET= tetracycline. Resistance is defined as ‘resistant’ or
‘intermediate’ based on CLSI breakpoints.

58

Table 2.4: Percentages of Isolates Exhibiting Antimicrobial Resistance
(Antimicrobials Showing No Significant Differences Among Groups).

 

TRI
AMI CEFR CEFX CIP CEFO SU
Septic
n=111 3.6 0.9 0.0 0.9 4.5 2.7
Sewage
n=109 1.8 0.9 0.9 0.9 0.9 1.8
Human
n=220 2.7 0.9 0.5 0.9 2.7 2.3
Beef
n=103 1.0 0.0 0.0 0.0 2.1 2.1
Dairy
n=107 0.0 0.9 0.9 0.0 0.9 0.9
Swine
n=98 0.9 0.0 0.0 0.0 0.0 1.8
Sheep
n=55 1.8 0.0 0.0 0.0 0.0 0.0
leestock
n=369 0.8 0.3 0.3 0.0 0.8 1.4
Dog
n=95 3.2 0.0 0.0 1.1 1.1 1.1
Cat
n=95 2.1 1.1 0.0 0.0 3.2 0.0
Pets
=190 2.6 0.5 0.0 0.5 2.1 0.5
Duck
n=93 1.1 1.1 0.0 0.0 2.2 2.2
Goose
n=119 0.8 0.0 0.0 0.0 0.8 0.0
Deer
n=103 1.0 0.0 0.0 0.0 1.0 0.0
Wildlife
n=315 1.0 0.3 0.0 0.0 1.3 0.6
River
n=329 0.9 0.0 0.0 0.0 1.5 1.5
ALL
n=1 423 1.4 0.4 0.1 0.2 1.5 1.2

Abbreviations for antimicrobial agents: AMI = arnikacin,
CEFR= ceftiofur, CEFX= ceftriaxone, CIP= ciprofloxacin,
CEFO= cefoxitin, GEN= gentamicin, TR/SU=
trimethoprim/sulfamethimazole. Resistance is defined as
‘resistant’ or ‘intermediate’ based on CLSI breakpoints.

59

Figure 2.4: MIC Values for Amikacin.

 

 

   
 
 
 
 
    

 

 

 

   

 

I Human
90 ~ 1:.) Livestock
80 7 Pets
3 23 4 I Deer
T“; 50 j EWaterfowI
1n .
“-5 40 _ 51 River
a: 30 T
20 - 7
10 ~ /
0 —1 lg I'I ,
0.5 (S) (8) >4 (NN)
MIC (pg/ml)

 

 

 

(S) represents MIC dilutions at which isolates are considered susceptible,
based on CLSI breakpoints. (NN) represents MIC dilutions at which
isolates are considered non-interpretable (N o breakpoint values available
for arnikacin).

Figure 2.5: MIC Values for Amoxicillin/Clavulanic Acid.

 

  

 

 

 

     

 

   

 

I Human
90 - Livestock
80 « Pets
8 70 A I Deer
«0 60 “ a
2 ¢ 5 Waterfowl
o 50 ~ /
2 d E River
9. 40 ¢
° /
3g 30 a f
A
20 - g
/
10 - g
0 2M 1.. A _-!-f 9.,Tm an T—h. -r-____,
1/0.5 (S) 2/1.0 (S) 4/2 (S) 8/4 (S) 16/8 (I) 32/16 (R) >32/16 (R)
MIC (pg/ml)

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

60

Figure 2.6: MIC Values for Ampicillin

 

 
    
    

  

 

 

 

 

 

I Human
7o _ Livestock
60 Pets
m 50 IDeer
1% 7 a Waterfowl
3 4° ‘ t
'5 3° - é
a
.\' 20 - g
10 - g
/
O _ flu _'.'. :' 4 ....'.‘
1 (S) 2(8) 4 (S) 8 (S) 16 (I) 32 (R) >32 (R
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

Figure 2.7: MIC Values for Ceftiofur

 

 

 

 

 

       

I Human
90 1 El Livestock
:3 I IPets
g 60 IDeer
T: 50 EWaterfowl
g 40 5 aRiver
° 301
.\°
20
‘10 E
o 7 _. . .. A, _ 7. _ _ _fi_ ‘fi—%
0.05 (S) 0.12 (S) 0.25 (S) 05(8) 1 (S) 2 (S) 4(I) 8 (R) >8 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

61

Figure 2.8: MIC Values for Ceftriaxone

 

 

I Human

El Livestock
Pets

I Deer

a Waterfowl
In River

 

 

 

% of Isolates

7
/
t
a
/
a
a
a
/
r
a
a
r
r

I

 

 

0.05(S) 025(5) 1(8) 4(3) 8(8) 32(1) 64(R)
MlC(pglmI)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

Figure 2.9: MIC Values for Cephalothin

 

 

  
 
  

 

 

 

  

 

 
   
 
    

 

I Human
60 - 12 Livestock
50 _ Pets
in I Deer
g 40 ' EWaterfowI
g 30 ' I River
.2. 20 —
1o —
0 _ _. . rm- 55.3. .” .
2 (S) 4(8) 8 (S) 16 (I) 32 (R) >32 (R)
MIC (pg/ml)

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

62

Figure 2.10: MIC Values for Chloramphenicol

 

 

   
 
   
 

 

 

 

 

I Human
70 - ta Livestock
60 _ Pets
85°— $32....
3 :3 I E River
.,_ _
a: 20 ~
10 a
0 _ . _:i:. -:
2 (S) 8 (S) 16 (I) 32 (R) >32 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

Figure 2.11: MIC Values for Ciprofloxacin

 

 

I Human

12 Livestock
Pets

I Deer

a Waterfowl
a River

 

 

 

% of Isolates

 

g
é
%—

(3) 0.0 (3) 0.05 (3) 011(3) 025(3) 0.5(3) 4(R) >4 (R)
MIC (pg/ml)

0.

O
_L
N

 

 

 

(S) represents dilutions at which isolates are considered susceptible and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

63

Figure 2.12: MIC Values for Cefoxitin

 

 

  
 
 
 
  

 

 

 

I Human

60 — in Livestock

50 - Pets
8 I Deer
:3; 4° “ eWaterrowI
g 30 1 nRiver
°m—
a!

10 .

o -

 

 

0.5 (3) 1 (3) 2(3) 4(3) 8(3) 16 (I) >16 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

Flgure 2.13: MIC Values for Gentamicin

 

  

 

 

 

     

 

 

I Human

90 IE Livestock

80 I‘ Pets
8 70 J I Deer
a 60 a EWaterfowI
6 50 A
at A III River
: 40 7
° 30 3
=2 r

20 g

10 a

0 :%T 2 g' "‘ “" =" r77‘7fl ' II
0.25 (S) 0.5 (S) 1 (S) 2 (S) 4 (S) 8 (I) 16 (R) >16 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

64

Figure 2.14: MIC Values for Kanarnycin

 

 

 

 

 

  

 

 

I Human
120 1 1:1 Livestock
100 2 Pets
g 80 I Deer
3 EWaterfowI
2 60 ‘ In River
"6
a! 40 -
20 -
0 < .. _ l _
16 (S) 32 (I) 64 (R) >64 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

Figure 2.15: MIC Values for Nalidixic Acid

 

 

 

 

 

  

 

I Human
80 Livestock
70 7 Pets
d
60 n I Deer
it 50 r
_ 4 aWaterfowl
3 40 ‘ A -
Z 1 Z a RIver
o 30 i g
3‘ 20 - Z
10 ‘ g
7 ; :;:
Q n“, 4 :1: In ILL %J‘Jj
0.5 (3) 1 (3) 2(3) 4(3) 8 (3) 16 (3) 32 (R) >32 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

65

Figure 2.16: MIC Values for Streptomycin

 

 

  
 
  
   

 

 

 

I Human

120 — I}: Livestock
100 ‘ Pets

1» I Deer

g 80 ~

_ E Waterfowl

E 60 « a River

0.-

o

=2

 

40 -
20 .
o , , _ , .-
32 (NN) 64 (R) >64 (R)
MIC (pg/ml)

 

 

 

(R) represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints. (NN) represents a dilution at which isolates are considered
non-interpretable, as the CLSI breakpoint between distinguishing isolates
intermediate and susceptible to streptomycin is below 32 rig/ml.

Figure 2.17: MIC Values for Sulfmethoxazole

 

 

I Human

n Livestock
Pets

I Deer

a Waterfowl
a River

 

 

 

% of Isolates

 
   

L, “v7fi____‘ ‘ I”) ”I

32 (S) 512 (R) >512 (R)
MIC (pg/ml)

 

 

 

 

(S) represents dilutions at which isolates are considered susceptible and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

66

Figure 2.18: MIC Values for Tetracycline

 

 

 

 

 

 

   

I Human
120 — I: Livestock
10° ~ 7 .32:
w /
E 80 i g aWaterfowI
g so - :::/ IRiver
'6 222/
40 — /
=2 :11/
:51/
20 ‘ :::/
0 ,, 515/ 15:1 W,,.
4 (S) 8 (I) 16 (R) 32 (R) >32 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible, (1)
represents dilutions at which isolates are considered intermediate, and (R)
represents dilutions at which isolates are considered resistant, based on
CLSI breakpoints.

Figure 2.19: MIC Values for Trimethoprin/Sulfamethoxazole

 

 

   
  

 

 

 

IHuman
120 1 mLivestock
100 Pets
at I Deer
{1 80
.5 1 aWaterfowI
2 60 ‘ aRiver
‘6 I
32 40 l
20
0 hW‘Tv v YvV”T_-l’_“l
0.12/2.4 (S) 0.25/4.8 (S) 0.5/9.5 (S) 1/19 (S) 4/76 (R) >4/76 (R)
MIC (pg/ml)

 

 

 

(S) represents dilutions at which isolates are considered susceptible and (R)
represents dilutions at which isolates are considered resistant, based on CLSI
breakpoints.

67

Chapter III: The Use of Antimicrobial Resistance Profiles of Escherichia coli from
Livestock, Pet, Wildlife, and Human Sources to Identify Origins of Fecal
Contamination in the Red Cedar River (Michigan)
Abstract

Discriminant analyses of antimicrobial resistance profiles of Escherichia coli
were performed to identify sources of fecal contamination into the Red Cedar River in
central Michigan. A total of 1094 E. coli isolates were obtained from separate samples of
human (septic and sewage), livestock (dairy cattle, beef cattle, swine, and sheep), pets
(dogs and cats), and wildlife (free-ranging ducks, geese, and deer), and minimum
inhibitory concentrations (MIC values) were obtained for each of 16 antimicrobials. MIC
values were also obtained for 329 river isolates collected over 8 months as part of a fecal
coliform monitoring program. Principal component analysis revealed that a large
percentage of variation in the data was due to differences in resistance to
sulfamethoxazole, tetracycline, and ampicillin, and that a considerable amount of overlap
was present between species groups. The best discriminant analysis model, as
determined by the cross-validation method, was obtained using the k-nearest neighbor
nonparametric method and 11 of the 16 initial antimicrobials. Correct classification rates
were 35.5% for human isolates, 44.4% for livestock, 39.0% for pets, and 64.1% for
wildlife, with an average rate of correct classification (ARCC) of 46%, indicating better
than chance classification. Positive predictive values for each species group were 41 %
for an isolate classified as human, 71% for an isolate classified as livestock, 37% for an
isolate classified as pet, and 43% for an isolate classified as wildlife. Classification of
river isolates using this model indicated 16.4% as human, 13.5% as livestock, 17.9% as

pets, and 52.3% as wildlife in origin. Considering the classification of river isolates in

68

light of the positive predictive values indicates that contamination is likely fiom mixed
sources. This study indicates that discriminant analysis of MIC values may be a cost

effective aid to identifying dominant sources of fecal contamination.

Introduction

Fecal contamination of surface water used for drinking, recreation, and agriculture
poses serious health risks. Sources of fecal contamination include point sources, such as
municipal effluent, and non-point sources, such runoff from agriculture, wildlife, and
malfunctioning septic systems (Wesikal et al., 1996; Mallin et al., 2000; Collins, 2004;
Shehane et al., 2005). Human and animal feces may contain pathogens that are
hazardous to human and animal health (Scott et al., 2002), and knowing the source of
contamination is necessary to assess the risks involved. Furthermore, in order to
implement management practices to reduce fecal contamination, one must be able to
determine the origins of the contamination.

Fecal indicator bacteria, such as Escherichia coli, are used to indicate and
quantify fecal contamination. However, these organisms are present in the intestines of
all warm-blooded and some cold-blooded animals (Harwood et al., 1999), and thus do
not indicate a specific source. A variety of phenotypic and genotypic microbiological
methods have been used to characterize fecal indicator bacteria and link them to a source
(Scott et al., 2002; Meays et al., 2004). Recent methods, collectively referred to as
microbial source tracking, include ribotyping (Parveen et al., 1999; Carson et al., 2001 ),
repetitive DNA sequences (Dombeck et al., 2000), amplified fragment length

polymorphism analysis (Guan et al., 2002) and antimicrobial resistance analysis (Parveen

69

et al., 1997, Hagedom et al., 1999; Harwood et al., 2000; Graves et al., 2002; Whitlock
et al., 2002; Carrol et al., 2005). Each of these methods relies on the creation of a
database of bacteria from known animal and human sources and the use of statistical
techniques to differentiate these sources based on the genotypic or phenotypic variables
under consideration. Isolates fiom unknown sources, such as those from surface water
samples, are then classified into source groups based on these variables. Molecular
methods have been considered superior for detecting specific differences between isolates
(Guan et al., 2002; Scott et al., 2002), but the creation of large databases using these
techniques can be very expensive and technically difficult. In contrast, antimicrobial
resistance analysis is a relatively simple and inexpensive method of microbial source
tracking.

Antimicrobial resistance analysis for microbial source tracking is based on the
idea that the bacterial flora in the intestines of humans and various animal species are
subjected to different types, concentrations, and amounts of antibiotics (Scott et al.,
2002). Exposure to antibiotics will result in selective pressure that will favor the
survival of resistant organisms. Differences in the resistance profiles of fecal bacterial
from human and various animal sources have been noted in recent studies (Whitlock et
al., 2002; Burnes et al., 2003; Carrol et al., 2005; Sayah et al., 2005). Based on their
limited exposure to antimicrobials, we would expect a low prevalence of resistance in
fecal bacteria from free-ranging wildlife (Routman et al., 1985; Guan et al., 2002; Burnes
et al., 2003). However, wildlife may come in contact with feeds containing

antimicrobials or with resistant bacteria in the environment (Livermore, 2001;

70

Selvaratnam and Kunberger, 2004), and, as a result, may harbor resistant intestinal
bacteria (Rolland et al., 1985 ; Middleton and Ambrose, 2005).

Discriminant analysis is a multivariate statistical technique that uses combinations
of variables to maximize the separation of groups and classify unknown samples into
groups based on the resulting classification equation. Although a number of studies have
used discriminant analysis of antimicrobial resistance profiles, the use of minimum
inhibitory concentrations (MIC values) from standard antimicrobial testing panels for
microbial source tracking has not been described. The present study is also unique in that
only one isolate from each individual sample was included in the analyses. This was
done in an attempt to maximize the diversity and representativeness of the known source
library, given our resources and the number of isolates we were able to test. Since
estimations of model accuracy have varied greatly among recent bacterial source tracking
studies using antimicrobial resistance analysis and discriminant analysis, additional
studies such as this will help to understand the applications and the shortcomings of these
methods.

The purpose of this study was to use a panel of 16 antimicrobial agents and the
statistical method of discriminant analysis to differentiate bacteria from humans (sewage
and septic samples), livestock (dairy cattle, beef cattle, swine, and sheep), pets (dogs and
cats) and free-ranging wildlife (geese, ducks, and deer) and to classify samples from the
Red Cedar River (Michigan) into one of these source groups. We hypothesized that
discriminant analysis would be able to separate and classify known-source isolates with
greater accuracy than would be expected by chance classification. That is, with the four

groups of human, livestock, pet, and wildlife, we would expect more than 25% correct

71

 

classification for each group. The objectives of this study were to 1) determine the
discriminant analysis model that best differentiates fecal E. coli isolates from known
sources, and 2) use this model to classify E. coli isolates from the Red Cedar River into

source categories.

Materials and Methods

Study Site:

All samples were collected within or near the Red Cedar Watershed between May
2004 and May 2006 (see chapter 11). During the time of river sample collection (May
through December, 2004), weekly counts done at the MDEQ indicated relatively low
numbers of colform bacteria (<600 cfu/ 100ml) for 22 of the 28 weeks of collection.
However, on six occasions average weekly counts rose above 1000 chI/ 100ml (Kline-

Robach and Witter, 2002).

Sample Collection and Bacterial Isolation:

Septic tank, sewage, deer, goose, duck, dog, cat, beef cattle, dairy cattle, swine,
sheep, and river samples were collected, as described in Chapter II. E. coli bacteria were
isolated from 111 septic, 109 sewage, 103 deer, 119 goose, 93 duck, 95 cat, 95 dog, 97
beef cattle, 107 dairy cattle, 110 swine, 55 sheep, and 330 river samples, as described in
Chapter 11. River isolates were cultured from samples collected at six locations along the

Red Cedar River, as described in Chapter II.

Susceptibility Testing:

72

The method for sensitivity testing using Sensititre technology (TREK Diagnostic
Systems, Inc., Cleveland, Ohio) was described in Chapter II. The CMV7CNCD
antimicrobial panel was chosen because it included antimicrobials commonly used in
human and animal medicine, and would thus be more likely to be able to detect
differences between human and various animal fecal bacteria. The types and
concetrations of antimicrobials included in the CMV7CNCD Sensititre panel are listed in

Table 2.1.

Descriptive Statistics:

Isolates were grouped as being from human (septic or sewage), livestock (beef
cattle, dairy cattle, swine, or sheep), pet (dog or cat), or wildlife (deer, duck, or goose)
sources, or river samples. Graphs of the percent of each group showing a given MIC
value for each antimicrobial agent are presented in Chapter 11 (Figures 2.4-2.19).
Differences between the four species groups for each antimicrobial were tested for using
Kruskal-Wallis tests on MIC values and chi-square or Fisher’s exact tests on resistance or
susceptible designations, as described in Chapter II. Results of these analyses were

presented in Chapter II.

Principal Component Analysis:

In order to conceptualize the variability in these data, principal component
analysis (PCA) was performed using SAS, version 9.1. The log converted MIC values
for each antimicrobial agent were used as variables. Plots were then made of the first two

principal components for each of the four species groups.

73

Discriminant Analysis:

Discriminant analyses were performed using SAS version 9.1. Multiple analyses
were performed in order to determine the parameters and variables that best separated the
species groups. The ability of each model to separate sources and correctly classify
known source isolates was determined using the cross-validation method. This method
involves removing each isolate and classifying that isolate based on the library of
remaining isolates. Since the isolate being classified is not included in the library used to
derive the classification rule, this method is considered less bias than the resubstitution
method. Application of the crossvalidation method results in source-by-source matrix in
which the percentages of correctly classified isolates (correct classification rates) are
noted along the diagonal. Calculations of the average rate of correct classification
(ARCC) for each model were made by averaging the percentages along the diagonal, as
described by Wiggins et al. (1996). The kappa statistic was used to test if correct
classification rates were greater than would be expected by chance classification.

Due to the nonparametric nature of the MIC distributions (see Figures 2.4-2.19),
nonparametric discriminant analyses were performed in addition to parametric methods.
These included k-nearest neighbor methods, as well as biweight, triweight, uniform,
Epanechnikov, and normal kemal methods using both pooled and individual within-
group covariance matrices.

Discriminant analyses were first performed using all variables and four species
groups (human, livestock, pets, and wildlife), and then different combinations of

antibiotics were included. Additionally, analyses considering each species individually

74

and analyses using various grouping of species were considered. For comparison,
analyses using resistant and susceptible designations rather than MIC values were also
performed.

For the four group (human, livestock, pet, and wildlife) model with the highest
ARCC, the positive predictive value (PPV) of the model was calculated for each of the
four species groups, using the equation: PPV = [# of (e. g.) human isolates classified as
(e. g.) human] / [Total # of isolates classified as (e. g.) human]. This calculation was based
on the classification table resulting from the cross-validation method of classifying
known source isolates. PPV was calculated to aid in interpreting the results of classifying
river isolates using this model.

This model (The four group model with the highest ARCC) was then used to
classify unknown source river isolates. Chi-square tests were used to test if river isolates
were classified differently depending on whether they were collected in East Lansing or
Williamston, whether they were collected in the summer (May through August) or fall
(September through December), and whether they were collected when fecal coliform

counts were high (>1000) or low (<600).

Results
Principal Component Analysis:
The first principal component had an eigenvalue of 2.34 and explained 36% of the
variation in the data. The variables with the greatest (all positive) weights were the log of
the MIC values for sulfamethoxine, tetracycline, and ampicillin. The second principal

component had an eigenvalue of 1.25 and explained 19% of the variation in the data. The

75

 

variables with the greatest (positive) weights were ampicillin, cephalothin, and
amoxicillin/clavulanic acid. Sulfarnethoxamine had a high negative weight for the second
principal component.

Plots of the first two principal components for each species group demonstrate
overlap among species within each species group (Figures 3.1-3.4). Furthermore,
superimposing these graphs for each species group shows that, while there are differences

in the patterns among groups, there is also a large amount of overlap.

Discriminant Analysis:

Model Selection:

With all 16 antimicrobials included in the analysis, the parametric models resulted
in slightly higher ARCC values than the nonparametric models, with ARCC values of
45% for the pooled covariance matrix normal model and 41% for the individual within
group covariance matrix normal model. Of the nonparametric models, the k-nearest
neighbor models produced the best ARCC values, with the ARCC ranging from 37% to
41 %, depending on the value of k. K denotes the number of nearest neighbors to be
considered in classifying each observation, and analyses using values for k between one
and four were attempted. Kernel methods resulted in ARCC values that ranged from
32% to 39%, depending on the specific model and whether pooled or individual within-
group covariance matrices were used. In general, models using a pooled covariance
matrix resulted in slightly greater ARCC values than models using individual within-

group covariance matrices.

76

Using the k-nearest neighbor model with k=4, the combination of antimicrobial
agents that resulted in the model with the highest ARCC value (46%) included 11 out of
the 16 initial antimicrobials. Tables 3.1 and 3.2 represent the classification tables from
the discriminant analyses including 16 and 11 antimicrobials, respectively. The 11
antimicrobials included were arnikacin, ampicillin, ceftiofur, ceftriaxone, cephalothin,
cefoxitin, gentamicin, nalidixic acid, streptomycin, sulrnethoxarnine, and tetracycline.
Amoxicillin/clavulanic acid, kanamycin, ciprofloxacin, chloramphenical, and
trimethoprim sulfa were excluded from the analysis. Analyses using binary coding of
resistance and susceptibility, rather than MIC values, for all antibiotics and for the 11

antibiotics listed above both resulted in ARCC values of 39%.

Model Evaluation:

With the k-nearest neighbor model (k=4) using MIC values for 11 antimicrobials,
individual group correct classification rates and the ARCC were all higher than would be
expected by chance alone. Correct classification rates were 35% for humans, 44% for
livestock, 39% for pets, and 64% for wildlife, and the ARCC was 46% (Table 3.2).
Random classification should result in about 25% correct classification for each of the
four groups. A kappa coefficient of 0.29, with 95% confidence limits between 0.25 and
0.33, supports our hypothesis that known source isolates are classified more accurately
than would be expected by chance classification. The test of the null hypothesis, that
kappa=0 and there is no agreement between the source of the isolates and how they are

classified by the model results, in a p value < 0.0001.

77

The positive predictive values for each species group, using this model, were 41 %
for an isolate being identified as human, 71% for livestock, 37% for pet, and 43% for

wildlife.

Resubstitution vs. Crossvalidation:

In general, classification rates were higher when the resubstitution method was
used, although for many models the difference between these methods was small. For the
k-nearest neighbor, k=4, 11 antimicrobial model with an ARCC of 46% based on the

crossvalidation method, the resubstitution methods resulted in an ARCC of 55%.

Alterations of Species Groupings:

Using the k-nearest neighbor, k=4, 11 antimicrobial model, analyses with
different combinations of species groups were conducted. The ARCC values and range
of individual group correct classification rates for these analyses are presented in Table
3.3. In general the ARCC values decreased when species were stratified into a greater
number of categories. The analysis considering each of the 11 sources separately resulted
in an ARCC of 20%. Decreasing the number of categories considered by grouping
species further resulted in an increase in the ARCC value. Grouping species into the
three groups of livestock, urban (humans and pets), and wildlife resulted in an ARCC of
53%, whereas the three groups of humans, domesticated animals (livestock and pets) and
wildlife resulted in an ARCC of 48%. Of the two-group analyses, wildlife versus all

others and livestock versus all others resulted in the highest ARCC values of 68% and

78

66%, respectively. The analyses of human versus all others and pets versus all others

both resulted in ARCC values of 59%.

Classification of River Isolates:

When river isolates were classified using the k-nearest neighbor method including
11 antimicrobial agents, 16% of isolates (54) were classified as human, 13% (44) were
classified as livestock, and 18% (59) were classified as domestic animals, and 52% (172)
were classified as wildlife (Table 3.4).

A comparison of river isolates by location, season, and fecal coliform counts
showed no significant associations between any these variables and the sources into
which river isolates were classified. All differences were within 2%, except slightly
more river isolates were classified as wildlife during summer months (55.8% as
compared with 48.8%) (Figure 3.5) and when fecal coliform counts were low (53.2% as
compared with 49.4%) (Figure 3.6). Additionally, 15.2% of river isolates collected
during the fall months were classified as livestock, as compared with 11.5% of river

isolates collected during the summer months (Figure 3.6).

Discussion
Both the PCA analysis and the discriminant analysis are consistent with there
being considerable variability in these data that is not due to differences between species
groups. The PCA analysis showed that the MIC values for the antimicrobial agents
sulfarnethoxamine, tetracycline, and ampicillin were the largest contributors to the

variability in these data. While Kruskal-Wallis tests did show significant differences in

79

MIC values between groups for these and other antimicrobials (see Chapter 11), there is a
large amount of overlap between groups. Plots of the first two principle components
reveal different patterns for different species groups, particularly livestock and wildlife,
but there is a large amount of overlap among the PCA graphs for all species groups.

For the discriminant analyses, we found that the highest ARCC values were
obtained using a subset of 11 of the antimicrobials tested, rather than all 16. Some
previous studies have demonstrated that ARCC values improve with increasing numbers
and concentrations of antimicrobials (Wiggins et al., 1996; Harwood et al., 2000).
However, Wiggins (1996) found that the best ARCC values were obtained using 4 out of
5 antimicrobials tested and Hagedom et al. (1999) found that the best ARCC values were
obtained using 6 out of 7 antimicrobials tested. The most useful antimicrobials for
discrimination between species have also varied between studies, probably a result of
differences in antimicrobial use and predominant species in different areas. For example,
Burnes et al., (2003) found that spectinomycin, streptomycin, kanamycin and ampicillin
were the most important measurement variables and gentamicin and tetracycline were
relatively unimportant, whereas in our study the inclusion of kanamycin actually
decreased the ARCC, while it was improved by the inclusion tetracycline and gentamicin.
In our study the inclusion or removal of ceftriaxone, an antimicrobial for which no
significant difference was found between species, did not appreciably alter the ARCC. In
contrast, removal of both kanamycin and trimethoprim/sulfamethoxazole, which were
significantly different between species at the p<0.0001 level, improved the ARCC. The
inconsistency between significance tests and discriminating variables is likely due to

large within-species variation and the differences in the statistical bases of these analyses.

80

The ARCC values in this study were low compared with other studies using
discriminant analysis of antimicrobial resistance profiles of E. coli or fecal coliforms.
For example, using antimicrobial resistance analysis of fecal coliforms, Harwood et al.
(2000) obtained ARCC values ranging between 64% and 74%, depending on how species
were grouped. Burnes et al. (2002) obtained ARCC values ranging between 70% and
96%, and Whitlock et al. (2002) obtained an ARCC of 69% for a four group model.
Using antimicrobial resistance analysis of E. coli, specifically, Guan et al., (2002)
obtained ARCC values ranging between 34-74%, and Carroll et al., (2005) obtained
ARCC values ranging between 78-85%.

One reason for the relatively low ARCC values in our study may be the use of
the more rigorous cross-validation method, along with the inclusion of only one isolate
per sample. All of the above studies except Carroll et al. (2005) used the resubstitution
method of calculating the ARCC, and all except Guan et al. (2002) included multiple
isolates from an individual sample in their analysis. Since the resubstitution method uses
the same isolates to develop and to test the discriminant analysis model, it may result in
an overestimation of the model’s ability to correctly classify non-library isolates.
Similarly, the inclusion of multiple isolates per sample may increase the ARCC, if we
assume that the resistance profiles of isolates from the same fecal sample may be more
similar to each other than isolates from different individuals of the same species (Wiggins
et al., 2003)

Another reason for the low ARCC values obtained in this study may be the high
percentage of isolates from all species that were not resistant to any antimicrobials. This

could also explain why much of the error was due to the classification of isolates not

81

originating from wildlife into the wildlife category, as wildlife had the smallest
percentage of isolates resistant to one or more antimicrobials (see Chapter II). In order to
reduce the classification error, additional variables need to be included that better
separate the species groups. Inclusion of additional antimicrobial agents may improve
correct classification rates, if antibiotics showing both higher percentages of resistance
and greater differences in resistance between species can be found. Alternatively,
inclusion of other variables, such as genetic analysis, may improve classification rates.
Differentiation of species using antimicrobial resistance analysis is based on the
principle that different species are subjected to different histories of antimicrobial
exposure, which will select for different resistance patterns of their microbial gut flora.
However, the development, persistence, and spread of resistance are complex processes,
often involving transferable plasmids containing multiple resistance genes (Summers,
2002). Both humans and animals are continuously exposed to exogenous bacteria from
other individuals, animals, fomites, and food. Re-inoculation of gut flora or simply the
transfer of resistance factors from such bacteria may result in alterations of resistance
patterns in commensal intestinal bacteria. The use of antimicrobials in animal agriculture
resulting in resistant bacteria being transferred to humans via food is a topic of great
concern and controversy (Mathers, 2001; Smith, 2002; Witte, 2006). Furthermore,
because resistance factors are often linked on transferable plasmids, exposure to one
antimicrobial may result in resistance to multiple antimicrobials. Recent studies have
demonstrated the presence of antimicrobial resistant bacteria in contexts where
antimicrobial agents have not been used (Rolland et al., 1985; Enne et al., 2001). The

complex processes involved in antimicrobial resistance of microbial gut flora may

82

explain some of the overlap between species in antimicrobial resistance patterns and the
difficulty of differentiating species based on these profiles.

In our study, the rates of correct classification are greater than they would be by
chance alone, and thus the argument can be made that these analyses may be applicable
for identifying major sources of fecal pollution. Based on our model, the majority (52%)
of river isolates were classified as being from wildlife. However, the PPV for wildlife
was only 43%, which implies that for over half of the isolates classified as wildlife, we
can not be confident that they are actually from wildlife. In contrast, while only 14% of
river isolates were classified as being from livestock, the PPV for livestock is 71%, which
means that we have a greater level of confidence that the majority of isolates classified as
livestock are actually from livestock. Thus, when classification of river isolates is
considered along with the PPV for each species group, it appears that the fecal
contamination in the Red Cedar River is coming from a mixture of sources. This result is
not surprising, given that fecal coliform counts were relatively low during the course of
this study. The relatively high error rates of this model suggest that it may be more
appropriate for identifying dominant contributors to contamination.

The differences in how river isolates collected in the summer and fall were
classified, although not significant, are consistent with other studies that have found
variations in predominant sources with different environmental factors (Whitlock et al.,
2002; Booth et al., 2003; Burnes 2003; Carroll et al., 2005; Shehane et al., 2005). Booth
et al. (2003) found a greater percentage of human sources during cooler months, and
suggested that seasonal conditions such as a high water table and less evaporation of

effluent could contribute to an increase in human loading rates during these months.

83

Limitations of our study, including high error rates and small numbers of tested river

isolates, make conclusions about small differences in sources difficult.

Conclusion

Discriminant analysis of MIC values may be a cost effective aid to identifying
major sources of fecal contamination. Even with an ARCC of only 46% and a large
amount of overlap in resistance profiles of fecal bacteria from different species groups,
this model can provide information regarding likely sources of fecal contamination.
While this tool does not have the ability to by itself identify the specific source of a fecal
bacterium cultured from the Red Cedar River, it can indicate which sources are more
likely the dominant contributors to fecal contamination within this study area. Like any
test, this analysis should be used in combination with other information, such as the
prevalence of different species, land-use patterns, and potential contamination events.
Additional source-tracking methods, such as genetic analysis, could also be used in
combination with this method in order to more specifically identify or narrow down the
sources indicated by discriminant analysis of antimicrobial resistance profiles to be the

more likely contributors.

84

Figure 3.1: Principal Components Plot of Human (Septic and Sewage)
Fecal E. coli Isolates.

 

+ Septic
D Sewage

 

 

 

7': +

'85

EE D
32 1+.” 1+ .
§8 4 U 6

m +

 

First Principal Component

 

 

 

Figure 3.2: Principal Components Plot of Livestock Fecal E. coli Isolates.

 

 

 

 

 

 

 

+ Beef Cattle
Ci Dairy Cattle

‘15 A Sheep

3 o Swine

o

E

8 +9

'«I

a.

'8

.E ‘

E 8

'u

C

o

8

to

First Principal Component

 

 

 

85

Figure 3.3: Principal Components Plot of Pet Fecal E .coli Isolates.

 

+ Cat
Ci Dog

 

 

 

Second Principal
Component

First Principal Component

 

 

 

Figure 3.4: Principal Components Plot of Wildlife Fecal E. coli Isolates.

 

 

 

 

 

 

+ Deer
CI Duck

E 6 _ aGoose

C

E]

g- -1

E 4

8 +

8

'5 _ '3 U

.5 1 LI 1 n m

a. 2 +4 6 8

'0 ii-

2

8

m '4 ‘ ‘

First Principal Component

 

 

 

86

 

 

Table 3.1: Classification Table for K-Nearest—Neighbor, K=4, 16 Antimicrobials.
Numbers in parentheses represent numbers of isolates. ARCC=39%.

 

 

 

 

 

 

 

Human Livestock Domestic Wildlife
Human (220) 28.6% (63) 21.4% (47) 23.2% (51) 26.8% (59)
Livestock (369) 17.1% (63) 48.0% (177) 11.7% (43) 23.3% (86)
Domestic (190) 15.8% (30) 13.7% (26) 31.1% (59) 39.5% (75)
Wildlife (315) 16.2% (51) 13.0% (41) 21.6% (68) 49.2% (155)

 

 

 

 

Table 3.2: Classification Table for K—Nearest-Neighbor, K=4, 11 Antimicrobials.
Numbers in parentheses represent numbers of isolates. ARCC=46%.

 

 

 

 

 

 

 

 

 

Human Livestock Domestic Wildlife
Human (220) 35.5% (78) 14.1% (31) 20% (44) 30.1% (67)
Livestock (369) 14.1% (52) 44.4% (164) 8.4% (31) 33.1% (122)
Domestic (190) 10.5% (20) 9.0% (17) 39.0% (74) 41.6% (79)
Wildlife (315) 12.4% (39) 6.7% (21) 16.8% (53) 64.1% (202)

 

 

Table 3.3: Comparison of Discriminant Analysis Models with Different Groupings

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

of Species.
# of Groups Range of Correct ARCC
Groups Classification Rates

11 Sewage, Septic, Beef, 9.3%-45.4% 20%
Dairy, Swine, Sheep,
Cat, Dog, Duck, Deer,
Goose

7 Human, Beef, Dairy, l7.5%-47.3% 29%
Swine, Sheep, Pets,
Wildlife

5 Human, Livestock, 29.5%-43.9% 37%
Pets, Wildbirds, Deer

4 Human, Livestock, 35.5%-64.1% 46%
Pets, Wildlife

3 Urban, Livestock, 40.0%-72.4% 53%
Wildlife

3 Human, Domesticated, 33.6%-70.2% 48%
Wildlife

2 Human, Animal 51.8%-66.7% 59%

2 Wildlife, Other 55.8%-79.7% 68%

2 Pets, Other 56.4%-62.1% 59%

2 Livestock, Other 52.8%-79.7% 66%

 

87

 

 

 

Table 3.4: Classification of River Isolates and Positive Predictive Values for
Classification as Human, Livestock, Pet, and Wildlife.
Human Livestock Pets Wildlife
% River Isolates 16.4% (43) 13.5% (36) 17.9% (59) 52.3% (172)
(n=329 isolates)
PPV 41 % 71% 37% 43%
PPV refers to the positive predictive value of a river isolate being classified into a
given source group, and is calculated by dividing the number of isolates originally
from a given group and correctly classified into that group, by the total number of
isolates classified into that group.

 

 

 

 

 

 

 

 

 

Figure 3.5: Classification of E. coli Isolates From River Samples Collected in
Summer and Fall.

60
I ISummer(n=165)
50 ' Fall (n=164)

 

    

% River Isolates
(at)
O

 

Human Livestock Pets Wildlife

 

Source (By Classification)

 

 

Summer isolates are from samples collected in May through August,
2004, and Fall isolates are from samples collected in September
through December, 2004.

88

 

 

Figure 3.6: Classification of E. coli Isolates From River Samples Collected When
Fecal Coliform Counts Were Low (<600 cfu/100ml) and High (>1000 cfu/100ml).

 

60 a

 

El Fecal Coliform Count
50 - <600 cfu/100ml (n=246)

4o 4 I Fecal Coliform Count
>1000 cfu/100ml (n=83)

 

 

 

°/o River Isolates
0.)
O

 

 

   

 

 

 

 

 

Human Livestock Pets Wildlife
Source (By Classification)

 

 

 

89

Overall Summary, Conclusions, and Recommendations

Summary:

The purpose of this study was to use discriminant analysis of antimicrobial
resistance profiles of fecal Escherichia coli from known sources and unknown—source
river isolates to determine the sources of fecal contamination in the Red Cedar River.
Fecal E. coli bacteria were cultured fi'om human sources (septic tanks and sewage
samples), livestock (beef cattle, dairy cattle, swine, and sheep), pets (dogs and cats),
wildlife (deer, ducks, and geese) and the Red Cedar River. Sensitivity testing against a
panel of 16 antimicrobial agents revealed that the overall percentage of resistant isolates
was greatest among livestock samples, followed by human, pet, river, and wildlife
samples, although percentages of resistance varied between groups for each
antimicrobial. There were significant differences between one or more species groups in
resistance for 10 antimicrobials and MIC values for 14 antimicrobials. A k-nearest
neighbor discriminate analysis using 11 antimicrobials had the highest average rate of
correct classification (ARCC) of all attempted analyses. Using the cross-validation
method, correct classification rates were 35.5% for human isolates, 44.4% for livestock
isolates, 39.0% for pet isolates, and 64.1% for wildlife isolates, with an ARCC of 46%.
Positive predictive values for each species group were 41% for an isolate classified as
from a human source, 71% for livestock, 37% for pet, and 43% for wildlife. Overall,
16.4% of river isolates were classified as being from a human source, 13.5% from

livestock, 17.9% from pet, and 52.3% from wildlife, and classification of river isolates

90

did not differ significantly based on location, season, or fecal coliform counts at the time

of collection.

Conclusions:

1.

Although the percentages of resistant bacteria were lowest in the wildlife and
river groups, the presence of resistant bacteria in these groups are of concern due
to the implications regarding the dissemination of resistant bacteria in wildlife and
the environment.

There are significant differences in both antimicrobial resistance and MIC values
among E. coli from human, livestock, pet, and wildlife sources, and MIC values
may be a more sensitive measure of differences between these sources.
Interpretation of the classification of river isolates using the classification
equation generated from our discriminant analysis model, in light of the positive
predictive values for each source group, indicate that fecal contamination in the
Red Cedar River during the course of this study was most likely from a mixture of
sources.

Overall, this study indicates that discriminant analysis of MIC values may be a
cost effective aid to identifying dominant sources of fecal contamination,

especially when used in combination with other information.

Recommendations:

1.

Monitoring of antimicrobial resistance of fecal bacteria should be continued. We

feel that information on antimicrobial resistance in wildlife species is particularly

91

 

scarce and has important implications for human, animal, and environmental
health.

Including different antimicrobial agents may improve the differentiation of
known sources and result in the ability to more accurately classify unknown-
source isolates.

Red Cedar River isolates should continue to be tested and subjected to
classification by discriminant analysis in order to detect temporal variation in
sources of fecal contamination and determine if dominant sources are indicated
during periods of high fecal loads.

Combining broad scale antimicrobial resistance analysis studies with more
specific genetic methods may be a way to maximize accuracy and cost-

effectiveness of microbial source tracking studies.

92

Literature Cited

93

Anderson, J. D. 1974. The effect of R-factor carriage on the survival of Escherichia coli
in the human intestine. J. Med. Microbiol. 7:85-90.

Adetosoye, A. I. 1980. Infective drug resistance among Escherichia coli isolated from
clinically healthy domestic livestock. Vet. Microbiol. 5: 333-342.

Ash, R. J ., B. Mauck, M. Morgan. 2002. Antibiotic resistance of gram-negative bacteria
in rivers, United States. Emerg. Infect. Dis. 8(7):713-716.

Barenkamp, S. J ., R. S. Munson, Jr., and D. M. Granoff. 1981. Subtyping isolates of
Haemophilus influenzae type b by outer-membrane protein profiles. J. Infect. Dis.
143: 668-676.

Bernhard, A.E. and K. G. Field. 2000a. Identification of non-point sources of
fecal pollution in coastal waters by using host-specific 16S ribosomal DNA

genetic markers from fecal anaerobes. Appl. Environ. Microbiol 66(4): 1587-
1594.

Bernhard, A. E., and K. G. Field. 2000b. A PCR assay to discriminate human and
ruminant feces based on host differences in Bacteroides-Prevotella 16S ribosomal
DNA. Appl. Environ. Microbiol. 66: 4571-4574.

Bischoff, K. M., D. G. White, M. E. Hume, T. L. Poole, D. J. Nisbet. 2005. The
Chloramphenicol resistance genes chA is disseminated on transferable plasmids
that confer multiple-drug resistance in swine Escherichia coli. FEMS
Microbiology Letters 243: 285-291.

Boerlin, P., R. Travis, C. Gyles, R. Reid-Smith, N. J anecko, H. Lirn, V. Nicholson, S. A.
McEwen, R. Friendship, and M. Archambault. 2005. Antimicrobial resistance
and virulence genes of Escherichia coli isolates from swine in Ontario. Appl.
Environ. Microbiol. 71(11): 6753-6761.

Bonjoch, X., E. Balleste, A.R. Blanch. 2005. Enumeration of bifidobacterial
Populations with selective media to determine the source of waterborne fecal
pollution. Water Research 39: 1621-1627.

Booth, A. M., C. Hagedom, A. K. Graves, S. C. Hagedom, K. H. Mentz. 2003. Sources
of fecal pollution in Virginiais blackwater river. J. Environ. Eng 129: 547-552.

Bower, R. J. 2001. Fecal source identification using antibiotic resistance analysis. Puget
Sound Notes 45:3-8.

Burkhardt, M. R. 1999. Determination of submicrograrn-per liter concentrations of
caffeine in surface water and groundwater samples by solid-phase extraction and
liquid chromatography. J. AOAC Int. 82: 161-166.

94

Bumes, BS. 2003. Antibiotic resistance analysis of fecal coliforms to determine fecal
pollution sources in a mixed-use watershed. Environ. Monit. Assess. 85: 87-98.

Carroll, 5., M. Hargreaves, and A. Goonetilleke. 2005. Sourcing faecal pollution from
onsite wastewater treatment systems in surface waters using antibiotic resistance
analysis. J. of Appl. Microbiol 99:471-482.

Carson, A. C., B. L. Shear, M. R. Ellersieck, A. Asfaw. 2001. Identification of fecal
Escherichia coli from humans and animals by ribotyping. Appl. Environ.
Microbiol. 67: 1503-1507.

Ghee-Sanford, J .C., R.I. Aminov, I.J. Krapac, N. Garrigues-JeanJean, R.I. Mackie, 2001.
Occurrence and diversity of tetracycline resistance genes in lagoons and
groundwater underlying two swine production facilities. Appl. Environ.
Microbiol 67(4): 1494-1502

Choi, S., W. Chu, J. Brown, S. Becker, V. J. Harwood, S. C. Jiang. 2003. Application
of Enterococci antibiotic resistance patterns for contamination source
identification at Huntington Beach, California. Mar. Poll. Bull. 46:

748-755

Cole, D., D.J.V. Drum, D.E. Stallknecht, D.G. White, M.D. Lee, S. Ayers, M. Sobsey,
and J .J . Maurer. 2205. Free-living Canada geese and antimicrobial resistance.
Emerg. Infect. Dis. 11(6): 935-938.

Collins, R. 2004. Wetlands and aquatic processes: Fecal contamination of pastoral
wetlands. J. Environ. Qual. 33: 1912-1918.

Crozier, J. B., B. Clark, H. Weber. 2002. Identifying sources of fecal pollution in the
Roanoke River, Roanoke County, Virginia. Virginia J. Sci. 53(3): 157-165.

Desmarais, T.R., H.M. Solo-Gabriele, and OJ. Palmer. 2002. Influence of soil on fecal
indicator organisms in a tidally influenced subtropical environment. Appl.
Environ. Microbiol. 68:1 165-1 172.

Dombek, P.E., L.K. Johnson, S. T. Zimmerley, M. J. Sadowsky. 2000. Use of repetitive

DNA sequences and the PCR to differentiate Escherichia coli isolates from
human and animal sources. Appl. Environ. Microbiol. 66(6): 2572-2577 .

Edberg, S.C., E.W. Rice, R.J. Karlin, M.J. Allen. 2000. Escherichia coli: the best
biological dinking water indicator for public health protection. J. Appl.
Microbiol. 88: 1068-1168.

Enne, V., D. Liverrnore, P. Stephens, L. Hall. 2001. Persistence of sulphonamide

95

resistance in Escherichia coli in the UK despite national prescribing restriction.
Lancet 357: 1325-1328.

Fallacara, D. M., T. Y. Monahan, R. F. Wack. 2001. Fecal shedding and antimicrobial
susceptibility of selected bacterial pathogens and a survey of intestinal parasites in
free-living waterfowl. Avian Dis. 45: 128-135.

Fong, T-T, D. W. Griffin, E. K. Lipp. 2005. Molecular assays for targeting human and
bovine enteric viruses in coastal waters and their application for library-
independant source tracking. Appl. Environ. Microbiol. 71(4): 2070-2078.

Geary, PM. and CM. Davies. 2003. Bacterial source tracking and shellfish
contamination in a coastal catchment. Water Sci. Technol. 47 (7-8): 95-100.

Gilliver, M.A., M. Begon, S.M. Hazel, C.A. Hart. 1999. Antibiotic resistance found in
wild rodents. Nature 401: 233-234.

Guan, S., R. Xu, S. Chen, J. Odumeru, C. Gyles. 2002. Development of a procedure for
discriminating among Escherichia coli isolates from animal and human sources.
Appl. Environ. Microbiol. 68: 2690-2698.

Grabow, W. 0. K., 0. W. Prozesky, and J. S. Burger. 1975. with transferable or non
transferable drug resistance. Water Res. 9:777-782.

Graves, A.K., C. Hagedom, A. Teetor, M. Mhal, A.M. Booth, R.B. Reneau. 2002.
Antibiotic resistance profiles to determine sources of fecal contamination in a
rural Virginia watershed. J. Environ. Qual. 31 : 1300-1308.

Griffin, D. W., C. J. Gibson, 1. I. 1., E. K. Lipp, K. Riley, J. H. Paul 111, and J. B. Rose.
1999. Detection of viral pathogens by reverse transcriptase PCR and of microbial
indicators by standard methods in the canals of the Florida Keys. Appl. Environ.
Microbiol. 65: 4118-4125.

Guan, S., R. Xu, S. Chen, J. Odumeru, C. Gyles. 2002. Development of a procedure for
discriminating among Escherichia coli Isolates from Animal and Human Sources.
Appl. Environ. Microbiol. 68(6): 2690-2698.

Hagedom, C., Roberson, S.L., Filtz, J .R., Grubbs, S.M., Angier,T.A., Reneau, RB.
1999. Determining sources of fecal pollution in a nrral Virginia watershed with
antibiotic resistance pattern in fecal streptococci. Appl. Environ. Microbiol.
65(12): 5522-5531.

Hand, D. J. 1981. Discrimination and Classification. John Wiley and sons, Inc., New
York.

Harwood, V. J ., J. Butler, D. Parrish, and V. Wagner. 1999. Isolation of fecal coliform

96

bacteria from the diamondback terrapin (Malaclemys terrapin centrata). Appl.
Environ. Microbiol. 65: 5522-5531.

Harwood, V.J., J. Whitlock, V.H. Withington. 2000. Classification of the antibiotic
resistance patterns of indicator bacteria by discriminant analysis: Use in
predicting the source of fecal contamination in subtropical Florida waters. Appl.
Environ. Microbiol. 66: 3698-3704.

Houndt, T., and H. Ochman. 2000. Long-tenn shifts in patterns of antibiotic resistance
in enteric bacteria. Appl. Environ. Microbiol. 66: 5406-5409.

Hoyle, D.V., C.M. Yates, M.E. Chase-Topping, E.J. Turner, S.E. Davies, J .C. Low, G.J.
Gunn, M.E.J. Woolhouse, S.G.B. Amyes. 2005. Molecular epidemiology of
antimicrobial-resistant commensal Escherichia coli strains in a cohort of newborn
calves. Appl. Environ. Microbiol. 71(11): 6680-6688.

Hsu, F-C., Y. S. C. Shieh, J. van Duin, M. J. Beekwilder, and M. D. Sobsey. 1995.
Genotyping male-specific RNA coliphages by hybridization with oligonucleotide
probes. Appl. Environ. Microbiol. 61: 3960-3966.

Hughes, V.M. and N. Datta. 1983. Conjugative plasmids in bacteria of the "pre
antibiotic" era. Nature (London) 302: 725-726.

Ishii, S., W.B. Ksoll, R.E. Hicks, M.J. Sadowsky. 2006. Presence and grth of
naturalized Escherichia coli in temperate soils from Lake Superior watershed.
Appl. Environ. Microbiol. 72(1): 612-621.

J iang, S. C., R. T. Noble, and W. Chu. 2001. Human adenoviruses and coliphage in
urban runoff-impacted coastal waters of southern California. Appl. Environ.
Microbiol. 67: 179-184.

J imenez-Clavero, M.A., C. Fernandez, J .A. Ortiz, J. Pro, G. Carbonell, J .V. Tarazona,
N. Roblas, V. Ley. 2003. Teschoviruses as indicators of porcine fecal
contamination of surface water. Appl. Environ. Microbiol. 69(10): 6311-6315.

J imenez-Clavero, M.A., E. Escribano-Romero, C. Mansilla, N. Gomez, L. Cordoba,
N. Roblas, F. Ponz, V. Ley, J -C. Saiz. 2005. Survey of bovine enterovirus in
biological and environmental samples by a highly sensitive real-time reverse
transcription-PCR. Appl. Environ. Microbiol. 71(7): 3536-3543.

J ofre, J ., M. Blasi, A. Bosch, and F. Lucena. 1989. Occurrence of bacteriophages
infecting Bacteroides fragilis and other viruses in polluted marine sediments.

Water Sci. Technol. 21: 15-19.

Johnson, L.K., M.B. Brown, E.A. Carruthers, J .A. Ferguson, P.E. Dombek, M.J.

97

 

Sadowsky. 2004. Sample size, library composition, and genotypic diversity
among natural populations of Escherichia coli from different animals influence
accuracy of determining sources of fecal pollution. Appl. Environ. Microbiol.
70: 4478-4485.

Jones, M. E., E. Peters, A. M. Weersink, A. Fluit, J. Verhoef. Widespread occurrence of
integrons causing multiple antibiotic resistance in bacteria. 1997. The Lancet
349: 1742-1743.

Jones, BL. 1999. Potential health risk associated with the persistence of Escherichia
coli 0157 in agricultural environments. Soil Use and Management. 15: 76-83.

Kasper, C. W., J. L. Burgess, I. T. Knight, R. R. Colwell. 1990. Antibiotic resistance
indexing of Escherichia coli to identify sources of fecal contamination in water.
Can J. Microbiol. 36: 891-894.

Kelch, W. J ., and J. S. Lee. 1978. Antibiotic resistance patterns of gram-negative
bacteria isolated from environmental sources. Appl. Environ. Microbiol.
36(3):450—456.

Kelsey, R. H., G. 1. Scott, D. E. Porter, B. Thompson, L. Webster. 2003. Using multiple
antibiotic resistance and land use characteristics to determine sources of fecal
coliform bacterial pollution. Environ. Monit. Assess. 81: 337-348.

Kline-Robach, R. and S. G. Witter. 2002. MSU-Water: Watershed action through
education and research, progress report, July 2001- September 2002. [Online]
http://www.msu-water.msu.edu/asp/reports.asp.

Krumpennan, P. H. 1983. Multiple antibiotic resistance indexing of Escherichia coli to
identify high-risk sources of fecal contamination of foods. Appl. Environ.
Microbiol. 46: 165-170.

Kumar, K., S.C. Gupta, S.K. Baidoo, Y. Chander, C.J. Rosen. 2005. Antibiotic uptake
by plants from soil fertilized with animal manure. J. Environ. Qual. 34: 2082-
2085.

Lachenbruch, P. A. 1975. Discriminant Analysis. Hafner Press, New York.

Lasald, C., R. Rodriguez, and G. A. Toranzos. 2005. Statistical analysis: possible reasons
for unreliability of source tracking efforts. Appl. and Environ. Microbiol. 71(8):
4690-4695. .

Lee, S.-H., and S.-J. Kim. 2002. Detection of infectious enteroviruses and adenoviruses
in tap water in urban areas in Korea. Water Res. 36: 248-256.

Livennore, D.M., M. Warner, L.M.C. Hall, V. Enne, S.J. Projan, P.M. Dunman, 2001.

98

Antibiotic resistance in bacteria fi'om magpies (Pica Pica) and rabbits
(Orycytolagus cuniculus) from west Wales. Environ. Microbiol. 3: 658-661

Mallin, M. A., K. E. Williams, E. C. Esharn, R. P. Lowe. 2000. Effect of human
development of bacteriological water quality in coastal watersheds. Ecol. Apll.
10: 1047-1056.

Mathers, KB. 2001. Antimicrobial drug use and veterinary costs in US. livestock
production. USDA Agriculture Information Bullletin 766.

McLachlan, G. J. 2005. Discriminant Analysis and Statistical Pattern Recognition. John
Wiley and sons, Inc., New Jersey.

Meays, C. L., K. Broersma, R. Nordin, and A. Mazumder. 2004. Source tracking fecal
bacteria in water: a critical review of current methods. J. Environ. Manage. 73:
71-79.

Middleton, J .H., and A. Ambrose. 2005. Enumeration and antibiotic resistance patterns
of fecal indicator organisms isolated fi'om migratory canada geese (branta
canadensis). J. Wild. Dis. 41(2): 334-341.

Miescier, J. J ., and V. J. Cabelli. 1982. Enterococci and other microbial indicators in
municipal wastewater effluents. J. Water Pollut. Control Fed. 54: 1599-1606.

Neu, HQ 1999. The crisis in antibiotic resistance. Science, New Series 257(5073):
1064-1073.

Olsen, J. E., D. J. Brown, D. L. Baggesen, and M. Bisgaard. 1992. Biochemical and
molecular characterization of Salmonella enterica serovar berta, and comparison
of methods for typing. Epidemiol. Infect. 108: 243-260.

Osterblad, M., K. Norrdahl, E. Korpimaki, and P. Huovinen. 2001. Antibiotic resistance.
How wild are wild mammals? Nature 409: 37-38.

Parveen, S., R.L. Murphree, L. Edmiston, C.W. Kasper, K.M. Portier, M.L. Tamplin.
1997. Association of multiple-antibiotic-resistance profiles with point and
nonpoint sources of Escherichia coli in Apalachicola Bay. Appl. Environ.
Micribiol. 63(7): 2607-2612.

Parveen, 8., KM. Portier, K. Robinson, L. Edmiston, M.L. Tamplin. 1999.
Discriminant analysis of ribotype profiles of Escherichia coli for differentiating
human and nonhuman sources of fecal pollution. Appl. Environ. Microbiol.
65(7): 3142-3147.

Parveen, 8., NC. Hodge, R.E. Stall, S.R. Farrah, M.L. Tamplin, 2001. Phenotypic and

99

genotypic characterization of human and nonhuman Escherichia coli. Wat. Res.
35(2): 379-386.

Prescott, J .F ., J .D. Baggot. 1993. Antimicrobial therapy in veterinary epidemiology, 2nd
ed. Iowa State University Press, Ames.

Prescott, J .F ., W.J.B. Hanna, R. Reid-Smith, K. Drost. 2002. Antimicrobial drug use
and resistance in dogs. Can Vet J. 43: 107-116.

Rhodes, M. W., and H. Kator. 1999. Sorbitol-ferrnenting bifidiobacteria as indicators of
diffuse human fecal pollution in estuarine watersheds. J. Appl. Microbiol. 87:
528-535.

Richmond, M. H., G. W. Jack, R. B. Sykes. 1971. The beta-lactamases of gram negative
bacteria including pseudomonads. Ann. N. Y. Acad. Sci. 182: 243-257.

Ritter, K. J ., E. Carruthers, C. A. Carson, R. D. Ellender, V. J. Harwood, K. Kingsley, C.
Nakatsu, M. Sadowsky, B. Shear, B. West, J. E. Whitlock, B. A. Wiggins, and J.
D. Wilbur. 2003. Assessment of statistical methods used in library-based
approaches to microbial source tracking. Journal of Water and Health 1(4): 209-
223.

Rolland, R.M., G. Hausfater, B. Marshall, S.B. Levy. 1985. Antibiotic-resistant
bacteria in wild primates: increased prevelence in baboons feeding on human
refuse. Appl. Environ. Microbiol. 49: 791-794.

Routman, E., D.M. Miller, J. Phillips-Conroy, D. Hart]. 1985. Antibioitc resistance and
population structure in Escherichia coli from free-ranging African yellow
baboons. Appl. Environ. Microbiol. 50(4): 749-754.

Sayah, RS. 2004. Identifying sources of surface water contamination by discriminant
analysis of antimicrobial resistance in E. coli. MS. Thesis, Michigan State
University, East Lansing, MI.

Sayah, R.S., J .B. Kaneene, Y. Johnson, R. Miller. 2005. Patterns of antimicrobial
resistance observed in Escherichia coli isolates obtained from domestic- and wild-
animal fecal samples, human septage, and surface water. Appl. Environ.
Microbiol. 71(3): 1394-1404.

Selvaratnam, S. and J .D. Kunberger. 2004. Increased frequency of drug-resistant
bacteria and fecal coliforms in an Indiana creek adjacent to farmland amended
with treated sludge. Can. J. Microbiol. 50(8): 653-656.

Scott, T. M., J. B. Rose, T. M. Jenkins, S. R. Farrah, and J. Lukasik. 2002. Microbial

source tracking: current methodology and future direction. Appl. Environ.
Microbiol. 68: 5796-5803.

100

 

Schroeder, C.M., C. Zhao, C. DebRoy, J. Torcolini, S. Zhao, D. G. White, D. D. Wagner,
P. F. McDermott, R. D. Walker, and J. Meng. 2002(a). Antimicrobial resistance
of Escherichia coli 0157 isolated from humans, cattle, swine, and food. Appl.
Environ. Microbiol. 68: 576-581.

Schroeder, C. M., J. Meng, S. Zhao, C. DebRoy, J. Torcolini, C. Zhao, P. F. McDermott,
D. D. Wagner, R. D. Walker, and D. G. White. 2002(b). Antimicrobial resistance
of Escherichia coli 0126, 0103, 01 1 l, 0128, and 0145 from animals and
humans. Emerg. Infect. Dis. 16:1409-1414.

Shehane, S. D., V. J. Harwood, J. E. Whitlock, and J. B. Rose. 2005. The influence of
rainfall on the incidence of microbial faecal indicators and the dominant sources
of faecal pollution in a Florida river. J. Appl. Microbiol. 98: 1127-1136.

Siegel, D., Huber, W.G., Drysdale, S. 1975. Human therapeutic and agricultural uses of
antibacterial drugs and resistance of the enteric flora of humans. Antimicrobial
Agents and Chemotherapy 8(5): 538-543.

Selvaratnam, S. and J .D. Kunberger. 2004. Increased frequency of drug-resistant
bacteria and fecal coliforms in an Indiana creek adjacent to farmland amended
with treated sludge. Can. J. Microbiol. 50(8): 653-656.

Simmons, G. M., Jr., S. A. Herbein, and C. M. James. 1995. Managing nonpoint fecal
coliform sources to tidal inlets. Water Resour. Update 100: 64-74.

Sinton, L. W., R. K. Finlay, and D. J. Hannah. 1998. Distinguishing human from animal
faecal contamination in water: a review. N. Zeal. J. Mar. Freshwater Res. 32: 323-
348.

Smith, P. R., E. Farrell, and K. Dunican. 1974. Survival of R' Escherichia coli in sea
water. Appl. Microbiol. 27:983-984.

Smith, D.L., Harris, A.D., Johnson, J .A., Silbergeld, E.K., Morris, J .G. 2002. Animal
antibiotic use has an early but important impact on the emergence of antibiotic
resistance in human commensal bacteria. Proc Natl Acad Sci USA 99(9): 6434-
6439.

Standley, L. J ., L. A. Kaplan, and D. Smith. 2000. Molecular tracer of organic matter
sources to surface water resources. Environ. Sci. Technol. 34: 3124-3130.

Summers, A. O. 2002. Generally overlooked fundamentals of bacterial genetics and
ecology. Clin. Infect. Dis. 34(Suppl 3): $85-92.

Tsubokurea, M.A., A. Matsumoto, K. Otsuki, S.B. Animas, T. Sanekata. 1995. Drug

101

resistance and conjugative R plasmids in Escherichia coli strains isolated from
migratory waterfowl. J. Wild. Dis. 31: 352-357.

USEPA (U .S. Environmental Protection Agency). 2001. Protocol for developing
pathogen TMDLs. EPA 841-R-00-002. Office of Water (4503F), United States
Environmental Protection Agency, Washington, DC, 132pp.

Webster, L.F., B.C. Thompson, M.H. Fulton, D.E. Chestnut, R.F. Van Dolah, A.K.
Leight, G.I. Scott. 2004. Identification of sources of Escherichia coli in South

Carolina estuaries using antibiotic resistance analysis. J. Exper. Mar. Biol. Ecol.
298: 179-195.

Weiskal, P. K., B. L. Howes, G. R. Heufelder. 1996. Coliform contamination of a coastal

embayment: sources and transport pathways. Environ. Sci. Technol. 30, 1872-
1 881 .

Whitlock, J .E., D.T. Jones, V.J. Harwood, 2002. Identification of the sources of fecal

coliforms in an urban watershed using antibiotic resistance analysis. Water Res.
36: 4273-4282

Wiggins, BA. 1996. Discriminant analysis of antibiotic resistance patterns in fecal
streptococci, a method to differentiate human and animal sources of fecal
pollution in natural waters. Appl. Environ. Microbiol. 62: 3483-3486.

Wiggins, B. A., R. W. Andrews, R. A. Conway, C. L. Corr, E. J. Dobratz, D. P.
Dougherty, J. R. Eppard, S. R. Knupp, M. C. Limjoco, J. M. Mettenburh, J. M.
Rinehardt, J. Sonsino, R. L. Torrijos, and M. E. Zimmerman. 1999. Use of
antibiotic resistance analysis to identify nonpoint sources of fecal pollution.
Appl. Environ. Microbiol. 65: 3483-3486.

Wiggins, B.A., P.W. Cash, W.S. Creamer, S.E. Dart, P.P. Garcia, T.M. Gerecke, J. Han,
B.L. Henry, K.B. Hoover, E.L. Johnson, K.C. Jones, J .G. McCarthy, J .A.
McDonough, S.A. Mercer, M.J. Noto, H. Park, M.S. Phillips, S.M. Pumer, B.M.
Smith, E.N. Stevens, A.K. Vamer. 2003. Use of antibiotic resistance analysis for

representativeness testing of multiwatershed libraries. Appl. Environ. Microbiol.
69(6) 3399-3405.

Witte, W. 1998. Medical consequences of antibiotic use in agriculture. Science 279:996
997.

Young, K. D., and E. L. Thackston. 1999. Housing density and bacterial loading in urban
streams. J. Environ. Eng. 125: 1177-1180.

102

       

  
 

 
   

MICHIGA

IIIIIIWII

3 12@@

  

U

III/III

L 15R

ill/Iii)

45 6840

III