PLACE N RETURN BOX to tomcat-thi- checkout mm your record.
TO AVOID FINES Mom on or baforo dd. duo.

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU IsAn Afflnnatlvo Action/Equal Opponunlty Initiation
Wanna-m

 

 

 

 

ASSOCIATIONS BETWEEN BACTERIA AND CONJUGATIVE PLASMIDS:
MODEL SYSTEMS FOR TESTING EVOLUTIONARY THEORY

By

Paul Eugene Turner

A DISSERTATION
Submitted to
Michigan State University

in partial fiilfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Zoology

1995

ABSTRACT

ASSOCIATIONS BETWEEN BACTERIA AND CONJUGATIVE PLASMIDS:
MODEL SYSTEMS FOR TESTING EVOLUTIONARY THEORY

By

Paul Eugene Turner

Interactions between bacteria and conjugative plasmids were used to address two
distinct questions. (1) The first question concerns the relationship between genetic
variability and rates of adaptation. Can plasmid-mediated recombination be used to
generate novel genotypes, and thereby accelerate the rate of adaptation in otherwise
asexual populations of bacteria? To test this hypothesis, genetically distinct Hfi' (high
frequency recombination) cells were periodically introduced into experimental populations
of Escherichia coli. Recombinant genotypes became numerous (or even fixed) in all
treatment populations, but, surprisingly, these recombinants showed no significant increase
in fitness relative to the recipient's ancestor, nor relative to non-recombining control
populations. One possibility is that fitness measurements do not accurately reflect
complex selection dynamics, such as frequency-dependent and even nontransitive
interactions, brought on by recombinant genotypes. In at least one recombination
treatment population, further experiments confirmed that frequency-dependent selection
allows recombinant genotypes to coexist. This stable coexistence of two genotypes on a
single limiting resource was promoted by a cross-feeding interaction, but some evidence of
a demographic tradeofi‘ was also found. (2) The second question concerns the evolution
of virulence in pathogens and other infectiously transmitted elements. Does host density
influence the evolution of plasmid virulence and mode of transmission? This hypothesis
was tested by allowing associations of E. coli and plasmid pBlS to evolve in replicated

environments with different inputs of susceptible hosts. The plasmids' effects on host

fitness and their conjugation rates were both observed to evolve, with a consistent tradeofl‘
between rates of horizontal and vertical transmission. However, manipulations of host
density had no effect on the evolution of plasmid virulence and mode of transmission. One
possible explanation is that conjugation of pBlS does not behave according to mass-

action, and some evidence in support of this possibility is presented.

Dedicated to
Sylvia Baskerville Turner
and
Eugene Turner

For their neverending love,
and confidence in my abilities.

iv

ACKNOWLEDGMENTS

I thank my parents, Eugene and Sylvia, and my siblings, Lennie and Peter, for
providing a loving environment in which to grow. I am extremely grateful to all those
friends who helped me escape fi'om the tedium of graduate school, especially the Boys of
Summer: Keith Hartley, Matthew Fitzgibbons, and Dennis Verrette, without whom I
would (literally) not be here today. I thank my fellow graduate students and colleagues at
University of California, Irvine, and Michigan State University for the discussion of ideas
that led to this dissertation. These pillars of thought include (but are not limited to) Kevin
Bailey, Tom Clark, Ananias Escalante, Jorge Santo Domingo, and the members of the
Lenski lab (who, unfortunately, are too numerous to mention). I thank my advisory
committee at MSU (Guy Bush, Dennis Fulbright, Donald Hall, and Andrew Jarosz) for the
discussion of my written work and difficult (but insightfirl) exam questions. A special
thanks goes to Joseph Graves for encouraging me to pursue graduate studies at UCI. I
owe a debt of gratitude to Albert Bennett and Richard Hudson for their patient provision
of laboratory and office space while I finished my thesis. In addition, I am forever
indebted to my fiiend and mentor Richard Lenski, for his demonstration of the qualities
necessary to excel in scientific research, and for his tolerance of my correspondence-
course style graduate career. Most of all, I would like to thank my dear fiiend Mary Beth
Decker, because graduate school would have been unbearable without the happiness and

good times we shared over the past few years.

TABLE OF CONTENTS

List of Tables
List of Figures

Introduction
Plasmid Biology
Thesis Overview

Chapter I - Effects of recombination with exogenous genotypes on the rate of
bacterial evolution
Introduction
Experimental overview
Materials and Methods
Bacterial strains
Culture conditions
Recombination treatment
Control treatment
Fitness assays
Screening for recombinant genotypes
Phenotypic markers
Allozyme-electrophoresis markers
Results
Genetic changes
Changes in fitness
Discussion
Potential for gene flow to have swamped adaptive evolution in the
treatment populations
Potential for complex selection dynamics to have obscured more
rapid adaptive evolution in the treatment populations
Conclusions

Chapter II - Tests of ecological mechanisms promoting the stable coexistence of
recombinant bacterial genotypes
Introduction
Cross-feeding
Demographic tradeofl‘
Materials and Methods

xii

~

10
10
13
13
14
16
17
l7
17
17
17
22
24

25

28
3O

33
33
35
36
39

Bacterial strains
Media and culture conditions
Fitness assay and definitions
Results
Demonstration of the stable equilibrium
Evidence for frequency-dependence
Evidence for stable equilibrium
Evaluation of the demographic tradeofi‘ hypothesis
Estimation of maximum growth rates
Estimation of affinity for limiting resource
Evaluation of the cross-feeding hypothesis
Effect of resource concentration on frequency-dependence
Evidence for cross-feeding after glucose has been depleted
Effect of potential metabolites on relative fitness
Discussion

Chapter III - Tradeofi‘ between horizontal and vertical modes of transmission in
bacterial plasmids
Introduction
Experimental system
Theoretical predictions
Experimental overview
Materials and Methods
Bacterial strains
Media and culture conditions
Plasmid electrophoresis and restriction
Experimental treatments: control populations
Experimental treatments: host density manipulations
Fitness assay
Conjugation rate
Cost of plasmid carriage
Results
Ancestral plasmid traits
Ancestral cost of plasmid carriage
Ancestral conjugation rate
Control populations
Evolutionary dynamics
Fitness changes
Host density manipulations
Evolutionary dynamics
Fitness changes
Changes in plasmid traits
Cost of plasmid carriage
Conjugation rate
Tradeofi‘ in modes of transmission

vii

39
39
42
43
43
43
43
45
45
47
50
50
53
58
6O

66
66
67
68
69
73
73
75
76
76
76
77
78
79
80
80
8O
81
82
82
82
85
85
88
91
91
94
96

Discussion 96

Chapter IV - Unexpected effect of host density on conjugation rate and

invasion of plasmid pBlS 101

Introduction 101
Materials and Methods 102
Results 103
Discussion 1 10
Appendix 1 12
List of References 116

viii

LIST OF TABLES

Table 1. Genetic differences between donor and recipient strains used in Chapter I.

Table 2. Summary of the temporal dynamics of genetic change in experimental
populations.

Table 3. Summary of the genetic changes in experimental populations at the end
of 1,000 generations.

Table 4. Mean fitness for each experimental population, and for the ancestral
genotypes.

Table 5. Analysis of the proportion of recipients obtaining the Tetr marker from
the donor strains during a standard mating cycle.

Table 6. Genetic markers for the two recombinant strains of E. coli used in
Chapter II, as well as their parental donor and recipient strains.

Table 7. ANOVA of effects of initial frequency and glucose concentration on
fitness of lLac+ relative to Lao“.

Table 8. ANOVA of effects of initial frequency and final sample time
(12 or 24 h) on fitness of Lac+ relative to Lac’.

Table 9. ANOVA of effects of initial fi'equency and supplemental acetate
concentration on fitness of Lac+ relative to Lac‘.

Table 10. ANOVA of effects of initial frequency and supplemental glycerol
concentration on fitness of Lac+ relative to Lac'.

Table 11. Summary of experimental and control treatment groups for Chapter III.
Table 12. Expected evolutionary changes in traits pertaining to plasmid virulence.
Table 13. Key bacterial strains used in Chapter III.

Table 14. Ancestral plasmid traits.

11

20

21

23

27

40

51

54

59

59

72

73

74

81

Table 15. Nested AN OVA to examine the effects of susceptible-host-density
treatment, and population within treatment, on fitness of evolved
populations relative to ancestor. 89

Table 16. Mixed-model two-way ANOVA to examine the effects of plasmid
genotype and assay environment on the change in cost of plasmid carriage. 92

Table 17. Estimates of log 10 conjugation rate (7) for pBlS at three different
glucose concentrations, and corresponding cell densities. 107

Table 18. AN OVA of the effect of glucose concentration on loglo conjugation
rate (7) of plasmid pBlS. 108

Table 19. Genotypes seen among ten isolates from each recombination treatment 112
population at generation 1,000.

LIST OF FIGURES

Figure 1. Dynamics of a conjugative plasmid and its bacterial host. 3

Figure 2. Mean fitness relative to the ancestor for the twelve E. coli populations
described by Lenski and Travisano (1994). 8

Figure 3. Survival of E. coli K12 donors, E. coli B recipients, and recombinant
genotypes during the five-day cycle of the recombination treatment. 15

Figure 4. Evolutionary dynamics in population Ara'l of the recombination
treatment. 18

Figure 5. Complex selection dynamics revealed by pairwise interactions among
three genotypes. 31

Figure 6. Numerical simulations showing stable coexistence of two strains
on a single resource in a seasonal environment, mediated by a
demographic tradeoff. 38

Figure 7. The fitness of recombinant E. coli strain Lac+, relative to strain Lac“,
is a decreasing function of its own frequency. 44

Figure 8. Starting from different initial frequencies, strains Lac+ and Lac‘
establish a stable polymorphism in DM25. 46

Figure 9. Numerical simulations of the fitness of Lac+ relative to Lac‘, as a
firnction of the initial frequency of Lac+, assuming only a demographic
tradeofl‘ between growth rates at high and low glucose concentrations. 49

Figure 10. Fitness of strain Lac+, relative to strain Lac‘, as a function of its initial
frequency, in DM media containing five different glucose concentrations. 52

Figure 11. Fitness of strain Lac+, relative to strain Lac‘, as a function of its initial
frequency, in DM25, calculated between 0 and 12 h and between
0 and 24 h. 56

Figure 12. Net rate of change in viable cell density for Lac+ and Lac' in DM25
between 12 and 24 h, after glucose has been exhausted from the medium. 57

Figure 13. Effect of supplemental glycerol concentration on the fitness of Lac+
relative to Lac', in medium also containing glucose at 2.5 ug mL'l.

Figure 14. Horizontal, vertical, and net rates of increase for two plasmid
genotypes, as a fimction of susceptible (plasmid-flee) host density.

Figure 15. Evolutionary dynamics in the plasmid-bearing control populations.

Figure 16. Fitness trajectories for plasmid-bearing and plasmid-free control
populations during evolution in the antibiotic-free environment.

Figure 17. Changes in the frequency of Ara+ immigrant backgrounds in plasmid-
bearing treatment populations subjected to medium or high levels of
immigration by plasmid-free cells.

Figure 18. Changes in the frequency of Tets plasmid variants in treatment
populations subjected to low, medium, or high levels of immigration
by plasmid-free cells.

Figure 19. Mean fitness in treatment populations subjected to low, medium,
and high levels of immigration by plasmid-free cells.

Figure 20. Change in cost of carriage to the host (Ac) for the eight evolved
plasmids that retained their ability to conjugate.

Figure 21. Conjugation rates (y) for the ancestral and ten evolved plasmids.
Figure 22. Genetic correlation between rate parameters governing horizontal
and vertical modes of plasmid transmission in pB 15 and its evolved

derivatives.

Figure 23. Densities of donors, recipients, and transconjugants of plasmid pBlS
during serial transfer at seven different concentrations of glucose.

Figure 24. Dynamics of one-day mating experiments with pBlS and E. coli B
at three glucose concentrations.

xii

61

70

83

84

86

87

9O

93

95

97

104

109

INTRODUCTION

Plasmid Biology

Plasmids are circular, extrachromosomal DNA molecules able to autonomously
replicate within a bacterial cell. Many characteristics exhibited by bacteria that are of
importance in medicine, agriculture, commerce, and the environment are, in fact, plasmid-
determined. Such characteristics include resistance to antibiotics, the ability of nitrogen-
fixing Rhizobium strains to nodulate roots of legumes, antibiotic production by
Streptomycetes, and the biodegradation of certain herbicides (Hardy 1986; Kinashi et al.
1987). Although plasmids are a nearly ubiquitous feature of naturally-occurring species of
bacteria, under most growth conditions they are dispensable to their host cells (Freifelder
1987).

All plasmids are able to control their own replication using the host's cellular
machinery and are transferred vertically across generations of the host. In the absence of
selection on the host for specific plasmid-encoded characters (such as antibiotic
resistance), most plasmids reduce the fitness of their hosts relative to isogenic plasmid-free
counterparts (Levin 1980; Dykhuizen and Hart] 1983; Lenski and Bouma 1987; Lenski
and Nguyen 1988; Nguyen et al. 1989); hence, they can be regarded as parasites under
these conditions (Levin and Lenski 1983). Many plasmids are also able to transfer
horizontally from an infected host (donor) to an uninfected host (recipient) through a
process called conjugation (Lederberg 1956). Although some of the details of the
conjugation process are still poorly understood, conjugation is initiated by contact
between donor and recipient cells via a plasmid-encoded protein appendage called a sex

pilus. Thus, conjugative plasmids are transmitted by two distinct modes: horizontal

(infectious) transmission occurs by conjugation, while vertical (intergenerational)
transmission occurs by host cell division (Figure 1).

Bacterial reproduction per se is strictly asexual. However, bacteria may undergo
sex via recombination when a plasmid integrates into the host chromosome and retains its
ability to transfer by conjugation. The best known example of chromosomal transfer is the
integration of an F plasmid into an Escherichia coli host, which converts the host into a
high frequency recombination (Hfi') cell. An Hfi' cell (donor) has the ability to conjugate
with a recipient cell lacking the F plasmid (F ') and to donate copies of its chromosomal
genes. During transfer of Hfr DNA to a recipient cell, the mating pair usually breaks apart
before the entire chromosome is transferred, but on average several hundred genes are
transferred before the cells separate (Freifelder 1987; Lloyd and Buckrnan 1995).
Separation almost always occurs before the final segment of F is transferred; thus, the
recipient usually remains F'. During or after Hfr transfer, regions of the transferred DNA
fragment are frequently exchanged with the recipient chromosome, thereby converting the

recipient into a recombinant genotype.

Thesis Overview

In this dissertation, I describe experiments using conjugative plasmids and their
bacterial hosts that test two distinct evolutionary hypotheses. First, I hypothesize that
plasmid-mediated recombination will generate novel genotypes and, thereby, accelerate
the rate of adaptation in otherwise asexual populations of E. coli. Second, I hypothesize
that the evolution of plasmid transmission and virulence is determined by the density of
uninfected bacterial hosts in the environment.

Chapter I examines the efi‘ects of plasmid-mediated recombination on rates of
fitness increase in experimental populations of bacteria. Earlier studies have described the
evolutionary dynamics in experimental populations of E. coli whose sole source of genetic

variation was spontaneous mutation (Lenski et a1. 1991; Lenski and Travisano 1995).

 

 

Q ‘
Segregation
65mm “Aw Plasmid /bea>

cell cell
multiplication ® multiplication ®

\0. / w/

@363

lConjugation

x Cf J

 

 

Figure 1. Schematic representation of the dynamics of a conjugative plasmid and
its bacterial host. Conjugative plasmids (o) are transmitted in two ways: vertically through
plasmid-bearing cell multiplication, and horizontally through conjugation. Modified from
Levin and Lenski (1983).

Over the course of thousands of generations, the rate of fitness increase in these
populations diminished substantially, apparently because the populations exhausted most
of the beneficial mutations. Chapter I describes a series of experiments to determine
whether the rate of fitness improvement in these populations could be re-accelerated by
plasmid-mediated recombination, which would increase the available genetic variation.

In one of these treatment populations, I observed coexistence between two
recombinant genotypes on a single limiting resource, glucose. This polymorphism does
not conform to a simple model of competitive exclusion (Gause 1934; Hardin 1960).
Chapter II demonstrates that this polymorphism is indeed stable, and then it describes a
series of experiments to test two alternative hypotheses that might explain the stable
polymorphism: (l) a demographic tradeofi‘, such that one genotype is competitively
superior when glucose is abundant whereas the other genotype is the better competitor for
sparse glucose; and (2) a cross-feeding interaction, whereby the superior competitor for
glucose excretes a metabolite for which the other genotype is the better competitor.

Chapter III examines the impact of susceptible host density on the evolution of
plasmid virulence and mode of transmission. When susceptible hosts are common, the
opportunity for infectious transfer is great and selection is hypothesized to favor increased
rates of horizontal transmission, even at the expense of reduced vertical transmission
caused by increased virulence. When available hosts are rare, however, vertical
transmission is the more fi'equent mode of transfer and selection should favor increased
rates of vertical transmission by reducing virulence, even at the expense of reduced
infectious transmission. I tested this hypothesis by allowing associations between E. coli
and a plasmid to evolve in replicated environments supplemented with different densities
of susceptible hosts.

A key assumption of the model for evolution of plasmid virulence is that bacterial
conjugation behaves according to mass-action. That is, the opportunity for horizontal

transfer by conjugative plasmids is supposed to increase in direct proportion to the density

of available hosts. In Chapter IV, I begin to explore the validity of this assumption in light
of some unexpected results in the preceding experiment. To do so, I tested the ability of
conjugative plasmids to invade bacterial populations at low, intermediate, and high
densities of plasmid-flee hosts, and I independently estimated their conjugation rates under

each of these conditions.

CHAPTER I
EFFECTS OF RECONIBINATION WITH EXOGENOUS GENOTYPES ON
THE RATE OF BACTERIAL EVOLUTION

INTRODUCTION

In contrast to most other organisms, reproduction and sexuath in bacteria are
discrete and independent functions (Dykhuizen and Green 1991). Bacterial reproduction
per se is strictly asexual, occurring by binary fission But if sex is defined as the exchange
of genetic material between organisms, then bacteria undergo sexual recombination
through the processes of transformation, viral-mediated transduction, and plasmid-
mediated conjugation (Levia 1988; Hopwood and Chater 1989; Maynard Smith 1990;
Dykhuizen and Green 1991; Maynard Smith et al. 1993). For example, an F plasmid
inserted into a bacterial chromosome converts the bacterium into an Hfi' (high fiequency
recombination) donor strain.

Population-genetic studies of bacteria isolated from nature have demonstrated that
some species, including Escherichia coli, have very high levels of linkage disequilibrium,
indicating clonal population structures (Selander and Levin 1980; Whittam et al. 1983;
Caugant et al. 1984; Maynard Smith et al. 1993; Whittam and Ake 1993). However,
there is also molecular evidence for occasional chromosomal recombination in natural
populations of even highly clonal bacteria such as E. coli (Milkman and McKane-Bridges
1990; Maynard Smith 1990; Bisercic et al. 1991; Dykhuizen and Green 1991; Maynard
Smith et al. 1991; Whittam and Ake 1993; Guttman and Dykhuizen 1994a, 1994b). In
addition, recent analyses of population genetic structure in Bacillus subtilis (Istock et al.
1992), Rhizobium etIi (Souza et al. 1992), and Neisseria gonorrhea (Maynard-Smith et al.

1993) suggest that recombination is much more fiequent in these species than in E. coli.

In all, the evolutionary significance of recombination in bacteria remains a contentious
subject (Maynard Smith 1990; Maynard Smith et al. 1993; Lenski 1993).

The effect of recombination (along with mutation and migration) in basic models
of population genetics is to increase genetic variation Natural selection may then act on
this variation to increase the mean fitness of an evolving population (Wright 1931, 1932;
Fisher 195 8; Roughgarden 197 9). However, recent analyses have shown that the
implications of increased genetic variation for mean fitness are not always so simple (see,
e.g., Frank and Slatkin 1992).

Bacteria provide an excellent experimental model to study evolutionary processes
such as mutation, natural selection, and adaptation (Luria and Delbruck 1943; Atwood et
al. 1951; Helling et al. 1987; Dykhuizen 1990; Lenski et al. 1991; Bennett et al. 1992;
Lenski 1992; Lenski and Travisano 1994; Travisano et al. 1995). Because of their large
population sizes and short generation times, bacteria may be propagated in defined
environments for hundreds and even thousands of generations, allowing evolutionary
changes and processes to be observed in detail. While a few experimental evolution
studies with bacteria have allowed recombination (Graham and Istock 1979, 1981), these
have not been directly concerned with quantifying the effect of recombination on the rate
of adaptive evolution. In this chapter, I present the results of a 1,000-generation
experiment that was designed to examine the effects of recombination on the rate of

adaptive evolution in otherwise asexual populations of E. coli.

Experimental overview. - The bacterial strains used in this study were isolated
from twelve populations of E. coli B maintained in the laboratory as part of a long-term
evolution experiment (Lenski et al. 1991; Lenski and Travisano 1994). During 10,000
generations of evolution, the mean fitness of these populations relative to a common
ancestor increased by about 50% on average (Figure 2). All twelve populations were

descended from a single clone, which lacked plasmids or functional phage. Hence,

 

 

 

LLJ
Z 0:
*—
S E
LLJ U)
Q: LLJ

0
a Z
LLJ <
Z O
t 1—
L1.

0.9 I l I I I
0 2000 4000 6000 8000 10000

TIME (generations)

Figure 2. Mean fitness relative to the ancestor for the twelve E. coli populations
described by Lenski and Travisano (1994). Symbols give grand mean and 95% confidence
interval for the twelve populations. The curve gives the best fit of a hyperbolic model to
the grand means, and it shows a significant deceleration in the rate of improvement over
time. The populations used in the present study were derived from isolates obtained after

7,000 generations of this earlier study.

mutation was the only source of genetic variation for adaptation by natural selection

Most of the increase in mean fitness occurred during the first few thousands of generations
of this experiment. Thus, by generation 5,000 or so, the rate of adaptation had slowed to
such an extent that the approach to a selective plateau was apparent (Figure 2).

The central question that I sought to address in this study is whether the rate of
adaptive evolution in these populations could be re-accelerated by providing an additional
source of genetic variation, by means of sexual recombination with distantly related
strains. To that end, a single clone was isolated fiom each of the twelve evolving
populations after 7,000 generations, by which time their rates of increase in fitness had
slowed substantially. Each clone was then used to found a new pair of populations, one of
which would experience sexual recombination and the other of which would serve as an
asexual control.

Recombination was achieved by periodically adding plasmid-bearing Hfr (high
fiequency recombination) E. coli K12 donor cells to the treatment populations. These
donors are genetically quite distinct from the E. coli B recipients, thus providing an
opportunity both to introduce substantial genetic variation by sexual recombination and to
score the movement of several marker alleles. To prevent the spread of the donor
genotypes by ecological competition -— in contrast to the spread of their genes by
recombination and subsequent selection - I chose to use donor genotypes that were
deficient in growth in the experimental environment owing to mutations in critical
metabolic genes. Thus, the donor genotypes could transmit their genes by conjugation,
but they could not propagate themselves asexually.

The twelve pairs of recombination treatment and asexual control populations were
then propagated for a further 1,000 generations, in the same environment in which their
asexual progenitors had been propagated for the previous 7,000 generations. While their
environments were identical, the control populations had only mutation as a source of

genetic variation, whereas the treatment populations had recombination with the Hfi'

10

donors as an additional source of variation. Thus, the experiment could address the
following questions:

(i) Did genetic markers from the Hfr donors enter the recipient populations,
indicating that the recombination treatment had the intended effect of increasing the
available genetic diversity?

(1i) Did this additional diversity, in fact, allow more rapid adaptive evolution in the
treatment populations than in the control populations? That is, did sexual recombination

re-accelerate the rate of increase in mean fitness?

MATERIALS AND METHODS

Bacterial strains. - A single clone was isolated at generation 7,000 from each of
the twelve experimental populations described by Lenski and Travisano (1994). Six of the
clones can grow on the sugar L-arabinose (Ara+) and six cannot (Ara'). Ara+ and Ara'
strains form white and red colonies, respectively, on tetrazolium-arabinose (TA) indicator
plates (Levin et a1. 1977). Each clone was used to found one population in each of the
recombination and control treatments described below. Each of these founder, or
baseline, strains was characterized in terms of nine phenotypic traits and five
electrophoretic loci (Table 1). There were no differences among the twelve baseline
recipients, except for the Ara marker. All baseline strains were stored in a glycerol-based
suspension at -80 °C to allow for direct comparisons between them and their evolved
derivatives at a later time.

The donor strains used in this study were four Hfi' strains of E. coli K12. Each of
these donors has an F plasmid inserted at a difl‘erent map position in its chromosome
(Table 1). Thus, each has a difi‘erent point of origin (OriT site) for conjugative transfer

and subsequent recombination. These E. coli K12 donors are rather distantly related to

ll

«Buchanan v5 qubouoa 832 - as + Ahmv £o>EoNqoba v5 Gob Eugen?“ 8quan 3 53% .8538.— 93

33:55... 8.82 N 93 a A: 838— 23 A83 agagé mama no 32» 8 53838.. 5599: 28 Dana 88%5 - 98 + _

 

 

 

a 3 N _ _ _ N a c + - - a a - + 83mg
4 N_ N a _ _ N a c + - - a a - + 83mm
4. N N _ _ _ N c a + + - m c - .. anqmm
e S N _ g _ N a m - + + m c + + «mam
«NE .a8 us 888
t _ N N N _ .— a + + + a a + 4+ 8:33
158$ 8883
$5.5 mmm E: ma meé me 3. NE. >= 34 ma hm Sn. 34 Ed. 5.9m
Nouocofioboo—m 1:50:23

 

N .8920 a 33 a?» ages. as .88 8038.. 888% 8280 ._ 2.3

ll

Sagan—a 93 335303 3885 - 93 + Ame £38395» 93 90.5 3526933 mouomauau 3 502838.. gamma 98

3329.5» 0302 a can a x36 823— 28 A83 80548.7..— flama co 32m 8 535638.. 5:59: v5 55% 3835 .. 98 + _

 

 

 

 

a v» N _ _ _ N N a + - - a a - + magma
a. N— N N _ ~ N a a + - - m .— - + 8N3
A. N N ~ _ _ N a c + + - m a - - ENE
e S N y _ _ N m m - + + m .— + + mwNAmM
was. :8 S 88.5
1 ~ N N N ~ .— m + + + a a + 4+ 8:85
3.6 a8 S 8833
«Bio mmm ES En? hand me 3. NS. >= :3 m2 hm Nom. 03 8< £93m
Nouocoaaoboo—m ~0Ebo=o§

 

N 8.50 5 Be. a?» .8382 as 8:8 803E 888% 8an ._ 2.3.

12

23:89.2 «Beam 8a .GNNBm .SGBm «235 3988“.

2385...: 803 N23. .b:mco>%D 03> Na 8850 :0on 0:80:00 :8 M 05 no 8883 gown .m 88.: 358:0 803 88% 22. n
8:98 «53:68 3:83 0283 05 macaw 8:88 865 N: «353% o: 803 805 6.505% E 05 com 33me v

2036288 .880 as: w:%:oomo% .8 w:%=oomu mo BPS .: .888 888%.:

a no 9 £8 «53:68 8:: 88m 88% no c8888 mo Euro 05 .5.“ 08808020 SM :8 .M 05 co «82:8 5 dogma an: m

Amman guaca— Ba AHA—.6 8.8088: 88:983.: 89:38 Ann—43 035m0%>:o%

3:3? Aamvmév game—v.33 cacaoawoaamofié Ame gowonvfiov 88:08: «083.8 com momma—o 5:508 88%: 838:2 N
.8 23 NE cows... 3 88888

.8888 as areas. 883 a as ... 5,5 agasafifi as as: 0582 .83 8am: £2 ages 8 $38888

.583 _ 2.3.

13

the E. coli B recipients and they differ at several genetic markers (Table 1; see also
Selander and Levin 1980). Each donor has a tetracycline marker located approximately
10 minutes from its On'T site (Table 1). These genetic difl‘erences between donors and
recipients allowed detection of recombinant genotypes generated over the course of the
experiment, by means of selective-plating and isozyme-electrophoresis. In addition, each
donor strain was auxotrophic for at least one amino acid (Table 1), so that it could not
grow in the minimal medium employed in this experiment. Hence, these Hfr strains could
donate genes via recombination but were themselves unable to proliferate and outcompete

the prototrophic E. coli B recipients.

Culture conditions. - The culture medium employed in all experiments was Davis
Minimal (DM) broth (Carlton and Brown 1981) supplemented with 2x10"6 g thiamine
hydrochloride and 25 ug glucose ml'l. This medium supports a stationary-phase bacterial
density of ~5x107 cells ml‘l. Culture volume was 10 ml, maintained in SO-ml Erlenmeyer
flasks and placed in a shaking incubator at 37 0C and 120 rpm. All cultures were serially
propagated each day by transferring 0.1 ml of each stationary-phase (24 h) culture into 9.9
ml of fresh medium. The resulting loo-fold re-growth represents ~6.64 generations of
binary fission each day. The Hfr strains used in this experiment were grown up separately
in Luria broth (LB), a rich medium which allows for a stationary-phase bacterial density of
~2x109 cells ml'l.

Recombination treatment. - A single clone of each baseline strain was used to
initiate each of the twelve populations (six Ara+ and six Ara“) in the recombination
treatment. Every day, for 150 days, 0.1 ml from the previous day's culture was transferred
into 9.9 ml of fresh DM. On day 0, and every fifth day (or 33 generations) thereafter, I
also added 0.01 ml of an equally-proportioned volumetric mixture of the four Hfi' strains,

which had been grown overnight in LB. This manipulation produced initial densities of

14

about 2x106 and 5x105 donor and recipient cells rnl'l, respectively, for a ratio of about
4:1. On those days when donors were added, the flasks were placed in a non-shaking
incubator at 37 0C for one hour to allow uninterrupted mating between donors and
recipients. Subsequently, the flasks were transferred to a shaking incubator at 37 0C for
23 hours. On all other days, the flasks were held in a shaking incubator for all 24 hours.
During the next four days, the cultures were propagated using serial transfer. Every 15
days (100 generations), just prior to the addition of donor cells, 10% glycerol was added
to a sample from each population, and the samples were placed into a freezer at -80 0C
for future study.

Preliminary experiments were conducted to ensure that the protocol for the
recombination treatment would work in two important respects. First, I wanted to make
sure that recombinants (transconjugants) did, in fact, occur at measurable frequencies.
Second, I wanted to make sure that the auxotrophic donors died out afler their addition to
experimental cultures. The dynamics occurring in the five-day cycle of the recombination
treatment are shown in Figure 3. As desired, recombinants possessing the donor's
tetracycline resistance and the recipient's streptomycin resistance were readily detected;
and the number of auxotrophic donors fell below the limit of detection (< 10 cells rnl‘l)
after three days (Figure 3).

Control treatment. -- The same twelve clones (six Ara+ and six Ara') used to
found the populations in the recombination treatment were also used to initiate the control
populations. On day 0, and every fifth day thereafter, each recipient population underwent
a ”sham” manipulation to duplicate the recombination treatment, but without allowing
genetic recombination. In particular, the control populations received 0.01 ml of LB
media in which the Hfr donor strains had been grown to stationary phase, but from which
all cells had been removed by filtration. Also, for the first hour after receiving this

"placebo”, the control populations were placed in the non-shaking incubator prior to being

 

 

 

9
85+ 4 o— r a
7_

26-

.135-

8

94"

u:

33F
2_ \
1_ _.:\_..
O l l I I I

O ‘I 2 8 4 5

Time (days)

Figure 3. Results of a preliminary experiment to determine the survival of E. coli
K12 donors (triangles), E. coli B recipients (circles), and recombinant genotypes (squares)

during the five-day cycle of the recombination treatment.

16

moved to the shaking incubator. Thus, the control populations experienced the same
selective environment as the recombination populations, except for the effects of adding

donors.

Fitness assays. - To assay relative fitness (W), two strains were placed in
competition under the culture conditions described above, where one competitor was
Ara'l’ and the other was Ara’. Each strain was grown separately for one day in DM, as a
preconditioning step to ensure that both competitors were in comparable physiological
states. The two competitors were mixed at a 1:1 ratio, then diluted 1:100 into fi'esh DM
and allowed to grow and compete during a 24-hour grth cycle. Initial and final
densities of each competitor were estimated by spreading them on TA plates, which
permitted the competitors to be distinguished by colony color.

Let the initial densities of the Ara+ and Ara‘ competitors be N1(0) and N2(0),
respectively; and let N 1(1) and N20) be their corresponding densities after one day. The
average rate of increase (or realized Malthusian parameter), mi, for either competitor is

then calculated as:

mi = lawn/lawn I (1 day).

The fitness of one strain relative to another (Wij) is estimated as the ratio of their
Malthusian parameters (Lenski et al. 1991):

Wij = mi/mj.
A fitness difference between the two competitors may reflect difi‘erences in their lag phase,

maximum or submaximum growth rates, survival at stationary phase, or some combination

thereof (e.g., Vasi et al. 1994).

17

Screening for recombinant genobpes
Phenobpic markers. - Every 100 generations, ten clones were isolated at random
from each population in the recombination and control groups. The phenotype of these
randomly-chosen isolates was determined using nine genetic markers that could be scored
on indicator plates: arabinose and lactose utilization, tetracycline and streptomycin
resistance, resistance to the bacteriophages TIX and T6, and auxotrophy for the amino

acids leucine, arginine and isoleucine-valine.

AIIozyme-electrophoresis markers. -- Lysates were prepared for the ten clones
chosen at random from each population at generation 1,000. Each clone was grown
overnight in 10 ml of LB, and the resulting culture was sonicated using the method
described by Pinero et al. (198 8). These clones were scored for five allozyme markers
using cellulose acetate (I-Iellena Laboratories) electrophoresis: alcohol dehydrogenase
(ADI-I), isocitrate dehydrogenase (IDH), mannose phosphate isomerase (MP1), peptidase
(PEP) and 6-phosphogluconate dehydrogenase (6-PGD). These five enzymes were
chosen (from 16 in preliminary screens) because their mobilities were found to difi‘er for
the E. coli K12 donors and E coli B recipients used in my experiment (Table 1). At least
three independent electrophoretic assays were performed on each isolate for each

allozyme tested.

RESULTS

Genetic changes. - Figure 4 shows the genetic changes that were seen in one of
the recombination treatment populations, based on scoring nine physiological traits for ten
clones at loo-generation intervals. Three distinct recombinant genotypes were detected,

and by generation 700 the ancestral genotype had fallen below the limit of detection.

l8

 

 

 

 

1.0 o o o o o I - - l
6 0.8 —
C \
g ‘\
g 0.6 ~ ‘- . ‘
g 3'
> 0.4 -
CC) /fi\\ 3.:
e /
0.2 — / _\
// \\
0.0 ‘ ' 4 a \e
0 200 400 800 1000

Time (generations)

Figure 4. Evolutionary dynamics in population Ara'l of the recombination
treatment. Population samples were obtained every 100 generations, and changes in
genotype fiequency are based upon the nine phenotypic markers described in Table 1.
The ancestral genotype (circles), and three distinct recombinant genotypes (squares,

diamonds, triangles) were detected over the course of the study.

19

Qualitatively similar dynamics were seen in the other recombination treatment populations,
as summarized in Table 2. New genotypes were first seen, on average, at about
generation 550, with 2.67 distinct new genotypes seen during the 1,000-generation
experiment. By contrast, using the same physiological traits and sampling scheme,
absolutely no variants were detected in any of the control populations (Table 2).

The above data must severely underestimate the extent of genetic diversity, and the
rate of genetic change, in the recombination treatment populations. Only nine
physiological markers were scored; three of these markers (those for amino acid
auxotrophy) were subject to strong selection for the ancestral state, and no variation was
seen in those traits. To flirther evaluate the efl‘ect of the recombination treatment, I also
scored ten clones taken from each population at generation 1,000 for the five
electrophoretic markers that distinguished the donor and recipient populations (Table 1).

Table 3 summarizes the diversity revealed by combining the electrophoretic and
physiological markers. On average, there were 6.08 distinct genotypes per recombination
treatment population, which is especially remarkable given the fact that only ten clones
were characterized for each population. The ancestral (recipient) genotype was absent
from 6 of the treatment populations, and the frequency of the ancestral type was 20% or
less in ten of the twelve populations. Again by contrast, none of the electrophoretic or
physiological markers showed any variation in any of the twelve control populations.
Table 19 (see Appendix) lists the various genotypes and their fi'equencies at 1,000
generations.

Evidently, the recombination treatment greatly increased the level of genetic
variation available for natural selection, and it led to much faster rates of evolutionary
change at the genetic level. In the next section, I examine the consequences of this
additional variation for the rate and extent of adaptive evolution, as measured by gains in
fitness.

20

Table 2. Summary of the temporal dynamics of genetic change in experimental

populations.

 

New genotype Ancestral type Number of distinct

 

Population first seen last seen new genotypes
Recombination treatment

Ara'l 600 gen. 600 gen. 3
Ara'2 600 gen. 1,000 gen. 1
Ara'3 300 gen. 200 gen. 2
Ara'4 200 gen. 1,000 gen. 1
Ara'S 500 gen. 500 gen. 5
Ara'6 200 gen. 1,000 gen. 3
Ara+l 800 gen. soo gen. 3
Ara+2 1,000 gen. 1,000 gen. 1
Arm 900 gen. 1,000 gen. 2
Ara+4 200 gen. 1,000 gen. 5
Ara+5 600 gen. 1,000 gen. 4
Ara+6 700 gen. 1,000 gen. 2
Control

All 12 Never 1,000 gen. 0

 

Note: Every 100 generations, ten clones were randomly chosen from each

population and then scored for the nine phenotypic markers listed in Table 1. The

dynamics for recombination treatment population Ara’l are shown in Figure 4.

21

Table 3. Summary of the genetic changes in experimental populations at the end of 1,000

 

 

generations.
Proportion of Number of distinct genotypes,
Population ancestral type including ancestral type
Recombination treatment
Ara’l 0% 5
Ara'2 20% 6
Ara'3 0% 6
Ara'4 80% 2
Ara'S 0% 5
Ara’6 0% 6
Ara+1 0% s
Ara+2 10% s
Ara+3 10% 9
Ara+4 40% 7
miss 0% 7
Ara+6 20% 7
Control
All 12 100% 1

 

Note: After 1,000 generations, ten clones were randomly chosen fiom each
population and then scored for the five electrophoretic and nine phenotypic markers listed
in Table l.

22

Changes in fitness. - At the end of the 1,000-generation experiment, the fitness of
each population in the recombination treatment and control groups was measured relative
to a common competitor. This common competitor (REL2543) was the ancestral
genotype for recombination treatment and control populations Ara'l in this study, having
been isolated at generation 7,000 in the experiment described by Lenski and Travisano
(1994). A selectively neutral Ara+ mutant (REL4190) of this clone was also obtained to
allow competition experiments with the Ara‘ populations. The genetically heterogeneous
populations in the control and recombination groups were competed against the common
competitor bearing the opposite Ara marker state, after each competitor was
preconditioned in DM for one day. I also estimated the fitness of all twelve ancestral
genotypes relative to the common competitor. Fitness assays were replicated four-fold in
blocks of 36 (corresponding to the twelve ancestral genotypes, and the twelve 1,000-
generation populations in each treatment group). The mean fitnesses relative to the
common competitor are shown in Table 4.

I first tested whether the variation provided by mutation alone had allowed the
populations in the control group to gain in fitness relative to the ancestor. A one-tailed
paired comparison between the means for each control population and the ancestor
indicates that the ~4% improvement is statistically significant (ts = 2.349, df = l 1, p =
0.03 9).

I then tested whether the additional variation provided by sexual recombination
allowed the twelve populations in the treatment group to gain fitness to a greater extent
than their counterparts in the control group. A one-tailed paired comparison between the
means for each treatment population and its control indicates no significant acceleration
due to the recombination treatment (ts = -0.801, df = 11, p = 0.780). Therefore, I must
conclude that recombination did not allow the treatment populations to gain fitness more
rapidly than the control populations, despite the fact that the recombination treatment

clearly accelerated the rate of genetic change (Tables 2 and 3).

23

Table 4. Mean fitness for each experimental population, and for the ancestral genotypes.

 

 

 

 

Treatment
Ancestor Control Recombination
Population
Ara'l 0.984 0.986 1.092
Ara’2 1.017 1.043 1.087
Ara'3 0.967 1.058 0.816
Ara‘4 1.000 1.015 0.984
Ara'S 0.881 0.968 0.985
Ara'6 0.954 1.005 1.032
Ara+1 0.965 1.120 1.028
Ara+2 1.016 1.064 1.050
Ara+3 1.104 1.037 1.064
Ara+4 1.049 1.082 1.031
Ara+5 1.042 1.021 1.024
Ara+6 1.009 1.051 1.039
Grand mean (:1; SE)
0.999 (0.016) 1.038 (0.012) 1.019 (0.021)

 

Note: Each population's fitness was assayed, with four-fold replication, relative to
the common competitor bearing the alternative Ara marker state. Standard errors for the

grand means are based on n = 12 replicate populations.

24

DISCUSSION

I sought to examine the efi‘ect of sexual recombination on the rate of evolution in a
bacterial model system The populations of E. coli that I studied had previously evolved
under a constant environmental regime for 7,000 generations. Twelve populations had
been founded by clones of an asexual strain, so that the populations depended entirely on
mutation as a source of genetic variation for adaptation by natural selection Over time,
the rate of adaptive evolution in these populations slowed considerably from an initially
rapid pace, based on changes in mean fitness (Figure 2).

For this study, I established two new sets of twelve populations each; one set
served as controls while the other set underwent sexual recombination. The control
populations were propagated for another 1,000 generations in the same environment as
their ancestors, and they continued to depend on mutation as their sole source of genetic
variability. The treatment populations were also propagated for 1,000 generations in the
same environment, but they were also periodically subjected to matings with a pool of Hfr
(high fi'equency recombination) donors. These donors were genetically distinct from the
recipient populations, and the donors themselves could not grow in the experimental
environment. However, the donors were able to transfer genes to the resident recipient
populations by conjugation. The resulting recombination provided the recipient
populations with an additional source of variation that might allow them to adapt more
quickly.

My study can be summarized by two major results, which appear to be
contradictory. On the one hand, the recombination treatment dramatically increased the
genetic diversity present in the experimental populations and, indeed, accelerated the rate
of their genetic change (Figure 4, Table 2). After 1,000 generations, ten individuals from
each population were scored at 14 loci. On average, the twelve recombination treatment

populations contained about six distinct genotypes, with the ancestral genotype

25

representing only 15% of the total (Table 3). But in the twelve control populations, every
individual still had the ancestral allele at each locus scored.

On the other hand, the recombination treatment had no measurable efi‘ect on the
rate of adaptive evolution, as measured by changes in mean fitness during 1,000
generations. The control populations improved, on average, by a few percent relative to
their ancestors; the treatment populations improved by about the same amount or perhaps
slightly less (Table 4). Thus, the dramatic increase in genetic diversity and rates of genetic
change brought on by recombination did not produce any fitness advantage.

How can these two results be reconciled? I cannot offer a definitive answer, but I
can suggest two possible explanations. One possibility is that my recombination treatment
was, in some sense, far too efi‘ective. That is, rather than merely providing an additional
source of potentially USCfill variation, the level of gene flow might have been so high that
the recipient populations were faced with an onslaught of deleterious alleles fiom the
donor strains. The other possibility is that complex selection dynamics may render invalid
the estimates of fitness obtained relative to a common competitor. For example, if
selection is fi'equency—dependent - or, in the extreme, if competitive interactions are
nontransitive - then there is not necessarily the expectation that more rapid adaptive
evolution would lead to higher ”final" fitness values relative to an arbitrary competitor.

In the sections that follow, I consider in more detail the plausibility of these

alternative explanations.

Potential for gene flow to have swamped adaptive evolution in the treatment
populations. - In general, recombination increases genetic variation in evolving
populations. Natural selection may then use this increased variability to increase mean
fitness (Wright 1932; Fisher, 1958). However, if recombination occurs at too high a rate,
and involves a gene pool that is not well adapted to the local environment, then

recombination pressure may actually decrease the mean fitness of a population because the

26

locally adapted resident genotypes are ”swamped out” by gene flow. For example, plant
populations that are locally adapted to living on soils contaminated by heavy metals fi'om
mining activities are subject to a high genetic load in the form of pollen flow fiom nearby
populations of metal-intolerant plants (McNeilly 1968; Bradshaw 1971; Ford 1975).
Interestingly, some plant populations living in such environments have evolved a high
degree of selfing, apparently to avoid the problem of gene flow (Antonovics 1968;
Macnair and Cumbes 1989). Thus, recombination pressure might, in principle, explain
how recombinant genotypes spread through treatment populations in this study, without
any concomitant increase in mean fitness.

The fact that, every fifth day, the recombination treatment populations were
subject to four donors for every recipient might suggest extreme recombination pressure.
However, bacterial conjugation is very difi‘erent from regularized genetic exchange due to
meiosis and fertilization, and it occurs at much lower rates than obligate outcrossing. To
explore the quantitative extent of gene flow in my recombination treatment, I performed a
series of mating experiments, with the four donors considered both individually and
pooled. These one-day mating experiments were performed, with three-fold replication,
under conditions identical to those used every fifth day in the evolution experiment proper.
I monitored the accumulation during a standard mating cycle of transconjugants that
carried both the Tetr marker from the donor strains and the Strr marker from the ancestral
recipient REL2545. In each of the four donor strains, the Tet" marker is located about 10
minutes fi'om the origin of transfer (approximately one-tenth of the distance along the
circular chromosome).

My data clearly show that two of the donor strains, REL288 and REL296, were
responsible for most of the gene transfer (Table 5). More importantly, it is also clear that
only a small proportion - in no case greater than 0.1% - of the recipient population
received the Tet’ marker. However, the Tet" marker is only one locus, and I am interested

in estimating the fraction of recipients that might have received any donor genes

27

96% .565 mos—g :88 05 55 6.8.085 38:92 83 >83 meta: zoom ”302

 

 

 

 

I: x 3m 42 4 8a so. x «3 52 x t: 9: x a: 9: x a: 359:8

50% =<
“.2 x a4 «2 x Se be u 3.x he a x: me x 86 9: x R.» ”84mm
1: u 8.". 1: x a: a: u as 52 x 92 a: x m3 9: x a... 83mm
we 9 an a: x 8a be u an 9: x can me x 2.” 2: x «3 33mm
1: x a: 42 x 3m 52 x 2x as x on." a: x «2. 02 9 Se 33mm

Q .26 \ a. 9 a a m a
assaeosm
8288a new as". was

 

is 23 8285 8333

 

.226 95.2: 235% a mate—e «£86 3:3 05 89a coo—HE each 05 €90 «30338 no newton—9a 05 mo mum—Si .m 033.

28

whatsoever (under the most extreme scenario of recombination pressure where almost all
donor genes would be harmful to the recipient). A recent study has shown that, following
an Hfr mating, the average fragment integrated into the recipient's chromosome is on the
order of ten minutes, or about 10% of the chromosome (Lloyd and Buckman 1995).
Therefore, one can estimate roughly that about 90% of all gene transfer events would not
have included the Tet’ marker. In that case, the proportion of recipients receiving any
genes fiom the Hfr donors would be about lO-fold higher than estimated from the Tetr
marker alone, but still less than 1%.

In fact, this number over-estimates the level of gene flow for three reasons. First,
the mating cycle was imposed only every fifth day, so the efl‘ective rate of recombination
per generation was much lower. Second, each Tetr recombinant cell is not the product of
an independent transfer event. For example, the transfer of a Tetr marker that occurred
two generations prior to the recipient population entering stationary phase would have left
about four granddaughter cells. Third, the matings were not allowed to proceed for the
full 100 minutes required to transfer the entire chromosome; instead, they were interrupted
by moving the mating culture to a shaking incubator after one hour.

In any case, it does not seem likely that more than a small fiaction of the cells in a
recipient population actually received genes from the Hfr donors. Hence, I reject the
hypothesis that an excessively high level of gene flow from maladapted donors may have
overwhelmed the populations subjected to the recombination treatment. It seems more
likely, therefore, that recombinants increased in frequency during the evolution experiment

proper owing to their selective advantage.

Potential for complex selection dynamics to have obscured more rapid adaptive
evolution in the treatment populations. - In this study, I estimated the fitnesses of the
derived populations relative to a common competitor (the ancestral genotype for one pair

of populations). These data gave no indication that the recombination treatment

29

populations had adapted to any greater extent than the asexual control populations,
despite much greater genetic diversity in the treatment populations. However, this method
of fitness estimation, by using a common competitor as a yardstick, implicitly assumes that
the dynamics of natural selection are frequency-independent and transitive. That is, it
assumes that the fitnesses of all genotypes can be ranked relative to one another based on
their ranking relative to a single competitor.

In previous experiments using the asexual progenitors of the populations in this
study, this assumption was tested in two different ways; in both respects, it seemed to
provide an accurate description of the selection dynamics. Fitnesses measured relative to
the common ancestor increased monotonically in these experimental populations (Lenski
et al. 1991; Lenski and Travisano 1994). Moreover, the magnitude of a derived
genotype's advantage relative to another could be accurately predicted from each one's
advantage relative to the ancestor (Lenski et al. 1991; Travisano et al. 1995).

However, more complex selection dynamics have been shown in some other
systems. An experiment with evolving populations of the yeast Saccharomyces
cerevisiae, provides an especially dramatic example of complex selection (Paquin and
Adams 1983). In that study, each successive genotype that came to dominate a
population was more fit than its immediate predecessor. However, the populations
sometimes declined in fitness relative to the original ancestral genotype owing to
nontransitive competitive interactions (in which B is more fit than A, C is more fit than B,
but C is less fit than A). Other studies with bacteria have shown various kinds of
frequency-dependent selection, including cases in which each of two genotypes has a
selective advantage only when it is suficiently rare (e.g., Rosenzweig et al. 1994) or
common (e.g., Chao and Levin 1981). Thus, it is plausible that the unidirnensional
estimates of fitness in my study may have obscured important frequency-dependent efi‘ects
on fitness, which (if fully understood) might show that adaptive evolution in the

recombination treatment populations was more rapid than in the control populations.

30

Testing for nontransitive interactions and other frequency-dependent efi'ects is a
daunting task, given the large number of replicate populations and distinct genotypes in
this study. I decided to focus on recombination treatment population Ara‘3. Afier 1,000
generations, the fitness of this population appeared to be below that of its ancestor (Table
4), suggesting nontransitivity may have been important.

Using genotypes from the samples stored at loo-generation intervals, I performed
additional competition experiments to look more closely for nontransitivity. These
experiments yielded results consistent with complex selection dynamics, but they did not
provide an absolutely clear case of nontransitivity. As shown in Figure 5, the dominant
genotype at generation 500 had a large disadvantage relative to the “usual" common
competitor (an Ara+ mutant of the ancestor for population Ara'l). The same genotype
had no discernible disadvantage (or advantage) relative to an Ara+ mutant of its own
direct ancestor. And the two ancestors themselves had very similar fitnesses, regardless of
which one was marked by the Ara+ mutation. Thus, fitness gains of certain genotypes
from this population appear to have been underestimated relative to the common
competitor.

I was unable, however, to identify a set of genotypes fiom within this population
for which the interactions were clearly nontransitive. Even so, these experiments strongly
suggest that fitnesses measured relative to a common competitor did not capture the full
complexity of the selection dynamics in this experimental system. In the next chapter, I
examine in some detail the ecological mechanisms that allow the stable fiequency-
dependent coexistence of two other recombinant genotypes that were also sampled from
population Ara'3.

Conclusions. - The results of my study demonstrate a very substantial impact of
Hfi plasmid-mediated recombination on the evolution of the E. coli chromosome.

However, the consequences of this recombination for the rate of adaptive evolution were

31

r"
O

 

Relatlve fltness

Q
(I)
I

 

 

 

 

.—

0.6 l l

 

 

Figure 5. Complex selection dynamics revealed by pairwise interactions among
three genotypes. REL43 49 was the dominant genotype in recombination treatment
population Ara'3 after 500 generations. REL2545 is its ancestral genotype; REL4322 is
identical except for the Ara+ marker. REL2543 is the ancestral genotype for another
population; REL4190 is identical except for the Ara+ marker. REL2543 and REL4190
served as common competitors for the fitness experiments summarized in Table 4. (A)
REL4349 is less fit than REL4190. (B) REL4349 is of equal fitness to REL4322. (C)
REL2543 and REL4322 are equally fit, as are (D) REL2545 and REL4190. Error bars

indicate 95% confidence intervals, based on five-fold replication of each fitness assay.

32

unclear. When fitnesses of derived genotypes were measured relative to a common
competitor, there was no evidence that the increased genetic variation led to more rapid
gains in fitness. However, supplementary experiments suggest that some genetic changes
that occurred in the recombinant populations may have been adaptive only in the context

of the particular milieu of genotypes in which they arose.

CHAPTER II
TESTS OF ECOLOGICAL MECHANISMS PROMOTING THE STABLE
COEXISTENCE 0F RECOMBINANT BACTERIAL GENOTYPES

INTRODUCTION

Frequency-dependent selection has long been hypothesized to maintain genetic
polymorphisms in natural populations (Fisher 195 8; Haldane and Jayakar 1963; Clarke
1964; Ayala and Campbell 1974; Levin 1988), and empirical evidence has been
accumulated to support this notion (e.g., Cain and Sheppard 1954; Hori 1993). However,
the complexity of the natural environment makes elucidation of the driving forces behind
fiequency-dependent selection extremely dimcult. Also, the long generation times of most
organisms often lead researchers to assume (rather than prove) that a stable polymorphism
exists.

Conclusive evidence that the fitness of a genotype can be related to its relative
fiequency has come from laboratory studies involving species of Drosophila (e. g., Wright
and Dobzhansky 1946; Kojima 1971; Van Delden et al. 1978). Due to controlled factors
in the selective environment and shorter generation times, laboratory studies would seem
to permit precise determination of the underlying causes of stable genetic polymorphisms.
However, this is not necessarily the case: fiequency-dependent selection involving the
ADH locus in laboratory populations of D. melanogaster has been clearly demonstrated,
but never fully explained (Van Delden 1982).

Because of their short generation times, very large population sizes, and general
ease of propagation, bacteria provide an excellent model to study resource-based
competition as well as other ecological and evolutionary processes (Levin 1972; Chao et
al. 1977; Levin et al. 1977; Hansen and Hubbell 1980; Helling et al. 1987; Dykhuizen

33

34

1990; Lenski and Travisano 1994). But reproduction in bacteria is strictly asexual and
laboratory environments typically provide only a single limiting resource. For these
reasons, researchers have often assumed that populations of bacteria will conform to the
classical model for the evolution of asexual organisms (Atwood et al. 1951; Moser 195 8;
Dykhuizen 1990). That is, experimental populations are subject to takeover by a single
genotype that harbors a mutation conferring some selective advantage. Atwood et al.
(1951) called this phenomenon ”periodic selection" because an advantageous mutant
periodically replaces its immediate predecessor. Thus, polymorphisms in bacterial
populations are only expected to exist transiently, while an advantageous mutant is
increasing in fi'equency relative to its ancestor. This notion is in accord with the
competitive exclusion principle (or Gause's axiom), which states that two competitors
cannot coexist on a single limiting resource (Gause 1934; Hardin 1960).

In apparent violation of this simple model leading to competitive exclusion, the
evolution and persistence of stable polymorphisms in laboratory populations of bacteria
has been clearly demonstrated. For instance, despite difi‘erences in maximum specific
growth rates and glucose transport, three clones of Escherichia coli were reported to
stably coexist in glucose-limited chemostat culture (Helling et al. 1987). It was later
shown that these three strains had evolved a complex method of cross-feeding that
involved difl‘erential patterns of secretion and uptake of two alternative metabolites,
acetate and glycerol, in addition to glucose (Rosenzweig et al. 1994). Other studies have
documented bacterial coexistence mediated by viruses (Chao et al. 1977), by
detoxification of toxins (Lenski and Hattingh 1986) and by habitat structure (Korona et al.
1994).

In theory, a "demographic tradeofi“ can also lead to stable coexistence between
two bacterial strains growing in a serial culture environment that contains only a single
limiting resource (Stewart and Levin 1973). Serial (batch) culture is analogous to a

seasonal environment, where resources are abundant at the beginning of the bacterial

35

growth cycle but become scarce as the bacterial population approaches its carrying
capacity. Levin (1972) observed that a ”demographic tradeofi“ led to coexistence
between strains of E. coli B and K12 in serial culture. One genotype was able to grow
better at high concentrations of glucose (apparently due to a shorter lag phase) while the
other genotype grew better at low concentrations (due to its greater capacity for growth in
transitional phase). This result provides an interesting example of contrasting life-history
strategies for bacteria that is analogous to the proposed r-K tradeofi‘ (MacArthur and
Wilson 1967; Pianka 1970). The "r—selected" strain was able to reproduce rapidly in an
uncrowded environment, while the "K-selected" strain made up for this early disadvantage
by maximizing fitness when the population was near carrying capacity. Although Levin
(1972) could not rule out the possibility that cross-feeding was also involved,
demographic tradeofi‘ alone is a viable mechanism for mediation of coexistence in serial
culture (Stewart and Levin 1973).

In a study to examine the efl‘ect of recombination on the dynamics of bacterial
evolution (Chapter I), I observed prolonged genotypic diversity when a population of E.
coli was propagated serially in an environment in which glucose was provided as the
limiting nutrient. Because the experimental environment was free of viruses and
antibiotics, I propose that either a cross-feeding interaction or a demographic tradeofi‘ in
growth rates may explain coexistence. These two hypotheses are not mutually exclusive,

so that coexistence might be explained by both mechanisms jointly.

Cross-fleeting. - Cross-feeding allows one strain to monopolize the primary
resource, while excreting some metabolite into the environment that disproportionately
enhances the growth of a second strain. Assuming the amount of metabolite produced is
proportional to the density of primary competitor, then a cross-feeder benefits from being
in the minority. Similarly, the primary competitor benefits when the population contains a
high density of cross-feeders due to its inherent growth advantage on the primary

36

resource. Thus, the relative fitnesses of both strains will be decreasing functions of their
own frequencies, so that each strain has a higher relative fitness when it is the minority
competitor. Hence, competition is fiercest among competitors of the same genotype and a

stable polymorphism may arise from the inability of either genotype to displace the other.

Demographic tradeofii - Coexistence mediated by a demographic tradeofl‘
requires that one strain is competitively superior when the sole limiting resource is at high
concentration, whereas the other strain is superior in competition when that resource is
scarce. Under the simple case of the Monod model (1949), each strain's growth rate is

given by:

dN/dt = N [Vmax S/ (S + K39],

where N is cell density, Vmax is maximum growth rate, S is resource concentration, and
K S is the concentration required to support growth at halfthe maximum rate. The rate of

resource depletion is given by:

dS/dt = - c (dN/dt),

where c is the conversion efficiency. A necessary (but insuflicient) requirement for
coexistence is that one strain has a higher Vmax, while the other strain has a higher ratio of
Vmax/Ks- Each strain will have an advantage at a difi‘erent stage of the grth cycle and
the two strains may stably coexist, each having an advantage when rare (Figure 6).

Here I present the results of a series of experiments to determine whether a cross-
feeding interaction or demographic tradeofi‘ best explains coexistence between two

recombinant strains of E. coli.

37

Figure 6. Numerical simulations showing stable coexistence of two strains on a
single resource in a seasonal environment, mediated by a demographic tradeoff. One
strain has Vmax = 0.4 h'1 and K s = 0.1 ug mL'l, whereas the other strain has Vmax =
0.59 h'1 and KS = 7.5 ug mL'l; c = 5x10‘7 ug for both. The vertical axis shows the
frequency of the first strain. At the beginning of the simulation, and after every 24 h
thereafter, the competing populations are diluted 1:100 into fresh medium that contains 25
ug mL"1 of the limiting resource. In each 24-h period, there are three phases during
which (1) the resource concentration is sufiiciently high that the strain with the higher
Vmax has an advantage, (2) the resource concentration has been reduced so much that the
strain with the lower K s has an advantage, and (3) the resource has been exhausted to the
point that neither strain can grow. The three curves differ only in the initial frequency of
the competing strains; each strain has an advantage when rare, and so their coexistence is
stable. Simulations were run with a time step of 0.001 h using SOLVER SWV (Blythe et
al. 1990). This program employs the fourth-order Runge-Kutta method, with a
modification that allows for a switch to be thrown (here, periodic dilution into fresh

medium).

38

1.0-

0.8 W

0.6 -

0.4 Wm
02 W

0.0 l l 1 l l l J g l l

0 24 48 72 96 120144168192 216240

HWEI<hfl

FREQUENCY

 

Figure 6.

39

MATERIALS AND NIBTHODS

Bacterial strains. - Chapter I describes the experiment from which the strains
used in this study were obtained. Briefly, twelve treatment populations were founded
fiom independently derived clones of E. coli B that had been serially propagated in the
selective environment for 7,000 generations (Lenski and Travisano 1994). These
populations (recipients) were serially propagated for another 1,000 generations (150
days), during which time they were allowed to undergo recombination with donor strains
approximately every fifih day. The donors were four Hfi (high frequency of
recombination) strains of E. coli K12, which are rather distantly related to E. coli B (Table
5; see also Selander and Levin 1980). These donors were all auxotrophic for at least one
amino acid. Thus, the Hfi donors were able to donate genes via recombination but were
unable to survive and out-compete the prototrophic E. coli B recipients.

For the purpose of this study, I isolated two recombinant genotypes (REL43 97
and REL43 98) at generation 1,000 from one of the treatment populations. Using nine
physiological and five electrophoretic markers, it is evident that both recombinant
genotypes contain a mixture of the markers present in E coli B and K12 (Table 5). The
two recombinants also differ fi'om one another at a number of loci, including in their ability
to utilize the sugar lactose. Thus, REL43 97 (Lac+) and REL43 98 (Lac') form white and
red colonies, respectively, when they are spread on tetrazolium-lactose (TL) indicator
plates (Levin et al. 1977). Both recombinants are prototrophic and so can grow without
supplemental amino acids. From here on, I refer to REL43 97 and REL43 98 simply as
Lac+ and Lac‘, respectively.

Media and culture conditions. - Unless otherwise noted, the culture medium
employed was Davis Minimal (DM) broth (Carlton and Brown 1981) supplemented with

2x10'6 g thiamine hydrochloride mL"1 and glucose at a specified concentration. For

40

 

 

 

 

a v» N _ _ _ N .— ._ + - - a .— .. + wag
._. a N _ _ _ N a a + - - ... a - + 33mm
4. N N _ _ _ N a a + + - a ._ .. - SNAmM
e S N _ _ _ N v. m .. + + a a + + 3%
£88
1. _ N N N _ a a + + + c a + - 934mm
“868:
I N N N N _ a .— + + + a u - .. «a 94mm
1. N N _ N N u m + + + c c + .. 3mg
snag
«.58 mm." :2 an? 8.3 :9 8. x; s: .3 m2 5 a... 63 8< 585
Nowoconqoboofi ~ofiboeonm

 

#5368 v5 fleece 3:88 :05 8 :03 an d 8320 5 33 N8 Mac «586 «55308.. 95 05 8m moo—38 3280 .w 033.

41

205000008 .0080 008 3880300 80 mama—.0000 .8 8080 8 80.85.; 008068

a .8 ._. 8:00 80802 38 00:0» 80:00 m0 80.858 .«0 83.5 05 8a 088080.30 SM .800 Q 05 00 .8888 8 803000 082 m

.890 88880 08 .986 080882 32968 9.888 .590 9.5000880

ES... .8043 8.806038 2886028686 .500 88806850 2868.2 88.68 8 88% .0588 8805 .8852 N
.8. 05. xi. 8068 a .8888.

6058008 0:0 332800 38:08 u 0:0 a 58 0:?080338 05 A086 08000— .Asv 0:888 0200 0880 50360002

53080080 ‘80 3008880 30088 - .80 + xme 80803088 05 30.5 8:03.083 00:03.80 0» 50600000.. .00an 08.
08088 3805 a as. . x80 888. 05 A88 865683 888. 8 360 2 568688. .0505 05 £8. 2805 - as. + _

.583 0 2.3.

42

example, DM25 indicates DM supplemented with 25 ug glucose mL’l, which yields ~
5x107 cells mL'1 at stationary phase. In all experiments, culture volume was 10 mL
maintained in SO-mL Erlenmeyer flasks; these flasks were placed in a shaking incubator at
37 °C and 120 rpm. As indicated, populations were propagated daily by transferring 0.1
mL of each stationary-phase (24 h) culture into 9.9 mL of flesh medium. The resulting
lOO-fold daily growth of each population represents ~6.64 generations of binary fission.

F imess assay and definitions. - Relative fitness was assayed by allowing Lac+ and
Lac‘ to compete under the culture conditions described above, except where other culture
conditions are specifically noted. Prior to competition assays, each strain was grown
separately for (at least) one day in the experimental medium as a preconditioning step to
ensure both competitors were in comparable physiological states. The two competitors
were then mixed at a 1:1 or other defined volumetric ratio, then diluted 1:100 into fiesh
medium and allowed to grow and compete during a standard one-day growth cycle. Initial
and final densities of each competitor were estimated by spreading cells on TL plates,
which distinguish the strains by colony color.

Let the initial densities of the Lac+ and Lac' competitors be N1 (0) and N2(0),
respectively; and let N 1(1) and N2(1) be their corresponding densities after one day. The
average rate of increase (or realized Malthusian parameter), m, for either competitor is

then calculated as:

mi = lnlM(1)/M(0)]/(l day).

The fitness of one strain relative to another (Wij) is estimated as the ratio of their
Malthusian parameters (Lensld et al. 1991):

Wij = mi/mj.

43

A fitness difl‘erence between the two competitors may reflect difl‘erences in their lag phase,
maximum or submaximum growth rate, survival at stationary phase, or some combination

thereof (e.g., Vasi et al. 1994).

RESULTS

Demonstration of the stable equilibrium

Evidence for frequency-dependence. - I first sought to establish whether
coexistence between recombinant strains Lac+ and Lac' was stable, such that each strain
could increase in frequency when it was initially rare. To that end, I performed fitness
assays in DM25 in which the two strains were mixed at thirteen difl‘erent initial frequencies
of Lac+z 0.99, 0.95, 0.9, 0.8, 0.75, 0.6, 0.5, 0.4, 0.25, 0.2, 0.1, 0.05, and 0.011. Assays
were replicated twice for each initial frequency.

Two aspects of the results (shown in Figure 7) are of particular interest. First,
there is compelling evidence that the relative fitness of each strain is a decreasing function
of its own initial frequency (slope = -0.338, ts = -7.855, df= 24, p < 0.001). Second,
these data allow me to predict the equilibrium fiequency of Lac+ and Lac‘ in DM25. By
definition, equilibrium will occur when the strains achieve fi'equencies in the population
where they are equally fit. Thus, the point where the regression line intersects a relative

fitness of 1.0 corresponds to a predicted equilibrium frequency for Lac+ of about 0.45.

Evidence for stable equilibrium. - To further prove the existence of a stable
polymorphic equilibrium, Lac+ and Lac’ were allowed to compete during daily serial
transfers in DM25 for 200 generations (30 days). Each strain was separately

preconditioned and then they were mixed at five initial fiequencies of Lac'l” (0.9, 0.75, 0.5,

 

Fitness of Lac+ relatlve to Leo-

 

 

0.6 1 1 1 1 I
0.0 0.2 0.4 0.6 0.8 1 .0

lnltlal frequency of Lac+

Figure 7. The fitness of recombinant E. coli strain Lac+, relative to strain Lac’, is
a decreasing firnction of its own frequency. Relative fitness was calculated as the ratio of
Malthusian parameters estimated during competition over one-day growth cycles in
DM25, starting with initial frequencies of Lac+ that ranged from 0.01 to 0.99. The line
indicates the least-squares regression These data imply a stable equilibrium frequency for
Lac+ of about 0.45, at which point relative fitness equals one.

45

0.25, 0.1), with three-fold replication. After 16 days, I temporarily suspended this
experiment by adding 3 mL glycerol to each culture flask, and storing the cultures at -80
0C. I restarted this experiment two weeks later by inoculating from each thawed fi'eezer
sample into Luria broth for one day. Cultures were then diluted into DM25 for one day of
re-acclimation before continuing the experiment proper for another 14 days in DMZS.
(Two replicates were terminated at days 19 and 27, due to contamination.) During the
entire 30-day competition experiment, I spread samples taken fi'om stationary-phase
cultures (i.e., at the end of the 24 hour growth-cycle) onto TL plates to determine the
density of Lac+ and Lac‘ strains.

As shown in Figure 8, the fi'equency of Lac+ converged upon an equilibrium value,
regardless of its initial frequency and despite the fact that the experiment was perturbed
midway by freezing and restarting the competition. Interestingly, the perturbation resulted
in a systematic benefit to the Lac+ competitor, but this effect was transient and
disappeared as the experiment continued. By day 30 of the competition experiment, the
mean fiequency of Lac+ was 0.509 (t 0.012 SE), which agrees well with the equilibrium
predicted fi'om the regression of relative fitness on initial fiequency (Figure 7). I conclude
that fi'equency-dependent selection is strong enough to drive the strains toward a stable

equilibrium ratio, at which point the two strains are of equal fitness.

Evaluation of the demographic tradeofl'lopothesis
Estimation of maximum growth rates. - For each strain, I measured the maximal
growth rate, Vmax: in DM1000, wherein glucose is well above the concentration at which
it limits growth rate. Vmax was estimated by regressing In N versus time during the period
of exponential-phase growth. Cell densities were estimated by counts obtained using a
Coulter electronic particle counter (model ZM and channeler model 256), and

experiments were performed with five-fold replication for each strain.

Frequency of Lac+

 

 

 

Tlme (days)

Figure 8. Starting from difi‘erent initial fi'equencies, strains Lac+ and Lac" establish
a stable polymorphism in DM25. After 30 days, observed frequencies agree rather well
with the equilibrium predicted from one—day competition experiments (Figure 7). The
asterisk (*) indicates a perturbation when the experiment was put in a freezer for

temporary storage.

47

Lac+ and Lac‘ gave mean estimates for Vmax of 0.9225 h'1 (1 0.0137 SD) and
0.9770 h"1 (1 0.0081 SD), respectively. This difl‘erence in Vmax was statistically
significant (t, = 7.651, 8 df, p < 0.001).

Estimation of afi'inity for limiting resource. - K s is the concentration of resource
at which a strain can grow at halfof its maximum rate (i.e., Vmax/Z), according to the
Monod (1949) model. As such, l/Ks provides a measure of a strain's afiinity for a limiting
resource. To determine whether Lac+ and Lac' difl‘er in their affinity for glucose, I sought
to estimate K, for each strain.

I followed the estimation procedure of Vasi et al. (1994), which uses the range of
concentrations over which a population can replace itself; given a set dilution rate, to
estimate K s Three replicates of each strain were removed fi'om the freezer and
propagated in DM25 to acclimate them to growth on glucose. I then transferred inocula
into DM containing 0.01, 0.025, 0.1, 0.25, or 1 ug mL'l glucose, with the inocula
adjusted proportionately to account for expected difi‘erences in final density. As a control
for growth on possible contaminating resources, I also inoculated cells into DM without
any glucose. Cultures were propagated by 1:100 serial dilution for seven days.
Immediately prior to each dilution, samples were spread onto TL plates to estimate cell
densities. I then regressed In N against time to determine the lowest glucose concentration
at which each strain could maintain a constant population size.

Lac+ was found to be able to replace itselfin the face of serial dilution at glucose
concentrations of 0.1 ug mL'1 and higher. Let V= (l/N)(dN/dt) be some submaximal rate
of increase. The loo-fold daily increase (to ofl‘set dilution) implies that V> ln(100)/24 h =
0.19 h'l. From the Monod (1949) model, and using Vmax for Lac+ as estimated above, it
follows that V> 0.19 h-1 can be sustained in DMO.1 only M, < 0.39 ug tnL-l. Similarly,
the inability ofLac+ to persist in DM0.025 implies that K, > 0.10 ug mL-l. In contrast,
Lac“ was found to be able to sustain itself in the face of 1:100 daily serial dilution only at

48

glucose concentrations of 0.25 ug mL'1 and higher. Using Vmax for Lac" as estimated
above, it follows similarly that K, for that strain lies between 1.04 and 0.41 ug mL'l.

My results are consistent with a demographic tradeofl‘ between Vmax and K s-
Lac' has a higher maximum growth rate than Lac+, whereas Lac+ has the greater amnity
for glucose and can evidently grow faster at very low glucose concentrations. However,
the existence of such a tradeofl‘ is not suficient to explain stable coexistence, which relies
on the input concenration of limiting resource and the dilution factor (Stewart and Levin
1973).

Thus, it was necessary to explore the parameter space (within the limits of
experimental uncertainty for each strain's Vmax and K s) to evaluate if the input resource
concentration (25 ug mL'l) and dilution factor (1: 100 d'l) that were imposed would
allow stable coexistence. To that end, numerical simulations using SOLVER (Blythe et al.
1990) were run. The criterion for coexistence was that each strain must be able to
increase in frequency when it was initially rare.

A summary of the results from many numerical simulations follows. First, the
mean estimates for each strain's Vm were used to consider the effects of difl‘erent K s
values within the bounds of uncertainty. Lac‘ was found to exclude Lac+ competitively
unless the difference in the K s values for the strains was rather close to the maximum
allowed by the experimental uncertainty. But if the difference in their K s values was too
large, then Lac+ would competitively exclude Lac‘. However, there was a small region of
stable coexistence in between these two extremes. For instance, coexistence resulted if
Lac+ had Vmax = 0.9225 h-1 and KS = 0.12 ug mL-l while Lac' had Vm = 0.9770 h-1
and K, = 0.9 ug m1:1 (Figure 9). Next, the efl‘ect of uncertainty in the estimate of the
difl‘erence in Vmax between the two strains was considered. Ifthe size of this difi‘erence
was increased, then the advantage shifted towards Lac‘; if it was decreased, then the
advantage shifled towards Lac+. In all cases, however, there was at most a small region

of coexistence in terms of K s values. More importantly, in those cases where stable

49

1.010

1

7

1.005

Relative to Lac

/

 

 

1.000
U ‘\
U
_.i
3 0.995 -
U)
U)
(D
g 0.990 1 I l l l
L'— 0.0 0.2 0.4 0.6 0.8 1.0

Initial Frequency of Lac

Figure 9. Numerical simulations of the fitness of Lac+ relative to Lac‘, as a
function of the initial fi'equency of Lac+, assuming only a demographic tradeofi‘ between
growth rates at high and low glucose concentrations. For Lac+, Vmax = 0.9225 h‘1 and
Ks = 0.12 ug mL-l. For Lac“, Vmax = 0.977 h-1 and K, = 0.9 ug mL-l. For both strains,
c = 5x10'7 ug. Simulations assume combined population is diluted 1:100 into medium
containing 25 ug mL'1 of the limiting resource. Simulations were run using SOLVER
(Blythe et al. 1990) with a time-step of 0.001 h. Relative fitness was calculated as the
ratio of Malthusian parameters, exactly as in the experiments. A demographic tradeofl‘
that is consistent with the estimates for each strain's Vmax and Ks (see text) allows stable
coexistence. However, the predicted frequency-dependence is much weaker than was

observed in experiments (see Figure 7).

50

coexistence was observed, the strength of the resulting fi'equency-dependence was very
weak, with fitness advantages for the rare strain on the order of only 1% (Figure 9). This
result contrasts with the much stronger fiequency-dependence observed in the actual
experiments, where each strain had a fitness advantage in excess of 10% when it was rare
(Figure 7).

These results strongly suggest that a demographic tradeofl‘ between growth rates
at high and low glucose concentrations contributes only slightly to the observed stable
coexistence between Lac+ and Lac’. Thus, I also sought to evaluate the possible
importance of cross-feeding interactions involving the difi‘erential secretion and utilization

of metabolic by-products.

Evaluation of the cross-feeding hwothesis

Eflect of resource concentration onfiequency-dependence. -- If cross-feeding is
an important factor promoting fi'equency-dependence, then one might expect the stable
coexistence of Lac+ and Lac' to break down at low glucose concentrations. In other
words, at low glucose concentrations, the population density of the metabolite reducing
strain would be reduced, with a concomitant reduction on the concentration of metabolite
and therefore a diminished opportunity for cross-feeding (see also Rosenzweig et al.
1994). To examine this possibility, I looked for an influence of resource concentration on
the frequency-dependence of relative fitness. I performed fitness assays in which the two
strains were mixed at five initial frequencies ofLac+ (0.9, 0.75, 0.5, 0.25, 0.1), and
allowed to compete in media supplemented with five different concentrations of glucose
(DMl, DM25, DM25, DM250, DM1000). In all cases, both strains were removed fi'om
the fieezer into DM1000, acclimated for two days in the competition medium, and then
competed for one day. Each treatment combination was replicated five-fold, but two

replicates were excluded due to contamination.

51

An AN OVA indicates that the interaction between glucose concentration and
initial fiequency has a highly significant efl‘ect on relative fitness (Table 7). As shown in
Figure 10A, the fitness of each strain is a decreaseing filnction of its own frequency when

strains are competed at glucose concentrations of 25, 250, and 1000 ug mL'l. At all three

Table 7. AN OVA of effects of initial fi'equency and glucose concentration on fitness of

 

 

Lac+ relative to Lac'.

Source SS df MS F P
Initial fi'equency 0.230 4 0.058 11.22 <0.001
Glucose concentration 0.211 4 0.053 10.28 <0.001
Interaction 0.297 16 0.019 3.62 <0.001
Error 0.503 98“ 0.005

 

* Treatment combinations were replicated five-fold, but there were two missing values.

of these concentrations, the one-tailed regression of relative fitness on initial frequency is
highly significant (DM25: slope = -0.300, ts = -10.349, df= 23, p < 0.001; DM250: slope
= -0.254, t, = -5.097, df= 23, p < 0.001; DM1000: slope = -0.171, ts = -3.992, df= 23, p
< 0.001). However, at the lower glucose concentrations of 1 and 2.5 ug mL'l, the
dependence of relative fitness on initial fiequency breaks down, with Lac+ having a small
advantage over Lac' regardless of frequency (Figure 10B). In contrast to the conditions

necessary for stable coexistence, there is no significant negative regression of fitness on

52

Fltness of Lac+ relative to Leo-

 

 

 

 

 

 

 

 

 

 

 

 

l

O

3 B

2 1.2 '—

Q)

-2

g 11 — l U ,E

\— #‘flfi’T

23 0 I ##1##.2? I fl

(U 1.0 " ____.—-——*"’p’— .

:3 ———-----r .

O

8 0.9 —

(D

.43

LL 0.8 l l l l l
0.0 0.2 0.4 0.6 0.8 1 .0

Initial frequency of Lac+

Figure 10. Fitness of strain Lac+, relative to strain Lac‘, as a function of its initial
fi'equency, in DM media containing five difi‘erent glucose concentrations. (A) In medium
containing glucose at concentrations of 25 (circles), 250 (squares), or 1000 (diamonds) ug
mL'l, each strain has an advantage when rare such that there exists a stable
polymorphism. (B) In medium containing glucose at concentrations of 1 (squares) or 2.5
(diamonds) ug mL'l, there is no evidence that the rarer strain has an advantage. Each
point is the mean of five replicates. Lines indicate least-squares regressions; see the text

for statistical analyses.

53

frequency (DMI: slope = -0.048, ts = -0.776, df= 21, p = 0.448; DM25: slope = 0.124,
is = 2.219, df= 23, p = 0.982). I also performed all pairwise comparisons using the slopes
of relative fitness on initial fi'equency obtained at the different glucose concentrations
(5x4/2 = 10 comparisons in all). I employed the sequential Bonferroni criterion (Rice
1989) to compute significance levels. None of the three higher concentrations (DMZS,
DM250, DM1000) yielded significantly difi‘erent slopes fi'om one another, nor were the
slopes fi'om the two lower concentrations (DMl, DM25) significantly difi‘erent. Yet the
slopes for all other pairs were significantly different at p < 0.05, with the marginally
nonsignificant exception of DM1 and DM1000 (p = 0.084). Evidently, fi'equency-
dependence favoring the rarer strain is manifest at glucose concentrations of 25 ug mL'1
and higher, but it breaks down at much lower concentrations.

The observation that Lac+ prevails in competition at low glucose concentrations is
consistent with the demographic tradeofl‘ documented above. Averaging over all initial
fiequencies, Lac+ has a fitness of 1.060 (-_l-_ 0.038 95% CL.) relative to Lac' when the two
strains compete in medium with glucose at 1 ug mL‘l. This relatively small fitness
differential, even at so low a concentration, implies a difference in K s values between the
two strains that is much smaller than was used in numerical simulations (Figure 9) to
obtain stable coexistence based on a demographic tradeofi‘ between Vmax and KS. This
discrepancy adds further support to my earlier claim that a demographic tradeofl‘ alone is
unable to explain the observed stable coexistence. Moreover, the finding that the
conditions for stable coexistence break down at low glucose concentrations supports the

hypothesis that cross-feeding is responsible for the stable coexistence of Lac' and Lac+.

Evidence for cross-feeding after glucose has been depleted -- In the absence of
frequency—dependent forces, it was observed that Lac' has an inherent growth advantage
over Lac+ owing to its higher Vmax- I therefore sought to determine in what phase of the

population growth cycle Lac+ makes up for this deficiency. To that end, I performed

54

fitness assays in which the two strains were mixed at three initial fi'equencies of Lac+ (0.9,
0.5, 0.1), with eleven-fold replication, in DM25. Treatments comprising all Lac+ and all
Lac“ cells with three-fold replication were also included. Samples were spread on TL
plates at 0, 12, and 24 h to determine the densities of Lao"’ and Lac'.

Figure 11 shows the fitness of Lac+ relative to Lac’ in DM25 calculated first using
the 0 and 12 h data and then using the 0 and 24 h data. Once again, the relative fitnesses

of these two strains are shown to be frequency-dependent. But Figure 11 also shows that

Table 8. ANOVA of efl‘ects of initial frequency and final sample time (12 or 24 h) on

 

 

fitness of Lac+ relative to Lac'.

Source SS df MS F P
Initial frequency 0.154 2 0.077 13.644 <0.001
Final sample time 0.091 1 0.091 16.113 <0.001
Interaction 0.003 2 0.002 0.286 0.752
Error 0.338 60 0.006

 

Note: The experiment was carried out in medium containing glucose at 25 ug mL'l.

Lac+ has a systematic advantage relative to Lac" between 12 and 24 h An AN OVA
reveals that the efl‘ects of initial frequency and sampling interval are highly significant
(Table 8).

55

According to numerical simulations using the values of Vmax and K s estimated for
Lac' and Lac+, as well as the conversion emciency c (leO‘7 ug), glucose should have
been thoroughly depleted fiom the culture medium within the first 10 h, even allowing for
a lag phase prior to growth of l or 2 h (see Vasi et al. 1994). Thus, the finding that Lac+
gains a significant advantage between 12 and 24 h suggests either that Lac+ is growing on
some metabolite (and at a faster rate than Lac') or that Lac' is dieing (and at a faster rate
than Lac+), or perhaps both.

To evaluate these two alternative explanations, I computed the absolute rate of
change in population density between 12 and 24 h for each strain. A positive or negative
value indicates net growth or death, respectively. Figure 12 shows that, in the absence of
the other strain (i.e., the initial fiequency of Lac+ equals either 0 or 1), each strain is
subject to some cell death, although Lac+ and Lac' do not difi‘er significantly in their death
rates (ts = 1.161, df= 4, p = 0.310). However, in the presence of Lac“, Lac+ experiences
net growth between 12 and 24 h (Figure 12). Lac’ also benefits fi'om the presence of
Lac+, in that Lac’ shows no net decrease due to death as it does when it is alone.
Therefore, between 12 and 24 h, each strain benefits in absolute terms from the presence
of the other strain (Figure 12), although Lac+ has the advantage in relative terms (Figure
11).

Based on these results, 1 conclude that cross-feeding interactions occur after
glucose has been depleted fiom the culture medium. These interactions favor Lac+, but
the fact that each strain benefits from the other's presence between 12 and 24 h suggests
that two (or more) metabolic by-products may be involved in the dynamics of this stable
polymorphism. Although the primary aims of this study were to distinguish between the
demographic-tradeofi‘ and cross-feeding hypotheses, in the next section I report some
preliminary experiments in which two potential metabolites were experimentally
manipulated.

Fltness of Lac+ relatlve to Leo-

 

 

 

 

 

lnltlal frequency of Lac+

Figure 11. Fitness of strain Lac+, relative to strain Lac', as a function of its initial
frequency, in DM25, calculated between 0 and 12 h (open bars) and between 0 and 24 h
(filled bars). The fitness of Lac+ relative to Lac' increases between 12 and 24 hours, long
after glucose has been exhausted (see text), implying either differential mortality or growth
on metabolic by-products. Each bar represents the mean (i SE) of 11 replicates; the
AN OVA is given in Table 8.

57

    

 

 

 

 

 

 

 

 

 

0.08 —
{I 0.04 —
5
CD
CD
c , 7
.2 0.00 --------------- - ...............
0
”5 t
(D
g -0.04 —
-0.08

0.0 0.1 0.5 0.9 1 .0

lnltlal frequency of Lac+

Figure 12. Net rate of change in viable cell density for Lac+ (open bars) and Lac‘
(filled bars) in DM25 between 12 and 24 h, after glucose has been exhausted from the
medium. In the absence of the other strain, each strain declines due to death. However,
Lac+ experiences net growth between 12 and 24 h when Lac‘ is present. Lac‘ also benefits
from the presence of Lac+ between 12 and 24 h, although Lac+ has the advantage in
relative terms (see also Figure 11). Each bar represents the mean (t SE) rate of change in
viable cell density (h‘ 1) between 12 and 24 h with ll-fold replication for initial frequencies
of 0.1, 0.5, and 0.9 and with 3-fold replication for initial frequencies of 0 and 1.

58

Eject of potential metabolites on relative fitness. - During either aerobic or
anaerobic growth on glucose, E. coli generates a complex mixture of metabolic by-
products (Neidhardt et al. 1990). A strain that has an enhanced ability to utilize one of
these metabolites might be able to persist, even if it was at a disadvantage in acquiring
glucose. In fact, Rosenzweig et al. (1994) recently demonstrated the stable coexistence of
E. coli strains in chemostat culture (in which the demographic-tradeofi‘ hypothesis is not
applicable, because the environment is temporally constant). Their physiological analyses
implicated acetate and glycerol as the relevant metabolites promoting coexistence, and
these are the two metabolites that I will also consider.

In particular, I sought to determine if the addition of either acetate or glycerol
would alter the outcome of competition between Lac+ and Lac‘ and might even allow
these two strains to stably coexist in a medium where otherwise one strain excluded the
other. I showed previously that stable coexistence broke down in DM containing only 2.5
ug mL'l (Figure 10B). Performing experiments at this low glucose concentration has the
further advantage that the concentration of metabolites produced by the bacteria should be
proportionately less, so that the effect of an added metabolite might be more clearly
elucidated.

I performed an experiment in which Lac+ and Lac" were allowed to compete at
two initial frequencies of Lac+ (0.1 and 0.9) in DM25 supplemented with the following
concentrations of acetate: 0, l, 2.5, and 10 ug mL'l. These assays were replicated five-
fold. I also performed an identical experiment, except using glycerol instead of acetate.
In all cases, competitors were removed from the fieezer into DM25, preconditioned for
one day in the competition medium, and then allowed to compete for one day.

An AN OVA (Table 9) shows no efl‘ect of either acetate concentration or initial
fiequency, nor any interaction between them, on relative fitness. It seems unlikely,
therefore, that acetate plays any important role in the stable coexistence between Lac‘ and
Lac+ that I have observed.

59

Table 9. AN OVA of efi‘ects of initial fiequency and supplemental acetate concentration

on fitness of Lac+ relative to Lac’.

 

 

Source SS df MS F P
Initial fiequency 0.002 1 0.002 0.348 0.559
Acetate concentration 0.038 3 0.013 2.341 0.092
Interaction 0.025 3 0.008 1.557 0.219
Error 0. 174 32 0.005

 

Note: Experiment was carried out in medium also containing 2.5 ug rrlL'1 glucose.

Table 10. AN OVA of effects of initial frequency and supplemental glycerol concentration

on fitness of Lac+ relative to Lac'.

 

 

Source SS df MS F P
Initial fi'equency 0.006 1 0.006 1.987 0.168
Glycerol concentration 0.036 3 0.012 4.171 0.013
Interaction 0.008 3 0.003 0.929 0.438
Error 0.092 32 0.003

 

Note: Experiment was carried out in medium also containing 2.5 ug mL"l glucose.

By contrast, glycerol concentration has a significant efl‘ect on the outcome of
competition between Lac+ and Lac', although this efl‘ect is independent of initial
frequency (Table 10). The efl‘ect of glycerol is complex, however, as shown in Figure 13.
Increasing the glycerol concentration fi'om 0 to 1 ug mL'l shifls the advantage towards
Lac+, while increasing the concentration fiom 1 to 10 ug mL"1 shifts the advantage back
towards Lac'. This finding suggests that glycerol could be an important metabolite in the
stable coexistence between Lac‘ and Lac+. That is, the addition of glycerol provides an
advantage to Lac+, which may ofi‘set its slower growth at high glucose concentrations,

provided that the concentration of glycerol is not too high.

DISCUSSION

In a study intended to examine the efl‘ects of genetic recombination on the
dynamics of bacterial evolution (Chapter I), I found that at least two distinct genotypes of
E. coli coexisted in one of the experimental populations. This observation was
unexpected since the populations were provided with only a single resource (glucose)
which limited population density. In this chapter, I sought to determine the ecological
mechanisms responsible for the coexistence of two of these recombinant strains. To that
end, I first demonstrated that the coexistence was dynamically stable by showing that each
strain possessed a competitive advantage when it was in the minority (Figures 7 and 8).

I then considered two mechanistic hypotheses to explain the stable coexistence.
One hypothesis relies on the seasonal nature of the experimental environment, such that
the glucose concentration changed temporally due to periodic transfers of the bacteria into
fresh medium. In a seasonal environment, two genotypes may stably coexist on a single
limiting resource (Figure 6) if one of them has an advantage when the resource is abundant

and the other has a sufficiently large opposing advantage when the resource has become

61

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.2 -
is
_l
.9- 1.1 _
9
a: T
a: 1 T
+ 1.0 --------- ~1- ------------ ----+--_l:-i ------------------
o J— T
3 .l.
“5 0.9 _
8
(D
.8
LL 0.8 l 1 J I
O 1.0 2.5 10

Glycerol concentration (09 ml")

Figure 13. Effect of supplemental glycerol concentration on the fitness of Lac+
relative to Lac‘, in medium also containing glucose at 2.5 ug mL' 1. Bars show the means
(1- SE) of ten replicates, five each with Lac+ at initial frequencies of 0.1 and 0.9. Data
were pooled across initial frequencies, which had no significant effect in this experiment
(see Table 10). The addition of small amounts of glycerol shifts the competitive advantage

to strain Lac+, but this advantage disappears at higher glycerol concentrations.

62

scarce (Stewart and Levin 1974; Tilman 1982). According to the other hypothesis, one or
more additional resources are introduced into the environment by the metabolic activities
of the genotypes themselves. Here, two or more genotypes may stably coexist if one of
them has an advantage on the exogenously supplied resource whereas the other has a
suficiently large opposing advantage in acquiring the metabolic by-product.

I demonstrated a demographic tradeofi‘ between growth rates at high and low
glucose concentrations. That is, strain Lac" was the better competitor for abundant
glucose whereas strain Lac+ was superior in competition for sparse glucose. But the
expected magnitude of each strain's advantage when rare, based on this tradeom was
insuficient to explain the observed strength of fiequency-dependence (Figure 9 versus
Figure 7) and, in fact, may even have been too weak to allow stable coexistence. Thus,
although a tradeoff between growth rates at high and low glucose concentrations exists, it
cannot be primarily responsible for the stable polymorphism that led me to look for the
hidden ecological mechanism.

I also demonstrated cross-feeding interactions between the two strains. Alter
glucose was exhausted by cell growth, each strain, when grown alone, experienced a net
decrease in population density due to death (Figure 12). However, when the two strains
were grown together, each strain fared better (after the glucose was depleted) than when
it was alone. In fact, strain Lac+ actually increased in density, indicating net population
growth, when Lac’ was abundant. Thus, Lac+ can ofl'set a disadvantage relative to Lac'
in competition for glucose by its greater ability to acquire and assimilate one or more
metabolic by-products of glucose utilization. However, the polymorphism becomes
unstable at low concentrations of glucose (Figure 10B), because population density is
reduced with a corresponding reduction in the concentration of metabolites. Evidently,
the stable coexistence between these two recombinant strains was mediated primarily by a

cross-feeding interaction.

63

In addition to the inherent interest in determining the ecological mechanisms
responsible for the stable coexistence of competitors in this simple model system, I believe
that my results have several other more general implications. First, this system shows how
organisms can, through their own biological activities, alter a simple environment into one
that is more complex In turn, this environmental complexity allows for diversity to be
stably maintained where it could not otherwise persist. As eloquently stated by
Rosenzweig et al. (1994, p. 915), "It seems clear that even starting with the simplest
possible genetic and environmental conditions, complexity is generated from uniformity,
allowing biodiversity to build upon itself.”

Second, because I studied recombinant strains bearing many easily discernable
genetic markers (Table 5), I had the opportunity to observe polymorphisms that might
otherwise have gone undiscovered. In a related study with populations that are evolving
in a strictly asexual fashion, and which lacked readily distinguished genetic markers,
Lenski et al. (1991) found no obvious evidence for complex ecological interactions of the
sort discerned in this study. It would be interesting to examine those asexual populations
more thoroughly to determine whether some stable polymorphisms might have been
overlooked. If such polymorphisms were not observed in those strictly asexual
populations, it raises the interesting question of whether the likelihood for stable
polymorphisms to evolve are greater in a system that allows for recombination between
strains that are already genetically divergent. 0f possible relevance is Levin's (1972)
observed stable coexistence between E. coli B and K12 strains in environmental
conditions similar to those that I employed, since the recombinants used in this study were
hybrids of E. coli B recipients and K12 donors (Table 5). Even though Levin did not
conclusively establish the ecological mechanism that mediated coexistence between E. coli
B and K12, it seems plausible that it might be the same mechanism observed here.

Third, many evolutionary ecologists seem to believe it is enough to show that there

exists a tradeofi‘ between two ecological traits (e.g., r versus K) in order to explain stable

coexistence of the corresponding genotypes or species. But while such tradeofi‘s are
generally necessary for stable coexistence, they may not be sufi‘icient. In this study, I
observed a demographic tradeofi’ between growth rates at high and low glucose
concentrations. Such a tradeofi‘ may in principle promote stable coexistence in a
temporally fluctuating environment (Stewart and Levin 1973; Tilman 1982), but whether it
does so depends on the size of the tradeofi‘ as well as the extent of fluctuations in
population density and resource concentration. In fact, a concentration-mediated tradeofi‘
for the two recombinant strains in this study was not an adequate explanation for the
observed strength of the fiequency-dependent advantage that each strain had when rare
(Figures 7 and 9). Therefore, when possible, experimental studies of the ecological
mechanisms that promote stable coexistence should move beyond merely demonstrating
that a required tradeofi‘ exists and look to establish the quantitative agreement between
observed and predicted dynamics. While able to fulfill this objective for the demographic-
tradeofi‘ model, I could not perform such an analysis for the cross-feeding hypothesis
because the identity of the relevant metabolites and the parameters governing their rates of
production and consumption were unknown.

Lastly, ecologists have long been concerned with the factors responsible for
maintaining greater or lesser diversity of species within communities (Council and Orias
1964; MacArthur and Wilson 1967; Connell 1978). One hypothesis suggests that
communities with greater primary production are more diverse than less productive
communities, which might contribute to the latitudinal cline in ecological diversity (Fischer
1960; Pianka 1966). Several possible mechanistic bases for such a relationship are
imaginable. For instance, more productive communities should support longer food
chains, thereby directly increasing diversity. Another possible mechanism is the
production by one species of a metabolite (or other secondary resource) that supports
another species, which is inferior in competition for the primary resource and would

otherwise be excluded. When the abundance of primary resource is diminished, then the

65

density of the producer species may be reduced to a level where the secondary resource is
not abundant enough to maintain the other species. In fact, the stable coexistence of
recombinant bacterial strains that I observed in more productive environments (i.e., with
higher inputs of glucose) was eliminated when productivity was reduced. Although this is
but one example in a highly simplified experimental system, it illustrates the value of
understanding the specific dynamical mechanisms responsible for maintaining ecological
diversity.

CHAPTER III
TRADEOFF BETWEEN HORIZONTAL AND VERTICAL MODES OF
TRANSMISSION IN BACTERIAL PLASMIDS

INTRODUCTION

Horizontal transmission occurs whenever a parasite is transmitted fi'om an infected
individual to an uninfected individual, whether by direct contact or via an infectious
particle. Vertical transmission occurs when an infected individual reproduces (either
sexually or asexually), giving rise to progeny which also harbor the infectious agent. For
certain parasites there exists a fundamental conflict between these two modes of genetic
transmission. Many activities of a parasite that increase its rate of infectious transmission
(e.g., greater intra-host production of infectious particles) are likely to lower host fitness.
This reduction in host fitness reduces the potential for vertical transmission of the parasite.
Thus, a tradeofl‘ between vertical and horizontal transmission is likely to exist.

Parasite virulence is a relative term, and it may be usefully defined in terms of the
reduction in host fitness due to infection (Levin and Lenski 1983; May and Anderson
1983; Bull et al. 1991; Herre 1993; Bull 1994; Ewald 1994). Group-selectionist
arguments once led to widespread acceptance of the notion that parasites always evolve
toward a state of attenuated virulence (Ewald 1994). However, recent theoretical and
empirical studies have contradicted the idea that parasites inevitably evolve toward benign
coexistence with their hosts (Levin and Pimentel 1981; Ewald 1983, 1987; May and
Anderson 1983; Herre 1993; Bull 1994; Lenski and May 1994). Many of these studies
implicate the availability of uninfected hosts in the environment as a key factor in
determining the evolution of parasite virulence and, in those parasites that can also be

transmitted vertically, which mode of transmission is selectively favored. The basic

66

67

argument is as follows. Consider the case of a parasite that is transmitted by direct
contact between infected and uninfected (susceptible) hosts. When the density of
uninfected hosts is high, the expected time to transmission of the parasite from an infected
host to an uninfected host is short. Consequently, it is relatively advantageous for the
parasite to maximize its transmission to new hosts, regardless of the effect on its current
host's fitness. Thus, selection favors more virulent forms of the parasite. However, when
the density of uninfected hosts is low, the time to transmission will be much greater. Here,
a parasite benefits from doing little damage to its present host because the rate of
transmission to new hosts is small, and a parasite that reduces the fitness of its present
host too much may actually reduce its own likelihood of infectious transmission. By the
same logic, if a parasite can be transmitted vertically, this mode of transmission becomes

increasingly important and selectively favored when uninfected hosts are rare.

Experimental system. - All plasmids are able to control their own replication using
the host machinery and are transferred vertically across generations of the host cell. In the
absence of selection on the host for specific plasmid-encoded characters (such as antibiotic
resistance), most plasmids reduce the fitness of their hosts relative to isogenic plasmid-free
counterparts (Levin 1980; Dykhuizen and Hartl 1983; Lenski and Bouma 1987; Lenski
and Nguyen 1988; Nguyen et al. 1989); hence, they can be regarded as parasites under
these circumstances (Levin and Lenski 1983; Bouma and Lenski 1988). Many plasmids
are also able to transfer horizontally from an infected cell (donor) to an uninfected cell
(recipient) through a process called conjugation (Lederberg 1956). Although some of the
details of the conjugation process are still poorly understood, conjugation is initiated by
contact between donor and recipient cells via a plasmid-encoded protein appendage
known as a sex pilus. Thus, conjugative plasmids can be transmitted by two distinct
modes: horizontal (infectious) transmission occurs by conjugation, whereas vertical

(intergenerational) transmission occurs by host cell division.

68

lheoretical predictions. - Here I define plasmid Virulence” as the magnitude of
the reduction in host fitness due to plasmid carriage, which concomitantly lowers the
plasmid's ability to be transmitted across generations of the host. Activities of a plasmid
that increase its horizontal transmission (such as production of more sex pili) should
generally increase virulence (i.e., lower the host's fitness), thereby reducing the plasmids
own rate of vertical transmission. Therefore, there exists a fundamental conflict between
the two modes of transmission available to conjugative plasmids. When susceptible hosts
are abundant, horizontal transmission is more important and selection should favor
increased rates of horizontal transfer (increased vinllence). When susceptible hosts are
scarce, vertical transmission is the more fi'equent mode of transmission and selection
should favor increased rates of vertical transmission (reduced virulence).

A simple model describes how different modes of plasmid spread should be
favored under difi‘erent densities of susceptible hosts. Four assumptions are implicit in the
model: (i) no dynamic feedbacks afi‘ect the density of susceptible hosts (i.e., host density is
treated as a constant); (ii) plasmid-bearing cells cannot be reinfected; (iii) horizontal
transfer depends on mass-action kinetics; and (iv) there is a genetically-determined
tradeofi‘ between the rates of horizontal and vertical plasmid transmission. Let H be the
density of plasmid-flee hosts (volume'l) and P be the density of plasmid-bearing hosts
(volume‘l). Ify is the rate at which a plasmid is able to conjugatively transfer (volume
time'l) and rn is the intrinsic growth rate of plasmid-bearing cells (time’l), then the rate of

change in plasmid-bearing cells is

dP/dt=mP+yHP,

and the per capita rate of change is

69

r=dP/Pdt=m+yH.

The individual components of r reveal that the vertical component of plasmid spread (m) is
independent of host density, while the horizontal component (y H) is directly proportional
to the density of potential recipients. In theory, a conjugative plasmid can be stably
maintained in a bacterial population if the rate of horizontal transfer is suficiently high to
ofi‘set losses due to its harmful effects on host fitness (Stewart and Levin 1977).

To further illustrate the tradeoff between horizontal and vertical transmission,
consider two plasmid genotypes A and B. Plasmid A conjugates at a rate of VA = 1 and
allows its host to grow at the rate m A = 1. Plasmid B conjugates at a rate of 73 = 1.5 but
reduces its host’s growth rate to m}; = 0.5, and it is therefore more virulent than A
Figure 14 depicts the horizontal, vertical and net rates of transfer for plasmids A and B
under difi‘erent densities of susceptible hosts, as predicted by the model. When H < 1, r A
> rB so that the less virulent plasmid A prevails by virtue of its greater vertical
transmission. But when H > 1, r A < r3 and the more virulent plasmid B prevails by virtue
of its superior horizontal transmission. Thus, whether a more or less virulent plasmid is

favored depends on the abundance of uninfected (susceptible) hosts.

Experimental overview. - Bacterial p0pulations provide an excellent means to
study evolutionary theory for several reasons. Bacteria have short generation times and
large population sizes, making relatively long-term evolutionary studies possible. Entire
populations of bacteria can be stored in a suspended state, so that direct comparisons
between ancestors and evolutionary descendants are possible. Lastly, bacteria such as
Escherichia coli have been widely used as an experimental model in genetics, molecular

biology and microbial physiology. Therefore, much is already known about the biology of

70

Horizontal transmission

Rate

 

 

Rate

 

 

 

Rate

 

 

 

Host density

Figure 14. Horizontal, vertical, and net rates of increase for two plasmid
genotypes, A and B, as a function of susceptible (plasmid-free) host density. The more
virulent plasmid (B) is favored only when susceptible hosts are sufficiently abundant. See

text for details.

71

E coli and its interactions with plasmids and other infectious elements. For these reasons,
evolution experiments with bacteria and their plasmids provide a powerfill system to study
host-parasite interactions and the evolution of virulence (Levin and Lenski 1983; Bouma
and Lenski 1988; Bull et al. 1991).

I conducted a SOD-generation (75-day) experiment to look at the influence of
susceptible host density on the evolution of plasmid virulence and mode of transmission.
For this study, I used plasmid pB15, a naturally occurring plasmid previously reported to
persist by a high rate of horizontal transmission in chemostat culture (Lundquist and Levin '
1986). I subjected three replicated treatments of the E coli/pBIS association to batch
culture environments that contained difi‘erent input densities of susceptible (plasmid-flee)
hosts. Two of the three treatments allowed a fixed immigration of plasmid-flee hosts into
resident plasmid-bearing populations at periodic intervals (every 20 generations), creating
immigrant-to-resident ratios in these populations of either 1:1 or 100: 1. In contrast, the
third treatment never received any plasmid-fiee immigrants. Therefore, the last treatment
was expected to favor vertical transmission exclusively, while the other two treatments
should favor horizontal transmission to different degrees. To ensure that the bacterial
populations retained their plasmids, all three treatment groups were periodically placed in
an antibiotic environment, to preclude over-growth by plasrnid-fi'ee immigrants or
segregants.

Two control groups were created for the experiment as well. One control
contained plasmid-free E. coli populations that served as an immigrant pool for the
treatments described above. The other control contained replicated E coli/pB15
associations and was designed to study the effects of spontaneous plasmid segregation
(loss) in the absence of antibiotic. All five experimental groups are summarized in Table
l 1.

To evaluate whether host density afi‘ected the plasmid's evolution, I looked for

changes in plasmid conjugation rate and the cost of plasmid carriage to the host. As

72

.002» :08 8 8088803 08880.. 00.8.. 803 000.3. ”0002

 

 

has: 33839. .50

32 am 03888 .33 a? 0883.053 :
sees 3388.. as

2 a3 033.88 §3§§3§6€33 2
has: 338.8: .8

we as 033.88 32558383333 a
Ammo—v 80880090 88083 30058030 .«0

me on some 8350 33.38 §e§3 . m

3 on 38 3388 80-8.83 a

88% 80.88% 00 80%—0m ~8qu 3:000

.5085 2353.? 82.35

 

.E 000980 .203 3000» 80880.: .8800 88 88088098 .«0 b0885m .: 030,—.

73

explained earlier, I expect both the cost of carriage and the conjugation rate to be higher
under conditions of high host density than under conditions of low host density (Table 12).
Cost of carriage and conjugation rate are both easily measured (Bouma and Lenski 1988;
Simonsen et al. 1990).

Table 12. Expected evolutionary changes in traits pertaining to plasmid virulence.

 

 

 

 

Susceptible host density
Trait Low High
Cost of carriage - +
Conjugation rate - +
MATERIALS AND METHODS

Bacterial strains. - Table 13 describes the pertinent features of the bacteria used
in this study. All strains were derived from a single clone of E. coli B (REL1206),
previously evolved for 2,000 generations in a glucose-limited environment (Lenski et al.
1991). Although this strain is prototrophic, it is unable to utilize the sugar L-arabinose as
a nutrient (Ara'). A spontaneous arabinose-utilizing (Ara"') mutant of the strain
(REL120’7) was obtained in a previous study (Bennett et al. 1992). The Ara+o and Ara'o

74

.382 3565.. 8.3.. v.32 9a 83a

3 as» 93%.. a 88%2 3335 .32 .3835 33350 $3: .5 as 83.2 3 ”scape“ 3: 2 5.3 as»? 35 _

 

 

8%z\¢+c< n _ m3 053 2 9mm 33 “50389.50 Emma
.3233 383.32% 333533238538” Enema
3323.5 Swim—angssaaazgogseoe 22.03
83+a< £3 35% was 82.55.. 50%; 330mm
83.3 382.2 3885 fie. egg was... 82323 35038355 324mm
c+e< 33% E88 “80592.. 35a: 35 38$an Suimm
Pe< :8 339.235 80-..?83 8203
5330.532 ~3§§0 053—3“ 5.9m

 

.E .330 a 33 a?» 3863 .3 .2 23

75

ancestors in this study form white and red colonies, respectively, on tetrazolium-arabinose
(TA) indicator plates (Levin et al. 1977).

The conjugative plasmid used in this study, pBlS, was obtained from Prof. B. R
Levin (Emory University, Atlanta, Georgia). This plasmid was originally isolated from an
E. coli strain sampled from a human under antibiotic treatment (B. R Levin, personal
communication). It is a large (~80 kb), low-copy number plasmid, that confers resistance
to the antibiotics kanamycin (Kant) and tetracycline (Tet') (Lundquist and Levin 1986).
Kanamycin binds to the 708 ribosomal subunit and is lethal to sensitive cells (bactericidal).
Tetracycline inhibits protein synthesis and stops sensitive cells from dividing
(bacteriostatic). I continued resistance encoded by pB15 to these antibiotics by spreading
plasmid-bearing and plasmid-free cells onto TA plates supplemented with 25 ug mL'l Kan
and 1 ug mL'1 Tet.

Plasmid pBIS was transferred to the REL1206 and REL1207 backgrounds by
mixing each strain with a donor, and then selecting for transconjugants generated in
overnight mating cultures. In this way, I obtained ancestral strains REL53 82 (Ara'o/po)
and REL53 84 (Ara+o/po), which harbored pB 15 but were otherwise isogenic to REL1206
and REL1207, respectively (Table 13).

Media and culture conditions. - The culture medium employed in all experiments
was Davis minimal broth (Carlton and Brown 1981) supplemented with 2 mg L'1 thiamine
hydrochloride and 1000 mg L-1 glucose (DM1000). A plasmid-selective medium of
DM1000 supplemented with 25 ug mL-l Kan (DMIOOO-t-Kan) was also used. Either
medium allows a stationary-phase bacterial density of ~109 cells mL'l. Culture volume
was 10 mL, maintained in 18x150 mm borosilicate glass tubes and placed in a non-shaking
incubator at 37 0C. All cultures were propagated daily by vortexing and then transferring
0.1 mL of each culture into 9.9 mL of fresh medium (serial transfer). During this 24-hour

76

cycle, bacterial populations attained stationary-phase densities. The resulting lOO-fold
daily growth of each bacterial population represents ~6.64 generations of binary fission

Plasmid electrophoresis and restriction - Plasmid DNA was extracted using the
method of Birnboirn and Doly (1979). The relative sizes of restriction-endonuclease-
digested plasmid fragments were examined by electrophoresis on 0.6% agarose gels.
Restriction endonucleases EcoRI and BamHI were used (New England Biolabs).
Restriction digests were prepared using the procedure described in Sambrook et al.
(1989).

Experimental treatments: control populations. - Three clones of Ara+o were used
to initiate the three populations in the plasmid-free (F) control. Each population
underwent serial transfer into DM1000 for 75 days. Every three days, each population
also served as an immigrant pool to manipulate susceptible host density (see below).

Three clones of Ara‘o/po were similarly used to initiate the three populations in the
plasmid-bearing (B) control. Each population also underwent serial transfer into DM1000
for 75 d. B control populations did not experience antibiotic selection for plasmid
maintenance, and therefore they were potentially subject to takeover by spontaneous

plasmid-free segregants.

Experimental treatments: host density manipulations. - Three clones of Ara'o/po
were used to found the three replicate populations in each experimental treatment. At the
start of the experiment (day 0), each population in the low opportunity for horizontal
transfer (L) treatment underwent serial transfer into DM1000. On day 1, each of these
populations underwent serial transfer into DM1000+Kan to ensure plasmid-maintenance
(i.e., to select against any plasmid-free segregants). On day 2, each population was
serially-transferred into DM1000. This three-day cycle was repeated for 75 d.

On day 0, each resident population in the medium opportunity for horizontal
transfer (M) treatment and a paired immigrant population from the F control were mixed
at a 1:1 volumetric ratio and diluted 1:100 into DM1000. The populations in the high
opportunity for horizontal transfer (H) treatment were initiated in the same way, but at a
100:1 immigrant-to-resident ratio. All populations were allowed to grow and conjugate
during a standard daily growth cycle. On day 1, all populations underwent serial transfer
into DM1000+Kan to remove any plasmid-flee immigrants or spontaneous segregants.
On day 2, all populations were serially-transferred into DM1000. On day 3, each
population experienced an identical immigration event, involving the same paired
immigrant population, as on day 0. This three-day cycle was repeated for 75 d.

Aside from differences in inunigrant-to-resident ratios, all aspects of the
environment (including the frequency of antibiotic selection) were kept constant between
the L, M, and H treatments. All five experimental treatment and control groups are
summarized in Table 11.

Every three days for the first 100 generations and periodically during the
subsequent 400 generations, a sample from each experimental population was spread on
TA and TA+Kan plates. Plate counts were used to determine the density of total cells and
of plasmid-bearing cells, respectively; the density of plasmid-free cells was obtained by
subtraction. Every 15 days (100 generations), a sample from each population was also
spread on a TA+Tet plate to determine whether plasmid-bearing lines still retained their
resistance to tetracycline. Every 15 days, after serial transfer had taken place, glycerol
was added to a sample from each population and the sample was stored in a freezer at -80
°C for future study. Single colonies were also chosen at random from each population at
generation 500 and stored at -80 0C.

Fitness assay. - To assay relative fitness (W), two strains were placed in

competition under the culture conditions described above, where one competitor was

78

Ara+ and the other was Ara'. Each strain was grown separately for one day in the
experimental medium, as a preconditioning step to ensure that both competitors were in
comparable physiological states. The two competitors were mixed at a 1:1 ratio, then
diluted 1:100 into fresh medium and allowed to grow and compete during a standard 24-
hour growth cycle. Initial and final densities of each competitor were estimated by
spreading samples on TA plates, which permitted the competitors to be distinguished by
colony color.

Let the initial densities of the Ara+ and Ara' competitors be N1(0) and N2(0),
respectively; and let N10) and N2(1) be their corresponding densities after one day. The
average rate of increase (or realized Malthusian parameter), m;, for either competitor is
then calculated as: 1

mi = InlIh(1)/M(0)1I(1 day).

The fitness of one strain relative to another (Wij) is estimated as the ratio of their
Malthusian parameters (Lenski et al. 1991):

Wij = mi/mj.

A fitness difl'erence between the two competitors may reflect difl‘erences in their lag phase,
maximum or submaximum growth rates, survival at stationary phase, or some combination

thereof (e.g., Vasi et a1 1994).

Conjugation rate. - In order to assay rates of plasmid transfer, I obtained mutants
of Ara+o and Ara'o that were resistant to nalidixic acid (Nalr). Thus, I obtained strains
PET318 and PET319 that were Nal’, but otherwise isogenic to Ara+o and Ara'o,
respectively (Table 13).

79

Conjugation rate (7) was assayed under the DM1000 culture conditions described
above by mating a plasmid-bearing donor with a plasmid-free Nalr recipient that bore the
opposite arabinose marker. Donors and recipients were grown to stationary phase in
DM1000+Kan and DM1000, respectively. Each strain was then preconditioned for one
day in DM1000. Donors and recipients were mixed at a 1:1 volumetric ratio, then diluted
1:100 into fresh medium and allowed to grow and mate during a standard daily growth
cycle. Afier 0 and 24 hours, the densities of donors (D), recipients (R) and
transconjugants (I) were determined by colonies formed on appropriate selective and
nonselective plates. I also estimated the growth rate (hrl) in exponential phase (111) of
mating cultures by regressing In [total cell density] versus time during the period of
exponential-phase growth. Total cell densities were estimated by counts obtained using a
Coulter electronic particle counter (model ZM and channelyzer model 256).

I estimated the rate of transfer (mL hrl) for matings between donor and recipient

cells in batch culture using the formula of Simonsen et al. (1990):

y = wln[l + (T /R)(N/D)]/(N - No),

where N = T + R + D and No is initial population size.

Cost of plasmid carriage. - Cost of carriage (c) was assayed by competing a
plasmid-bearing strain with a plasmid-flee strain that differed only by a neutral marker.
The plasmid-bearing and plasmid-free competitors were grown separately in
DM1000+Kan and DM1000, respectively. The two competitors were then preconditioned
for one day in DM1000. The strains were then mixed at a 1:1 ratio and allowed to
compete during a standard daily growth cycle, as previously described. The fitness of the
plasmid-bearing strain relative to the plasmid-fi'ee strain over that cycle was then
calculated fi'om plate counts obtained after 0 and 24 hours.

The change in cost of plasmid carriage (Ac) was assayed by allowing a strain
bearing the ancestral plasmid to compete against a neutrally marked isogenic strain
harboring an evolved plasmid. To do so, I moved p0 onto the Ara+o/Nal" background to
create strain PET3 54, Arato/Nalr/po (Table 13). An evolved plasmid of interest was
moved to an Ara‘o/N a1r background, and the resulting strain was allowed to compete
against PET3 54. In all assays, competitors were grown separately in DM1000+Kan, and
then preconditioned for one day in the competitive environment. The strains were then
mixed at a 1:1 ratio and allowed to compete during a standard daily growth cycle. The
fitness of the strain bearing the evolved plasmid relative to the strain bearing the ancestral

plasmid was estimated, and Ac was calculated by subtracting this fitness estimate from 1.0.

RESULTS

Ancestral Plasmid Traits

Ancestral cost ofplasmid carriage. - Experiments were performed to determine
the fitness of the ancestral plasmid-bearing strain (Ara'o/po) relative to the plasmid-free
ancestor (Ara+o). To examine whether segregation and conjugation might complicate
estimation of this quantity, I sampled fi'om competition cultures to determine the
proportion of segregants and transconjugants generated. I then computed relative fitness
with and without adjusting for segregants and transconjugants. In ten replicate fitness
assays, the estimated mean fitness of Ara'olpo relative to Ara+o, adjusting for plasmid
losses and gains, was W= 0.981; mean fitness estimated without such adjustments was W
= 0.979. These two estimates were not significantly different (ts = 0.077, 18 df, p =
0.939). Because monitoring plasmid losses and gains provided no better estimate of
relative fitness in these one-day competition experiments, I performed two more blocks of

fitness assays without such monitoring.

81

The combined results from three blocks of fitness assays showed that the mean
fitness of Ara'o/po relative to Ara+o was W= 0.989. There was no significant efi‘ect of
block on fitness (p = 0.655), and mean fitness did not differ significantly fi'om 1.0 (ts =
1.226, 29 df, p = 0.230). 1 - W= 0.011 provides an estimate of the cost of carriage for
pB15 in the ancestral background. I conclude that pBlS imposes only a slight (and

statistically insignificant) fitness cost in Ara'o.

Table 14. Ancestral plasmid traits.

 

 

Mean 95% Confidence
Estimate Interval
Cost of plasmid carriage, c 0.011 1 0.019
Loglo conjugation rate, 1 -12. 161 1; 0.085

 

Fitness assays with ten-fold replication also showed that the plasmid-bearing
resident (Ara‘o/po) and its plasmid-bearing Ara+ counterpart (Ara+o/po) were equally
competitive in both the presence of kanamycin (W= 1.003 1- 0.025 SE; ts = 0.113, 9 df, p
= 0.912) and in its absence (W= 0.996 1 0.014 SE; t3 = 0.256, 9 df, p = 0.804). Thus, the

arabinose marker has no significant efl‘ects on fitness under my experimental conditions.

Ancestral conjugation rate. -- Twenty replicate mating assays between the

ancestral plasmid-bearing strain, Ara‘o/po, and an Ara‘fiVNalr recipient were conducted in

82

the antibiotic-flee evolutionary environment (DM1000). The conjugation rate coemcient,
7, was estimated using the endpoint method of Simonsen et al. (1990), as summarized in
the Materials and Methods. The mean estimate of loglo y for the ancestral plasmid was
-12.161 (i 0.040 SE) mL hrl. Means and 95% confidence intervals for the ancestral

plasmid traits are given in Table 14.

Control Populations
Evolutionary dynamics. - Plasmid-bearing (B) control populations were
potentially subject to takeover by spontaneous plasniid-fi'ee segregants. I observed that,
on average, segregants never reached an average frequency of more than about one-third
of the total population (Figure 15). This result suggests that horizontal transfer occurred
at a high enough rate to reinfect segregants, so that they always remained in the minority.

Fitness changes. - The ancestral plasmid-free resident (Ara'o) and the ancestral
immigrant (Ara"’o) were competitively equivalent in the antibiotic-flee evolutionary
environment (DM1000). Twenty replicate fitness assays yielded a mean fitness of Ara'o
relative to Ara+o of W= 1.008, which did not differ significantly fiom 1.0 (t, = 1.068, 19
df; p = 0.299). Using these two strains as equivalent baseline competitors, I sought to
determine whether the plasmid-bearing controls and plasmid-free controls difi‘ered in their
rates of adaptation to the antibiotic-free environment (DM1000). To do so, I measured
the fitness, relative to either Ara'o or Ara+ , at 100 generation intervals for each
population in the F and B controls, respectively. Fitness measurements were replicated
twice for each population, and the grand mean fitness over the three replicate populations
in each control group was calculated at each time point. The F and B controls underwent
nearly parallel fitness improvements during the first halfof the experiment, but the B
controls later fell behind in their rate of adaptation (Figure 16). The relevance of this

result will become apparent in the next section.

83

 

 

9.0 -

8.5 ifi/W
7— ,A .....
E “— 4,. .4/ k\ / \k ..—— —"' ‘
g \ /A/
8 8.0 — *
6'5
0
_l

7.5 -

70 J l 1 l 1

O 100 200 300 400 500

Tlme (generatlons)

Figure 15. Evolutionary dynamics in the B control populations (see Table 11), in
which the plasmid-bearing resident populations received neither plasmid-free immigrants
nor antibiotics. Each point represents the mean of three replicate populations. Plasmid-
free segregants (triangles) were detected as a minority population, but they were unable to

replace the plasmid-bearing genotypes (circles). Points are averages based on the three
replicate B populations.

1.0 — ‘

 

0.9 l l l 1 l l
O 1 00 200 300 400 500

Time (generations)

 

Fitness relative to plasmld-free ancestor
_L
_L
I

Figure 16. Fitness trajectories for plasmid-bearing and plasmid-free control
populations during evolution in the antibiotic-free environment. Plasmid-free (F) control
populations (triangles) continued to adapt to the environment throughout the 500
generations, whereas plasmid-bearing (B) control populations (circles) did not. Fitness
was measured relative to the plasmid-free ancestor. Each point represents the grand mean
(.1 SE) of three populations.

85

Host Density Manipulations

Evolutionary dynamics. - Plasmid-bearing populations in the medium (M) and
high (H) opportunity for horizontal transfer treatments were allowed to conjugate with,
and be transmitted to, immigrant genotypes at regular intervals. The immigrants could not
become established, however, unless they acquired a plasmid from the resident population,
owing to my imposition of antibiotic selection every third day. Every 100 generations, I
sampled from these plasmid-bearing populations to determine the fi'equency of Ara+
immigrants. My results showed that novel associations between immigrant hosts and
resident plasmids were rare until relatively late in the experiment (Figure 17). This result
may coincide with the fitness trajectories for the F and B controls, which diverged late in
the experiment (Figure 16). Apparently, there was strong selection for the resident
plasmids to reach the immigrant genetic background only alter the fitness gains for the
plasmid-free immigrants had begun to outpace the gains for the plasmid-bearing lines. I
also observed that the higher immigration rates in the H treatment did not lead to a higher
final proportion of immigrant hosts. Rather, all three M populations contained nearly
100% immigrant hosts, whereas the three H populations showed roughly 0%, 50% and
100% immigrant hosts (Figure 17).

All populations in the low (L), medium (M), and high (H) opportunity for
horizontal transfer treatments were founded by an ancestral plasmid that conferred
resistance to kanamycin and tetracycline (Kant Tet"). Every 100 generations, I sampled
from treatment populations to determine whether plasmid-bearing cells had become
sensitive to tetracycline (Tets), an unselected marker. I observed that the first novel
genotype to be detected in each treatment population was an ancestral-type (resident) host
bearing a Tet“ plasmid; the final frequency of Tets plasmids in each population was highly
variable within treatments (Figure 18). Interestingly, Tets plasmids were never detected in

the B controls, even though they were founded by the same ancestral plasmid used to

86

1.0 -

0.8 -

0.4 r

0.2 -

Frequency of Ara+ Immigrants

 

0.0

 

 

 

0.8 -

0.6 -

0.4 -

l

0.2

0.0 A c M
o 100 200 300 400 500

Frequency of Ara+ immigrants

 

 

Time (generations)

Figure 17. Changes in the frequency of Ara+ immigrant backgrounds in plasmid-
bearing treatment populations subjected to medium (M: panel A) or high (H: panel B)
levels of immigration by plasmid-free cells. The final frequency of Ara+ immigrants
showed no significant association with treatment (Mann-Whitney test, p > 0.2). Each

curve represents an independent replicate population.

87

1.0 P

0.8 -

0.6 _

02-

Frequency of Tat“ plasmids

 

0.0 ' 4 f +
O 1 00 200 300 400 500

 

0.8 -

0.6 -

0.2 -

Frequency of Tat“ plasmlds

 

 

0.0 I 1
O 1 00 200 300 400 500

 

1.0 r

0.8 -

0.4 r

0.2 -

  
   

Frequency of Tat“ plasmlds

  
 

 

L if
400 500

 

0.0 _
O 1 00 200 300

Time (generations)

Figure 18. Changes in the frequency of Tets plasmid variants in treatment
populations subjected to low (L: panel A), medium (M: panel B), or high (H: panel C)
levels of immigration by plasmid-free cells. The final frequency of Tets plasmids showed
no significant association with treatment (Kruskal-Wallis test, p > 0.5). Each curve

represents an independent replicate population.

found treatment populations. Why was the loss of tetracycline resistance common in all
three treatment groups but not in the controls?

During the experiment, treatment populations, but not controls, were periodically
subjected to kanamycin in order to ensure plasmid maintenance (see Materials and
Methods). Although the original plasmid encoded resistance to both kanamycin and
tetracycline, kanamycin was chosen as the selective agent because it is lethal to plasmid-
free cells, whereas tetracycline merely prevents these cells from growing. Only resistance
to kanamycin determined survival in the treatment environments, and my data suggest that
the loss of tetracycline resistance may have been a response to kanamycin selection. More
importantly, there was no overall pattern in the final fiequency of Tets plasmids in relation
to the susceptible-host-density treatments (e.g., the final frequency of Tet-‘3 plasmids
ranged from near 0% to _>_ 50% in L, M, and H treatments).

Fitness Changes. —- I sought to determine whether host density manipulations in
treatment lines led to difi‘erential improvements in fitness. To do so, I allowed a
heterogeneous sample fi'om each of the nine evolved populations to compete against the
reciprocally marked plasmid-free ancestor, Ara+o or Ara'o, in DM1000. To ensure
accurate estimates of relative fitness, I randomly sampled from competition cultures to
track the possible movement of plasmids. The proportion of segregants and
transconjugants generated in each competition was used to adjust my estimates of the
number of plasmid-bearing and plasmid-free competitors. Similarly, in competitions
involving an evolved population that was heterogeneous for the Ara marker, the
proportion of Ara+ and Ara' plasmid-bearing cells was used to adjust the number of
evolved competitors. I then computed the fitness of each mixed population relative to the
plasmid-free ancestor. The mean of five fitness assays was obtained for each final
treatment population, and the grand mean for the three populations in each experimental
group was calculated. The grand mean fitness relative to the plasmid-flee ancestor

89

increased in all three treatment lines (Figure 19), indicating adaptation to the antibiotic-
free environment. I performed a nested AN OVA to examine the variation in relative
fitness within and between treatment groups (Table 15). There was no significant

variation in mean fitness either within treatments or between treatments.

Table 15. Nested AN OVA to examine the efi‘ects of susceptible-host-density treatment,

and population within treatment, on fitness of evolved populations relative to ancestor.

 

 

Source SS df MS F P
Treatment 0.0016 2 0.0008 0.1 19 0.890
Population 0.0405 6 0.0067 0.787 0.586
Error 0.3082 36 0.0086

 

I also performed fitness assays by allowing single-colony isolates (instead of
heterogeneous samples) from each L, M, and H treatment population to compete against
the plasmid-free ancestor. In close accord with the heterogeneous samples, the single-
colony isolates had relative fitnesses averaging 1.167 (: 0.012 SE) and there was no
significant efi‘ect of treatment (data not shown). I conclude that differences in susceptible
host density did not lead to difi‘erential increases in fitness relative to the plasmid-flee
ancestor.

The preceding fitness measurements reflect a composite of effects due to

evolutionary changes in both the host and plasmid. A more sensitive measure of changes

 

 

 

 

 

 

 

 

 

 

 

*3 1.3 —

(D

O

C

03

8 1.2 ~ T

s— T

i: l * I
‘g 1.1 -

(D

_(Q

Q.

,9 1.0 ----------------------------------------------------
(D

a;

g 0.9 -

8

g 0.8 l I 1
a LOW MEDIUM HIGH

Figure 19. Mean fitness in treatment populations subjected to low, medium, and
high levels of immigration by plasmid-flee cells. Each bar represents the grand mean (1
SE) of three replicate populations. Fitnesses were measured relative to the plasrnid-fi'ee

ancestor in an antibiotic-free environment. See Table 15 for statistical analysis.

91

in plasmid efi‘ects would be to place each evolved plasmid on the same - ancestral - host
background, and then estimate the cost of carriage for each evolved plasmid (see Bouma
and Lenski 1988). To do so requires plasmid transfer and, hence, competitions involving
clonal isolates instead of heterogeneous populations. In fact, I can allow an ancestral host
carrying an evolved plasmid to compete against a marked but otherwise isogenic ancestral
host carrying the original plasmid, and thereby estimate directly the change in cost of
plasmid carriage. Such competitions may be carried out in each of the two environments
(DM1000 and DM1000+Kan) that plasmids experienced during their evolution (see
Materials and Methods). I may then ask: Do the changes in cost of plasmid carriage (if
any) difi‘er among susceptible-host-density treatments? Alternatively, do any such changes

correlate with the loss or retention of plasmid-encoded tetracycline resistance function?

Changes in Plasmid Traits

For each population in the L, M, and H treatments, I isolated the majority
genotype present at generation 500 and stored it in the hem. Population M1 was
polymorphic for two equally common genotypes that difl‘ered in their resistance to
tetracycline (Figure 188), and so two genotypes were used for that population. Thus, a
total of ten clonal isolates were obtained for the nine populations. From here on, I refer to
the evolved plasmid present in each majority genotype by the population from which it
was obtained (with the Tetr and Tets plasmids fi'om population M I referred to as er and
M1 3, respectively).

Cost of plasmid cam’age. - I sought to determine whether evolved plasmids had
undergone changes in the cost of their carriage to the ancestral host. To do so, I moved
each evolved plasmid into strain Ara'olNalr to create eight new host-plasmid associations.
As shown in the next section, two of the ten evolved plasmids were unable to conjugate

and so could not be transferred to a new host. Then, I allowed each of these eight

92

constructs to compete against Ara+o/Nalrlpo in antibiotic (DM1000+Kan) and in
antibiotic-free (DM1000) environments. Fitness assays were replicated three-fold, and the
fitness of the evolved association relative to the ancestral association was determined.

The difi‘erence of this fitness value fiom 1.0 gives a direct estimate of the change in cost of

carriage, Ac. I observed that, in both environments, cost of carriage had decreased for the

Table 16. Mixed-model two-way AN OVA to examine the effects of plasmid genotype and

assay environment on the change in cost of plasmid carriage.

 

 

Source SS df MS F P
Genotype 0.4366 7 0.0624 8.321 <0.001
Environment 0.0003 1 0.0003 0.258 0.627
Interaction 0.0087 7 0.0012 0.166 0.990
Error 0.2399 32 0.0075

 

Note: Plasmid genotype is a random efl‘ect and assay environment is a fixed effect.

evolved plasmids which were Tets, but had typically increased for those that were Tet‘
(Figure 20). I performed a two-way ANOVA to evaluate the effects of plasmid genotype
and assay environment on Ac (Table 16). This test confirmed that Ac was indeed
heterogeneous among the evolved plasmids, whereas neither the environment nor the
genotype-environment interaction had any significant efi‘ect on this trait. The lack of

correspondence between susceptrble-host-density treatment and Ac is illustrated by the

93

 

 

 

   

 

 

0.3—
% 02 — A
a
O 0.1—

“5

E

a)

9

2

O
_O.3 I I I I I I I I
L1 L2 L3 M1r M2 M3 H2 H3

o.3[
B
Ea
0 0'1 _
B :In
goo"- -- "=9“ -- --
E
a, —o.1 —
2’
(fig—0.2—

-0.3 I I I I I I I I

L 1 L2 L3 M1 r M2 M3 H2 H3

Figure 20. Change in cost of carriage to the host (Ac) for the eight evolved
plasmids that retained their ability to conjugate. To estimate Ac, each evolved plasmid
was transferred to the ancestral host background, which was then allowed to compete
against a neutrally marked, isogenic host carrying the ancestral plasmid. A Ac < 0
indicates that an evolved plasmid was less costly than the ancestral plasmid, and may even
have become beneficial. Open and filled bars represent Tetr and Tets evolved plasmids,
respectively. (A) Competition assays performed in an antibiotic-free environment. (B)
Assays performed in medium containing kanamycin. Error bars show SE. See Table 16

for statistical analysis.

94

fact that treatment L plasmids, for example, show both significant increments and
decrements in cost (Figure 20). I also computed the grand mean Ac in each assay
environment for the three evolved Tets plasmids and the five evolved Tetr plasmids.

These data indicate a statistically significant relationship between the tetracycline marker
and Ac (DM1000: ’3 = 4.382, 6 df, p = 0.005; DM1000+Kan; ‘s = 4.414, 6 df, p = 0.005).
I conclude that the evolved Tets plasmids substantially reduced the cost of their carriage in
relation to the ancestral plasmid, whereas the evolved Tet’ plasmids are actually more
costly than the ancestral plasmid.

Conjugation rate. - I measured the conjugation rate (1) for each of the ten
evolved plasmids. Mating assays were performed, with three-fold replication, between
each single-colony isolate and a recipient bearing the opposite arabinose marker
(Ara'*'()/Nalr or Ara'olNal‘). All five evolved plasmids that still expressed the tetracycline-
resistance function (Tet') conjugated at rates slightly higher than that of the ancestral
plasmid (Figure 21). Conversely, all five of the evolved plasmids that lost resistance to
tetracycline (T ets) conjugated at rates much lower than the ancestral plasmid; in fact, two
showed a complete inability to conjugate. The lack of any correspondence between
susceptible-host-density treatment and conjugation rate of the evolved plasmids is
illustrated by the tremendous variation in y for plasmids within each treatment. Treatment
M, for example, yielded one plasmid with y greater than the ancestor, one with 7 reduced
somewhat, and one in which 7 equals 0. Although the treatment efi‘ect was not significant,
a Mann-Whitney test indicates that the difference in conjugative rates between the 'l‘ets
and Tetr derived plasmids is highly significant (Us = 0.0, n1 = 5, n2 = 5, p = 0.008). I
conclude that the evolved Tet-‘3 plasmids show a deficiency in their ability to conjugate, or
an inability to transfer at all, in comparison to the ancestor and the evolved Tet" plasmids.

95

 

 

 

 

 

 

 

 

 

 

 

 

—11.5 —
_12.o - {'1 {I *l {I j
.03
E *I
C -12.5 —
2
13
3 —13.0 -
C
O
0
2 —13.5 -
U)
_I gm; //’/3
44-0 — %
’Y:

 

 

A L2 L3 M18 M1r M2 M3 H1 H2 H3

Figure 21. Conjugation rates (7) for the ancestral (A) and ten evolved plasmids.
Each bar represents the mean (i SE) of three measurements, except for the ancestor,
which is based on 20 assays. Two plasmids (M Is and H1) had conjugation rates at or
near 0, for which transconjugants were not detected. Open and filled bars represent Tetr
and 'I‘ets plasmids, respectively. 7 shows no significant association with immigration

treatments L, M, and H (Kruskal-Wallis test, p > 0.2).

96

T radeofl' in modes of transmission - Figure 22 shows the correlation between
loglo 1 and Ac (in DM1000) for the eight evolved plasmids that could be transferred to
the common background. The correlation is highly significant (r = 0.867, p - 0.005),
which clearly shows that a plasmid's conjugation rate is tightly coupled with the cost of its
carriage to the host bacterium. In other words, I observed the predicted genetic tradeofl‘
between opportunities for horizontal and vertical transmission of the plasmid. In addition,
I observed that retention versus loss of the tetracycline resistance function served (for an
unknown reason) as a proxy phenotypic marker for greater or lesser plasmid virulence,

respectively.

DISCUSSION

Many parasites can be transmitted either horizontally (infectiously) or vertically
(via host reproduction). In such cases, it is commonly assumed that there exists a
genetically determined tradeofi‘ between a parasite's potential to be transmitted
horizontally and vertically (Levin and Lenski 1983, May and Anderson 1983, Bull 1994,
Ewald 1994). This tradeofi‘ presumably occurs because activities of a parasite that
increase its infectiousness will be generally harmfirl to its host. Ifone assumes also that
hosts can only be infected by a single genotype of the parasite, then within-host
competition between parasites can be ignored. With these two assumptions, a simple
model predicts that the density of uninfected (susceptible) hosts should determine whether
a parasite evolves to become more or less infectious and, concomitantly, more or less
virulent in terms of its efi‘ect on host fitness (Figure 14). That is, when uninfected hosts
are common, the opportunity for infectious transfer is large and selection should favor
increased rates of horizontal transmission (increased virulence). But when uninfected

hosts are scarce, then vertical transmission is more frequent and selection should favor

97

 

 

-11.5 '-
. O C

9 -12.0 - . .
E’
c “12.5 "
_g
“6
3 -13.0 -
C
8
2 -13-5 '-
8 I I I
—‘ —14.0 -

_14.5 1 1 1 A

-O.2 -O.1 0.0 O. 1 0.2

Change In cost of carriage

Figure 22. Genetic correlation between rate parameters governing horizontal and
vertical modes of plasmid transmission in pB15 and its evolved derivatives. "A" is the
ancestral plasmid. The circles show five evolved plasmids that retained the ancestral
expression of tetracycline resistance. The squares show three evolved plasmids that
became tetracycline sensitive. The correlation between conjugation rate (see Figure 21)

and change in cost of carriage (see Figure 20A) for the eight evolved plasmids is highly
significant (r = 0.867, p = 0.005).

98

reduced virulence. I tested these predictions by allowing a conjugative plasmid that infects
the bacterium E. coli to evolve for 500 generations in replicated environments in which I
manipulated the densities of uninfected hosts.

I can summarize the main findings of my study as follows. (1) I observed a clear
tradeoff between evolved plasmids' conjugation rates and their efi‘ects on host fitness,
demonstrating an inability for these plasmids to simultaneously maximize both horizontal
and vertical modes of transmission (2) However, the susceptible-host-density treatments
had no systematic effect on the evolution of plasmid transmission and virulence, contrary
to the predictions of the epidemiological model. Instead, I observed that more and less
virulent plasmid genotypes took over the treatment populations with almost equal success,
irrespective of difi‘erences in susceptible-host-density treatments. Several successfirl
genotypes have a high conjugation rate coupled with a high cost of plasmid carriage, while
several others have a low conjugation rate and low cost of carriage. Taken together, these
two main findings present something of a puzzle. The epidemiological model's key
assumption - that there exists a firndamental tradeofi‘ between the two modes of plasmid
transmission - was firlfilled, but the prediction that susceptible host density would mediate
the evolution of virulence failed to materialize.

Bull et al. (1991) have also performed experiments to test essentially the same
evolutionary model that I tested. Using bacteria as hosts and filamentous viruses as
parasites, they manipulated the opportunity for vertical versus horizontal transmission.
Consistent with theory, Bull et al. (1991) saw that the parasites became less virulent (more
"benevolent") when there was no opportunity for horizontal transmission to uninfected
hosts. Thus, this earlier study supported the evolutionary predictions of the
epidemiological model, whereas my study did not. What might be the explanation for this
difi‘erence in evolutionary outcomes?

The two studies obviously differ in the bacterial parasites that were studied.

Perhaps the simplest explanation would be if the assumed genetic tradeoff between rates

99

of horizontal and vertical transmission did not hold for the plasmid that I looked at. But
this is not the case. Although I found no evidence for an efi‘ect of susceptible host density,
I did observe substantial genetic variation in both cost of carriage and conjugation rate
and, moreover, these two plasmid traits were highly correlated as predicted by theory
(Figure 22). Another simple explanation might be that the duration of my experiment (500
generations) was too short to observe the predicted evolution. But this also seems
unlikely, since I observed evolutionary changes with large efi‘ects on the relevant
phenotypes; these changes were simply not associated with my susceptible-host-density
treatments.

Presumably then, the theory which describes the evolution of plasmid virulence in
my system is wrong, or at least incomplete in some respect. Another key assumption of
the model shown in Figure 14 is that transmission dynamics behave according to mass-
action. That is, the opportunity for horizontal transfer by conjugative plasmids is
supposed to increase in direct proportion to the density of susceptible hosts (H). In
Chapter IV of this dissertation, I begin to examine the validity of this assumption. I have
observed that, contrary to the mass-action assumption, the conjugation rate parameter, 7,
is not constant but instead declines with increasing H. Thus, the product 7 H, which
reflects the opportunity for horizontal transmission does not scale proportionately with H.
In some experiments, the rate of horizontal transmission appeared to be highest at
intermediate host densities. This failure to fulfill the assumption of mass-action dynamics
might explain the failure to observe the predicted evolutionary efi‘ect of host density,
despite the clear genetic tradeofi‘ between horizontal and vertical modes of transmission.

The ancestral plasmid in this study, pB15, is resistant to the antibiotics kanamycin
and tetracycline. An interesting finding was the unexpected association between the
tetracycline-resistance function and the tradeofl‘ between cost of carriage and conjugation
rate in the evolved plasmids. In particular, every evolved plasmid that retained its

resistance to tetracycline was both more virulent and more transmissible than every

100

plasmid that had become tetracycline sensitive. One possible explanation for this
association might be that the firnctions for tetracycline resistance and conjugal transfer are
located adjacent to one another on the plasmid. Thus, deletion mutations might
simultaneously afl‘ect expression of both firnctions. At present, there is no genetic map for
pB15. However, I explored this possibility of deletions by running restriction digests on
the ancestral and derived plasmids using EcoRI and BamHI endonucleases. I saw no
obvious changes in plasmid size in any of the derived plasmids, either Tetr or Tets (data
not shown). Another possibility might be that the tetracycline resistance and conjugal
transfer functions are transcribed fi'om the same or overlapping promoters, in which case
point mutations in the promoter region could simultaneously increase or decrease
expression of both functions. However, I have not yet tested this hypothesis.

Because of their short generation times, large population sizes, and general ease of
culture, bacterial populations provide powerfirl experimental systems to test evolutionary
hypotheses. However, some individuals might worry that bacteria behave too much like
computer simulations and are not firll of complications and surprises like ”real" organisms.
My results show otherwise. For reasons that are not obvious, my study failed to support
the predicted effect of susceptible host density on the evolution of virulence, despite a
clear demonstration of an underlying genetic tradeofi‘ between horizontal and vertical
modes of transmission. And while the observed tradeofl‘ conformed to theoretical
expectations, it is not at all clear how or why this tradeofi' should involve the gene for
tetracycline resistance, and yet it does. Evidently, despite their seeming simplicity,

plasmid-bacterium interactions are sometimes unexpectedly complex

CHAPTER IV
UNEXPECTED EFFECT OF HOST DENSITY
ON CONJUGATION RATE AND INVASION OF PLASMID p315

INTRODUCTION

In the absence of selection on the host bacterium for specific plasmid-encoded
characters such as antibiotic resistance, most plasmids reduce the fitness of their hosts
relative to isogenic plasmid-free counterparts and so can be regarded as parasites (Levin
1980; Dykhuizen and Hartl 1983; Lenski and Bouma 1987; Lenski and Nguyen 1988;
Nguyen et al. 1989). However, a conjugative plasmid can invade a plasmid-flee
population of bacteria (and be stably maintained) if the rate of horizontal transfer by
conjugation is sufl'rciently high to ofl'set losses due to harmful efi‘ects on host fitness
(Stewart and Levin 1977). The opportunity for a conjugative plasmid to invade a
population of susceptible hosts is expected to increase in proportion to the density of the
hosts (Levin et al. 1979; Simonsen et al. 1990). This expectation is based on the
assumption that the kinetics of plasmid transfer behave according to mass-action; i.e., the
rate at which newly infected hosts are formed depends on the product of plasmid-bearing
and plasmid-free host densities (cell ml'l) and on the rate constant of conjugation, 7 (ml
cell'l hrl).

Plasmid pH 15 is a naturally occurring plasmid, and it has been previously reported
to persist by horizontal transmission in chemostat culture (Lundquist and Levin 1986). In
chapter III of this dissertation, pB 15 was used to test the hypothesis that higher densities
of susceptible hosts would favor plasmids with higher conjugation rates and greater
vimlence (i.e., more deleterious efi‘ects on host fitness). The evolved plasmid variants

fulfilled the assumption of a genetic correlation between conjugation rate and virulence.

101

102

However, I found no evidence for the predicted efl‘ect of susceptible-host-density
treatments on the direction of plasmid evolution. One possible explanation for this failure
to confirm the evolutionary prediction, despite fulfilling the key genetic assumption, is that
plasmid transfer may not have obeyed mass-action kinetics. Therefore, in this chapter, I
examine the efi‘ect of host density on the conjugation rate of p315 and on its ability to
invade bacterial populations growing in batch culture.

MATERIALS AND METHODS

The Escherichia coli B host strain used in this study had evolved for 2,000
generations in a glucose-limited environment (Lenski et al. 1991). Plasmid pB15 is a large
(~80 kb), low-copy-number plasmid that confers resistance to kanamycin and tetracycline
(Kant Tet“). Bacterial populations were cultured by serial transfer in Davis minimal (DM)
broth (Carlton and Brown 1981) supplemented with 2x10'5 ug ml"l thiamine
hydrochloride and glucose at a specified concentration. For example, DM25 indicates DM
with 25 ug glucose ml-l, which yields ~5x107 cells ml-1 at stationary phase. Culture
volume was 10 ml, maintained in 50-ml Erlenmeyer flasks placed in a non-shaking
incubator at 37 0C.

To assay conjugation rate (1), donor and recipient strains were mixed at a 1:100
ratio, then diluted 1:100 into fresh medium and allowed to grow and mate during a
standard 24-hour growth cycle in a non-shaking incubator at 37 0C. After 24 h, the final
densities of donors (D), recipients (R), and transconjugants (I) were determined by the
number of colonies formed on selective and nonselective plates. I also estimated the
growth rate (h'l) in exponential phase (\v) of mating cultures by regressing ln [total cell
density] versus time during the period of exponential-phase growth. The rate of conjugal

103

plasmid transfer (ml cell-1 h-l) for matings in batch culture may be estimated using the
formula of Simonsen et al. (1990):

y = wln[l + (T /R)(N/D)]/(N - No),

where N = T + R + D and No is initial population size.

RESULTS

I first performed a series of experiments to examine the influence of susceptible
host density on the ability of plasmid pH 15 to invade a population of plasmid-fiee hosts.
Plasmid-bearing (donor) and plasmid-flee (recipient) cells were mixed at a ~1:500 ratio, in
each of seven glucose concentrations: DM12.5, DM25, DM50, DM100, DM200, DM400,
and DM800. Donors and recipients in this study differ in their ability to utilize the sugar
L-arabinose as a nutrient; thus the Ara+ donors and Ara" recipients form white and red
colonies, respectively, on tetrazolium-arabinose (TA) indicator plates (Levin et al. 1977).
Populations were propagated by serial transfer for ten days. Samples from experimental
populations were plated daily on TA and TA containing 25 ug kanamycin ml'1 to
determine total cell density and density of plasmid-bearing cells, respectively. The density
of plasmid-flee cells was determined by subtraction [At the end of the experiment,
population samples were also tested on agar containing 1 ug tetracycline ml‘l. No
dissociation between the two antibiotic resistance markers was observed] The
expectation was that the rate of increase for pB15-bearing cells would increase in direct
proportion to the density of plasmid-free hosts (manipulated by changing the glucose
concentration). As expected, the population of plasmid-bearing donors and

transconjugants declined at the lowest density of susceptible hosts, ~3.05 x 107 cell ml"l

104

 

 

107 A
8W
E 6-
_rg
8
O
.1
2l- W
\—_——
O l l l J l
O 2 4 6 8 1O
10— B

Log1o cells ml’1
is
{

 

Time (days)

Figure 23. Densities of donors (squares), recipients (circles), and transconjugants
(diamonds) of plasmid pB15 during serial transfer at seven difi‘erent concentrations of
glucose: 12.5 (A), 25 (B), so (C), 100 (D), 200 (E), 400 (F), and 800 (G) ug ml-l. The
plasmid-bearing cell population could increase when rare only at intermediate cell densities

(panels C and D). Dashed lines indicate that a cell population fell below the limit of
detection.

105

10

 

P _
6 4

o—
.-_E 260 a3

p—

 

1O

1O

6 4

LE 2.8 290..

_

 

1O

10

 

6 4

LE 260 063

 

10

Time (days)

Figure 23 (cont'd).

Log,o cells ml“

Log,0 cells ml‘1

10

106

 

 

 

>—
_—
ln-
h
—

10
Time (days)

Figure 23 (cont'd).

107

(Figure 23, panel A). Also as expected, the population of plasmid-bearing cells declined
more slowly, and even began to increase, as the density of plasmid-free cells was raised to
5.66 x 107, 9.31 x 107, and 1.74 x 108 by raising the glucose concentration (Figure 23,
panels B-D). But unexpectedly, at still higher densities of plasmid-free cells -- 3.71 x 108,

Table 17. Estimates of loglo conjugation rate (7) for pB15 at three different glucose

concentrations, and corresponding cell densities.

 

 

Glucose Final cell Mean estimate 95% Confidence
concentration density“ of loglo y Interval
5 ug ml-l 1.65 x 107 -9972 1 0.083
50 ug ml-1 8.81 x 107 -10.342 i 0.206
500 ug mI'l 8.76 x 108 -11.227 1 0.112

 

‘Total density of donors, recipients, and transconjugants at the end of the 24-hour growth

cycle, averaged over the six replicates.

6.94 x 108, and 1.06 x 109 — the plasmid no longer could invade and become established
(Figure 23, panels E-G). These data suggest that the rate of horizontal transfer actually
declined at higher densities of susceptible hosts, contrary to the mass-action expectation.
To further explore this unexpected efi‘ect of host density on horizontal transfer, I
estimated the rate constant of conjugative transfer (7) for pB 15 at three concentrations of
glucose: DM5, DM50, and DMSOO. Plasmid-bearing and plasmid-free cells were mixed at

108

a ~1: 100 ratio, in each concentration of resource. The densities of donors, recipients, and
transconjugants were determined as above, and 1 was then estimated using the formula of
Simonsen et a1. (1990). Six blocks of assays were performed. As shown in Table 17, 1
does in fact decline with increasing susceptible host density (achieved by manipulating
glucose concentration). An AN OVA confirmed that the effect of glucose concentration

Table 18. ANOVA of the efl'ect of glucose concentration on loglo conjugation rate (7) of

 

 

plasmid pB15.

Source SS df MS F P
Glucose concentration 4.991 2 2.495 103.674 <0.001
Block 0.243 5 0.049 2.021 0.161
Error 0.241 10 0.024

 

on loglo y is highly significant (Table 18). These data further indicate that horizontal
transfer opr 15 does not behave according to simple mass-action kinetics.

Figure 24 shows the dynamics for the donor, recipient, and transconjugant
populations in one representative block of the conjugation rate experiment (Table 17).
Interestingly, the number of transconjugants increased substantially between 8 and 10 h of
the growth cycle, under all three density treatments. By this time in the cycle, the donors
and recipients had evidently made the transition from exponential growth to stationary
phase as the medium was depleted of glucose. These observations suggest that significant

109

10— A

 

 

in

Log“, cells ml"
o>
\\ I
a l

 

 

 

 

 

101' B

8%
'E
26 .
'5
o
§4_ A —¢
_1

2_

0 [III] I l I 111 J

 

 

 

Log“, cells ml“
0)
fi
b
D

 

O l l l M 1 1 l l L l l _J
O 2 4 6 8 10 12 14 16 18 20 22 24

Figure 24. Dynamics of one-day mating experiments with p315 and E. coli B at
three glucose concentrations: 5 (A), 50 (B), and 500 (C) ug ml'l. Circles: plasmid-flee

recipients. Squares: donors harboring plasmid pB15. Diamonds: transconjugants.

110

plasmid transfer may occur during the transition to stationary phase. By contrast, most
models of plasmid population dynamics (Stewart and Levin 1977; Simonsen et al. 1990)

assume that the conjugation rate is proportional to growth rate.

DISCUSSION

In Chapter III, I used a simple evolutionary model to predict that plasmids should
evolve higher conjugation rates, and become more deleterious to their hosts, when
susceptible hosts are common. However, an experiment in which the density of
susceptible hosts was manipulated for plasmid pBlS did not support this prediction. A
possible explanation for the model's failure to predict the evolutionary response of p315
to host density is that its conjugal transfer (horizontal transmission) does not obey simple
mass-action kinetics.

In this chapter, I performed two experiments to test whether pB 15 transmission
conforms to mass-action kinetics. Both experiments indicated significant deviations fiom
mass-action In the first experiment, I observed that the rate of increase in the number of
pB15-infected hosts (when pB 15 was introduced at a low frequency) was greater at
intermediate hosts densities than at either low or high densities (Figure 23). In the second
experiment, I estimated the conjugation rate, 7, using the endpoint method of Simonsen et
al. (1990). Using plasmid R1, Simonsen et al. found that 1 was independent of host
density, as expected for simple mass-action kinetics. But for pBlS, y declined by about an
order of magnitude as host density was increased by an equivalent amount (Table 17, rows
2 and 3). Thus, for pB15, the product of y and susceptible host density (which gives the
per capita rate of conjugative transmission) was approximately constant over this range,

rather than increasing in direct proportion to host density as was expected.

111

At this juncture, it seems clear that pBlS's conjugation rate declines substantially
at higher host densities. However, I cannot explain mechanistically why this should be 80.
Perhaps the simplest explanation is that the conjugation process somehow becomes
”saturated" at higher host cell densities, much as bacterial growth reaches a maximum rate
that cannot be raised by increasing the concentration of a limiting resource (Monod 1949).
However, saturation kinetics cannot explain the apparent maximum rate of plasmid
increase at intermediate host densities, as seen in Figure 23. Another possibility is that
conjugation rate is not a firnction of host density per se but instead responds to the
concentration of glucose, which was varied in order to manipulate host density. That is,
higher concentrations of glucose may somehow inhibit the conjugal transfer of p815. The
observation that the number of transconjugants increases unexpectedly during the
transition from exponential growth to stationary phase (Figure 24), irrespective of cell
density, may be consistent with this explanation, because it suggests that pH 15
conjugation is somehow stimulated by the depletion of glucose.

Whatever the precise explanation for these efi‘ects, they may explain the failure of
the simple mass-action model to predict the evolutionary response opr 15 to
experimental manipulations of the density of susceptible, plasmid-free hosts. Although the
genetic assumption of a tradeofi‘ between rates of horizontal and vertical transmission was
firlfilled (Chapter III), the ecological assumption that the rate of horizontal transmission is
simply proportional to susceptible host density was evidently not satisfied (this Chapter).
More generally, models of phenotypic evolution depend on both genetic and ecological
assumptions, and the predictions of these models may fail as a consequence of violating

either type of assumption.

APPENDIX

112

 

 

 

no N N N_ N N m c m c - + N 03—
mo _ N N N _ a a m e - + _ 8M «.92
_d N N N N N c .e. a m + t m 03—
_.o N N _ N ~ .— m u m + + v 8M
_d _ N _ N N_ c m .— m + t m 8%
md N_ N N N _ c a. .— m + .. N 8M
Nd _ N N N N_ u m a a + t _ 8M
Nd _ N N N ~ ._ m u m + . cam—08m Ntfl<
m6 _ N N N _ a c m .e. .. t m. 8am
#0 _ N _ N N a c a. a. - t v 8M
_d _ N N N N .e. c m m - t n 8x
_d N N N N N_ m e m m + t N 8d
To _ N N N N_ m a m m + t _ 03— 72¢.
Name—ace Ewan—808M
mmm E: mg 9.56 :9 3. XS. hm $8 8..— 8< 090280
mono—60h 25280
gauche—.3585 _ofiboeonm

 

68.— 2.3825» 3 noun—anon 825.85 noun—mag :08 89¢ 838m :8 98:3 :08 8.5050 .m— 033.

figuring.

Table 19(cont'd).

0.3

Rec3

0.1

Rec4

0.1

Recs

Rec6

0.1

0.8

Ara'4 Base' e

0.2

12

Reel

0.3

12
12

Ara‘5 Rec 1

0.1

Rec 2

Rec3

0.1

0.1

12

Rec4

113

0.4

12

Rec5

0.3

12
12
12
12

Ara'6 Rec 1

0.2
0.1

Rec2

Rec3

0.2
0.1

12

Rec 4

Rec 5

0.1

Rec6

0.1

Ara+] Rec 1

0.5

Rec2

0.1

12
12

12
12

Rec3

0.2

Rec 4
Rec 5

0.1

Table 19(cont'd).

0.1

Ara+2 Baseline

0.1

12

Reel

0.1

Rec2

0.1

12

Rec 3

0.2
0.2

0.1

Rec4

12

12
12

RecS

Rec6

0.1

Rec7

0.1

Ara+3 Baseline

0.1

12

12

Reel

0.1

Rec2

114

12 12 0.1
12

12

12

Rec3

0.1

Rec4

0.1

Recs

0.1

Rec 6

0.1

12
12

Rec7

0.2

Rec8

0.4
0. 1

Ara+4 Baseline

Recl

0.1

Rec2

0.1

12

Rec 3

0.1

12

Rec4

0.1

Rec5

12 0.1

12

Rec6

115

$0380nt
gig +8< Ba .5 no @3383 have gown—smog 808305 08:00 o+3< 5:85 ~ +02 23 6.02. swag: 70.2 N
.+>= #3 .+wu< 3.0 803 8.50:0» =< .30qu 0u0co€ohn§0 93 0:50:23 .«o coverage 0 Eu _ 030,—. 00m _

_d N N N_ N N c o. a m + + e 00M
_d _ N N N N .— m c m + + m 00¢
N6 2 N _ N N_ a .e. a m + + v 03—
N6 ~ N N_ N N_ u m ._ m + + n 005m
3 ~ N N N N_ c a .— o. - + N 8M
_d N_ N N_ N N a a a m - + _ 00M
Nd _ N N N _ a .e. u a. + + 03.08M— 0+5
3 _ N N N N_ a m c m + + N 8M
_d N_ N N N N c .e. m a. + + o 8M
_d N N N N _ c m m m + + m 00M
_.o _ N _ N _ a a. a m + + v 00M
to N_ N N N N c a. a a + + m 8M
_d N N N_ N N_ a a .. m - + N 00M
_d N N N N N_ u c c m - + _ 001m n+8<

.3880 2 easy

LIST OF REFERENCES

116

LIST OF REFERENCES

Antonovics, J. 1968. Evolution in closely adjacent plant populations. V. The evolution of
self-fertility. Heredity 23:219-23 8.

Atwood, K. C., L. K Schneider, and R J. Ryan. 1951. Periodic selection in Escherichia
coli. Proceedings of the National Academy of Sciences USA 37: 146-155.

Ayala, F. 1., and C. A. Campbell. 1974. Frequency-dependent selection Annual Review of
Ecology and Systematics 5:115-138.

Bennett, A F., R E. Lenski, and J. E. Mittler. 1992. Evolutionary adaptation to
temperature. 1. fitness responses of Escherichia coli to changes in its thermal
environment. Evolution 46: 16-30.

Birnboim, H C., and J. Doly. 1979. A rapid alkaline extraction procedure for screening
plasmid DNA. Nucleic Acids Research 7: 1513-1523.

Bisercic, M., J. Y. Feutrier, and P. R Reeves. 1991. Nucleotide sequences of the gut!
genes fi'om nine natural isolates of Escherichia coli: evidence of intragenic
recombination as a contributing factor in the evolution of the polymorphic gnd
locus. Journal of Bacteriology 173: 3894-3900.

Blythe, S. P., W. S. C. Gurney, P. Mass, and R M Nisbet. 1990. Program and model
building guide for SOLVER (Rev. 4). Applied Physics Industrial Consultants,
Glasgow.

Bouma, J. E., and R E. Lenski. 1988. Evolution of a bacteria/plasmid association Nature
335:351-352.

Bradshaw, AD. 1971. Plant evolution in extreme environments. Pp. 20-50 in E. R Creed,
ed. Ecological Genetics and Evolution. Appleton-Century-Crofi, New York.

Bull, I. I. 1994. Perspective: virulence. Evolution 48: 1423-1437.

Bull, .1. 1., I. J. Molineux, and W. R Rice. 1991. Selection of benevolence in a host-
parasite sytem Evolution 45:875-882.

Cain, A. J., and P. M. Sheppard. 1954. Natural selection in Cepaea. Genetics 39:89-116.

117

Carlton, B. C., and B. I. Brown. 1981. Gene mutation. Pp. 222-242 in P. Gerhardt, ed.
Manual of methods for general bacteriology. American Society for Microbiology,
Washington DC.

Caugant, D. A, B. R Levin, and R K. Selander. 1984. Genetic diversity and temporal
variation in an E. coli population of a human host. Genetics, 98:467-490.

Chao, L., and B. R Levin 1981. Structured habitats and the evolution of anticompetitor
toxins in bacteria Proceedings of the National Academy of Sciences USA 78:
6324-6328.

Chao, L., B. R Levin, and F. M. Stewart. 1977. A complex community in a simple
habitat: an experimental study with bacteria and phage. Ecology 58:369-378.

Clarke, B. 1964. Frequency-dependent selection for the dominance of rare polymorphic
genes. Evolution 18:364-369.

Crow, 1. F. and M Kimura. 1965. Evolution in sexual and asexual populations. American
Naturalist 99:439-450.

Dykhuizen, D. E. 1990. Experimental studies of natural selection in bacteria. Annual
Review of Ecology and Systematics 21:373-398.

Dykhuizen, D. E., and L. C. Green. 1991. Recombination in Escherichia coli and the
definition of biological species. Journal of Bacteriology 22:7257-7268.

Dykhuizen, D. E., and D. L. Hartl. 1983. Selection in chemostats. Microbiological
Reviews 47:150-168.

Ewald, P. W. 1983. Host-parasite relations, vectors, and the evolution of disease severity.
Annual Review of Ecology and Systematics 14:465-485.

Ewald, P. W. 1987. Transmission modes and evolution of the parasitism-mutualism
continuum. Annals of the New York Academy of Sciences 503:295-306.

Ewald, P. W. 1994. Evolution of infectious disease. Oxford University Press, New York,
New York

Fisher, R A 1958. The Genetical Theory of Natural Selection. Dover Press, New York.
2nd ed.

Ford, E. B. 1975. Ecological Genetics, 4th ed. Chapman and Hall, London.

118

Frank, S. A, and M Slatkin 1992. Fisher‘s firndamental theorem of natural selection.
Trends in Ecology and Evolution 7:92-95.

Freifelder, D. 1987. Microbial Genetics. Jones and Bartlett Publishers, Inc., Portola
Valley, CA

Gause, G. F. 1934. The Struggle for Existence. Dover Press, New York

Guttman, D. S., and D. E. Dykhuizen 19948. Clonal divergence in Escherichia coli as a
result of recombination, not mutation. Science 266: 1380-1383.

Guttman, D. S., and D. E. Dykhuizen. 1994b. Detecting selective sweeps in naturally
occurring Escherichia coli. Genetics 138:993-1003.

Graham, J. B., and C. A Istock. 1979. Gene exchange and natural selection cause
Bacillus subtilis to evolve in soil culture. Science 204:637-639.

Graham, J. B., and C. A Istock. 1981. Parasexuality and microevolution in experimental
populations of Bacillus subtilis. Evolution 35:954-963.

Haldane, J. B. S., and S. D. Jayakar. 1963. Polymorphism due to selection depending on
the composition of a population. Journal of Genetics 58:318-323.

Hansen, S. R and S. P. Hubbell. 1980. Single-nutrient microbial competition: qualitative
agreement between experimental and theoretically forecast outcomes. Science
207:1491-1493.

Hardin, G. 1960. The competitive exclusion principle. Science 131:1292-1297.

Hardy, K. G. 1986. Bacterial plasmids, 2nd ed. Van Nostrand Reinhold Co., Berkshire,
England.

Helling, R B., T. Kinney, and J. Adams. 1981. The maintenance of plasmid-containing
organisms in populations of Escherichia coli. Journal of General Microbiology
123: 129-141.

Helling, R B., C. N. Vargas, and J. Adams. 1987. Evolution of Escherichia coli during
grth in a constant environment. Genetics 116:349-358.

Herre, E. A 1993. Population structure and the evolution of virulence in nematode
parasites of fig wasps. Science 259: 1442-1445.

Hopwood, D. A, and K F. Chater (eds). 1989. Genetics of bacterial diversity. Academic
Press, London

119

Hori, M 1993. Frequency-dependent natural selection in the handedness of scale-eating
cichlid fish. Science 260:216-219.

Istock, C. A, K E. Duncan, N. Ferguson, and X Zhou. 1992. Sexuality in a natural
population of bacteria-Bacillus subtilis challenges the clonal paradigm. Molecular
Ecology 1:95-103.

Kinashi, H., H. Shimaji, and A Sakai. 1987. Giant linear plasmids in Streptonlyces which
code for antibiotic biosynthesis genes. Nature 328:454—456.

Kojirna, K 1971. Is there a constant fitness value for a given genotype? No! Evolution
25:281-285.

Korona, R, C. H. Nakatsu, L. J. Forney, and R E. Lenski. 1994. Evidence for multiple
adaptive peaks fiom populations of bacteria evolving in a structured habitat.
Proceedings of the National Academy of Sciences USA 912903 7-9041.

Lederberg, J. 1956. Conjugal pairing in E coli. Journal of Bacteriology 71:497-498.

Lenski, R E. 1992. Experimental evolution Pp. 125-140 in Encyclopedia of
ll/ficrobiology, volume 2. Academic Press Inc.

Lenski, R E. 1993. Assessing the genetic structure of microbial populations. Proceedings
of the National Academy of Sciences USA 90:4334—4336.

Lenski, R E., and J. E. Bouma. 1987. Effects of segregation and selection on instability of
plasmid pACYC184 in Escherichia coli B. Journal of Bacteriology 169:5314-
53 16.

Lenski, R E., and S. E. Hattingh. 1986. Coexistence of two competitors on one resource
and one inhibitor: a chemostat model based on bacteria and antibiotics. Journal of
Theoretical Biology 122:83-93.

Lenski, R E., and T. T. Nguyen 1988. Stability of recombinant DNA and its efi‘ects on
fitness. Pp. S18-20 in J. Hodgson and A. M Sugden, eds. Planned release of
genetically engineered organisms, (Trends in Biotechnology/Trends in Ecology and
Evolution Special Publication). Elsevier Publishers, Cambridge, England.

Lenski, R E., and R M. May. 1994. The evolution of virulence in parasites and
pathogens: reconciliation between two competing hypotheses. J oumal of
Theoretical Biology 169:253-265.

Lenski, R E., M R Rose, S. C. Simpson, and S. C. Tadler. 1991. Long-term
experimental evolution in Escherichia coli. L Adaptation and divergence during
2,000 generations. American Naturalist 138: 1315-1341.

120

Lenski, R E., and M Travisano. 1994. Dynamics of adaptation and diversification: a
10,000-generation experiment with bacterial populations. Procwdings of the
National Academy of Sciences USA 91:6808-6814.

Levin, B. R 1972. Coexistence of two asexual strains on a single resource. Science
175:1272-1274.

’ ""'Levin, B R 1980. Conditions for the existence of R-plasmids in bacterial populations. Pp.

197-202 in S. Mitsuhashi, L. Rosival, and V. Krcmery, eds. Antibiotic resistance:
transposition and other mechanisms. Springer-Verlag KG, Berlin.

2 Levin, 1B: R 19888. Frequency-dependent selection in bacterial populations. Philosophical
Transcripts of the Royal Society of London B 319:459-472.

Levin, B. R 1988b. The evolution of sex in bacteria. Pp. 194-211 in R E. Michod, and B.
R Levin, eds. The Evolution of Sex. Sinauer Associates, Sunderland, Mass.

Levin, B. R, and R E. Lenski. 1983. Coevolution in bacteria and their viruses and
plasmids. Pp. 99-127 in D. J. Futuyma and M. Slatkin, eds. Coevolution. Sinauer
Associates, Sunderland, Mass.

Levin, B. R, F. M. Stewart, and L. Chao. 1977. Resource-limited growth, competition,
and predation: a model and experimental studies with bacteria and bacteriophage.
American Naturalist 111:3-24.

Levin, B. R, F. M. Stewart, and V. A Rice. 1979. The kinetics of conjugative plasmid
transmission: fit of a simple mass action model. Plasmid 2:247-260.

Levin, S., and D. Pimentel. 1981. Selection of intermediate rates of increase in parasite-
host systems. American Naturalist 117:308-315.

Lloyd, R G., and C. Buckrnan 1995. Conjugational recombination in Escherichia coli:
genetic analysis of recombinant formation in Hfi' x F' crosses. Genetics 139:1123-
1 148.

Lundquist, P. E., and B. R Levin. 1986. Transitory derepression and the maintenance of
conjugative plasmids. Genetics 113:483-497.

Luria, S. E., and M Delbruck 1943. Mutations of bacteria fiom virus sensitivity to virus
resistance. Genetics 28:491-511.

MacArthur, R H, and E. O. Wilson. 1967 . The Theory of Island Biogeography.
Princeton Univ. Press, Princeton, NJ, USA

121

Macnair, M R, and Q. J. Cumbes. 1989. The genetic architecture of interspecific
variation ianulus. Genetics: 122:211-222.

May, R M, and R M Anderson 1983. Epidemiology and genetics in the coevolution of
parasites and hosts. Proceedings of the Royal Society of London 219:281-313.

Maynard Smith, J. 1990. The evolution of prokaryotes: does sex matter? Annual Review
of Ecology and Systematics 21: 1-12.

Maynard Smith, J., G. Dowson, and B. G. Spratt. 1991. Localized sex in bacteria Nature
349: 29-31.

Maynard Smith, J., N. H Smith, M. O'Rourke, and B. G. Spratt. 1993. How clonal are
bacteria? Proceedings of the National Academy of Sciences USA 90:43 84-4388.

McNeilly, T. 1968. Evolution in closely adjacent plant populations. III. Agrostis tennis on
a small copper mine. Heredity 23:99-108.

Milkman, R, and M. McKane-Bridges. 1990. Molecular evolution of Escherichia coli
chromosome. III. Clonal fiames. Genetics 126: 505-517.

Monod, J. 1949. The growth of bacterial cultures. Annual Review of Microbiology 3:371-
394.

Moser, H. 1958. The Dynamics of Bacterial Populations Maintained in the Chernostat.
Carnegie Inst., Washington DC.

Muller, H J. 1932. Some genetic aspects of sex. American Naturalist 68:118-138.

Neidhardt, F. C., J. L. Ingraharn, and M. Schaechter. 1990. Biosynthesis and fireling.
Chap.5 in Physiology of the Bacterial Cell. Sinauer Assoc, Sunderland, MA,
USA

Nguyen, T. N. M., Q. G. Phan, L. P. Duong, K. P. Bertrand, and R E. Lenski. 1989.
Efi‘ects of carriage and expression of the Tn10 tetracycline-resistance 0peron on
the fitness of Escherichia coli K12. Molecular Biology and Evolution 6:213-225.

Paquin, C. E., and J. Adams. 1983. Relative fitness can decrease in evolving asexual
populations of S. cerevisiae. Nature 306:368-371.

Pianka, E. R 1970. On r- and K-selection American Naturalist 1042592-597.
Pinero, D., E. Martinez, and R K. Selander. 1988. Genetic diversity and relationship

among isolates of Rhizobium leguminosarurn biovar phaseoli. Applied and
Environmental Microbiology 54:2825-2832.

122

Rice, W. R 1989. Analyzing tables of statistical tests. Evolution 43:223-225.

Rosenzweig, R F., R R Sharp, D. S. Treves, and J. Adams. 1994. Microbial evolution in
a simple unstructured environment: genetic difi‘erentiation in Escherichia coli.
Genetics 137 :903-917.

Roughgarden, J. 1979. Theory of population genetics and evolutionary ecology: an
introduction. Macmillan, New York

Sambrook, J., E. F. Fritsch, and T. Maniatis. 1989. Molecular cloning: a laboratory
manual. 2nd ed. Cold Spring Harbor Laboratory Press, Plainview, New York

Sanno, Y., T. H. Wilson, and E. C. C. Lin 1968. Control of permeation to glycerol in
cells of Escherichia coli. Biochemical and Biophysical Research Communications
32:344-349.

Selander, R K., and B. R Levin. 1980. Genetic diversity and structure in Escherichia coli
populations. Science 210:545-547.

Simonsen, L., D. M. Gordon, F. M Stewart, and B. R Levin 1990. Estimating the rate of
plasmid transfer: an end-point method. Journal of General Microbiology 136:2319-
2325.

Sokal, R R, and F. J. Rohlf. 1981. Biometry. W.H Freeman and Co., New York

Souza, V., T. T. Nguyen, R R Hudson, D. Pinero, and R E. Lenski. 1992. Hierarchical
analysis of linkage disequilibrium in Rhizobium populations. Procwdings of the
National Academy of Sciences USA 89: 83 89-8393

Stewart, F. M., and B. R Levin 1973. Partitioning of resources and the outcome of
interspecific competition: a model and some general considerations. American
Naturalist 107: 171-198.

Stewart, F. M, and B. R Levin 1977. The population biology of bacterial plasmids: a
priori conditions for the existence of conjugationally transmitted factors. Genetics
87:209-228.

Stolp, H 1988. Microbial metabolism and regulation. Pp. 24-46 in Microbial Ecology:
Organisms, Habitats, Activities. Cambridge Univ. Press, Cambridge.

Sweet, 6., C. Gandor, R Voegele, N. Wittekindt, J. Beuerle, V. Truniger, E. C. C. Lin,
and W. Boos. 1990. Glycerol facilitator of Escherichia coli: cloning of glpF and
identification of the glpF product. Journal of Bacteriology 172:424-430.

123

Tilman, D. 1982. Resource Competition and Community Structure. Princeton Univ. Press,
Princeton, NJ.

Travisano, M., J. A Mongold, A F. Bennett, and R E. Lenski 1995. Experimental tests
of the roles of adaptation, chance, and history in evolution Science 267:87-90.

Travisano, M., F. Vasi, and R E. Lenski. 1995. Long-term experimental evolution in
Escherichia coli. III. Variation among replicate populations in correlated
responses to novel environments. Evolution 49:189-200.

Van Delden, W., A C. Boerema, and A Kamping. 1978. The alcohol dehydrogenase
polymorphism in populations of Drosophila melanogaster. 1. Selection in difi‘erent
environments. Genetics 90: 161-191.

Vasi, F., M. Travisano, and R E. Lenski. 1994. Long-term experimental evolution in
Escherichia coli. 11. Changes in life history traits during adaptation to a seasonal
environment. American Naturalist 144:432-456.

Whittam, T. 8., and S. E. Ake. 1993. Genetic polymorphisms and recombination in natural
populations of Escherichia coli. Pp. 223-245 in N. 'Takahata and AG. Clark, eds.
Molecular paleopopulation biology. Japan Scientific Society Press.

Whittam, T. S., H. O. Ochman, and R K. Selander. 1983. Multilocus genetic structure in
natural populations of Escherichia coli. Proceedings of the National Academy of
Sciences USA 80: 1751-1755.

Wright, S. 1931. Evolution in Mendelian populations. Genetics 16:97-159.

Wright, S. 1932. The roles of mutation, inbreeding, crossbreeding and selection in
evolution Proceedings VIth International Congress of Genetics 1:3 56-3 66.

Wright, S., and T. Dobzhansky. 1946. Genetics of natural populations. XII Experimental
reproduction of some of the changes caused by natural selection in certain
populations of Drosophila pseudoobscura. Genetics 31:125-156.

"‘11111111111111ES