ArticlePDF Available

Ecological Specialization and Adaptive Decay in Digital Organisms


Abstract and Figures

The transition from generalist to specialist may entail the loss of unused traits or abilities, resulting in narrow niche breadth. Here we examine the process of specialization in digital organisms--self-replicating computer programs that mutate, adapt, and evolve. Digital organisms obtain energy by performing computations with numbers they input from their environment. We examined the evolutionary trajectory of generalist organisms in an ecologically narrow environment, where only a single computation yielded energy. We determined the extent to which improvements in this one function were associated with losses of other functions, leading to organisms that were highly specialized to perform only one or a few functions. Our results show that as organisms evolved improved performance of the selected function, they often lost the ability to perform other computations, and these losses resulted most often from the accumulation of neutral and deleterious mutations. Beneficial mutations, although relatively rare, were disproportionately likely to cause losses of function, indicating that antagonistic pleiotropy contributed significantly to niche breadth reductions in this system. Occasionally, unused functions were not lost and even increased in performance. Here we find that understanding how the functions were integrated into the genome was crucial to predictions of their maintenance.
Content may be subject to copyright.
Ecological Specialization and Adaptive Decay in Digital Organisms.
Author(s): ElizabethA. Ostrowski, Charles Ofria, RichardE. Lenski
The American Naturalist,
Vol. 169, No. 1 (January 2007), pp. E1-E20
Published by: The University of Chicago Press for The American Society of Naturalists
Stable URL: .
Accessed: 28/02/2011 15:09
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact
The University of Chicago Press and The American Society of Naturalists are collaborating with JSTOR to
digitize, preserve and extend access to The American Naturalist.
vol. 169, no. 1 the american naturalist january 2007
Ecological Specialization and Adaptive Decay
in Digital Organisms
Elizabeth A. Ostrowski,
Charles Ofria,
and Richard E. Lenski
1. Ecology, Evolutionary Biology, and Behavior Program, Michigan
State University, East Lansing, Michigan 48824;
2. Department of Computer Science and Engineering, Michigan
State University, East Lansing, Michigan 48824
Submitted August 31, 2005; Accepted June 23, 2006;
Electronically published December 4, 2006
abstract: The transition from generalist to specialist may entail
the loss of unused traits or abilities, resulting in narrow niche
breadth. Here we examine the process of specialization in digital
organisms—self-replicating computer programs that mutate,
adapt, and evolve. Digital organisms obtain energy by performing
computations with numbers they input from their environment.
We examined the evolutionary trajectory of generalist organisms
in an ecologically narrow environment, where only a single com-
putation yielded energy. We determined the extent to which im-
provements in this one function were associated with losses of
other functions, leading to organisms that were highly specialized
to perform only one or a few functions. Our results show that
as organisms evolved improved performance of the selected func-
tion, they often lost the ability to perform other computations,
and these losses resulted most often from the accumulation of
neutral and deleterious mutations. Beneficial mutations, although
relatively rare, were disproportionately likely to cause losses of
function, indicating that antagonistic pleiotropy contributed sig-
nificantly to niche breadth reductions in this system. Occasion-
ally, unused functions were not lost and even increased in per-
formance. Here we find that understanding how the functions
were integrated into the genome was crucial to predictions of
their maintenance.
Keywords: adaptation, digital evolution, mutation accumulation, plei-
otropy, regressive evolution, specialization.
* Corresponding author. Present address: Department of Ecology and Evo-
lutionary Biology, Rice University, Houston, Texas 77005; e-mail: ostrowski@
Am. Nat. 2007. Vol. 169, pp. E1–E20. 2007 by The University of Chicago.
0003-0147/2007/16901-41271$15.00. All rights reserved.
Many theories about the origins and maintenance of bi-
ological diversity involve specialization and adaptive decay.
Specialization describes the process by which organisms
become highly adapted to a narrow range of environ-
mental conditions and may be associated with adaptive
decay—the loss of other traits, functions, or abilities that
results in the evolution of narrow niche breadth. A ten-
dency toward increased specialization is a defining feature
of adaptive radiations, as it forms the underpinnings for
niche partitioning and character displacement, which pro-
mote diversification (Simpson 1953; Schluter 2000).
The process of specialization can result in the loss of
other functions in environments in which they are no
longer useful, termed “adaptive decay.” For example, the
transition from a free-living to a parasitic lifestyle is
thought to involve not only adaptations that enable host
exploitation but also extensive decay of other unused func-
tions presumably necessary for survival outside the host,
with parasites showing reduced or streamlined genomes
relative to their free-living relatives (Andersson et al. 1998;
Shigenobu et al. 2000; Cole et al. 2001; Ochman and
Moran 2001).
Both specialization and adaptive decay have been doc-
umented in natural populations, but the underlying ge-
netic mechanism remains unclear. Some have hypothe-
sized that there are trade-offs, such that adaptation to one
environment inevitably results in loss of adaptation to oth-
ers (antagonistic pleiotropy hypothesis). Trade-offs may
result from an energetic burden associated with maintain-
ing or expressing unused functions or because improve-
ments to a selected trait directly interfere with the func-
tioning of an unselected trait. An alternative hypothesis is
that the loss of specialized features results from relaxed
selection, enabling mutations to accumulate in the por-
tions of the genome that encode unused functions (mu-
tation accumulation hypothesis). Which of these mecha-
nisms predominates is important, insofar as they lead to
different expectations as to the frequency of specialization
and the types of circumstances that promote it. For in-
stance, mutation accumulation requires that the genes that
contribute to increased adaptation in alternative environ-
E2 The American Naturalist
ments be distinct and that the environment be hetero-
geneous (in time or space), so as to give rise to the periods
of relaxed selection that enable mutations to accumulate.
Alternatively, antagonistic pleiotropy results in constraints
that prevent organisms from being simultaneously adapted
to many niches and does not require environmental het-
erogeneity, although it may be aided by it. To see why this
is true, imagine a generalist organism that is adapted to
use both resource A and resource B. If at some point all
further improvements on resource A cause a fitness re-
duction on resource B and vice versa, then specialization
can arise in the absence of any change in the environment.
By contrast, mutation accumulation requires that selection
be relaxed and thus implies that something previously un-
der selection no longer is; for example, some resource may
no longer be available. If fitness levels in two environments
are considered to be two different traits (Falconer 1952),
then antagonistic pleiotropy occurs when the same mu-
tation improves fitness in one environment and reduces
it in another. By contrast, in the case of mutation accu-
mulation, the mutations that increase fitness in the novel
environment and reduce it in the ancestral environment
are distinct (Cooper et al. 2001). For this reason, if an-
tagonistic pleiotropy is common, then the process of spe-
cialization will be closely tied to that of adaptive decay.
Despite long-standing interest in these two hypotheses,
it has been difficult to distinguish between them. A large
body of work on adaptive decay (also called regressive
evolution) has focused on cave organisms (Jones et al.
1992; Jernigan et al. 1994; Jeffrey et al. 2003). These or-
ganisms often exhibit highly convergent and distinctive
phenotypes, characterized by loss of pigmentation and re-
duced visual systems but with other sensory structures
being highly developed, such as antennae. Mutation ac-
cumulation hypotheses suggest that the lack of light in
caves resulted in relaxed selection and the accumulation
of mutations that eventually led to the losses of pigmen-
tation and visual sensory structures. Alternatively, antag-
onistic pleiotropy hypotheses posit that adaptation to low
light levels led to highly specialized sensory structures and
that the losses of other traits were a direct result of this
adaptation, possibly because the maintenance of unused
functions was costly. For example, Darwin (1859) hy-
pothesized that eyes are costly to burrowing rodents be-
cause they are prone to infection and thus that their evo-
lutionary loss may have been aided by natural selection.
Although the extent to which regressive phenotypes reflect
mutation accumulation or antagonistic pleiotropy has
been a subject of great debate, a recent study of cave fish
demonstrated linkage for quantitative trait loci (QTL) as-
sociated with both a regressive (eye size) and a constructive
(body weight) trait, suggesting that either antagonistic
pleiotropy or hitchhiking was responsible (Borowsky and
Wilkens 2002). In general, increased knowledge of the ge-
netic basis of traits, as well as their evolutionary dynamics,
is expected to shed light on the processes of mutation
accumulation and antagonistic pleiotropy.
Experimental approaches provide an alternative to com-
parative approaches, allowing direct examination of the
process of specialization and adaptive decay. A recent study
of evolving populations of Escherichia coli examined the
consequences of long-term adaptation to a simple envi-
ronment for the evolution of catabolic niche breadth (Coo-
per and Lenski 2000). Replicate populations of E. coli were
propagated for 20,000 generations in a medium containing
only a single available carbon source, glucose. While the
evolved populations were found to have increased ability
to compete for and catabolize glucose relative to their
ancestor, they also consistently evolved reduced ability to
catabolize other resources. Moreover, the identities of these
carbon sources were similar across independently evolved
populations. This parallelism suggested that the losses re-
sulted from antagonistic pleiotropy, that is, that the re-
duction in diet breadth had traded off with the improved
ability to use glucose. Populations that evolved elevated
mutation rates during the course of the experiment also
did not show significantly greater losses, contrary to the
expectations under mutation accumulation, further indi-
cating that antagonistic pleiotropy was the primary mech-
Here we address the process of specialization in a very
different medium—an evolving system composed of self-
replicating computer programs that mutate, compete, and
evolve in a computational environment. We examine the
digital equivalent of diet breadth—the ability of these or-
ganisms to perform complex computations that enable
them to garner energy from their environment. In this
system, we can not only observe the pattern of evolution
associated with specialization and adaptive decay, but we
can also examine in detail the underlying genetic processes;
that is, we can identify the specific mutations that result
in losses of function and determine their fitness effects.
We use this knowledge to distinguish between antagonistic
pleiotropy and mutation accumulation by asking whether
losses of function were the result of neutral or beneficial
mutations. Whereas losses that result from mutations that
are neutral in the selective environment constitute ex-
amples of mutation accumulation, those that result from
mutations that are beneficial in the selective environment
constitute examples of antagonistic pleiotropy.
Below, we give a brief introduction to the digital life
system, Avida. Additional information is provided in ap-
pendixes A and B, including a schematic of a digital or-
ganism and a glossary of terms. A more detailed description
of the system is available elsewhere (Wilke and Adami 2002;
Specialization in Digital Organisms E3
Lenski et al. 2003; Ofria and Wilke 2004), and documen-
tation is available online (
The Avida System
The Avida system is a software platform wherein self-
replicating computer programs (digital organisms) adapt
and evolve in a computational environment. Each digital
organism has a genome composed of a series of computer
instructions, which, by default, are executed sequentially
by a virtual central processing unit (CPU). However, some
instructions permit jumps or loops; for example, repli-
cation generally involves the execution of a copy loop.
Execution of a viable genome results in an organism copy-
ing itself, instruction by instruction, and on completion,
dividing by binary fission to produce two organisms. If
no empty space is available in the population, replication
results in the replacement of another organism in the pop-
ulation. Thus, the faster a given organism produces off-
spring, the more likely its genotype is to persist and spread
in the population over time.
Evolution occurs because the copy process is subject to
random mutations, at a rate specified by the experimenter.
Mutations can be point mutations, whereby one instruc-
tion in the genome is randomly replaced with another, or
they can be insertions or deletions, which enable the ge-
nome to grow or shrink in length. Mutations are normally
deleterious because they reduce the speed or efficiency with
which an organism replicates; in the extreme, they are
effectively lethal if they prevent an organism from being
able to replicate altogether. Mutations that are beneficial
increase the replication rate of the organism, either by
improving the efficiency with which it produces copies of
itself or else by enabling it to receive additional CPU cycles,
which allow it to execute more instructions. CPU cycles
can thus be thought of as energy in Avida: every instruction
executed burns a CPU cycle, but organisms must execute
instructions in order to replicate or perform other func-
tions. Digital organisms will thus generally adapt in one
of two ways. First, they may evolve to reduce the number
of CPU cycles they require to produce an offspring, that
is, to reduce their generation time. Alternatively, they may
evolve to obtain more CPU cycles (more energy), which
may allow them to produce more offspring.
Digital organisms can obtain additional CPU cycles by
performing bitwise logic functions on numbers they input
from their environment. A correct computation provides
an organism with CPU cycles above its initial allotment,
which can then be put toward further execution of the
genomic instructions, potentially resulting in an increased
rate of replication. In our experiments, organisms receive
an initial allotment of CPU cycles equal to their genome
length. This makes evolution neutral with respect to ge-
nome size and prevents shorter organisms from having an
inherent fitness advantage (for more details, see Lenski et
al. 2003; Ofria et al. 2003). Thus, whether a given organism
replicates faster depends only on whether the CPU cycles
obtained offset the additional CPU cycles required to per-
form the computation. Organisms therefore evolve not
only to perform computations but also to perform them
as efficiently as possible.
The performance of computations represents a kind of
a metabolism, in that the conversion of one or two num-
bers into another number provides the organism with en-
ergy. In Avida, the genomic instruction set includes one
that is called nand; this instruction can allow digital or-
ganisms to perform the NAND (not and) logic function,
provided that the instruction is properly coupled to input-
output instructions to obtain the numbers and output the
result. All other computations can be constructed by com-
bining multiple nand instructions with various other in-
structions. In this way, digital organisms also resemble real
computers, in that all computations performed by com-
puters can be built out of combinations of NAND func-
tions (sometimes referred to as nand gates).
The Equals Function
For the current study, we employed nine possible logic
functions, eight of which require two inputs, that is, two
binary numbers input from the environment. These nine
bitwise logic functions are as follows: NOT, NAND, AND,
putation of these functions has been described elsewhere
(Lenski et al. 2003), but for purposes of illustration, we
describe in greater detail one of these functions, equals
(EQU), which is the focus of the current study. EQU is a
computation where, for any two inputs, the correct output
contains a 1 (true) at every site where the bits are identical
and a 0 (false) at every site where the bits are not identical.
For example,
Input A: 010110111001
Input B: 001101011101
Thus, in an environment where EQU is rewarded, an
organism that inputs A and B and outputs the above num-
ber would receive additional CPU cycles. Because most
computations require the coordination of multiple steps,
digital organisms must store and manipulate intermediates
or partial results. For example, the performance of EQU
requires combining a minimum of five NAND functions
and at least 19 instructions in total (Lenski et al. 2003).
Finally, CPU cycle rewards are determined simply by com-
paring an organism’s inputs with its output, such that
selection is based on the phenotype, not the genotype.
E4 The American Naturalist
Figure 1: The evolution experiment had two phases: an initial period when replicate populations evolved in a complex environment with all
functions rewarded, followed by a period of evolution in a specialized environment with only EQU rewarded. Evolved organisms represent the final
dominant organism in each evolved population. Numbers shown beneath each organism specify its phenotype in terms of the number of times it
performs each function during its life cycle, in the following order: NOT, NAND, AND, OR-NOT, OR, AND-NOT, NOR, XOR, and EQU. For
clarity, the number that corresponds to an organism’s performance of EQU has been shaded.
In a recent article, Lenski et al. (2003) examined the
evolutionary origins of the EQU computation from an
ancestral organism that could perform no functions. They
found that the EQU function has the properties of a com-
plex feature: its performance required the coordinated ex-
ecution of numerous interacting parts. Moreover, its evo-
lutionary emergence required that other, simpler functions
also be rewarded; these simpler functions could then serve
as building blocks for such a complex function. Here we
expand on this work by examining specialization of the
EQU function. Starting from generalist organisms (those
that could perform a variety of computations, in addition
to EQU), we examine their evolution in a narrow envi-
ronment, where only EQU generates extra CPU cycles. We
examine how these digital organisms adapt to their novel
environment, the extent to which they evolve to be highly
specialized, and the evolutionary processes that govern
their transition from generalists to specialists. Finally, we
describe an unexpected result: that some unrewarded func-
tions were not lost and even increased in performance,
despite evolution in an ecologically narrow environment.
Here, we take advantage of the remarkable transparency
of digital systems, as well as the ability to probe the un-
derlying genetic architecture of the traits in question, to
develop a general understanding of the factors that drive
the balance between niche breadth reduction and conser-
vatism in this system.
Experimental Design
Two-Stage Evolution Experiment. In the first stage of the
experiment, replicate populations evolved from a single
handwritten ancestor that could self-replicate but that
could not perform any logic functions (fig. 1). These pop-
ulations evolved in an environment where the perfor-
mance of any of the nine computations provided CPU
cycles as rewards. These rewards were limited, however,
to once per gestation cycle, such that organisms generally
evolved to perform each function only once. (The gesta-
tion cycle is defined as the time from when an organism
Specialization in Digital Organisms E5
executes the first instruction in its genome to when it
produces an offspring.) Following 100,000 updates of evo-
lution, an arbitrary unit of time in Avida corresponding
to an average of 30 instructions executed per organism
(see glossary, table B1), the dominant genotype was iso-
lated from each population. These genotypes served as the
generalist ancestors (subsequently denoted Ancestor1–
Ancestor10) in the main experiment.
In the second stage of the experiment, replicate popu-
lations were founded from each generalist ancestor and
evolved in an environment where only EQU yielded extra
CPU cycles. The experiment consisted of 10 replicate pop-
ulations for each of the 10 ancestors, for a total of 100
populations. The ancestors were generalists in that they
could perform a wide variety of different logic functions,
though they differed in the number and identity of the
exact functions they performed ( ,average p 7.3 range p
–9 of nine possible logic functions). All ancestors, how-6
ever, performed EQU exactly once per gestation cycle. The
ancestors also varied in the number of instructions com-
prising their genomes, with the shortest having 59 in-
structions and the longest having 124 instructions
( ). All populations evolved for 100,000 up-average p 99.7
dates, during which time they received a reward only for
the EQU computation. In this new EQU-only environ-
ment, however, organisms received rewards every time
they performed the EQU computation and output the
appropriate result. Insertion, deletion, and point muta-
tions occurred at rates of 0.01, 0.01, and 0.08 mutations
per genome per replication, respectively. Population size
was limited to a maximum of 3,600 organisms, and off-
spring were placed randomly in the population, such that
the population was well mixed.
Examining the Line of Descent. To assess specialization and
adaptive decay following evolution in the EQU-only en-
vironment, the most abundant genotype from each pop-
ulation was saved and assayed at the end of each exper-
iment for its ability to perform each of the nine
computations, including EQU. For each of these geno-
types, we also determined its line of descent, which is the
sequence of all genotypes leading back to the original an-
cestor. By looking along the line of descent, we identified
pivotal genotypes where mutations arose that produced a
loss of function. We then classified each mutation ac-
cording to its fitness effect relative to the immediately
preceding parent genotype:
11 was beneficial, 1 was neu-
tral, and
!1 was deleterious. For the purposes of distin-
guishing between antagonistic pleiotropy and mutation
accumulation, we henceforth lump deleterious mutations
with neutral ones and refer to them collectively as non-
beneficial. The reason for doing so is that the antagonistic
pleiotropy hypothesis specifically concerns beneficial mu-
tations, whereas mutation accumulation could encompass
not only neutral mutations but also deleterious mutations
that drift or hitchhike to fixation.
Generally, organisms along the line of descent differed
from their immediate predecessors by a single mutation.
Occasionally, however, they differed by two or more mu-
tations. By default, we classified multiple mutations ac-
cording to their fitness effect in combination. However,
to ensure that these multiple mutational steps did not
influence our results, we also analyzed our data without
these multiple mutations. We also performed several ad-
ditional sets of experiments to determine how our results
were affected by our choice of particular parameter values.
First, we repeated our experiments at higher and lower
genomic mutation rates of 0.3 and 0.01, respectively, equal
to threefold higher and 10-fold lower than those in our
original experiments. To control for the effects of differ-
ential mutation supply, we performed additional experi-
ments where we scaled the length of the experiments in-
versely to the mutation rate. Thus, experiments at a
genomic mutation rate of 0.3 were run for 33,000 updates,
and those at the 0.01 genomic mutation rate were run for
1,000,000 updates. Second, to determine whether the pres-
ence of energetic constraints on genome size could alter
the balance between antagonistic pleiotropy and mutation
accumulation, we repeated our experiments in an envi-
ronment where there was strong selection to reduce ge-
nome length. To do so, we changed the baseline allotment
of an organism’s CPU cycles to be a constant (in this case,
equal to 100) rather than proportional to its genome
length. Under these conditions, organisms face strong se-
lection to reduce their genome size to a minimum. Al-
though there is little direct evidence that energetic con-
straints on genome size are common in biological
organisms (Petrov and Hartl 2000; Sliwa and Korona
2005), the reduced genomes of some intracellular bacteria
(Cole et al. 2001; Ochman and Moran 2001; Go´ mez-Valero
et al. 2004) nevertheless suggest the potential importance
of such contraints in the evolution of ecological spe-
Statistical Analyses
Performance of the EQU Function. To determine how fit-
ness and the performance of EQU varied depending on
the ancestor, we performed two one-way ANOVAs. These
analyses were performed using PROC GLM in SAS, with
ancestor designated as a random effect. In the first
ANOVA, we used the log relative fitness of evolved pop-
ulations as the response variable, where each evolved pop-
ulation’s fitness is relative to that of its own ancestor. In
the second ANOVA, the response variable was the number
of times EQU was performed in the numerically dominant
E6 The American Naturalist
Figure 2: Average number of times EQU is performed per life cycle by
evolved organisms, arranged from highest to lowest for each of the 10
ancestors. The ancestors all performed EQU only once, and each bar
represents the mean across 10 replicate populations evolved from that
ancestor. Values are plotted on a logarithmic scale, and error bars rep-
resent 1 SE. For clarity, the average value is also written above each bar.
genotype isolated from each evolved population. Because
variances were heterogeneous across ancestors, we per-
formed the ANOVAs as nonparametric Kruskal-Wallis
tests. These tests were performed in SAS using PROC
Antagonistic Pleiotropy versus Mutation Accumulation. To
examine the relative contributions of antagonistic pleiot-
ropy and mutation accumulation, we totaled the number
of beneficial and nonbeneficial mutations per ancestor
across the 10 replicate experiments at those steps where
some unused function was lost. Because nonbeneficial mu-
tations are typically more common than beneficial mu-
tations, even in the line of descent (Lenski et al. 2003),
we also assembled a baseline calculation of the relative
proportion of these two mutation types over the course
of evolution by totaling their numbers over the line of
descent as a whole, irrespective of whether they were as-
sociated with a loss of function. To determine whether the
ratio of beneficial to nonbeneficial mutations was signif-
icantly higher among those mutations that caused a loss
of function (which would provide support for the antag-
onistic pleiotropy hypothesis), we performed a Fisher’s
exact test that compared the number of beneficial versus
nonbeneficial mutations causing a loss of function with
the number that did not. To assess the statistical signifi-
cance of the contingency tables, we used the right-tailed
P value of a Fisher’s exact test, where a low P value would
indicate that beneficial mutations were significantly over-
represented among mutations causing losses of function.
The analyses were performed in SAS, using PROC FREQ
and the Fisher Exact option.
Specialization and Adaptive Decay in the
EQU-Only Environment
We consider three components of specialization. First, we
examine the extent to which populations evolve increased
performance of EQU, where the performance is deter-
mined as the total number of times an organism outputs
the result of the EQU function per reproductive cycle.
Second, because organisms can make improvements in the
efficiency of their EQU performance without increasing
the number of times it is performed, we also consider the
degree to which fitness increased in the EQU-only envi-
ronment. Third, we examine the extent of adaptive decay,
that is, the extent to which unrewarded functions were
lost during evolution in the EQU-only environment, lead-
ing to the evolution of narrow niche breadth.
With regard to the first of these criteria, we find that
evolved populations had greatly improved performances
of EQU. Whereas all ancestors performed EQU only once
per reproductive cycle, most evolved organisms performed
it tens or even hundreds of times (fig. 2). Interestingly,
the magnitude of this improvement depended strongly on
the ancestor (Kruskal-Wallis , ,
x p 42.41 df p 9 P !
). There was also variation in the performance of
EQU among replicate populations evolved from the same
ancestor. For example, in five of 100 populations (three
derived from Ancestor1 and one each from Ancestor9 and
Ancestor10), the performance of EQU did not increase
above the ancestral level. However, when averaged over
the 10 replicate populations, organisms evolved from An-
cestor1 had the third highest performance of EQU overall
(fig. 2). Although EQU performance did not increase in
these five populations, at least one beneficial mutation
fixed in every population, indicating that some adaptation
to the EQU-only environment occurred in all populations.
In these five populations, fitness improved by reductions
to generation time, although the fitness increases in these
populations were small compared with populations that
evolved increased performance of the EQU function. Such
variation in the extent of fitness improvement among rep-
licate populations evolved from the same ancestor indi-
cates that the chance occurrence of different mutations in
replicate populations was an important component of spe-
cialization in this system. Similarly, because the generalist
ancestors themselves evolved from the same handwritten
ancestor (fig. 1), differences in outcome that were contin-
Specialization in Digital Organisms E7
Figure 3: A, Fitness trajectory of populations in the EQU-only environment. Each line represents the average of 10 replicate evolution experiments
for each of 10 different generalist ancestors. Fitness is expressed as the log ratio of values for evolved organisms relative to their own ancestor, such
that all populations start at 0. B, Reduction in niche breadth during evolution in the EQU-only environment. Niche breadth was calculated as the
proportion of organisms performing each function at a given time and summed over all functions. Each line represents the average of 10 replicates
for each of the 10 ancestors.
gent on the generalist ancestor also demonstrate the im-
portance of chance events at an earlier stage of evolution.
While EQU performance did not increase in every pop-
ulation, fitness universally improved in the EQU-only en-
vironment (fig. 3A). Once again, the magnitude of the
increase depended greatly on the ancestor. Figure 3A shows
the fitness trajectory of the populations over time, averaged
over the 10 replicates for each ancestor. While all popu-
lations increased in fitness, there was substantial hetero-
geneity in the magnitude of this improvement, and again
the ancestor had a highly significant effect (Kruskal-Wallis
,, ).
x p 32.45 df p 9 P p .0002
Evolution of Niche Breadth Reductions
The loss of unrewarded functions was not universal and
also varied across ancestors. Retention of a particular func-
tion is indicated by a black cell in figure 4. Only seven of
100 populations retained only EQU and lost all unused
functions; these seven populations were distributed across
six different ancestors (fig. 4). As expected, there was a
significant correlation between EQU performance and fit-
ness in the EQU-only environment (Spearman’s r p
, ). However, there was no overall relation-0.81 P p .005
ship between increased EQU performance and the mag-
nitude of niche breadth reductions (Spearman’s r p
, ). For example, populations evolved from0.33 P p .35
Ancestor3 tended to maintain a relatively broad niche (fig.
4) and yet were the second-highest performers of EQU
(fig. 2). The decline in niche breadth over time is plotted
in figure 3B. Each line represents the average for a different
ancestor, and the colors for each population are the same
as those used for the fitness trajectories in figure 3A.
Population Genetic Processes Underlying the Evolution
of Reduced Niche Breadth
Where functions were lost, we were interested in deter-
mining whether losses were caused by mutation accu-
mulation or antagonistic pleiotropy. To address this ques-
tion, we determined the fitness effect of every mutation
that resulted in a loss of function along the line of descent.
If mutations causing functions to be lost were neutral or
deleterious in the EQU-only environment, it would in-
dicate that mutation accumulation was responsible for
losses of function. Similarly, if the mutations leading to
losses of function were beneficial, it would indicate that
antagonistic pleiotropy was responsible. Note that there
Figure 4: Outcome of evolution in the specialized EQU-only environment for all 100 populations, arranged by ancestor. Each row specifies the
final dominant genotype from one evolved population, and the colors indicate which functions were lost and the type of mutation (beneficial,
neutral, or deleterious) that caused the loss of function. In some cases, a single mutation led to the loss of multiple functions at once, such that
the colors of the blocks are not necessarily independent in every row.
Specialization in Digital Organisms E9
Table 1: Comparison of mutations associated with losses of func-
tion relative to their overall proportion along the line of descent
or not
Number of
PLoss Not Loss Not
Anc1 Beneficial 12 301 .308 .176 .034
Not 27 1,407
Anc2 Beneficial 15 592 .306 .424 .966
Not 34 803
Anc3 Beneficial 20 496 .714 .370
Not 8 844
Anc4 Beneficial 19 233 .396 .261 .032
Not 29 661
Anc5 Beneficial 22 303 .386 .205 .002
Not 35 1,178
Anc6 Beneficial 14 268 .326 .183 .020
Not 29 1,195
Anc7 Beneficial 10 282 .200 .208 .610
Not 40 1,076
Anc8 Beneficial 28 201 .431 .172
Not 37 968
Anc9 Beneficial 10 204 .213 .168 .265
Not 37 1,010
Anc10 Beneficial 17 394 .386 .232 .017
Not 27 1,302
Note: Results of Fisher’s exact tests comparing the loss of functions due to
beneficial versus nonbeneficial (not; neutral or deleterious) mutations. The
left-hand side shows the contingency table for each of the 10 ancestors. In
each case, the number of mutations was summed over 10 replicate popula-
tions. Loss and not categories refer to the number of mutations that were
associated with a loss of function or not, respectively. P
indicates the pro-
portion of mutations that were beneficial. P is the probability associated with
the right tail of a Fisher’s exact test—in other words, the probability of seeing,
by chance alone, as much or more overrepresentation of beneficial mutations
among loss-of-function mutations.
were at least two ways in which a mutation causing a loss
of function could be beneficial. First, it may be beneficial
because the instructions that encode EQU also encode
other functions, such that mutations that enhance EQU
performance interfere with these other functions; this
would constitute a classic example of pleiotropy. Second,
mutations causing losses of function could also be bene-
ficial because they reduce the energy spent performing
useless tasks, thereby increasing fitness in the EQU-only
environment. This also constitutes antagonistic pleiotropy,
in the sense that a single mutation improves fitness in one
environment but reduces it in another. For our purposes,
we did not distinguish between these two explanations: all
mutations that were simultaneously beneficial in the EQU-
only environment and resulted in a loss of function were
interpreted as support for the pleiotropy hypothesis. Our
classification scheme thus captures two categories of ex-
planation: those mutations that fix via selection for im-
proved performance in an EQU-only environment and
those mutations that fix by genetic drift or by hitchhiking
alongside beneficial mutations.
The mutations leading to losses of function are shown
for all populations in figure 4, arranged by ancestor. Be-
cause a single mutation occasionally led to the simulta-
neous loss of multiple functions, cells are not necessarily
independent of one another. In addition, because we are
interested in understanding the niche breadth of the final
derived organisms and the mutations that led to that state,
we do not consider cases where a function was lost but
subsequently regained. Thus, we examine only the mu-
tations causing losses of function if the function was absent
at the end of the experiment. In cases where a function
was lost, regained, and subsequently lost again, we consider
only the final loss of function. This methodology is most
likely conservative with respect to detecting antagonistic
pleiotropy, as earlier mutations (when adaptation is most
rapid) are more likely to be beneficial than later mutations.
For eight of 10 ancestors, the beneficial to nonbeneficial
ratio was higher among mutations causing losses of func-
tion than among those that did not. In seven of these eight
cases, the Fisher’s exact test was highly significant (table
1). In the two cases where the mutations causing losses of
function were predominantly neutral or deleterious (An-
cestor2 and Ancestor7), the differences were quite small.
In these cases, the left-hand P values of the Fisher’s exact
tests, which would test for overrepresentation of neutral
or deleterious mutations among mutations causing losses
of function, were nonsignificant ( and .53 for An-P p .07
cestor2 and Ancestor7, respectively).
To determine overall support for the antagonistic plei-
otropy hypothesis, we combined the P values from the 10
separate tests into a single P value, using a method de-
veloped by Fisher (Sokal and Rohlf 1995, p. 194). This
overall test is also highly significant ( ). For all ofP
! .001
our tests, we used a right-tailed P value, where a significant
result indicates that beneficial mutations were dispropor-
tionately represented among mutations causing losses of
function. Significant overrepresentation of neutral or del-
eterious mutations is not expected because, unlike bene-
ficial mutations, which can become overrepresented as a
result of selection, there is no analogous mechanism to
cause the overrepresentation of neutral or deleterious mu-
tations. Consistent with this expectation, none of the P
values based on the left tail was significant (data not
shown). Also, using a two-tailed P value (and noting that
the two tails are not symmetric in a Fisher’s exact test)
produced the same statistically significant results. The only
exception was the test for Ancestor1; in this case, the P
value increased from .034 (one tailed) to .053 (two tailed).
To determine how energetic constraints on genome size
might affect the balance between antagonistic pleiotropy
and mutation accumulation, we repeated our experiments
E10 The American Naturalist
in an environment where organisms received a baseline
CPU cycle allotment that was constant rather than pro-
portional to genome length. In such an environment, all
else being equal, decreases in genome size will confer a
fitness benefit; however, organisms also receive additional
CPU cycles for performing rewarded computational func-
tions. These experiments show that, as expected, genomes
are significantly shorter in organisms evolved in this en-
vironment (Kruskal-Wallis , ,
x p 46.6 n p 200 P !
). In addition, the presence of energetic constraints.0001
on genome size shifted the balance further in favor of
antagonistic pleiotropy; the right tail of the Fisher’s exact
test was significant for nine out 10 ancestors, indicating
that beneficial mutations were disproportionately repre-
sented among mutations causing losses of function (data
not shown). Although selection to reduce genome size in
this system is probably stronger than that operating in
natural systems (since a single instruction deletion con-
stitutes a proportionately larger decrease in genome size
than does a single base pair deletion), these findings nev-
ertheless confirm that energetic constraints on genome size
can increase the importance of ecological specialization
arising from antagonistic pleiotropy.
Finally, although we expect that most mutations that
cause losses of function will be deleterious in the ancestral
environment, we verified this expectation by measuring
the fitness effect of each loss-of-function mutation in the
ancestral environment. Of the 470 mutations that caused
a loss of function in our experiments (i.e., fig. 4, all blue,
red, or green cells), 448 of them were indeed deleterious
in the ancestral environment. Moreover, excluding those
few mutations that did not reduce fitness in the ancestral
environment had no effect on the statistical significance
of our results. Overall, our findings indicate that, where
losses of function occurred, they were disproportionately
likely to be caused by a beneficial mutation, and therefore,
antagonistic pleiotropy was an important factor in driving
the decay of unrewarded functions.
Steps with Multiple Mutations
Because some steps along the line of descent occasionally
included multiple mutations (i.e., a derived genotype dif-
fered from its immediate parent by more than a single
mutation), we sought to determine whether these multiple
mutations made a significant contribution to losses of
function. We found that multiple mutations accounted for
approximately 6.2% of all genotypic steps along the line
of descent and for approximately 4.1% of mutations caus-
ing losses of function. Thus, multiple mutations were, if
anything, underrepresented among the mutations causing
losses of function. The Fisher’s exact tests comparing losses
of function due to beneficial versus nonbeneficial muta-
tions (table 1) were largely unaffected by the exclusion of
multiple mutations (data not shown). The statistical sig-
nificance of the results differed only for Ancestor1, which
became nonsignificant once these mutational steps were
Niche Breadth Reductions at Higher
and Lower Mutation Rates
Our initial experiments were performed at a genomic mu-
tation rate of 0.1 for 100,000 updates. To assess the gen-
erality of these results, we repeated our experiments at
significantly higher and lower mutation rates of 0.3 and
0.01, respectively. As expected, niche breadth usually de-
clined more rapidly with increasing mutation rate (fig. 5).
However, it was not obvious whether the faster decay of
niche breadth was a result of the greater overall mutation
supply or whether mutation rate disproportionately af-
fected losses of function by altering the relative importance
of beneficial and nonbeneficial mutations. For example,
in asexual organisms, increasing the mutation rate is ex-
pected to increase the fixation of nonbeneficial mutations
to a greater extent than beneficial mutations because, at
high mutation rates, beneficial mutations will more often
arise in different lineages that interfere with each other’s
fixation, a phenomenon termed “clonal interference”
(Muller 1932; Gerrish and Lenski 1998; de Visser et al.
1999; Orr 2000). The relative roles of antagonistic plei-
otropy and mutation accumulation might thus be altered
by changes to the mutation rate.
To address this issue, we repeated our experiments but
this time scaled their duration inversely to the mutation
rate. Because our initial experiments were run at a 0.1
genomic mutation rate for 100,000 updates, we reran the
high mutation rate (0.3) experiments for 33,000 updates
and the low mutation rate (0.01) experiments for 1,000,000
updates. Scaling the runs in this way should control for
differential mutation supply; this prediction was verified
by examining the number of genotypes along the line of
the descent, which was found to be similar across treat-
ments (mean: , ,low p 147.1 medium p 141.9 high p
; all pairwise comparisons not statistically significant).139.4
Our results show that increasing the mutation rate while
holding mutation supply constant tends to decrease the
number of beneficial mutations that cause losses of func-
tion per experiment (least squares means: ,low p 1.92
, ). A two-way parametricmedium p 1.67 high p 1.38
ANOVA based on the number of beneficial mutations
causing losses of function found a significant effect of
ancestor ( , , ), mutationF p 12.41 df p 9, 270 P ! .0001
rate ( , , ), and their interac-F p 5.60 df p 2, 18 P p .013
tion ( , , ). As expected,F p 1.66 df p 18, 270 P p .046
losses of function resulting from nonbeneficial mutations
Specialization in Digital Organisms E11
Figure 5: Decline in average niche breadth over time as a function of mutation rate. Average niche breadth was calculated as the mean niche breadth
of the 10 replicate populations derived from each ancestor. Niche breadth was calculated as the proportion of organisms performing each function
at a given time point and summed over all functions. , 0.01 genomic mutation rate; , 0.1 genomic mutation rate;Green p low blue p medium
, 0.3 genomic mutation p high
showed the opposite pattern (least squares means:
, , ). Ancestor andlow p 2.76 medium p 3.03 high p 3.20
mutation rate were again both statistically significant
( , , and ,F p 20.09 df p 9, 18 P
! .0001 F p 3.70 df p
, , respectively). The interaction between2, 18 P p .045
mutation rate and ancestor, however, was nonsignificant
( , , ) for these mutations.F p 1.08 df p 18, 270 P p .376
Finally, a paired t-test comparing the proportion of loss-
of-function mutations that were beneficial at low and high
mutation rates showed the proportion to be significantly
higher at lower mutation rates, as hypothesized (two-tailed
, based on table 2).P p .002
We can also ask whether beneficial mutations remain
overrepresented among mutations causing losses of func-
tion at higher and lower mutation rates. These data are
presented in table 2, which shows the proportion of ben-
eficial mutations, relative to the total, that were associated
or not associated with a loss of function. At all three mu-
tation rates, beneficial mutations were usually present in
greater proportions among mutations causing losses of
function than among those that did not. For all mutation
rates, the proportion of beneficial mutations causing losses
of function was higher in descendents of eight out of 10
ancestors, although the identities of these eight ancestors
varied across the treatments. Many of these differences
were significant when we performed the Fisher’s exact tests
to examine the number of beneficial versus nonbeneficial
mutations causing losses of function or not, as we also
saw for our earlier analysis at the genomic mutation rate
of 0.1 (table 1). While the pattern at higher and lower
mutation rates is qualitatively similar to that for the me-
dium mutation rate, somewhat fewer tests were significant
at both extremes. In general, however, the pattern was
similar, despite large changes to the mutation rate, indi-
cating that antagonistic pleiotropy was an important con-
tributor to niche specialization at all three mutation rates.
As before, we also analyzed our results to determine
whether they were affected by the presence of multiple
mutations in some steps. At the low mutation rate, mul-
tiple mutations comprised 2.3% and 0% of steps along the
line of descent and losses of function, respectively. The
conclusions of the Fisher’s exact tests that compared the
E12 The American Naturalist
Table 2: Proportion of mutations along the line of descent that
were beneficial as a function of mutation rate
Genomic mutation rate
Low (.01) Medium (.1) High (.3)
Loss Not P Loss Not P Loss Not P
1 .162 .195 NS .308 .176 * .175 .190 NS
2 .321 .333 NS .306 .424 NS .288 .252 NS
3 .862 .321 *** .714 .370 *** .500 .256 *
4 .596 .484 NS .396 .261 * .426 .199 ***
5 .545 .267 *** .386 .205 ** .354 .363 NS
6 .362 .251 NS .326 .183 * .341 .200 *
7 .192 .186 NS .200 .208 NS .192 .175 NS
8 .431 .259 ** .431 .172 *** .441 .148 ***
9 .396 .210 ** .213 .168 NS .227 .226 NS
10 .381 .249 * .386 .232 * .250 .242 NS
Note: Comparison of the proportion of mutations substituted on the line
of descent that were beneficial among those causing losses of function versus
those that did not, at three different mutation rates. Notice that beneficial
mutations are generally present in higher proportions among mutations caus-
ing losses of function. Asterisks indicate the significance of the associated
Fisher’s exact test, which compared the number of beneficial versus nonbe-
neficial (neutral or deleterious) mutations that caused losses of function (loss)
versus those that did not (not). significant.NS p not
! .05
** .P
! .01
*** .P
! .001
fitness effects of mutations causing losses of function to
those that did not were generally unaffected by these mu-
tations, with the exception of Ancestor4, for which the
test became significant once steps with multiple mutations
were excluded. At the high mutation rate, multiple mu-
tations comprised 7.1% of all mutations along the line of
descent and 10.9% of mutations causing losses of function.
The statistical significance of all of the Fisher’s exact tests
at the high mutation rate were unchanged by the exclusion
of these mutations.
Functional Genetic Explanations for Niche Conservatism
A striking feature of these experiments is the extent to
which some functions were repeatedly retained across rep-
licate populations started from the same ancestor (fig. 4,
columns of black cells). For example, all 10 populations
evolved from Ancestor1 invariably retained OR, while
those evolved from Ancestor3 always kept both AND and
OR. There are at least two explanations for the mainte-
nance of unrewarded functions. One possibility is that
there may have been insufficient mutational pressure to
cause losses of function. While this effect may be expected
to be random with respect to functions, some functions
present larger targets for mutations because they require
more instructions to encode, and thus they may be lost
more consistently. To test whether mutational pressure was
strong enough to lead to decay of functions, we ran ad-
ditional experiments with one ancestor, Ancestor3, for
which derived populations had decayed the least on av-
erage over the course of their evolution. These experiments
were identical to the original experiment, except that no
functions—including even EQU—were rewarded. In 10
replicate experiments starting from Ancestor3, every func-
tion was lost, showing that insufficient mutational pressure
could not explain the failure for losses of function to occur.
A second possibility is that these functions were main-
tained because their performance was coupled to that of
EQU—in other words, due to pleiotropy. One line of evi-
dence that pleiotropy was often responsible for the main-
tenance of some functions is that their performance, de-
spite not being rewarded, often increased during evolution
in the EQU environment and, in many cases, in proportion
to that of EQU. Figure 6 shows the phenotype of evolved
organisms from three different ancestors. Correlations
comparing the performance of retained functions to that
of EQU are consistently significant (Ancestor10, OR-NOT:
, , ; Ancestor1, OR: ,r p 0.72 df p 6 P p .044 r p 0.81
, ; Ancestor2, NOT: , ,df p 8 P p .005 r p 0.98 df p 7
; Ancestor2, OR-NOT: , ,P
! .0001 r p 0.90 df p 6 P p
Given that different functions appeared to be coupled
in their performance, we wanted to see whether we could
understand the mapping between genotype and phenotype
that gave rise to these correlations. In other words, rather
than merely observing that some genotypes retained more
unused functions than others, we wanted to understand
the origins of this evolutionary pattern by investigating
the relationship among different functions in the ancestral
genome. To do so, we systematically assessed all 10 an-
cestral genomes for the extent to which knocking out a
given instruction affected the performance of each func-
tion. The resulting “genotype-phenotype map” allowed us
to infer the regions of the genome that encoded each func-
tion an organism performs, as well as the overlap in these
An example of a genotype-phenotype map, constructed
for Ancestor1, is shown in figure 7. Each row of the map
represents one of the instructions in the genome, starting
from the first (top row) to the last instruction (bottom
row). Taking each site in the genome in turn, we replaced
the instruction present at that site with a null instruction,
called nop-X, and then tested the ability of the resulting
knockout mutant to perform logic functions. Organisms
were tested only for functions that the unmutated “wild-
type” organism had itself performed. Each column of the
map denotes a different logic function that could be per-
formed by the unmutated organism, and the cells are col-
ored as follows. White means that when the instruction
in the corresponding row is replaced with a null instruc-
Specialization in Digital Organisms E13
Figure 6: Functions that were not lost were correlated with the perfor-
mance of EQU. Each row represents the phenotype of a different evolved
organism (one from each of the 10 replicate populations) for three il-
lustrative ancestors. Numbers show the number of times the organism
performs each logic function per life cycle. While many functions were
lost (indicated by 0), those that were not lost show a correlated increase
in their performance with EQU. Correlation coefficients for the perfor-
mance of these functions with EQU are indicated below each table. Cor-
relations were calculated only between pairs of data where the function
in question had been maintained (i.e., if the value in the table was
in over half the replicate populations, as indicated by shading.
tion, there is no effect on the function in the corresponding
column. Colored cells indicate that replacing the corre-
sponding instruction with a null instruction resulted in a
loss of that function, and thus these cells correspond to
the areas of the genome that encode the different func-
tions. Among the colored cells, red cells represent the sub-
set of the instructions required for any given function that
are also required for the EQU function. For any other
function, a mixture of red and blue therefore indicates
only partial overlap with the instructions that encode EQU.
Logic functions that lack any blue coloring, such as the
function OR, indicate that there are no sites in the genome
that can be mutated to cause the loss of that function and
still maintain the EQU function. We therefore expect that
functions such as these might rarely be lost in an EQU-
only environment, owing to the absence of any mutational
target that does not also affect EQU. Consistent with this
expectation, populations evolved from this ancestor in an
EQU-only environment never lost the ability to perform
OR but usually lost all other unnecessary functions (fig.
4, upper left). Analysis of the genotype-phenotype map for
this ancestor thus implies that the differential overlap in
the encoding of the various functions with that of the EQU
function is at least partly responsible for their differential
maintenance during evolution in an EQU-only
To assess this relationship more generally, we used these
genotype-phenotype maps for all the ancestors to identify
the genomic regions that corresponded to each logic func-
tion. We then assessed the extent to which each of these
functions overlapped with the EQU function and calcu-
lated the number of nonoverlapping instructions. We then
determined how many times (out of a possible 10) each
function was actually lost during evolution in an EQU-
only environment. The relationship between these two
measures is presented in figure 8 and shows that functions
that overlap completely with EQU (i.e., those with 0 non-
overlapping instructions) were most likely to be main-
tained, but the probability of maintenance drops rapidly
as the number of nonoverlapping instructions increases.
This result provides compelling support for our hypothesis
that the integration of these functions in the genome
played an important role in maintaining certain unused
functions during evolution in the EQU-only environment.
Specifically, it demonstrates that niche breadth evolution
in this system was driven not only by the selective envi-
ronment in which these organisms evolved but also by the
way in which their genotypes mapped onto their pheno-
types, that is, by their genetic architecture.
Two distinct population genetic mechanisms are thought
to promote the evolution of ecological specialization, as
reflected in a narrow niche breadth. One entails the ac-
cumulation of mutations that are neutral or deleterious
in a novel environment owing to relaxed selection on un-
used functions. In the other, fitness improvements in a
novel environment may come at the expense of other
traits, leading to trade-offs and losses of function. Trade-
offs can occur if the same genes contribute to two or more
Figure 7: Genotype-phenotype map showing the instructions that encode each function in Ancestor1. Each row in the map represents a single
instruction, starting from the first instruction in the organism’s genome (top row) to the final instruction (bottom row). Each column represents a
different function performed by the ancestor, and the coloring indicates what happens to the performance of that function when a given instruction
is knocked out (replaced by a null instruction). out the instruction does not affect performance of the function; blue orWhite p knocking
out the instruction causes the function to be lost. Red blocks indicate the subset of instructions required for a given function that,red p knocking
when knocked out, also cause the loss of EQU. Note that every instruction in this organism that knocks out OR also knocks out EQU, whereas
this pattern does not hold for NAND, OR-N, AND-N, or XOR.
Specialization in Digital Organisms E15
Figure 8: Association between the number of nonoverlapping instruc-
tions (required to perform some function but not required for the EQU
function) and the average proportion of times (out of a possible 10) that
the function was maintained during evolution in the EQU-only envi-
ronment. Error bars represent 1 SE.
traits, such that mutations that improve one may worsen
others. Trade-offs can also arise, even when traits do not
share a genetic basis, if the maintenance of unselected or
weakly selected traits entails an energetic burden.
Here, we describe the evolution of ecological speciali-
zation in digital organisms. Starting from a set of generalist
ancestors, each of which could perform a wide variety of
logic computations, we examined their adaptation to a
novel environment where only a single computation was
directly selected. A benefit of examining the process of
specialization in digital organisms is that we can precisely
trace the mutational steps leading from the generalist an-
cestor to the evolved specialist, which allows us to examine
in detail the mutations that lead to losses of function along
the way.
Our results revealed significant heterogeneity in the
magnitudes of fitness improvements in the novel EQU-
only environment, with different populations evolving to
perform EQU to different extents depending on the an-
cestor used to initiate the experiment. All of the ancestors
performed EQU once per gestation cycle at the start of
the experiment, and, in a few cases (five out of 100),
evolved organisms did not increase their performance of
EQU above the ancestral level. In most cases, however,
organisms evolved to perform the function tens or even
hundreds of times per gestation cycle. The evolved organ-
isms also varied in the extent to which their niche breadth
became narrower, with very few populations (only seven
out of 100) evolving to become pure EQU specialists that
could not perform any other functions.
Examination of the mutations that led to losses of func-
tion allowed us to quantify the relative importance of mu-
tation accumulation and antagonistic pleiotropy. These
data showed that, in absolute terms, more losses of func-
tion were caused by neutral or deleterious mutations than
by beneficial mutations. Yet, when we standardized for the
greater numbers of nonbeneficial substitutions along the
lines of descent, beneficial mutations were disproportion-
ately associated with losses of function. Although the pro-
portion of losses of function that could be attributed to
beneficial mutations was generally higher at lower muta-
tion rates, the proportion of beneficial substitutions overall
was also higher. For this reason, changes to the mutation
rate had little effect on the results of the Fisher’s exact
tests. Nevertheless, functions were more often lost due to
the accumulation of neutral and deleterious mutations,
and this effect increased with increasing mutation rate.
The finding that lower mutation rates permit the fix-
ation of proportionally more beneficial mutations suggests
that some kind of interference is occurring at the higher
mutation rates, although it is not clear whether the in-
terference arises from deleterious or beneficial mutations.
At high mutation rates, beneficial and deleterious muta-
tions may often arise on the same background, limiting
fixation to those beneficial mutations of large effect (Peck
1994; Orr 2000; Johnson and Barton 2002). High mutation
rates can also lead to interference among beneficial mu-
tations that arise in different clonal lineages (Gerrish and
Lenski 1998; de Visser et al. 1999). Distinguishing between
these two alternatives in evolving digital populations is a
subject for further study.
One surprising result of these experiments was how, in
particular ancestors, certain functions were often main-
tained in the absence of direct selection for their perfor-
mance. Examination of the genetic architecture of the an-
cestors revealed that overlap in the genetic instructions
that encode the different functions was a good predictor
of their maintenance during evolution in the EQU-only
environment. Not only were these functions maintained
but also their performance often increased in parallel with
that of EQU, resulting in unexpected positive correlations
between certain traits across populations evolved from the
same ancestor. Because we know that there was no direct
selection on these functions, their maintenance is more
analogous to that of a vestigial trait rather than the out-
come of selection operating on two traits simultaneously.
Wright (1977, p. 428) and Lande (1978) both suggested
that useless or even slightly detrimental functions might
be retained over long periods of time, owing to their pleio-
tropic relationships to characters under direct selection.
Nevertheless, it would be interesting to examine the con-
sequences of these genetically integrated traits in the event
that selection were to operate on them in opposing di-
E16 The American Naturalist
rections (Beldade et al. 2002), and we will examine this
possibility in future work.
Our results show that there was no single function that
was always retained with EQU; rather, the identity of the
retained functions varied depending on the particular an-
cestor. For example, organisms evolved from Ancestor3
failed to lose AND and OR, whereas organisms evolved
from Ancestor2 often failed to lose NOT and OR-NOT.
Moreover, because the ancestors all shared the same his-
torical environment, these differences in outcome reflect
stochasticity in the origins of each ancestor’s unique ge-
netic architecture—a genetic architecture that influenced
the subsequent trajectory of evolution in the EQU-only
environment. Where multiple functions were maintained
(e.g., Ancestor3), it would be interesting to explore
whether they had been built on each other sequentially.
One could imagine, for instance, that EQU evolved from
AND and that AND evolved from OR, and so on. Of
course, the construction of complex functions out of sim-
pler ones—a process that contributes to the emergence of
pleiotropy in this system—also occurs in natural systems
(Jacob 1977; Nilsson and Pelger 1994; Mele´ndez-Hevia et
al. 1996; Chen et al. 1997; Dean and Golding 1997). Thus,
investigations into the form and direction of pleiotropy in
nature might be informed through a consideration of the
evolutionary history of the traits in question.
The importance of genetic integration for the mainte-
nance of unrewarded functions in highly specialized en-
vironments led to substantial variation in the niche
breadth of evolved organisms, with some organisms evolv-
ing very narrow specialization and others maintaining
their niche breadth at about half their ancestral levels. This
result raises interesting questions about the relative long-
term success of these organisms in a fluctuating environ-
ment, where antagonistic pleiotropy and mutation accu-
mulation may continually degrade functions that might
become necessary again at some later time (Kawecki 2000).
Organisms with highly integrated genetic architectures
would potentially prosper in such environments, whereas
those with greater modularity might do better in a more
stable environment, particularly if genetic correlations
among traits were found to constrain the optimization of
each trait individually. These predictions do not differ from
existing theories about the kinds of environments that
select for generalist versus specialist species, with the for-
mer predicted to emerge in a temporally heterogeneous
environment and the latter when there is environmental
constancy (Levins 1968; Futuyma and Moreno 1988; Kas-
sen 2002). However, this perspective emphasizes the role
of genetic architecture in mediating these transitions rather
than selection as the sole determinant of niche breadth.
Although our results show that most losses of function
resulted from neutral or deleterious mutations, we note
that several factors are expected to alter the relative con-
tributions of antagonistic pleiotropy and mutation accu-
mulation to ecological specialization. Our results indicate
that an important consideration is the relative likelihood
that a given mutation will be beneficial or not: increases
to the mutation rate, in combination with asexual repro-
duction, permitted the fixation of more neutral and del-
eterious mutations, increasing losses of function that result
from these mutation types. Several authors have recently
considered the role of deleterious mutations in adaptation,
and this work has led to a reevaluation of how mutation
rate alters the rate of adaptation in asexual organisms (Orr
2000; Johnson and Barton 2002; Wilke 2004). One of the
difficulties encountered by this work is the complexity of
the process, which requires modeling many competing lin-
eages and evaluating nonequilibrium conditions, making
it difficult to derive exact solutions. Digital systems may
prove to be a suitable testing grounds for some of the
hypotheses generated by this work, particularly because of
the ease of estimating parameters that are difficult to mea-
sure in biological systems, such as the rate of occurrence
and fixation of beneficial and deleterious mutations (see
also Rozen et al. 2002; Sanjuan et al. 2004). Moreover,
mutation-accumulation explanations for specialization of-
ten assume that the relevant mutations are either condi-
tionally or weakly deleterious because unconditionally del-
eterious mutations have difficulty attaining fixation except
in small populations (Kawecki 1994; MacLean et al. 2004).
In asexual organisms, however, deleterious mutations can
hitchhike to fixation alongside beneficial mutations. Given
that many extreme examples of adaptive decay involve
bacteria (Cole et al. 2001; Ochman and Moran 2001; Wer-
negreen et al. 2002), the potential role of deleterious mu-
tations needs to be considered more carefully. In asexual
organisms, niche breadth reductions could be occurring
by both antagonistic pleiotropy (fixation of beneficial mu-
tations) and mutation accumulation (via increased fixation
of deleterious mutations), with the fixation of the former
predisposing that of the latter through hitchhiking. This
result suggests that ecological specialization, by this cou-
pling of mutation accumulation and antagonistic pleiot-
ropy, may occur more readily in asexual organisms.
Ecological theories of niche specialization predict that
organisms will evolve to match the heterogeneity of their
environment (Levins 1968; Via and Lande 1985; Scheiner
1993; Kassen 2002). Our results show that environmental
constancy can, in fact, drive the evolution of niche breadth
reductions, with all organisms evolving niche breadths that
were narrower than that of their ancestors. However, sub-
stantial diversity in niche breadth was observed among
independently evolved organisms despite their evolution
in identical environments. While the extent to which traits
are encoded by the same or different genes has sometimes
Specialization in Digital Organisms E17
been taken into consideration when predicting the relative
importance of antagonistic pleiotropy and mutation ac-
cumulation in driving specialization (Futuyma and Mo-
reno 1988; Fry 1993; Kawecki 1994, 1998), the degree to
which suites of traits may be genetically integrated has not
often been considered with regard to the maintenance of
functions across environments (but see Rausher 1988).
Although trade-offs are widely thought to promote the
evolution of ecological specialization, the requisite negative
correlations have often not been forthcoming (Jaenike
1990; Fry 1996; Agrawal 2000). The failure to detect neg-
ative correlations has led to a developing body of work
that focuses on alternative explanations for the evolution
of ecological specialization (Kawecki 1994, 1998; Whitlock
1996). In addition, Rausher (1988) suggested that trade-
offs may not always be detected. For example, studies of
diet breadth in phytophagous insects often employ host
species that are already part of the natural diet. If severe
trade-offs exist, then those host species are more likely to
have been excluded from the diet previously, and the ob-
served niche breadth will consist only of hosts for which
there was either little or no conflict. Our results are con-
sistent with this hypothesis, with trade-offs often quickly
leading to losses of function, leaving mostly positive cor-
relations among the remaining functions.
Ultimately, of course, experiments with digital organ-
isms cannot tell us what processes are actually at work in
any given natural system—that is an empirical question
that cannot be addressed by any model system, digital or
otherwise. However, digital systems provide a novel way
of assessing the logic that underlies many evolutionary
theories, especially where complex interactions limit the
opportunity for purely theoretical analysis. Our results
show that ecological specialization occurs in digital or-
ganisms and, moreover, that some of the same patterns
that have complicated simple theories of niche breadth in
natural systems, such as the apparent paucity of trade-offs
and an excess of positive correlations, also emerge here.
Finally, digital systems offer the ability to connect patterns
to processes and thus allow investigations of causal mech-
anisms more directly than is possible in any other system,
enabling tests of existing hypotheses as well as the devel-
opment of new ones that can in turn be tested in other
We thank J. Conner, T. Cooper, K. Gross, A. Jarosz, and
two anonymous reviewers for helpful comments on an
earlier version of this manuscript. This worked was sup-
ported by grants from the National Science Foundation
to R.E.L. and C.O. and by support from the Quantitative
Biology and Modeling Initiative at Michigan State Uni-
versity to E.A.O.
E18 The American Naturalist
Schematic of a Digital Organism in Avida
Figure A1: A digital organism consists of a genome (computer program), three registers, two stacks, and four heads (one of which is the instruction
pointer [IP]). Execution of the program requires central processing unit cycles, and the point of execution is indicated by the location of the IP.
An input-output instruction enables an organism to input binary numbers into the registers and output the results of computations. Most genomic
instructions operate directly on the numbers in the registers, although the push and pop instructions cause numbers in the registers to be pushed
onto the stack or popped off of the stack, respectively. The stacks are thus primarily used for storing numbers, whereas the registers are used to
manipulate them.
Glossary of Terms
Table B1: Glossary of terms
Term Definition
CPU Central processing unit. All organisms have the instructions in their genomes exe-
cuted by a virtual CPU. A mutation that causes an organism to have more CPU
cycles (i.e., to have its genome executed faster than others) is generally beneficial.
Digital organism A virtual computer, consisting of a genome (a computer program) and its associ-
ated hardware. The hardware consists of the CPU, which processes the instruc-
tions in the genome, two stacks, and three registers, which are used for storing,
retrieving, and manipulating numbers. Each organism also has an instruction
pointer (IP), which points to the next instruction to be executed in the genome,
and read, write, and flow heads, which are used to specify positions in memory,
such as in the copy process or for jumping and looping.
Specialization in Digital Organisms E19
Table B1 (Continued )
Term Definition
EQU A logic function, where two binary inputs are compared and the correct output is a
1 if the input bits are the same and a 0 if the bits are different. In this system,
EQU is actually a bitwise EQU, in that the correct output is the computation of
EQU across all 32 bits for the two inputs. Performance of EQU requires, at a
minimum, combining the outputs of five different NAND statements, in coordi-
nation with various other instructions.
Genome Sequence of instructions that may contain information for making duplicate copies
of the genome and for interactions with the environment. Execution of the in-
structions in a properly functioning genome leads to the production of an
Gestation time Number of instructions executed, and hence CPU cycles required, to produce an
offspring. Gestation time is generally a multiple of genome length but varies as a
function of the efficiency of the copy process and the number of loops in
Instruction Units that comprise the genome. Each site in the genome is one of 26 possible in-
structions. Instructions not present in an ancestral genome may be introduced
into the genomes of descendents via mutation.
Logic function Computations based on binary inputs. Organisms may evolve to perform bitwise
logic functions based on numbers they input from the environment.
Mutations Mutations can be point mutations, where one instruction is randomly replaced
with another during the copy process. Mutations can also be insertion or dele-
tion mutations, causing genomes to grow or shrink in length. The rates of point,
insertion, and deletion mutations are specified by the experimenter.
NAND One of the 26 possible instructions in the genome. Also a core logic function; all
other logic functions can be built from combinations of NANDs.
Literature Cited
Agrawal, A. A. 2000. Host range evolution: adaptation and trade-
offs in fitness of mites on alternative hosts. Ecology 81:500–508.
Andersson, S. G. E., A. Zomorodipour, J. O. Andersson, T. Sicheritz-
Ponten, U. C. M. Alsmark, R. M. Podowski, A. K. Naslund, A. S.
Eriksson, H. H. Winkler, and C. G. Kurland. 1998. The genome
sequence of Rickettsia prowazekii. Nature 396:133–143.
Beldade, P., K. Koops, and P. M. Brakefield. 2002. Developmental
constraints versus flexibility in morphological evolution. Nature
Borowsky, R., and H. Wilkens. 2002. Mapping a cave fish genome:
polygenic systems and regressive evolution. Journal of Heredity
Chen, L., A. L. DeVries, and C.-H. C. Cheng. 1997. Evolution of
antifreeze glycoprotein gene a trypsinogen gene in Antarctic no-
tothenioid fish. Proceedings of the National Academy of Sciences
of the USA 94:3811–3816.
Cole, S. T., K. Eiglmeier, J. Parkhill, K. D. James, N. R. Thomson,
P. R. Wheeler, N. Honore, et al. 2001. Massive gene decay in the
leprosy bacillus. Nature 409:1007–1011.
Cooper, V. S., and R. E. Lenski. 2000. The population genetics of
ecological specialization in evolving Escherichia coli populations.
Nature 407:736–739.
Cooper, V. S., A. F. Bennett, and R. E. Lenski. 2001. Evolution of
thermal dependence of growth rate of Escherichia coli populations
during 20,000 generations in a constant environment. Evolution
Darwin, C. 1859. On the origin of species by means of natural se-
lection. J. Murray, London.
Dean, A. M., and G. B. Golding. 1997. Protein engineering reveals
ancient adaptive replacements in isocitrate dehydrogenase. Pro-
ceedings of the National Academy of Sciences of the USA 94:3104–
de Visser, J. A. G. M., C. W. Zeyl, P. J. Gerrish, J. L. Blanchard, and
R. E. Lenski. 1999. Diminishing returns from mutation supply rate
in asexual populations. Science 283:404–406.
Falconer, D. S. 1952. The problem of environment and selection.
American Naturalist 86:293–298.
Fry, J. D. 1993. The general vigor problem: can antagonistic plei-
otropy be detected when genetic covariances are positive? Evo-
lution 47:327–333.
———. 1996. The evolution of host specialization: are trade-offs
overrated? American Naturalist 148(suppl.):S84–S107.
Futuyma, D. J., and G. Moreno. 1988. The evolution of ecological
specialization. Annual Review of Ecology and Systematics 19:207–
Gerrish, P. J., and R. E. Lenski. 1998. The fate of competing beneficial
mutations in an asexual population. Genetica 102/103:127–144.
Go´mez-Valero, L., A. Latorre, and F. J. Silva. 2004. The evolutionary
fate of nonfunctional DNA in the bacterial endosymbiontBuchnera
aphidicola. Molecular Biology and Evolution 21:2172–2181.
Jacob, F. 1977. Evolution and tinkering. Science 196:1161–1166.
Jaenike, J. 1990. Host specialization in phytophagous insects. Annual
Review of Ecology and Systematics 21:243–273.
Jeffrey, W. R., A. G. Strickler, and Y. Yamamoto. 2003. To see or not
to see: evolution of eye degeneration in Mexican blind cavefish.
Integrative and Comparative Biology 43:531–541.
Jernigan, R. W., D. C. Culver, and D. W. Fong. 1994. The dual role
E20 The American Naturalist
of selection and evolutionary history as reflected in genetic cor-
relations. Evolution 48:587–596.
Johnson, T., and N. H. Barton. 2002. The effect of deleterious alleles
on adaptation in asexual populations. Genetics 162:395–411.
Jones, R., D. C. Culver, and T. C. Kane. 1992. Are parallel mor-
phologies of cave organisms the result of similar selection pres-
sures? Evolution 46:353–365.
Kassen, R. 2002. The experimental evolution of specialists, generalists,
and the maintenance of diversity. Journal of Evolutionary Biology
Kawecki, T. J. 1994. Accumulation of deleterious mutations and the
evolutionary cost of being a generalist. American Naturalist 144:
———. 1998. Red Queen meets Santa Rosalia: arms races and the
evolution of host specialization in organisms with parasitic life-
styles. American Naturalist 152:635–651.
———. 2000. The evolution of genetic canalization under fluctuating
selection. Evolution 54:1–12.
Lande, R. 1978. Evolutionary mechanisms of limb loss in tetrapods.
Evolution 32:73–92.
Lenski, R. E., C. Ofria, R. T. Pennock, and C. Adami. 2003. The
evolutionary origin of complex features. Nature 423:139–144.
Levins, R. 1968. Evolution in changing environments. Princeton Uni-
versity Press, Princeton, NJ.
MacLean, R. C., G. Bell, and P. B. Rainey. 2004. The evolution of a
pleiotropic fitness trade-off in Pseudomonas fluorescens. Proceed-
ings of the National Academy of Sciences of the USA 101:8072
Mele´ndez-Hevia, E., T. G. Waddell, and M. Cascante. 1996. The
puzzle of the Krebs citric acid cycle: assembling the pieces of
chemically feasible reactions, and opportunism in the design of
metabolic pathways during evolution. Journal of Molecular Evo-
lution 43:293–303.
Muller, H. J. 1932. Some genetic aspects of sex. American Naturalist
Nilsson, D.-E., and S. A. Pelger. 1994. A pessimistic estimate of the
time required for an eye to evolve. Proceedings of the Royal Society
B: Biological Sciences 256:53–58.
Ochman, H., and N. A. Moran. 2001. Genes lost and genes found:
evolution of bacterial pathogenesis and symbiosis. Science 292:
Ofria, C., and C. O. Wilke. 2004. Avida: a software platform for
research in computational evolutionary biology. Journal of Arti-
ficial Life 10:191–229.
Ofria, C., C. Adami, and T. C. Collier. 2003. Selective pressures on
genomes in molecular evolution. Journal of Theoretical Biology
Orr, H. A. 2000. The rate of adaptation in asexuals. Genetics 155:
Peck, J. R. 1994. A ruby in the rubbish: beneficial mutations, dele-
terious mutations and the evolution of sex. Genetics 137:597–606.
Petrov, D. A., and D. L. Hartl. 2000. Pseudogene evolution and
natural selection for a compact genome. Journal of Heredity 91:
Rausher, M. D. 1988. Is coevolution dead? Ecology 69:898–901.
Rozen, D. E., J. A. G. M. de Visser, and P. J. Gerrish. 2002. Fitness
effects of fixed beneficial mutations in microbial populations. Cur-
rent Biology 12:1040–1045.
Sanjuan, R., A. Moya, and S. F. Elena. 2004. The distribution of
fitness effects caused by single-nucleotide substitutions in an RNA
virus. Proceedings of the National Academy of Sciences of the USA
Scheiner, S. M. 1993. Genetics and evolution of phenotypic plasticity.
Annual Review of Ecology and Systematics 24:35–68.
Schluter, D. 2000. The ecology of adaptive radiation. Oxford Uni-
versity Press, Oxford.
Shigenobu, S., H. Watanabe, M. Hattori, Y. Sakaki, and H. Ishikawa.
2000. Genome sequence of the endocellular bacterial symbiont of
aphids Buchnera sp. APS. Nature 407:81–86.
Simpson, G. G. 1953. The major features of evolution. Columbia
University Press, New York.
Sliwa, P., and R. Korona. 2005. Loss of dispensable genes is not
adaptive in yeast. Proceedings of the National Academy of Sciences
of the USA 102:17670–17674.
Sokal, R. R., and F. J. Rohlf. 1995. Biometry. Freeman, New York.
Via, S., and R. Lande. 1985. Genotype-environment interaction and
the evolution of phenotypic plasticity. Evolution 39:505–522.
Wernegreen, J. J., A. B. Lazarus, and P. H. Degnan. 2002. Small
genome of Candidatus Blochmannia, the bacterial endosymbiont
of Camponotus, implies irreversible specialization to an intracel-
lular lifestyle. Microbiology 148:2551–2556.
Whitlock, M. C. 1996. The Red Queen beats the jack-of-all-trades:
the limitations on the evolution of phenotypic plasticity and niche
breadth. American Naturalist 148(suppl.):S65–S77.
Wilke, C. O. 2004. The speed of adaptation in large asexual popu-
lations. Genetics 167:2045–2053.
Wilke, C. O., and C. Adami. 2002. The biology of digital organisms.
Trends in Ecology & Evolution 17:528–532.
Wright, S. 1977. Evolution and the genetics of populations. Vol. 3.
Experimental results and evolutionary deductions. University of
Chicago Press, Chicago.
Associate Editor: Richard Gomulkiewicz
Editor: Michael C. Whitlock
... En effet, leur interaction très étroite avec leur hôte entraîne la sélection de phénotypes comportementaux et physiologiques qui peuvent être spécifiques de certaines espèces hôtes (Dupas & Carton 1999;, voire de certaines souche d'hôtes ne différent que par leur endosymbiote . Par ailleurs, la coévolution avec l'hôte a longtemps été considérée comme induisant un désavantage des généralistes par rapport aux spécialistes ce qui sélectionnerait pour une augmentation de la spécialisation des parasitoïdes (Kawecki 1994;Whitlock 1996;Ostrowski et al. 2007). Ce désavantage aurait deux principales causes : ...
... ii) une évolution plus lente des généralistes qui pourrait les désavantager si les hôtes s'adaptent rapidement. Cette évolution plus lente d'une espèce généraliste serait due au fait que les différentes pressions de sélection s'exerçant sur ses capacités à parasiter différents hôtes avec succès (en supposant ces capacités indépendantes) seraient chacune plus faibles que la pression de sélection s'exerçant sur une espèce spécialiste (Kawecki 1994;Whitlock 1996;Ostrowski et al. 2007). Les pressions de sélections indépendantes sur chacun des hôtes sont principalement attendues si les facteurs de virulence sont spécifiques de chacun des hôtes. ...
... Si Le premier cas est lié à la notion de réseaux trophiques individuels consistant à regarder, non pas quelles espèces « mangent » (ou parasitent) quelles espèces, mais quels individus (génotypes) d'une espèce « mangent » quels individus (génotypes) d'une autre espèce (Melián et al. 2011 Le deuxième type d'espèce généraliste, composée d'individus intrinsèquement généralistes, a été prédit comme désavantageux par rapport aux stratégies spécialistes. Le ralentissement de l'évolution associé à l'utilisation d'une large gamme d'hôtes, diminuerait en effet leur capacité d'adaptation sur chacun des hôtes (Kawecki 1994;Whitlock 1996;Ostrowski et al. 2007). Ce coût peut cependant être contrebalancé par différents phénomènes tels la variabilité spatio-temporelle de disponibilité des différents types d'hôtes, ce qui est particulièrement attendue pour les dynamiques hôtesparasitoïdes (Hassell 2000;Xu & Boyce 2005;Klemola et al. 2010). ...
Les guêpes endoparasitoïdes effectuent leur développement dans un hôte arthropode, entraînant sa mort. Parmi les stratégies assurant leur succès parasitaire, la plus commune est l’injection de venin dans l’hôte lors de l’oviposition, provoquant la suppression de l’immunité de l’hôte et la régulation de sa physiologie. La composition du venin a été caractérisée dans un nombre croissant de familles de parasitoïdes et des études récentes suggèrent que la virulence des parasitoïdes peut évoluer rapidement en réponse à la sélection. Cependant, la variation intraspécifique de cette composition et la capacité du venin à évoluer à court-terme n’avaient pas été étudiées. Cette information est pourtant essentielle pour comprendre l’évolution de la gamme d’hôte des parasitoïdes et elle peut avoir des implications importantes en terme de lutte biologique.Cette thèse s’est intéressée à l’analyse de la variabilité de la composition du venin en utilisant deux couple d'espèces, l'un du genre Psyttalia, utilisé en lutte biologique contre la mouche de l’olivier et l'autre du genre Leptopilina, des parasitoïdes de la Drosophile. Après avoir démontré l’existence de variabilité inter-individuelle du venin, j’ai développé une méthode sans a priori basée sur d’analyser des profils d’électrophorèse 1D à l’aide de fonctions R, permettant la comparaison statistique des quantités des différentes protéines à partir de nombreux individus. En utilisant des approches d’évolution expérimentale, j’ai ainsi pu (i) analyser les changements du venin dans des populations naturelles de Psyttalia lounsburyi élevées en dans des conditions de laboratoire sur un hôte de substitution (ii) intégrer les résultats avec les données vénomiques obtenues sur deux espèces de Psyttalia (iii) analyser l’évolution du venin de Leptopilina boulardi en réponse à la résistance/sensibilité.Les données montrent une variabilité importante des composants du venin chez les parasitoïdes à tous les niveaux biologiques. Ils démontrent aussi pour la première fois que la composition de ce venin peut changer rapidement, confirmant son fort potentiel évolutif. Les conditions d’élevage ainsi que la résistance de l’hôte sont des paramètres qui affectent fortement le contenu du venin ce qui peut avoir des conséquences importantes en lutte biologique. Enfin, la composition du venin affecte la probabilité d’extinction des petites populations. Les mécanismes à l’origine de sa variabilité restent à présent à étudier.
... Digital evolution has been used to address ecological questions (Dolson and Ofria), involving pairwise competitive interactions (Cooper and Ofria, 2003); mutualism (Johnson and Wilke, 2004;Rocabert et al., 2017;Vostinar et al.), predator-prey (Shao and Ray, 2010), host-parasite (Zaman et al., 2011(Zaman et al., , 2014; Acosta and Zaman) and even entire ecological networks (Fortuna et al., 2013). The transition from generalist to specialist shows that antagonistic pleiotropy reduces niche breadth (Ostrowski et al., 2007). Coevolution among mutually dependent organisms reduces the amplitude of the oscillations of species abundances compared to purely ecological scenarios (Johnson and Wilke, 2004). ...
... In line with our observations, widespread species would be expected to maintain high levels of nucleotide variation due to ongoing gene flow, high effective population sizes and ongoing heterogeneous selection pressures (Star & Spencer, 2013;Willi et al., 2006). On the other hand, species that have become restricted to a relatively narrow niche may experience DNA decay and loss of function of some previously important genes, further limiting their capacity to expand (Dworkin & Jones, 2009;Hoffmann & Willi, 2008;Ostrowski et al., 2007). Intriguingly, we found that the population size differences between species were themselves associated with differences in the numbers of genes which they retained that originated in the phylostratum overlapping the origin of panarthropods. ...
Full-text available
Many Drosophila species differ widely in their distributions and climate niches, making them excellent subjects for evolutionary genomic studies. We have developed a database of high-quality assemblies for 46 Drosophila species and one closely related Zaprionus. Fifteen of the genomes were newly sequenced, and 20 were improved with additional sequencing. New or improved annotations were generated for all 47 species, assisted by new transcriptomes for 19. Phylogenomic analyses of these data resolved several previously ambiguous relationships, especially in the melanogaster species group. However, it also revealed significant phylogenetic incongruence among genes, mainly in the form of incomplete lineage sorting in the subgenus Sophophora but also including asymmetric introgression in the subgenus Drosophila. Using the phylogeny as a framework and taking into account these incongruences, we then screened the data for genome-wide signals of adaptation to different climatic niches. Firstly, phylostratigraphy revealed relatively high rates of recent novel gene gain in three temperate pseudoobscura and five desert-adapted cactophilic mulleri subgroup species. Secondly, we found differing ratios of non-synonymous to synonymous substitutions in several hundred orthologs between climate generalists and specialists, with significant higher trends for those in tropical and lower trends for those in temperate-continental specialists respectively than those in the climate generalists. Finally, re-sequencing natural populations of thirteen species revealed tropics-restricted species generally had smaller population sizes, lower genome diversity and more deleterious mutations than the more widespread species. We conclude that adaptation to different climates in the genus Drosophila has been associated with large-scale and multifaceted genomic changes.
... A third alternative mechanism is gene decay. If environmental stability relaxes selection on genes that would be needed in extreme conditions, they may rapidly become nonfunctional (Hoffmann, 2010;Maughan et al., 2007;Ostrowski et al., 2007). If gene decay has occurred, genetic covariance will not be detected (García-Robledo & Horvitz, 2012). ...
Tropical ectotherms are particularly vulnerable to global warming because their physiologies are assumed to be adapted to narrow temperature ranges. This study explores three mechanisms potentially constraining thermal adaptation to global warming in tropical insects: 1. tradeoffs in genotypic performance at different temperatures (the jack‐of‐all‐trades hypothesis) 2. positive genetic covariance in performance, with some genotypes performing better than others at viable temperatures (the ‘winner and ‘loser genotypes hypothesis) or 3. limited genetic variation as the potential result of relaxed selection and the loss of genes associated with responses to extreme temperatures (the gene decay hypothesis). We estimated changes in growth and survival rates at multiple temperatures for three tropical rain forest insect herbivores (Cephaloleia rolled‐leaf beetles, Chrysomelidae). We reared 2746 individuals in a full‐sibling experimental design, at temperatures known to be experienced by this genus of beetles in nature (i.e., 10‐35°C). Significant genetic covariance was positive for 16 traits, supporting the ‘winner and ‘loser genotypes hypothesis. Only two traits displayed negative cross‐temperature performance correlations. We detected a substantial contribution of genetic variance in traits associated with size and mass (0‐44%), but low heritability in plastic traits such as development time (0‐6%) or survival (0‐4%). Lowland insect populations will most likely decline if current temperatures increase between 2 ‐ 5°C. It is concerning that local adaption is already lagging behind current temperatures. The consequences of maintaining the current global warming trajectory would be devastating for tropical insects. However, if humans can limit or slow warming, many tropical ectotherms might persist in their current locations, and potentially adapt to warmer temperatures.
... The discrepancy in slope among genotypes represents responsiveness (unequal variances on different substrates); and intersection of the lines represents inconsistency grown at higher temperatures (Figs. 1, 2). It is known that accumulation of conditionally neutral mutations is a major mechanism underlying fitness trade-offs, and thus local adaptation, among environments [16,[47][48][49]. If the temperature effects for resource dependence of MA effects observed here are generalizable to other environmental factors (e.g., substitutable nitrogen resources, or compound resources of different qualities), this constitutes an explanation for greater population divergence, and overall greater genetic diversity, in warmer regions [50][51][52]. ...
Full-text available
Background Mutation accumulation (MA) has profound ecological and evolutionary consequences. One example is that accumulation of conditionally neutral mutations leads to fitness trade-offs among heterogenous habitats which cause population divergence. Here we suggest that temperature, which controls the rates of all biochemical and biophysical processes, should play a crucial role for determining mutational effects. Particularly, warmer temperatures may mitigate the effects of some, not all, deleterious mutations and cause stronger environmental dependence in MA effects. Results We experimentally tested the above hypothesis by measuring the growth performance of ten Escherichia coli genotypes on six carbon resources across ten temperatures, where the ten genotypes were derived from a single ancestral strain and accumulated spontaneous mutations. We analyzed resource dependence of MA consequences for growth yields. The MA genotypes typically showed reduced growth yields relative to the ancestral type; and the magnitude of reduction was smaller at intermediate temperatures. Stronger resource dependence in MA consequences for growth performance was observed at higher temperatures. Specifically, the MA genotypes were more likely to show impaired growth performance on all the six carbon resources when grown at lower temperatures; but suffered growth performance loss only on some, not all the six, carbon substrates at higher temperatures. Conclusions Higher temperatures increase the chance that MA causes conditionally neutral fitness effects while MA is more likely to cause fitness loss regardless of available resources at lower temperatures. This finding has implications for understanding how geographic patterns in population divergence may emerge, and how conservation practices, particularly protection of diverse microhabitats, may mitigate the impacts of global warming.
... In extreme cases, targeting one trait can lead to the loss of others Ostrowski et al., 2007), and reduced genetic diversity in crop plants can increase the likelihood of such losses. Examples of traits that are threatened by modern crop breeding include: (1) the ability to interact with mycorrhizae (Zhu et al., 2001), (2) symbiotic nitrogen fixation (Kiers et al., 2007), (3) inhibition of nitrification (Subbarao et al., 2006), and (4) the ability to benefit from earthworm activities (Noguera et al., 2011). ...
... Un environnement constant est attendu comme sélectionnant plutôt pour une spécialisation des individus en lien avec cet environnement (Remold, 2012). Cependant, la sélection des traits les mieux adaptés à cet environnement se fera souvent au détriment d'autres traits conduisant ainsi à l'existence de trade-off (Ostrowski, 2007). À l'inverse, un environnement changeant ou instable favoriserait les généralistes qui seraient mieux adaptés à divers environnements possibles que les spécialistes (Remold, 2012). ...
Full-text available
Les larves des insectes endoparasitoïdes se développent aux dépens d’un hôte, conduisant à sa mort. Leur réussite dépend de leur capacité à neutraliser la réponse immunitaire de l’hôte (formation d’une capsule mélanisée) et réguler sa physiologie, via l’injection de venin lors de l’oviposition. Bien que les parasitoïdes subissent des changements importants de leur environnement, son impact sur la composition protéique de leur venin a rarement été étudié. Dans cette thèse, j’ai étudié (i) l’impact de la température sur l’interaction immunitaire Drosophile – Leptopilina boulardi et la composition de son venin et (ii) le potentiel évolutif du venin en réponse à différents hôtes. La température influe sur le succès parasitaire de L. boulardi via la capacité de l’hôte à encapsuler et/ou celle du parasite à s’échapper de la capsule. De plus, il y a une plasticité de la composition du venin avec la température. Par ailleurs, une 1ère évolution expérimentale a montré une évolution rapide de composition du venin en réponse à la résistance/sensibilité de l’hôte et un coût à la présence de certains facteurs du venin. La 2nd évolution expérimentale sur des espèces hôtes différentes suggère (i) une spécialisation des parasitoïdes sur l’hôte d’élevage et (ii) une évolution différentielle de la réussite parasitaire (virulence) et de la composition du venin. Le venin assurant la réussite sur différents hôtes contiendrait donc une combinaison de protéines « spécialisées » pour chaque hôte et des protéines à effet « large-spectre » sur certains hôtes testés. Les résultats suggèrent un fort potentiel adaptatif des parasitoïdes en réponse à différents paramètres biotiques et abiotiques.
Experimental evolution with microbial model systems has transformed our understanding of the basic rules underlying ecology and evolution. Experiments leveraging evolution as a central feature put evolutionary theories to the test, and modern sequencing and engineering tools then characterized the molecular basis of adaptation. As theory and experimentations refined our understanding of evolution, a need to increase throughput and experimental complexity has emerged. Here, we summarize recent technologies that have made high-throughput experiments practical and highlight studies that have capitalized on these tools, defining an exciting new era in microbial experimental evolution. Multiple research directions previously limited by experimental scale are now accessible for study and we believe applying evolutionary lessons from in vitro studies onto these applied settings has the potential for major innovations and discoveries across ecology and medicine.
In this article, we summarize our research results on the topic of spontaneous emergence of intelligence. Many agents are sent to an artificial world, which is arbitrarily parametrizable. The agents initially know nothing about the world. Their only ability is the remembrance, that is, the use of experience which comes from the events that happened to them in the course of their operation. With this individual knowledge base they attempt to survive in this world, and getting better and better knowledge for further challenges: a completely random wandering in the world; wandering in possession of growing personal knowledge; wandering, where evolutionary entities exchange experiences when they accidentally meet in a particular part of the world. The results are not surprising, but very convincing: without learning, the chances of survival are the worst in a world with a given parametrization. When experiences are organized into a knowledge base through individual learning, the chances of survival are obviously better than in the case of a completely random walk. And finally when creatures have the opportunity to exchange experiences when they meet in a certain field, they have the greatest chance of surviving.
Full-text available
Many ideas about the evolution of specialization rely on trade-offs—an inability for one organism to express maximal performance in two or more environments. However, optimal foraging theory suggests that populations can evolve specialization on a superior resource without explicit trade-offs. Classical results in population genetics show that the process of adaptation can be biased toward further improvement in already productive environments, potentially widening the gap between superior and inferior resources. Here I synthesize these approaches with new insights on evolvability at low recombination rates, showing that emergent asymmetries in evolvability can push a population toward specialization in the absence of trade-offs. Simulations are used to demonstrate how adaptation to a more common environment interferes with adaptation to a less common but otherwise equal alternative environment. Shaped by recombination rates and other population-genetic parameters, this process results in either the retention of a generalist niche without trade-offs or entrapment at the local optimum of specialization on the common environment. These modeling results predict that transient differences in evolvability across traits during an episode of adaptation could have long-term consequences for a population’s niche.
A review of fossil evidence on the rates of limb loss in tetrapods indicates that millions of years are required for complete loss of external traces of limbs. Morphological series of intermediate stages of limb loss within genera in the lizard families Scincidae and Teiidae show that there are evolutionary pathways of body elongation and reduction of limb size relative to skull length, accompanied by loss of limb elements beginning distally. Evolutionary elongation of the lizard body occurs through an increase in the coefficient of allometric growth of the body with respect to the skull, which precedes structural reduction of the limbs. A review of embryological studies of limbed and limbless forms suggests that in amniotes the distal to proximal sequence of structural reduction evolves by the progressively earlier onset of cell death which usually occurs to form the spaces between the digits, in conjunction with the normal proximo-distal sequence of determination of mesodermal elements in limb development. Following a presentation of data on digital variation in lizards, genetic studies of digit loss and polydactyly are summarized which indicate a polygenic influence on structural variation. Using these data, mathematical models are constructed which show that weak selection pressures can produce geologically rapid structural changes. The mechanism of reexpression of long lost structures of the limb (such as cetacean pelvic limbs and atavistic digits in the horse, dog, and guinea pig) is considered in a view of information on appendicular mutations in the mouse.
The patterns of genetic correlations between a series of eye and antenna characters were compared among two sets of spring-dwelling and cave-dwelling populations of Gammarus minus. The two sets of populations originate from different drainages and represent two separate invasions of cave habitats from surface-dwelling populations. Matrix correlations, using permutation tests, indicated significant correlations both between populations in the same basin and from the same habitat. The technique of biplot, which allows for the simultaneous consideration of relationships between different genetic correlations and different populations, was used to further analyze the correlation structure. A rank-3 biplot indicated that spring and cave populations were largely differentiated by eye-antennal correlations, whereas basins were differentiated by both eye-antennal and antennal-antennal correlations. Eye-antennal correlations, which are likely to be subject to selection, were most similar within habitats, which are likely to have similar selective regimes.
The amphipod Gammarus minus is present in both caves and springs, with cave populations showing elaborated (size and antennae) and reduced (eye) characters relative to spring populations. Earlier studies have shown that cave populations resulted from independent invasions of hydrologically isolated subterranean drainages and that there is genetic variation for both elaborated and reduced characters. In this study we tested the hypothesis that a similar pattern of selection on isolated cave populations is responsible for the parallel evolution of cave morphologies. We used variation in mating success and fecundity to test for the presence of directional selection on eye, antennal, and body size characters in a set of cave and spring populations during a series of seasonal cross-sectional samplings. We found significant directional selection for smaller eyes in caves and larger eyes in springs, which supports the hypothesis that selection is responsible for reduced eye size in cave populations. We also found selection for larger body and antennal size in cave populations, which is consistent with the hypothesis that parallel patterns of selection in caves are responsible for the elaboration of body and antennal size. However, we found selection for larger body and antennal size in spring populations that is not consistent with the observed divergence of spring and cave populations. We suggest that unmeasured components of viability selection could be more important in springs than in caves and may act against the selection for larger size found in spring populations.