How repeatable is adaptive evolution? The role of geographical origin and founder effects in laboratory adaptation.

Abstract

The importance of contingency versus predictability in evolution has been a long-standing issue, particularly the interaction between genetic background, founder effects, and selection. Here we address experimentally the effects of genetic background and founder events on the repeatability of laboratory adaptation in Drosophila subobscura populations for several functional traits. We found disparate starting points for adaptation among laboratory populations derived from independently sampled wild populations for all traits. With respect to the subsequent evolutionary rate during laboratory adaptation, starvation resistance varied considerably among foundations such that the outcome of laboratory evolution is rather unpredictable for this particular trait, even in direction. In contrast, the laboratory evolution of traits closely related to fitness was less contingent on the circumstances of foundation. These findings suggest that the initial laboratory evolution of weakly selected characters may be unpredictable, even when the key adaptations under evolutionary domestication are predictable with respect to their trajectories.

Full text

Page 1

ORIGINAL ARTICLE
doi:10.1111/j.1558-5646.2008.00423.x
HOW REPEATABLE IS ADAPTIVE EVOLUTION?
THE ROLE OF GEOGRAPHICAL ORIGIN AND
FOUNDER EFFECTS IN LABORATORY
ADAPTATION
Pedro Sim˜ oes,1,2Josiane Santos,1,3Inˆ es Fragata,1,4Laurence D. Mueller,5,6Michael R. Rose,5,7
and Margarida Matos1,8
1Centro de Biologia Ambiental, Departamento de Biologia Animal, Faculdade de Ciˆ encias da Universidade de Lisboa,
Campo Grande, 1749-016 Lisboa, Portugal
2E-mail: pmsimoes@fc.ul.pt
3E-mail: jmssantos@fc.ul.pt
4E-mail: if@fc.ul.pt
5Department of Ecology and Evolutionary Biology, University of California, Irvine, California 92697
6E-mail: ldmuelle@uci.edu
7E-mail: mrrose@uci.edu
8E-mail: mmatos@fc.ul.pt
Received June 20, 2007
Accepted April 28, 2008
The importance of contingency versus predictability in evolution has been a long-standing issue, particularly the interaction be-
tween genetic background, founder effects, and selection. Here we address experimentally the effects of genetic background and
founder events on the repeatability of laboratory adaptation in Drosophila subobscura populations for several functional traits.
We found disparate starting points for adaptation among laboratory populations derived from independently sampled wild pop-
ulations for all traits. With respect to the subsequent evolutionary rate during laboratory adaptation, starvation resistance varied
considerably among foundations such that the outcome of laboratory evolution is rather unpredictable for this particular trait,
even in direction. In contrast, the laboratory evolution of traits closely related to fitness was less contingent on the circumstances
of foundation. These findings suggest that the initial laboratory evolution of weakly selected characters may be unpredictable,
even when the key adaptations under evolutionary domestication are predictable with respect to their trajectories.
KEY WORDS: Adaptation, Drosophila subobscura, evolutionary contingency, founder effects, genetic background, life-history
traits, repeatability.
Evolutionary contingencies can be a source of differentiation
among populations (Travisano et al. 1995; Joshi et al. 2003).
In particular, differences in adaptive dynamics have been shown
betweenpopulationswithdifferentancestorsthatshareacommon
environmentinwhichtheyundergosubsequentadaptation(Cohan
1984a; Cohan and Hoffmann 1986, 1989). Several factors may be
involved in such contingent differentiation. Small differences in
the course of selection might result in substantially different evo-
lutionary outcomes, particularly in populations studied without
good environmental controls. Different genetic backgrounds may
1817
C ?2008 The Author(s). Journal compilation C ?2008 The Society for the Study of Evolution.
Evolution 62-8: 1817–1829

Page 2

PEDRO SIM˜OES ET AL.
explain disparate adaptive responses among populations exposed
to the same selective pressures (Cohan and Hoffmann 1989; de
Brito et al. 2005), particularly in carefully controlled laboratory
studies.
Such differences among genetic backgrounds may specifi-
cally involve differences in additive genetic components of vari-
anceandcovariance,nonadditiveinteractionsamongloci,orlink-
age disequilibrium (Lande 1980; Cheverud and Routman 1995;
Falconer and Mackay 1996; Roff 2000; Wade et al. 2001; Step-
pan et al. 2002). These genetic differences may be due both to the
past selective history and to random allele frequency changes as-
sociated with sampling processes, particularly with bottlenecked
populationsizes(Bryantetal.1986;Cheverudetal.1999;Naciri-
Graven and Goudet 2003; Zhang et al. 2004). Furthermore, in-
teractions between directional selection and genetic drift during
the evolutionary process might increase the impact of different
genetic backgrounds (Cohan 1984b; Cohan and Hoffmann 1989;
de Brito et al. 2005).
Experimentalevolutionstudiesareparticularlywellsuitedto
testtherelevanceofgeneticbackgroundeffectsinadaptiveevolu-
tion, because they allow us to reproducibly measure the adaptive
response of replicated populations to defined environmental con-
ditions (Rose et al. 1996; Lenski 2004; Chippindale 2006). The
evolutionary responses to a laboratory selection regime can be
measured relative to control populations already adapted to the
laboratory.Severallaboratoryexperimentshaveaddressedtheim-
portance of genetic background effects in Drosophila laboratory
evolution (Cohan and Hoffmann 1986; Teot´ onio and Rose 2000;
Teot´ onio et al. 2002; Joshi et al. 2003; see reviews by Prasad and
Joshi 2003; Rose et al. 2004). Overall, these studies have found
some cases in which different genetic backgrounds clearly lead
to different evolutionary patterns during adaptation to a common
environment(e.g.,CohanandHoffmann1986;Teot´ onioandRose
2000).
Our team has focused on the study of adaptation to the lab-
oratory in Drosophila subobscura, studying the evolutionary tra-
jectoriesofseverallife-historytraitsinpopulationscollectedfrom
differentnaturallocations(seeMatosetal.2000a,2002).Wehave
found evidence of variation in adaptive response during the first
generations of laboratory adaptation between two sets of popu-
lations founded from the same natural location six years apart
(Matos et al. 2002). In a subsequent study, we analyzed the evolu-
tionary dynamics of two different sets of laboratory populations
derived from synchronous foundations obtained in 2001 from
two different natural sites: “AR” populations derived from a wild-
caught sample from Arr´ abida, Portugal, and “TW” populations
from Sintra, Portugal (see Sim˜ oes et al. 2007). This study re-
vealed a clear adaptive response in fecundity-related traits as well
as significant differences in the evolutionary dynamics of these
twosetsofpopulationsovertheirfirst14generationsoflaboratory
culture, most likely as a result of different initial genetic compo-
sition of the populations. In spite of the clear results observed in
that study, we could not specifically determine whether the differ-
ences in evolutionary dynamics observed were due to the genetic
composition of the wild source populations or to sampling effects
arising at the time of sample collection. Differences in genetic
variation due to a limited number of field-collected individuals
are expected to affect the subsequent evolutionary response of
populations (James 1971; Powell and Richmond 1974; Reznick
and Ghalambor 2005). Nevertheless, the importance of possible
sampling effects for variation in the laboratory adaptation process
is not easily predictable.
In this study we aim to test the repeatability of an adap-
tive process and the impact of contingent factors, such as chance
eventsaffectingthe initial genetic backgroundthat mayoccurdue
to either geographical or temporal shifts in the genetic composi-
tion of populations. For this purpose, we present a broad analysis
of the initial stages of laboratory evolution combining data from
different sets of populations obtained from collections across dif-
ferent years (1998, 2001, and 2005) and different geographical
locations (Arr´ abida and Sintra). We also compare the sensitivity
of adaptive processes to differences arising from the geographi-
cal location of wild source populations (Arr´ abida vs. Sintra) with
the impact of sampling effects among populations derived from
the same wild source, in synchronous evolving populations (2005
populations).
The main questions addressed in this study are:
(1) Is there repeatability during laboratory evolution?
(2) Do contingent factors associated with the foundation process
affect the evolutionary dynamics observed?
(3) If so, are these contingencies derived from temporal and/or
spatial differences in the genetic composition of the populations?
Do genetic sampling effects play a role in the evolutionary differ-
ences between populations?
Materials and Methods
FOUNDATION AND MAINTENANCE OF THE
LABORATORY POPULATIONS
This study includes data from seven different sets of wild-caught
samples of D. subobscura. These different sets of populations
were founded in the calendar years 1998 (NW populations;
see Matos et al. 2002), 2001 (AR and TW populations; see
Sim˜ oes et al. 2007), and 2005 (FWA, FWB, NARA, and NARB
populations, the new data presented here)—see Table 1 for an
overview of the experimental populations analyzed. Both NW
and TW populations were collected from a pinewood near Sintra,
Portugal whereas AR populations were collected from Arr´ abida,
Portugal—see Sim˜ oes et al. 2007.
1818
EVOLUTION AUGUST 2008

Page 3

REPEATABILITY OF ADAPTIVE EVOLUTION
Table 1. General description of the experimental populations
used in this study.
Foundation Location Year
of
sampling sampling
Sub-
samples founder population
females sizes
No. of Average
of
NW
TW
AR
NARA
NARB
FWA
FWB
Sintra
Sintra
Arr´ abida 2001
Arr´ abida 2005
Arr´ abida 2005
Sintra
Sintra
1998
2001
–
–
–
a
b
a
b
300
110
59
55
68
60
75
895.3
870.5
795.6
952.5
928.3
926.7
856.7
2005
2005
The additional foundations reported here were performed
in April 2005 and consisted of two independent collections
from each of the two previously sampled natural sites: Sintra,
Portugal—“FWA”and“FWB”populations;Arr´ abida,Portugal—
“NARA” and NARB.”
All populations were maintained under the same conditions:
discrete generations of 28 days, reproduction close to peak fe-
cundity, controlled temperature of 18◦C, with a 12-h L: 12-h D
photoperiod. Flies were kept in vials, with controlled adult den-
sities of around 50 individuals per vial and larval densities of
around 80 per vial. At each generation, emergences from the
several vials within each replicate population were randomized
using CO2anesthesia. Adult population sizes ranged, in general,
between 600 and 1200 individuals.
Two generations after foundation, each population was split
into three replicate populations (e.g., FWA1–3 designating the
three populations of the regime “FWA”), except the NW founda-
tion, split into five replicates. A set of longer established labora-
tory populations “NB” was used as a control for all the experi-
mental populations referred above. NB populations were at their
90thlaboratory generation when NW populations were founded
(Matos et al. 2002). At the time the 2001 AR and TW populations
were introduced into the laboratory, the NB populations were at
their 136thgeneration (Sim˜ oes et al. 2007). As for the 2005 col-
lections, the NB populations were at their 181stgeneration at the
time of their initial foundation. The early adaptation data ana-
lyzed in this study corresponds to the first 15 generations of NW
cultureandthefirst20generationsofARandTWlaboratorypop-
ulations. All new populations cultured from the 2005 foundations
were cultured for 21 generations, during which their adaptation
to laboratory conditions was also studied by means of phenotypic
assays.
LIFE-HISTORY TRAIT ASSAYS
In each generation assayed an additional egg collection was made
for the phenotypic assay, using the same basic procedure as de-
scribedaboveforthestandardmaintenanceofpopulations.Mated
pairs of flies used individuals emerging in the same day, and were
formed less than 6 h after eclosion started (stimulated by the
light phase). These pairs were transferred daily to vials contain-
ing fresh medium, and the total number of eggs laid per female
was counted daily for the first 12 days. After the fecundity assay
wasperformed,eachpairofflieswastransferredtoavialcontain-
ing plain agar medium where the number of hours of starvation
resistance was measured. Four characters were analyzed: age of
first reproduction (number of days between emergence and the
day of first egg laying), early fecundity (total number of eggs laid
during the first week), peak fecundity (total number of eggs laid
between days 8 and 12) and female starvation resistance (number
of hours until death, registered every 6 h).
Assays were performed at generations 4, 8, 13, and 15 of
NW laboratory culture and at generations 3, 4, 6, 7, 12, 14, 18,
and 20 of AR and TW populations. Phenotypic assays on the
2005 populations were carried out during generations 3, 6, 10,
12, 15, and 21. Sample sizes ranged between 14 and 21 pairs per
replicate population. All assays involved synchronous analyses
with NB populations.
STATISTICAL METHODS
Thesamplingmethodsusedinthisstudyresultedinanaturalhier-
archy of effects. At the very top of this hierarchy are samples that
were collected in different years, 1998, 2001, and 2005 (Fig. 1).
Within years are samples of flies from two different geographic
locations, Sintra and Arr´ abida (except for 1998, with a founda-
tion from Sintra only). In 2005 two collections were done in each
location (designated as subsamples in each location: FWA and
FWB from Sintra; and NARA and NARB from Arr´ abida). Every
sample from a given location or a given subsample was replicated
in the laboratory giving rise to three independent evolving lab-
oratory populations, except in the case of the 1998 foundation
that was fivefold replicated. The common replicates from a sin-
gle subsample or a single sample from one location are called a
foundation.
Thedatausedinvolvedphenotypicassaysforeachtrait(using
information at the individual level) concerning the initial adaptive
process of the following laboratory foundations: NW (foundation
in 1998; generations 4–15); AR and TW (foundations in 2001;
generations 3–20); FWA, FWB, NARA and NARB (foundations
in 2005; generations 3–21)—see also Figure 1. Differences rel-
ative to the average of the same-numbered NB replicate popula-
tion(assayedsynchronouslywithexperimentalpopulations)were
used as input data for all the analyses.
Differences to a reference population (NB populations) were
used to remove any inadvertent evolutionary change arising inde-
pendentlyfromtheprocessoflaboratoryadaptationandtoreduce
the confounding effects of environmental heterogeneity between
temporallyspacedphenotypicassaysthatmightobscuretheactual
EVOLUTION AUGUST 2008
1819

Page 4

PEDRO SIM˜OES ET AL.
Figure 1. The sampling hierarchy used in this study. The circle and
square symbols indicate the location of the Sintra and Arr´ abida lo-
cales in Portugal. Levels of hierarchy include year of foundation
(1998, 2001, and 2005), location (Sintra and Arr´ abida) and subsam-
ples (two collections from each location, in 2005). All foundations
were threefold replicated, except the 1998 foundation (NW) which
had five replicates. In both 2001 (AR from Sintra and TW from
Arr´ abida) and 2005 (FWA, FWB from Sintra and NARA, NARB from
Arr´ abida), foundations from the different geographical sources
were carried out synchronously, using the same pinewood at each
location.
evolutionary trends (vid. Matos et al. 2002; Sim˜ oes et al. 2007).
In fact, we have detected signs of environmental trends during
the time period of each study, reinforcing the importance of us-
ingreferencepopulationsalreadystabilizedintermsoflaboratory
adaptation.Inanycase,theaverageNBvalueswererelativelysta-
ble among studies being, respectively for 1998, 2001, and 2005
(average ± SE for each study): (1) Age of first reproduction:
3.0477 ± 0.0464; 2.9881 ± 0.1454; 3.0357 ± 0.1166; (2) Early
Fecundity: 122.6216 ± 4.0033; 121.0092 ± 5.0288; 128.5320
± 9.4313; (3) Peak Fecundity: 172.5627 ± 4.2603; 142.0682
± 2.0639; 165.0211 ± 6.8953; (4) Female starvation resistance:
43.8457 ± 1.3466; 42.5735 ± 1.1314; 39.3593 ± 1.3300.
Bootstrap Techniques: To study the effects of sampling hi-
erarchy on the slope of the selection response we used bootstrap
resampling techniques. Let the phenotype for year-i, location-
j, subsample-k, replicate-l, and individual-v be yijklv. Because
the goal is to predict these phenotypes from the generations
of selection we let xijklv be the generation in which pheno-
type yijklv was measured. In a particular replicate population
standard linear regression techniques can be used to estimate
the regression coefficients (ˆβijkl). From these regression coeffi-
cients residual error terms are determined from the relationship,
ˆ εijklv= yijklv−ˆβ1ijkl− xijklvˆβ2ijkl, v = 1, 2, ···, nijkl. These error
termswereusedtocreatethebootstrapsamples.Foreachreplicate
population an empirical distribution of error terms was estimated,
ˆ εijkl=?ˆ εijkl1,···,ˆ εijklnijkl
ing nijklsamples with replacement from ˆ εijkl. From the bootstrap
sample of error terms a vector of phenotypes is generated by,
y∗
strap intercept (ˆβ∗
regression techniques. This process was repeated to generate B
bootstrap intercept (ˆβ∗b
Bootstrappingresidualsrequiresthatthevarianceoftheerror
terms be the same over all generations (Efron and Tibshirani
1993). For the composite phenotype this assumption appears to
bewellmet.Asanexampletheresidualerrorsforthefirstreplicate
Sintra population in 1998 is shown in Figure 2.
Variation across levels of the sampling hierarchy: To deter-
mine the importance of the sampling hierarchy to variation in the
adaptive response we created different samples from the basic
bootstrap intercepts and slopes. Because the techniques are the
same with intercepts and slopes we illustrate the methods with
slopes. A random sample of 1000 bootstrap slopes,ˆβ∗b
chosen from each of the 23 different replicate populations with
each replicate (l = 1,2, ···, 23), and slope (b = 1, ···, 1000) hav-
ing an equal chance of being chosen. With this particular sample
we then computed the mean of the slopes for each collection from
the same replicate population. We then subtracted the mean from
the slopes within each replicate to obtain a centered slope value.
This allows us to compare all slopes on the same scale.
?.
Abootstrapsample,ε∗=
?
ε∗
1,···,ε∗
nijkl
?
,wascreatedbytak-
v=ˆβ1ijkl+ xijklvˆβ2ijkl+ ε∗
v. From the vector of y∗’s a new boot-
1) and slope (ˆβ∗
2) is estimated by standard linear
1ijkl) and slope estimates,ˆβ∗b
2ijkl, b = 1,···,B.
2ijkl, were
Figure 2. Residual errors from the bootstrap resampling analysis
for the NW1replicate population (Sintra, 1998 foundation).
1820
EVOLUTION AUGUST 2008

Page 5

REPEATABILITY OF ADAPTIVE EVOLUTION
To compute the variation over levels of foundations a sample
of 1000 slopes (or intercepts) was taken from all 23 replicates.
The mean of those slopes that were from a common foundation
were then computed and subtracted from the appropriate slopes.
Inasimilarfashionthesubsamplevariationwascomputedby
taking a sample of 1000 slopes from all the replicate populations
in 2005. Then the mean of those slopes that originated from the
samelocationwascomputedandthensubtractedfromeachslope.
Similar methods were followed to compute the centered slopes
and intercepts for locations and time.
To compute the centered slopes and intercepts for different
locationswedidnotuseSintra1998intheanalysis.Consequently,
there were 23 − 5 = 18 different replicate populations represent-
ing Sintra and Arr´ abida in 2001 and 2005. For each of these 18
populations there were 1000 bootstrap slopes. To estimate the
variances we sampled at random 1000 slopes from these 18 pop-
ulations, thus each replicate population had an equal chance of
being included in this sample. Not all of the bootstrap replicates
wereusedbecauseothercomparisonshadmore(e.g.,foundations
and replicate variation) or less (subsamples) than 18 populations,
so we wanted each level to be based on the same number of
observations.
These 1000 randomly chosen slopes were divided into two
groups based on the population they were sampled from: (1)
Sintra 2001 and Arr´ abida 2001 and (2) Sintra 2005 and Arr´ abida
2005. The mean of the first group was subtracted from all the
observations in the first group and likewise we subtracted the
mean of the second group from the observations in the second
group. These centered slopes then reflect the variation among
locations with a common year of sampling. The variance in the
figures is the variance of these 1000 centered values.
For variation among years the same 18 populations were
used but the two groups were (1) Sintra 2001 and 2005, and (2)
Arr´ abida 2001 and 2005. So now the variation around the mean
reflected variation among different years.
Effectsofhierarchyonratesofevolution:Thebootstrapsam-
ples can also be used to test for significant differences in the slope
of the selection response and in the y-intercept. Our procedure
follows the methods used in the previous section. For instance to
testfordifferencesbetweensubsampleswesampledonebootstrap
slopeorinterceptfromeachreplicatewithineachsubsamplefrom
acommonlocation.Themeanofslopesfromeachsubsamplewas
then computed and their average was taken and saved. This pro-
cess was repeated 1000 times. A statistical test consists of com-
puting the fraction of these 1000 differences that are greater than
zero. Two times this fraction or 1 minus this fraction (whichever
is less) constitute the P-value for significant differences.
Composite phenotype: The effects of laboratory adaptation
on the phenotypes measured in this study are almost certainly
not independent. It is likely that there are genetic correlations be-
tween traits like early fecundity and peak fecundity. Alternatively
each of these traits may contribute to fitness in these laboratory
environments to different degrees and thus all would be affected
by natural selection. It is worthwhile to consider a composite of
all the life-history traits and follow the change in this composite
trait. This composite trait would also reduce the number of vari-
ables tested and simplify the analysis of the basic evolutionary
questions.
Onewaytodeduceanappropriatecompositetraitistoempir-
ically determine what trait is changed by selection to the greatest
extent. We have done this by using all four female phenotypes
(standardized as difference relative to the NB population) from
each of the 23 replicate samples. In each replicate we used phe-
notype measurements at the earliest generation, called the start
of selection, and phenotype measurements at the last generation,
called the end of selection. Using linear discriminant analysis,
the linear combination of all four traits that produced the largest
separation between the start and the end of selection was found
(Morrison 1976).
A composite phenotype is defined from the coefficients in
Table 2 as
y =
4
?
i=1
pili,
(1)
where piis the ith phenotype and liis the ith linear discriminant
coefficient. There are some general trends that fall out of the sign
and magnitude of the discriminant functions. Age of first repro-
duction and female starvation resistance are roughly on the same
scale and they have the same negative sign so the composite func-
tion will increase as these two phenotypes decrease. However,
the magnitude of the coefficient for age of first reproduction is
about 1000 times larger and thus the value of the composite phe-
notype depends to a much greater degree on this phenotype than
onfemalestarvationresistance.Thetwofecundityphenotypesare
roughly on the same scale but of opposite sign. Thus, the com-
posite phenotype increases with increasing early fecundity and
decreasing peak fecundity. However, because the coefficient for
early fecundity is roughly three times the peak fecundity coeffi-
cient it contributes much more to the final value of the composite
Table 2. The linear discriminant coefficients for the four stan-
dardized female life-history traits obtained from the discriminant
analysis.
Phenotype Linear discriminant
coefficient (li)
Age of first reproduction
Early fecundity
Peak fecundity
Female starvation resistance
−0.409
0.0214
−0.00883
−0.000510
EVOLUTION AUGUST 2008
1821

Page 6

PEDRO SIM˜OES ET AL.
Figure 3. Distribution of the composite female phenotype. This
distribution follows from applying equation (1) to data collected
at the start and at the end of laboratory selection over all replicate
populations.
phenotype. Overall, age of first reproduction and early fecundity
contribute the most to the numerical value of the discriminant
function. The distribution of the composite phenotype shows sub-
stantial differentiation between the start and end of laboratory
selection (Fig. 3).
Results
Theresponsetolaboratoryadaptationoftheindividuallife-history
traits is variable between traits and levels of sampling. Early
fecundityandpeakfecunditygenerallyincreaseovertimewhereas
age of first reproduction typically declines over time (see Fig. 4).
These changes are in accord with our general expectations for
improvements in fitness. On the other hand female starvation
resistance may increase, decrease, or not change at all, depending
on the source population and year (see Fig. 4D).
The progression of the composite phenotype shows consis-
tent increases over all locations and years (Fig. 5) suggesting that
this measure is responding consistently to laboratory adaptation
(see also Fig. 3). We next examine if the response to selection
varies across subsamples (within each location), locations, or
years.
By looking at variation within each sampling level we can
get an impression of the magnitude of variation that originated
with each hierarchical level of sampling. One of the most vari-
ably evolving single phenotypes is female starvation resistance
(Fig. 6). The slope for this character shows a large change from
theindividualreplicatestothedifferencesamongreplicateswithin
foundations and then smaller increases at the higher levels of sub-
sample and time (Fig. 6A). The changes are still more dramatic
for the intercept of female starvation resistance (Fig. 6B). There
is a considerable increase in the variation of this parameter pro-
ceeding from the individual replicate level to differences among
replicates and then a further, although smaller, increase from that
level to the higher levels of the analysis, with the exception of
effects of location.
The composite phenotype shows much less variation due
to the different sampling levels (Fig. 7). Variation in slopes in-
creases roughly threefold between the individual replicate and
among replicates within the same foundation (Fig. 7A). How-
ever, the other sampling levels (subsamples, location and year)
have no more variation than is seen at the level of the foun-
dation. This would suggest that most of the variability among
these inferred evolutionary trajectories arises between indepen-
dent replicate populations founded from a common wild-caught
sample.
Regression intercepts for the composite phenotype are more
affectedbylevelsofsamplinghierarchy(Fig.7B).Variationatthe
foundation level presents also a roughly threefold increase rela-
tive to the variation of individual replicates. However, variation is
further increased by both subsamples and spatially distinct sam-
ples (Fig. 7B). The trend in Figures 6 and 7 and for the other traits
not shown is that higher sampling levels add greater variation to
the intercept than they do to the slope of the selection response.
A summary of the tests for significant differences in slopes
at the different sampling levels is given in Table 3. The individual
phenotypes show significant effects at 1 to 5 of the 7 categories.
Among the individual phenotypes there is at least one example of
significant effects from each level of hierarchy. These significant
effects are more frequently associated with female starvation re-
sistance,whichshowedsignificanteffectsatalllevelsofhierarchy
except location. In contrast, fecundity-related traits as well as the
composite phenotype only presented significant differences for 1
or 2 comparisons (see Table 3). We found three significant effects
of location out of five tests, all involving fecundity-related traits
as well as the composite phenotype. On the other hand, for these
traits there was no significant effect of the subsample level, in
contrast with the results obtained for female starvation resistance.
Theinterceptrepresentsthephenotypevalueatthestartofthe
selection experiments. The evolutionary intercepts for individual
phenotypes show significant effects at 3 to 4 of the 7 categories,
withnumeroussignificanteffectsofsamplinghierarchy(Table4).
This indicates that temporal and spatial sampling produces popu-
lations whose initial phenotypes are significantly different.
Discussion
Our results clearly support the importance of contingency for the
starting point of adaptive evolution. Contingent effects can appar-
ently arise from both temporal and/or spatial effects during the
process of sampling from the wild, changing among populations
1822
EVOLUTION AUGUST 2008

Page 7

REPEATABILITY OF ADAPTIVE EVOLUTION
even within the same geographical region. This variation occurs
predominantly in the initial performance of populations and, to a
lesser extent, in their subsequent evolutionary trajectory. Female
starvation resistance shows clearly a more contingent pattern of
evolution both in the differences of initial state and early adap-
tive rates between foundations, even changing sign. The more
consistent general response to laboratory adaptation of fecundity
characters is expected because of their relevance to fitness. In
Figure 4. Evolutionary response of individual life-history traits to laboratory selection. (A) Age of first reproduction; (B) Early fecundity;
(C) Peak fecundity; (D) Female starvation resistance. The symbols show the average values in each location and year. For 2005 the
averages for each location included both subsamples. The lines show the linear fit to the phenotypic change in populations derived from
each location.
the same vein, the composite phenotype shows generally parallel
evolution across temporal foundations.
As a whole, these results clearly highlight the prominent
role of fecundity characters during laboratory adaptation that has
also been observed in other recent studies (see also Hercus and
Hoffmann 1999; Sgr` o and Partridge 2000; Matos et al. 2002;
Sim˜ oes et al. 2007). On the other hand, the importance of stress
resistance characters is apparently equivocal. In fact, it is not
EVOLUTION AUGUST 2008
1823

Page 8

PEDRO SIM˜OES ET AL.
Figure 4. Continued.
possible to generalize or predict the evolutionary patterns during
laboratory adaptation for this particular trait, given its varying
evolutionary responses to the year of sampling from the wild (see
Tables 3 and 4 and Fig. 4).
The differences in the relative impact of these evolutionary
contingenciesamongtraitsappearthustodependontherelevance
of the specific trait for the overall performance of a population
in this new environment. In particular, the relationship between
starvation resistance and fitness is probably complex or weak
relative to that of fecundity traits (Hoffmann et al. 2001; Matos
et al. 2002; Sim˜ oes et al. 2007). This would explain our find-
ing that the evolutionary response of resistance traits is more
contingent on the initial composition of the laboratory popu-
lations, leading to significant variation in evolutionary patterns
across foundations. This might also in turn explain disparities
between laboratories with respect to changes in starvation resis-
tance during captivity (e.g., Hoffmann et al. 2001, Griffiths et al.
2005).
1824
EVOLUTION AUGUST 2008

Page 9

REPEATABILITY OF ADAPTIVE EVOLUTION
Figure 5. The response of the composite phenotype to labora-
tory selection. The symbols show the mean composite phenotype
calculated from equation (1) in each location and year. For 2005,
the averages for each location included both subsamples. The lines
show the linear fit to the phenotypic change in populations de-
rived from each location.
CONSISTENCY OF ADAPTATION DURING
LABORATORY EVOLUTION
This study reveals substantial variation in the adaptive response
to a new environment, both at the start of adaptation as well
as, for weakly selected characters, in initial evolutionary rates
among foundations. Disparities in rates of adaptation have been
inferred in all our previous studies, which have separately treated
temporal (e.g., Matos et al. 2002) and spatial (e.g., Sim˜ oes et al.
2007) effects. The present study involves both further replication
of that earlier work and an overall analysis of the entire series of
experiments, including seven independent foundations stretching
over almost 10 years. As such, our analysis provides the most
complete analysis of the adaptive evolution of a species in re-
sponse to the same selective regime, when multiple foundations
from wild populations at different locations and different times
are employed.
The results of the present study also provide a useful contrast
with those of other authors who started their laboratory evolution
experiments from different source populations. For example, Co-
han and Hoffmann (1986) found that Drosophila melanogaster
populations derived from collections obtained from wild popula-
tions at very different latitudes along the west side North Amer-
ican coast showed different correlated responses to selection for
knockdown resistance to ethanol. Teot´ onio and his collaborators
(Teot´ onio and Rose 2000; Teot´ onio et al. 2002) performed a re-
verse evolution experiment, in which genetically differentiated
populations were returned to their common ancestral environ-
ment and then allowed to evolve in parallel for 50 generations.
They found significant heterogeneity among the evolutionary tra-
jectories of these populations in the same environment. Teot´ onio
et al. concluded that past selective history along with a variable
relationship between life-history characters and fitness were re-
sponsible for the heterogeneity that they observed. Both of their
studies indicated effects of genetic background on evolutionary
response.
It is possible that the differences that we observe in the initial
state and, to some extent, in the subsequent process of adapta-
tion between our laboratory populations are reduced in the long
term. In fact, our previous analyses of the 2001 foundations from
Arr´ abida and Sintra showed significantly different evolutionary
rates of adaptation in the short term no longer obtained with more
generations (see Sim˜ oes et al. 2007). In this sense, our results
are more in keeping with other experimental evolution studies in
Drosophila, which have shown that the effects of evolutionary
history are transient (e.g., Joshi et al. 2003).
The long-standing experimental evolution work of Lenski
and colleagues using E. coli has also addressed the effects of his-
torical contingencies in evolution (Travisano et al. 1995; Elena
and Lenski 2003). In one particular experiment, E. coli lines that
had previously evolved in glucose for 2000 generations were
placed in a maltose environment for anot her 1000 generations.
Theevolutionaryresponseinthisnewenvironmentwasmeasured
in 36 E. coli populations—as the design employed three repli-
cate populations from each of the 12 populations that previously
evolvedinglucose(seeTravisanoetal.1995).Theresultsshowed
that the replicate derivatives of these populations evolving in the
new maltose environment achieved similar fitness levels despite
prior history and/or subsequent chance events. On the other hand,
EVOLUTION AUGUST 2008
1825

Page 10

PEDRO SIM˜OES ET AL.
Figure 6. The distribution of the parameters slope and intercept over levels of hierarchy for female starvation resistance: (A) distribution
of slopes; (B) distribution of intercepts. The distributions are normalized so their area sums to 1. Slopes within a given level of hierarchy
are centered about their mean. The variance of each level is given as σ2. The “replicates” level corresponds to the variation obtained
for each individual replicate whereas the “foundations” level refers to the variation between replicates within the same foundation,
“subsample” level refers to the variation between replicate populations from different subsamples (within each location), and the same
rationale applies to the two higher hierarchical levels (location and time).
Figure 7. The distribution of the slope and intercept parameters over levels of hierarchy for the composite phenotype: (A) distribution of
slopes; (B) distribution of intercepts. The distributions are normalized so that their area sums to 1. Slopes within a given level of hierarchy
are centered about their mean. The variance of each level is given as σ2. The “replicates” level corresponds to the variation obtained
for each individual replicate whereas the “foundations” level refers to the variation between replicates within the same foundation,
“subsample” level refers to the variation between replicate populations from different subsamples (within each location), and the same
rationale applies to the two higher hierarchical levels (location and time).
1826
EVOLUTION AUGUST 2008

Page 11

REPEATABILITY OF ADAPTIVE EVOLUTION
Table 3. The P-values for tests of significant differences between slopes for different levels of sampling using bootstrap analysis. Tests
with P-values less than 5% are shown in bold face type. The comparisons between locations involved the 2001 and 2005 data.
Trait P-values
LevelTest AgeEarly
fecundity
Peak
fecundity
Female starvation
resistance
Composite
SubsampleArr´ abida
Sintra
Arr´ abida-Sintra
Arr´ abida, 2001–2005
Sintra, 1998–2001
Sintra, 2001–2005
Sintra, 1998–2005
0.31
0.33
0.004
0.85
0.026
0.27
0.16
0.68
0.33
0.18
0.022
0.010
0.098
0.11
0.12
0.082
0.01
0.16
0.28
0.75
0.18
0.022
0.95
0.74
0.008
0.018
0.006
<0.001
0.30
0.25
0.024
0.092
0.98
0.31
0.47
Location
Year
chance and historical events were found to have a clear impact
on traits less correlated with overall fitness (e.g., cell size), lead-
ing to differentiation among populations in these particular traits
(Travisano et al. 1995). Our data similarly suggests that evolu-
tionary contingencies have more impact on traits less relevant to
fitness. Specifically, starvation resistance, a trait expected to have
lessdirectimpactonfitnessinthelaboratoryenvironmentrelative
tofecunditytraits,showsanevolutionarypatternthatismorecon-
tingent on the effects of foundation. Nevertheless, comparisons
of studies in sexual populations with those involving asexual pop-
ulations should be made with caution, given that the former do
not allow us to disentangle clearly the effects of prior evolu-
tionary history from other sources of initial genetic differences
between founder populations (e.g., sampling effects upon foun-
dation). Studies involving more systematic multiple samplings
from the same natural populations—using replicate foundations
within each given year of sampling (as done for the 2005 data)—
may improve our capacity to distinguish between these several
sources of variation.
Table 4. The P-values for tests of significant differences between intercepts for different levels of sampling using bootstrap analysis.
Tests with P-values less than 5% are shown in bold face type. The comparisons between locations involved the 2001 and 2005 data.
Trait P-values
LevelTestAgeEarly
fecundity
Peak
fecundity
Female starvation
resistance
Composite
Subsample Arr´ abida
Sintra
Arr´ abida-Sintra
Arr´ abida, 2001–2005
Sintra, 1998–2001
Sintra, 2001–2005
Sintra, 1998–2005
0.34
0.14
0.010
0.088
<0.001
0.20
<0.001
0.53
0.026
0.11
0.10
0.12
0.012
<0.001
0.76
0.002
0.11
0.44
0.004
0.002
0.69
0.35
0.60
0.47
<0.001
<0.001
<0.001
<0.001
0.29
0.066
0.016
0.66
<0.001
<0.001
<0.001
Location
Year
WHAT CAN WE SAY ABOUT THE WILD POPULATIONS
THAT WE SAMPLE AND SUBJECT TO LABORATORY
EVOLUTION?
The present study is different from those of Teot´ onio and col-
leagues (e.g., Teot´ onio and Rose 2000; Teot´ onio et al. 2002) or
Lenski and colleagues (Travisano et al. 1995) in that the ge-
netic differentiation that our experimental populations start with
comes from nature. This naturally raises the issue of the relation-
shipbetweenlaboratorystudiesofexperimentalevolutionandthe
properties of populations in the wild.
Some authors have argued that the use of long-established
laboratory populations limits inferences about evolutionary pro-
cesses (Harshman and Hoffmann 2000; Linnen et al. 2001). This
view assumes that experimental evolution studies aim at extrapo-
lating specific evolutionary patterns shown by laboratory popula-
tions to populations in the wild. However, this is not necessarily
the goal of such studies. Experimental evolution often focuses
primarily on tests of predictions regarding evolutionary processes
in general. In our view, any particular laboratory environment is
EVOLUTION AUGUST 2008
1827

Page 12

PEDRO SIM˜OES ET AL.
just another environment featuring a particular selection regime
to which populations may adapt (Matos et al. 2000b). Moreover,
we propose that the laboratory environment can in fact be an
ideal setting to address the potential for adaptive responses and
to test general predictions concerning evolutionary patterns, such
as convergence. The analysis of the laboratory adaptation of re-
cently wild-collected samples allows experimenters to study in
detail the evolutionary response of populations with high starting
genetic variability as well as to address issues such as the impact
of different genetic backgroundson evolution, or the repeatability
of evolutionary patterns across temporally and spatially sampled
populations, as we have here.
OurfindingssuggestthatD.subobscurapopulationscanvary
significantlyintheirinitialperformanceinanovellaboratoryenvi-
ronment even when sampled from a relatively small geographical
area over a relatively short period of time. Our long-term research
program provides a powerful and novel window into the potential
for adaptive evolution of populations in the wild, a window very
different from that provided by the collection of data pertaining
solely to standing genetic variation in the wild. The latter has
been a traditional research topic within population genetics, from
thepioneeringstudiesofDobzhanskyandLewontin(Dobzhansky
1937; Lewontin and Hubby 1966) to the recent attempts to detect
selection in natural populations (see Ford 2002, for a review). It is
also a different angle on this question from that afforded by stud-
ies of the phenotypics of selection in nature (e.g., Lande 1979;
Lande and Arnold 1983; Arnold and Wade 1984a,b; Grant and
Grant 1995; Reznick et al. 1997).
What we are suggesting is that, although the laboratory evo-
lution of a sample collected from a wild population is necessarily
limited to the particular selection regime(s) chosen by an exper-
imenter, it provides one of the most readily interpretable assays
of the potential for adaptation of a population. As such, this par-
ticular type of assay should perhaps be more common among the
experimental designs used in evolutionary research.
ACKNOWLEDGMENTS
This study was partially financed by “Fundac ¸˜ ao para a Ciˆ encia e a Tec-
nologia” (FCT) project no. POCTI/BSE/33673/2000 and by FCT and
POCI 2010 project no. POCI-PPCDT/BIA-BDE/55853/2004 (both with
co-participationofFEDER).JShadaBTIgrantandPShadaPh.D.grant
(SFRH/BD/10604/2002) from FCT.
LITERATURE CITED
Arnold, S. J., and M. J. Wade. 1984a. On the measurement of natural and
sexual selection: theory. Evolution 38:709–719.
———. 1984b. On the measurement of natural and sexual selection: applica-
tions. Evolution 38:720–734.
Bryant, E. H., S. A. McCommas, and L. M. Combs. 1986. The effect of
an experimental bottleneck upon quantitative genetic variation in the
housefly. Genetics 114:1191–1211.
Cheverud, J. M., and E. J. Routman. 1995. Epistasis and its contribution to
genetic variance components. Genetics 139:1455–1461.
Cheverud, J. M., T. T. Vaughn, L. S. Pletscher, K. King-Ellison, J. Bailiff,
E. Adams, C. Erickson, and A. Bonislawski. 1999. Epistasis and the
evolution of additive genetic variance in populations that pass through a
bottleneck. Evolution 53:1009–1018.
Chippindale, A. K. 2006. Experimental Evolution. Pp. 482–501 in C. Fox and
J. Wolf, eds. Evolutionary genetics: concepts and case studies. Oxford
Univ. Press, London.
Cohan, F. M. 1984a. Can uniform selection retard random genetic divergence
between isolated conspecific populations? Evolution 38:495–504.
———. 1984b. Genetic divergence under uniform selection. I. Similarity
among populations of Drosophila melanogaster in their responses to
artificial selection for modifiers of ciD. Evolution 38:55–71.
Cohan, F. M., and A. A. Hoffmann. 1986. Genetic divergence under uniform
selection.II.Differentresponsestoselectionforknockdownresistanceto
ethanol among Drosophila melanogaster populations and their replicate
lines. Genetics 114:145–163.
———.1989.Uniformselectionasadiversifyingforceinevolution:evidence
from Drosophila. Am. Nat. 134:613–637.
de Brito, R. A., L. S. Pletscher, and J. M. Cheverud. 2005. The evolution of
geneticarchitecture.I.Diversificationofgeneticbackgroundsbygenetic
drift. Evolution 59:2333–2342.
Dobzhansky, T. 1937. Genetics and the origin of species. Columbia Univ.
Press, New York.
Efron, B., and R. J. Tibshirani. 1993. An introduction to the bootstrap. Chap-
man & Hall, New York.
Elena, S. F., and R. E. Lenski. 2003. Evolution experiments with microor-
ganisms: the dynamics and genetic basis of adaptation. Nat. Rev. Genet.
4:457–469.
Falconer, D. S., and T. F. C. Mackay. 1996. Introduction to quantitative ge-
netics. 4th ed. Longman, Harlow.
Ford, M. J. 2002. Applications of selective neutrality tests to molecular ecol-
ogy. Mol. Ecol. 11:1245–1262.
Grant, P. R., and B. R. Grant. 1995. Predicting microevolutionary re-
sponsestodirectionalselectiononheritablevariation.Evolution49:241–
251.
Griffiths, J. A., M. Schiffer, and A. A. Hoffmann. 2005. Clinal variation and
laboratory adaptation in the rainforest species Drosophila birchii for
stress resistance, wing size, wing shape and development time. J. Evol.
Biol. 18:213–222.
Harshman, L. G., and A. A. Hoffmann. 2000. Laboratory selection experi-
ments using Drosophila: what do they really tell us? Trends Ecol. Evol.
15:32–36.
Hercus, M. J., and A. A. Hoffmann. 1999. Desiccation resistance in inter-
specific Drosophila crosses: genetic interactions and trait correlations.
Genetics 151:1493–1502.
Hoffmann, A. A., R. Hallas, C. Sinclair, and L. Partridge. 2001. Rapid loss
of stress resistance in Drosophila melanogaster under adaptation to
laboratory culture. Evolution 55:436–438.
James, J. W. 1971. The founder effect and response to artificial selection.
Genet. Res. Camb. 16:241–250.
Joshi,A.,R.B.Castillo,andL.D.Mueller.2003.Thecontributionofancestry,
chance, and past and ongoing selection to adaptive evolution. J. Genet.
82:147–162.
Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, ap-
plied to brain: body size allometry. Evolution 33:402–416.
———. 1980. The genetic covariance between characters maintained by
pleiotropic mutations. Genetics 94:203–215.
Lande,R.,andS.J.Arnold.1983.Themeasurementofselectiononcorrelated
characters. Evolution 37:1210–1226.
1828
EVOLUTION AUGUST 2008

Page 13

REPEATABILITY OF ADAPTIVE EVOLUTION
Lenski, R. E. 2004. Phenotypic and genomic evolution during a 20,000-
generationexperimentwiththebacteriumEscherichiacoli.PlantBreed-
ing Rev. 24:225–265.
Lewontin, R. C., and J. L. Hubby. 1966. A molecular approach to the study
of genic heterozygosity in natural populations. II. The amount of varia-
tion and degree of heterozygosity in natural populations of Drosophila
pseudoobscura. Genetics 54:595–609.
Linnen, C., M. Tatar, and D. Promislow. 2001. Cultural artifacts: a com-
parison of senescence in natural, laboratory-adapted and artificially
selected lines of Drosophila melanogaster. Evol. Ecol. Res. 3:877–
888.
Matos, M., M. R. Rose, M. T. Rocha Pit´ e, C. Rego, and T. Avelar. 2000a.
Adaptation to the laboratory environment in Drosophila subobscura. J.
Evol. Biol. 13:9–19.
Matos, M., C. Rego, A. Levy, H. Teot´ onio, and M. R. Rose. 2000b. An
evolutionary no man’s land. Trends Ecol. Evol. 15:206.
Matos, M., T. Avelar, and M. R. Rose. 2002. Variation in the rate of conver-
gent evolution: adaptation to a laboratory environment in Drosophila
subobscura. J. Evol. Biol. 15:673–682.
Morrison, D. F. 1976. Multivariate statistical methods. 2nd ed. McGraw-Hill,
New York.
Naciri-Graven, Y., and J. Goudet. 2003. The additive genetic variance af-
ter bottlenecks is affected by the number of loci involved in epistatic
interactions. Evolution 57:706–716.
Powell, J. R., and R. C. Richmond. 1974. Founder effects and linkage dise-
quilibria in experimental populations of Drosophila. Proc. Natl. Acad.
Sci. USA 71:1663–1665.
Prasad, N. G., and A. Joshi. 2003. What have two decades of laboratory
life-history evolution studies on Drosophila melanogaster taught us? J.
Genet. 82:45–76.
Reznick, D. N., and C. K. Ghalambor. 2005. Selection in nature: experimen-
tal manipulations of natural populations. Integr. Comp. Biol. 45:456–
462.
Reznick, D. N., F. H. Shaw, F. H. Rodd, and R. G. Shaw. 1997. Evalua-
tion of the rate of evolution in natural populations of guppies (Poecilia
reticulata). Science 275:1934–1937.
Roff, D. 2000. The evolution of the G matrix: selection or drift? Heredity
84:135–142.
Rose, M. R., T. J. Nusbaum, and A. K. Chippindale. 1996. Laboratory evolu-
tion: the experimental Wonderland and the Cheshire Cat syndrome. Pp.
221–241 in M. R. Rose and G. V. Lauder, eds. Adaptation. Academic
Press, San Diego, CA.
Rose, M. R., H. B. Passananti, and M. Matos, eds. 2004. Methuselah Flies:
a case study in the evolution of aging. World Scientific Publishing,
Singapore.
Sgr` o,C.M., andL.Partridge. 2000.Evolutionaryresponses ofthelifehistory
of wild-caught Drosophila melanogaster to two standard methods of
laboratory culture. Am. Nat. 156:341–353.
Sim˜ oes, P., M. R. Rose, A. Duarte, R. Gonc ¸alves, and M. Matos. 2007.
Evolutionary domestication in Drosophila subobscura. J. Evol. Biol.
20:758–766.
Steppan, S. J., P. C. Phillips, and D. Houle. 2002. Comparative quantitative
genetics: evolution of the G matrix. Trends Ecol. Evol. 17:320–327.
Teot´ onio, H., and M. R. Rose. 2000. Variation in the reversibility of evolution.
Nature 408:463–466.
Teot´ onio, H., M. Matos, and M. R. Rose. 2002. Reverse evolution of fitness
in Drosophila melanogaster. J. Evol. Biol. 15:608–617.
Travisano, M., J. A. Mongold, A. F. Bennett, and R. E. Lenski. 1995. Experi-
mental tests of the roles of adaptation, chance, and history in evolution.
Science 267:87–90.
Wade, M. J., R. G. Winther, A. F. Agrawal, and C. J. Goodnight. 2001.
Alternative definitions of epistasis: dependence and interaction. Trends
Ecol. Evol. 16:498–504.
Zhang, X.-S., J. Wang, and W.G. Hill. 2004.Redistribution of gene frequency
and changes of genetic variation following a bottleneck in population
size. Genetics 167:1475–1492.
Associate Editor: M. Travisano
EVOLUTION AUGUST 2008
1829