Adaptation to a new environment allows cooperators
to purge cheaters stochastically
Adam James Waitea,b,1and Wenying Shoub,1
aMolecular and Cellular Biology Program, University of Washington, Seattle, WA 98195; andbDivision of Basic Sciences, Fred Hutchinson Cancer Research
Center, Seattle, WA 98109
This Feature Article is part of a series identified by the Editorial Board as reporting findings of exceptional significance.
Edited by Herbert Levine, University of California at San Diego, La Jolla, CA, and approved September 5, 2012 (received for review June 14, 2012)
Cooperation via production of common goods is found in diverse
life forms ranging from viruses to social animals. However, natural
selection predicts a “tragedy of the commons”: Cheaters, benefit-
ing from without producing costly common goods, are more fit
than cooperators and should destroy cooperation. In an attempt
to discover novel mechanisms of cheater control, we eliminated
known ones using a yeast cooperator–cheater system engineered
to supply or exploit essential nutrients. Surprisingly, although less
fit than cheaters, cooperators quickly dominated a fraction of co-
cultures. Cooperators isolated from these cocultures were supe-
rior to the cheater isolates they had been cocultured with, even
though these cheaters were superior to ancestral cooperators. Re-
sequencing and phenotypic analyses revealed that evolved co-
operators and cheaters all harbored mutations adaptive to the
nutrient-limited cooperative environment, allowing growth at a
much lower concentration of nutrient than their ancestors. Even
after the initial round of adaptation, evolved cooperators still sto-
chastically dominated cheaters derived from them. We propose the
“adaptive race” model: If during adaptation to an environment,
the fitness gain of cooperators exceeds that of cheaters by at least
the fitness cost of cooperation, the tragedy of the commons can be
averted. Although cooperators and cheaters sample from the same
pool of adaptive mutations, this symmetry is soon broken: The best
cooperators purge cheaters and continue to grow, whereas the
best cheaters cause rapid self-extinction. We speculate that adap-
tation to changing environments may contribute to the persistence
of cooperative systems before the appearance of more sophisti-
cated mechanisms of cheater control.
evolution of cooperation and cheating|experimental evolution|
genetic hitchhiking|synthetic biology
volunteers contribute their time to build Wikipedia, which can
be used by anyone with Internet access. Throughout the animal
kingdom, alarm calls are produced by individuals to warn others
of danger, even though producing the call makes the caller more
conspicuous (1). Microbes excrete a plethora of costly com-
pounds that can be used by the producers and their neighboring
cells to acquire nutrients that are hard to obtain, access favorable
environments, or improve antibiotic resistance (2, 3). In biolog-
ical systems, publicly available goods are generally “common
goods,” because consumption by one individual reduces their
availability to others. “Cheaters” use the common good without
paying a cost to produce it. Thus, because the common good is
equally accessible to all members of a population, cheaters, in-
troduced through migration or mutation, will be more fit than
cooperators, increase in frequency, and eventually exhaust the
common good, leading to the “tragedy of the commons” (4). For
instance, although cooperative viruses produce diffusible shared
proteins required for viral reproduction, selfish viruses synthesize
less but sequester more of these proteins and thereby displace
cooperative viruses, lowering overall infectivity (5). Cancers, a
leading cause of death globally (6), cheat by exploiting the common
he cooperative act of paying a cost to produce a publicly
available good is a common biological phenomenon. Human
good produced by normal cells that cooperate to form a func-
tional human body.
Despite exploitation of common goods by naturally arising
cheaters, cooperation persists (7–13). How does cooperation
survive cheating? We first summarize mechanisms known to sta-
to discover novel mechanisms of cheater control by excluding
known ones from an engineered yeast cooperator–cheater system.
Frequently, through unexpected genetic or physical processes,
what initially appear to be cheaters or common goods are not;
thus, the tragedy of the commons does not apply. For instance,
a gene required for cooperation can have pleiotropic effects,
such that a cell defective in paying the cost of cooperation is
also incapable of enjoying the cooperative benefit. In this case,
cheaters will end up suffering a net fitness cost. This situation has
been found to occur in the social amoeba Dictyostelium dis-
codium. D. discodium responds to starvation by aggregating and
forming a fruiting body. During fruiting body formation, some
cells form a self-sacrificing, nonreproductive stalk, which lifts
other cells that differentiate into reproductive spores. The gene
encoding the receptor necessary for differentiation into stalk
cells is also necessary for proper spore formation; thus, cheaters
trying to avoid the stalk fate cannot become spores (14). Another
possibility is that what appears to be a common good is actually
partially privatized by its producer. For instance, the budding yeast
Saccharomyces cerevisiae secretes invertase to hydrolyze the
disaccharide sucrose into glucose and fructose, which can be
metabolized more efficiently. These monosaccharides were ini-
tially thought to be strictly common goods (15), although it was
later found that ∼1% are retained by the producing cell (16). Even
such a seemingly insignificant level of privatization can allow co-
operators to invade a population of cheaters (16, 17). This could
explain the coexistence of invertase-producing cooperative cells
with nonproducing cheating cells in wild populations (7). The
benefits of privatization can only be realized when all cooperators
produce and consume the same common good (homotypic co-
operation). In mutualism, privatization is pointless because each
individual requires a common good produced only by its partners.
Even if cheaters have a net fitness advantage over cooperators
that produce true common goods, several mechanisms can avert
the tragedy of the commons. First, when individuals interact
through the production and/or consumption of inexpensive com-
mon goods in randomly formed groups that assemble and
Author contributions: A.J.W. and W.S. designed research; A.J.W. performed research;
A.J.W. contributed new reagents/analytic tools; A.J.W. and W.S. analyzed data; and
A.J.W. and W.S. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
See Commentary on page 19037.
1To whom correspondence may be addressed. E-mail: firstname.lastname@example.org or
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| November 20, 2012
| vol. 109
| no. 47
disassemble cyclically, as long as an increase in the availability of
the common good leads to a less than proportional increase in
the fitness of its consumers, a stable equilibrium between coop-
erators and cheaters is expected (18, 19). This “diminishing
return” of the common good (16, 18) can account for the sur-
prising observation that in the yeast invertase system, maximum
group size is attained with a mixture of cheaters and cooperators:
Cooperators produce more invertase than they can use, and
cheaters convert this excess benefit into additional biomass (20).
The cost-to-benefit ratio can be kept low if the common good is
produced facultatively (i.e., only when needed), which is the case
for most organisms, or if the durability of common goods is high,
as found in siderophore production in Pseudomonas aeruginosa
(21, 22). Second, for cooperation based on scarce common goods,
mechanisms of “positive assortment” that increase the frequency
of interactions between cooperators (23) can facilitate the per-
sistence of cooperation. Positive assortment can involve specifi-
cally directing benefits to other cooperators and excluding or
punishing cheaters based on recognition or previous experience
(24–26). This can occur even in organisms lacking nervous sys-
tems. For instance, microbes can achieve “recognition” through
cell adhesion and chemical communication (27), and legumes
“reward” and “punish” beneficial and cheating rhizobia, re-
spectively (28, 29). Another mechanism of positive assortment is
“population viscosity,” brought about by limited dispersal in
spatially structured environments, which keeps cooperators
clustered with their relatives in homotypic cooperation (24, 30–
32), or with their partners in heterotypic cooperation (33). Thus,
natural cooperative systems, whether homotypic or heterotypic,
use many different mechanisms to mitigate the tragedy of the
commons, allowing cooperation via common goods to be a suc-
cessful evolutionary strategy.
Existing cooperative systems may have evolved for millions of
years, and can now deploy mechanisms, such as pleiotropy, fac-
ultative production of common goods, and cheater recognition,
to prevent destruction by cheaters. However, what about cheater
control at the origins of these systems? A spatially structured
environment can stabilize cooperation against cheating; however,
under certain circumstances, it can hinder cooperation (34–36)
through enhancing competition among cooperators (37–39).
Furthermore, motile species may behave as if they were locally
well-mixed even if they live in a spatially structured environment.
To search for novel cheater-control mechanisms that might sta-
bilize nascent cooperative systems, we extended a synthetically
engineered model of cooperation (40) to include cheating (Fig.
1A). In this system, which is based on the yeast S. cerevisiae, co-
operation occurs between the red-fluorescent R←L
requireslysine (←L) andprovides adenine(→A), andtheyellow-
vides lysine (→L). The presence of both R←L
essary for growth in minimal media lacking adenine and lysine
(SD) (40). The cheater is a cyan-fluorescent C←Lstrain that
requires lysine but does not provide any nutrients.
This system allows us to remove the mechanisms currently
thought to avert the tragedy of the commons in a systematic man-
ner. Engineering the cooperative interactions through mutations
that constitutively overproduce metabolites eliminates the pos-
sibility of pleiotropy or facultative production. Using a two-partner
heterotypic cooperative system renders privatization of common
goods futile, because neither strain can directly use the common
good it produces. After an initial period of asymmetrical nutrient
release and cell death (40), our cooperative cocultures approach
a stable doubling time (∼12 h) that is much slower than those of
the corresponding monocultures supplemented with the neces-
sary nutrient (∼2 h). Thus, common goods are very limited. Addi-
tionally, because cooperation is based on metabolic manipulations
not found in the ancestor strain, it is unlikely that the cooper-
ating partners could recognize one another or exclude the
→Lstrain, which requires adenine (←A) and pro-
→Aand Y← A
cheater. Because all strains are derived from S288C, which does
not flocculate (41), growth in well-mixed liquid culture prevents
spatial clustering. Using a simple, well-defined model (42) free of
known mechanisms of cheater control, we hoped to discover new
mechanisms that may stabilize cooperation against cheating.
Cheaters Are Fitter Than Cooperators. Many biosynthetic pathways
are regulated by end-product feedback inhibition (e.g., 43, 44),
which suggests that metabolite overproduction carries an evo-
lutionarily significant cost. To quantify the cost of adenine over-
production in our cooperative yeast system, we competed R←L
with C←Lin SD supplemented with excess (164 μM) lysine and
found that the latter carried a fitness advantage of 1.8% [95%
confidence interval (CI): 1.6–1.9%; Fig. 1B]. In the absence of
lysine, which occurs during the prolonged delay in lysine release
difference in death rate: −0.003 to 0.0003 h−1; Fig. S1). In both
conditions, the fitness difference between expressing CFP and
DsRed is small. Given the overall fitness advantage of C←Lover
and the lack of genetic mechanisms for cheater control, we pre-
dicted that in a well-mixed environment, C←Lwould increase in
frequency and eventually destroy the cooperative system.
→Aand C←Ldied at similar rates (95% CI of the
→A, the futility of privatization and paucity of common goods,
Cheaters Are Stochastically Purged from Cocultures. We experimen-
tally tested the prediction of deterministic cheater dominance by
This master mix was split into replicate cocultures that were
monitored for growth and diluted to ensure that nutrients other
Surprisingly, within 50 generations, we observed a bifurcation in
growthrates (Fig.2B):Somecocultures weregrowingslowlyornot
at all (Fig. 2A, gray triangles), whereas others continued to grow
(Fig. 2A, orange triangles) at rates very similar to cheater-free
cocultures (Fig. 2A, black circles). At ∼450 h, we quantified the
frequency of each population in each coculture using flow cytom-
etry (Fig. 2C). The slow-growing cocultures (Fig. 2C, gray) con-
tained mostly dead or dying cells that had lost membrane integrity,
and therefore reacted with the nucleic acid dye TO-PRO-3. The
remaining live cells were predominantly C←L(“cheater-domi-
nated”). Furthermore, cheater takeover took much less time than
→L, and C←Lat a ratio of 1:1:1 in SD (Fig. 2).
tem is composed of two nonmating yeast strains. The red-fluorescent R←L
strain (WY950) requires lysine and overproduces adenine, whereas the yel-
strain (WY954) requires adenine and overproduces
lysine. The cheater is a cyan-fluorescent C←Lstrain (WY962) that requires
lysine and does not overproduce adenine. (B) Cheaters have a fitness ad-
vantage over cooperators. R←L
SD supplemented with nonlimiting (164 μM) lysine. Cocultures were peri-
odically diluted into fresh supplemented medium to maintain exponential
growth. The ratio of C←Lto R←L
ratio was fit using weighted nonlinear least-squares regression to the form
AexpðrtÞ, where A is the initial ratio, r is the cheater growth rate minus the
cooperator growth rate, and t is time in generations of R←L
arithmic y axis. C←Lhas a 1.8% (95% CI: 1.6–1.9%) advantage over R←L
(A) Yeast model of cooperation and cheating. Our cooperative sys-
→Aand C←Lstrains were mixed and competed in
→Awas determined using flow cytometry. The
→A. Note the log-
| www.pnas.org/cgi/doi/10.1073/pnas.1210190109 Waite and Shou
predicted: In fewer than 40 generations, the average C←L=R←L
ratio was 280:1, as opposed to the 2:1 ratio predicted by the 1.8%
fitness advantage of C←L(Fig. 1B). Even more surprising were
the fast-growing cocultures (Fig. 2C, orange). These cocultures
contained mostly live R←L
subsequent experiments with a C←Lstrain differentially marked
with drug resistance that C←Lcan be driven extinct in cooperator-
Extremely Fit Mutations Drive Stochastic Cooperator Dominance. To
understand what had caused rapid divergence in population
growth and stochastic cheater outcomes, we investigated the
detailed population dynamics of these cocultures by frequently
sampling replicate cocultures using flow cytometry. All R←L
C←Lpopulations, regardless of whether they eventually became
cooperator- or cheater-dominated (Fig. 3A; solid or dashed lines,
respectively), were nearly identical for the first ∼60 h of growth,
suggesting that the vast majority of cells were behaving identi-
cally and, most likely, ancestrally. Afterward, R←L
diverged rapidly (Fig. 3A). On the other hand, Y←A
operator-dominated and cheater-dominated cocultures behaved
identically until after the divergence between R←L
(>80 h; Fig. 3A), suggesting that whatever was occurring in the
One possible explanation for the rapid population divergence
was the presence of variants with large fitness advantages relative
to their ancestor. Consider the cooperator-dominated coculture
(Fig. 3A, solid lines). If the R←L
variant (R!) by the time of population divergence (∼60 h), R!
must have proliferated enough to influence the growth kinetics
of the red-fluorescent population visibly. To estimate the fitness
advantage of R! over non-R!, we would need to know their rel-
ative abundance over time. We could not distinguish the two
subpopulations directly from flow cytometry. However, because
we expected both to be influenced by the presence of R! in a
similar way, we assumed that the observed behavior of C←Lwas
similar to that of the non-R! portion of the red-fluorescent pop-
ulation. Thus, the population size of R! could be approximated by
the difference between the red-fluorescent and cyan-fluorescent
populations. Assuming that all R! cells were descendants of a
single cell present at the beginning of the experiment, this cell must
→Aand C←Lpopulations preceded any changes in the Y←A
→Apopulation obtained the most fit
→Aand C←Lpopulations initially behaved similarly, and
have doubled, on average, every ∼3.4 h to achieve its estimated
abundance at 80 h, a large improvement from the ancestral av-
erage doubling time of ∼18.5 h (Materials and Methods). The same
argument can be applied to the cheater-dominated coculture,
which suggests that cooperation was destroyed by similar, highly
adaptive variants that arose in the C←Lpopulation. These variants
must not have been rare (>1 in 106cells), because, otherwise, in
many cocultures, R←L
variant and population divergence would have been slow.
race” to obtain the most fit variant. If the same pool of variants
→Aand C←Lwould have failed to sample any
→Aand C←Lthus appeared to be engaged in an “adaptive
cocultures. Exponential cultures of R←L
cells/mL per strain. Twelve plus-cheater (triangles) and three minus-cheater (black circles) cocultures were propagated, with total cell density maintained in
the subsaturation range of ∼106cells/mL to 1.5 × 107cells/mL by dilution into fresh SD when necessary. (B) After ∼400 h, the growth rates of plus-cheater
cocultures were bimodally distributed into fast-growing (orange) and slow-growing (gray) groups. (C) Fast-growing cocultures are dominated by cooperators.
At ∼450 h, the frequencies of dead cells reacting to the nucleic acid dye TOPRO-3 (Dead), C←L, R←L
(gray) cocultures were quantified using flow cytometry. Boxes extend from the first quartile to the third quartile of the data, “whiskers” extend to the most
extreme observations, and the thick bar inside each box represents the median.
Stochastic cheater outcome in initially identical cooperator–cheater cocultures. (A and B) Distinct growth behavior in replicate cooperator–cheater
→L, and C←Lwere washed to remove residual supplements and mixed as indicated to a final density of 4.2 × 105
→A, and Y←A
→Lcells in fast-growing (orange) and slow-growing
cooperator-dominated and cheater-dominated cocultures. Cocultures were
initiated at a density of 1.7 × 105cells/mL per strain. Cell densities of the
three subpopulations were measured by flow cytometry by adding fluores-
cent beads of a known concentration to each sample. After initially fol-
lowing similar trajectories, rapid divergence resulted in dominance of R←L
cells (triangles connected by solid lines) or C←Lcells (circles connected by
dashed lines). A representative example of each class is shown. (B) Frequency
of cooperator-dominated cocultures is determined by the initial frequency
of cooperators. Cocultures were prepared as in A, except the initial density of
initial population size influenced how quickly cocultures diverged but not
the eventual frequency of cooperator-dominated cocultures. Error bars
represent 95% CIs, assuming that the number of viable cocultures followed
a binomial distribution with parameters n and p, where n is the total
number of cocultures and p is the probability of being cooperator-domi-
nated. The slope of the logistic regression fitting line, log?
suggests that the odds of being a cooperator-dominated coculture are
nearly equal to the initial proportion of R←L
system, the fitness cost of cooperation is small.
(A) Rapid divergence in the growth of cooperators and cheaters in
→Awas varied from 8.3 × 105to 4.2 × 106cells/mL per strain. Total
where c is the initial cooperator frequency, is 1.0 (95% CI: 0.7–1.3). This
→A. This occurs because, in our
Waite and ShouPNAS
| November 20, 2012
| vol. 109
| no. 47
was sampled by both populations, the final ratio of cooperator-
dominated/cheater-dominated cocultures should be determined
by the initial ratio of R←L
these two ratios should be approximately equal because the fit-
ness advantage of C←Lover R←L
the fitness gain of a successful variant. We tested this idea by
setting up cocultures at different initial ratios of R←L
was 1.0 (95% CI: 0.7–1.3). Thus, consistent with our hypothesis
iants, the frequency of being cooperator-dominated was de-
termined by the initial frequency of R←L
We next examined whether the improved fitness of these
variants in the context of cooperation and cheating was a heri-
table phenotype. We isolated Y←A
a cooperator-dominated coculture; grew them in monoculture in
SD supplemented with adenine or lysine; washed them free of
supplements; and mixed them 1:1:1. Like the parental coculture,
all six cocultures became cooperator-dominated (Fig. 4A, ii),
which was significantly more than what was observed for 1:1:1
all-ancestor cocultures (Fig. 4A, i; 6 of 6 vs. 31 of 64 cooperator-
dominated; Fisher’s exact test, P < 0.03). The isolated Y←A
clone did not contribute to cooperator dominance (Fig. 4A, iii; 3
of 6 vs. 31 of 64 cooperator-dominated; Fisher’s exact test, P >
0.9). This is consistent with the idea that, short of partner-spe-
cific recognition, any changes in Y←A
from each of seven independent cooperator-dominated cocul-
tures and found that they dominated ancestral C←Lin a nearly
deterministic fashion (Fig. 4A, iv; 41 of 42 vs. 31 of 64 co-
operator-dominated; Fisher’s exact test, P < 3 × 10−8). We then
tested five isolates of C←Lfrom three independent cheater-
dominated cocultures and found that they deterministically
dominated ancestral cooperators (Fig. 4A, vii; 0 of 48 vs. 31 of 64
cooperator-dominated; Fisher’s exact test P < 10−9). Thus,
changes occurring in R←L
for dominating ancestral C←Land R←L
during the adaptive race, then even the “losing” types (i.e., R←L
from cheater-dominated and C←Lfrom cooperator-dominated
cocultures) may nevertheless have improved relative to their
respective ancestors. We tested two losing C←Lisolates from
cooperator-dominated cocultures, and both were significantly
→A=C←L. Furthermore, in our system,
→Ais insignificant compared with
→Lwas always 1:1). The slope relating the two ratios
→Aand C←Lwere sampling from the same pool of var-
→A, and C←Lclones from
would affect C←Land
→Aequally. On the other hand, we tested one R←L
→Aand C←Lwere heritable and sufficient
→Aand C←Lsampled from the same pool of mutations
better than ancestral C←Lat dominating ancestral R←L
v; 0 of 12 vs. 31 of 64 cooperator-dominated; Fisher’s exact test,
P < 0.002). We tested one losing R←L
than ancestral R←L
6 vs. 31 of 64 cooperator-dominated; Fisher’s exact test, P < 0.03).
In addition to dominating ancestral C←Land the evolved C←L
that they had raced against, evolved R←L
sense that they lowered the minimal cell density required to
initiate a viable cooperative coculture (40). We tested a panel of
dominated cocultures. We mixed each isolate 1:1 with ancestral
ancestral cocultures, cocultures initiated with evolved R←L
quired less than one-third of the initial cell density to achieve
growth in 50% of replicate cocultures (Fig. 4B). Cocultures ini-
tiated with evolved Y←A
ment (Fig. S2).
→Aisolate, and it was better
→Aat dominating ancestral C←L(Fig. 4A, vi; 6 of
→Astrains improved in the
→Astrains isolated from cooperator-dominated or cheater-
→Land serially diluted each coculture into SD. Compared with
→Lshowed a similar degree of improve-
The Adaptive Race Is Fueled by Mutations in a Small Set of Genes
Involved in Nutrient Transport. We used whole-genome rese-
quencing to identify the mutations underlying the rapid evolu-
tion in cooperators and cheaters. Some strains contained only one
unambiguous, high-quality, nonsynonymous SNP. These muta-
tions defined a set of genes, exactly one of which was mutated in
every sequenced strain that contained any nonsynonymous muta-
tions. We therefore reasoned that mutations in these genes were
responsible for improvements in cooperation and cheating rel-
ative to the ancestor strains. All alleles of these genes and the
strains that contain them are reported in Table 1 (a complete list
of sequenced strains and their SNPs can be found in Dataset S1).
Of the 17 unique alleles found, ECM21 and DOA4 had 6 each,
accounting for 70% of the total. A simple estimate based on the
Poisson distribution suggests that we found ∼97% of the genes
responsible for the initial fitness increase. The small number of
genes could still allow for rapid divergence between populations
if each of the many alleles conferred a different fitness gain,
because it would be unlikely for two populations to sample the
same allele of the same gene.
Most of the identified genes suggested that evolved R←L
C←Lenhanced import of the limiting lysine provided by Y←A
by decreasing transporter turnover. Ecm21p is an arrestin-like
adaptor protein that allows the E3 ubiquitin ligase Rsp5p to
ubiquitinate and degrade the high-affinity lysine permease Lyp1p
“losing” strains. Cocultures consisting of ancestral strains (A) and strains isolated from cooperator-dominated (orange) or cheater-dominated (gray) cocultures
(E) were formed as indicated at equal initial densities. Percentages of cooperator-dominated cocultures were measured from replicate cocultures whose
numbers (n) are indicated. Error bars represent the 95% CI calculated as in Fig. 3B. (B) Improved R←L
ancestor. Ancestral R←L
direction to aid visualization.
Evolved cooperators and cheaters are superior to ancestors. (A) Improved cooperation and cheating are heritable and occur in both “winning” and
→Acan initiate cooperation at lower densities than its
→A(black) and R←L
→Aisolated from cooperator-dominated (orange) and cheater-dominated (gray) cocultures were mixed 1:1 with an-
→Land serially diluted into a microtiter plate. After 1 mo at 30 °C, wells showing visible growth were scored. Data are jittered slightly in the vertical
| www.pnas.org/cgi/doi/10.1073/pnas.1210190109Waite and Shou
in response to stress (45). All but one of the ECM21 mutations
introduced a premature stop codon before the PY-motif neces-
sary for interaction with Rsp5p (45). Doa4p is a deubiquitination
protein required to maintain free ubiquitin pools (46) and is
specifically implicated in deubiquitination of plasma membrane
proteins (47). Significantly, Doa4p represses expression of the
general amino acid permease Gap1p (48), which can import ly-
sine (49, 50). Less frequently observed were mutations in RSP5
itself; in BRO1, which is necessary for the function of Doa4p by
recruiting it to endosomes (51); and in the high-affinity lysine
permease gene LYP1 (50, 52).
To test whether nutrient transport was enhanced in evolved
strains, we compared the ability of evolved and ancestral cells
to grow into microcolonies (defined as >5 cells) under limiting
concentrations of lysine. Although ancestral R←L
could grow on 4 μM but not 1 μM lysine (Fig. 5A, Anc.), every
as they did on 4 μM lysine (Fig. 5A, Evo.). This improvement
came at a cost, because almost all the evolved cooperators and
cheaters grew slower in nonlimiting lysine than their ancestors
(Fig. 5B), which is consistent with their low frequency at the start
of our experiments. Thus, adaptation to the nutrient-limited
cooperative environment involved a tradeoff in fitness in non-
To confirm that the observed mutations were sufficient to im-
prove growth on limiting lysine, we replaced full-length ECM21 in
cells were able to grow on media containing 1 μM lysine (Fig.
→Aand C←Lisolate could grow as well on 1 μM lysine
→Aand C←Lwith the truncated version found in the
→Astrain CT22 (Table 1). Indeed, nearly 100% of these
Improved Cooperators Can Stochastically Defeat Newly Arising
Cheaters. If during adaptation to the nutrient-limited cooper-
ative environment, the maximum fitness gain of cooperators
exceeds that of cheaters by at least the cost of cooperation,
cooperators will win the race (Fig. 2A), and eventually purge the
inferior cheaters. However, a mutation could turn an evolved
cooperator into a cheater, which would now be as well-adapted
to the nutrient-limited cooperative environment as the cooper-
ators, and would therefore rise in frequency. Because successive
fitness gains during adaptation to the same environment are
expected to decrease over time (53), we tested whether evolved
cooperators could stochastically dominate otherwise isogenic
cheaters during a subsequent round of the adaptive race. We
chose two improved cooperator strains (CT22 and CT75; Table
1) carrying different truncation alleles of ECM21 and derived,
via allele replacement, matched cooperators and cheaters marked
by different drug resistances (2 independent pairs from CT22
and 1 pair from CT75; Table 1 and Table S1). As a control,
similarly matched cooperators and cheaters were derived from
As before, R←L
→A(Table S1, WY1356-WY1359).
→Aand C←Lpairs were mixed with ancestral
→Lat a ratio of 1:1:1. The frequencies of each cell type were
nutrient-limited cooperative environment
Mutations potentially sufficient for the observed large fitness advantage in the
GeneAllele StrainCell typeCoculture dominated byMutant frequency
Ins ‘A’ → C735*
Ins, indicates an insertion.
*Indicates STOP codon.
†Indicates the presence of additional high-quality SNPs in this strain.
to grow at low-lysine concentrations and the maximum growth rate achieved
at high-lysine concentrations. (A) Evolved strains grow better at low-lysine
concentrations than their ancestors. Exponentially growing ancestral (Anc.) or
evolved (Evo.) cooperator (red) and cheater (blue) strains were washed free of
lysine and starved for 16–20 h to deplete vacuolar storage of lysine. Cells were
plated at ∼200 cells/cm2onto minimal medium agar containing 4 μM (solid bar)
or 1 μM (shaded bar) lysine. After 1–3 d, clusters of >5 cells were counted as
microcolonies. Percentages and SEs were estimated using a generalized linear
mixed-effect model (Materials and Methods). The ecm21Δ143allele from
evolved cooperator CT22 was cloned into the ancestral cooperator and cheater
strains (Materials and Methods), and their growth on low-lysine plates was
measured (ecm21). The percentage is based on scoring >100 cells. Error bars
indicate the 95% CI,assuming a binomial distribution with parameters n and p,
as in Fig. 3B. (B) Evolved cooperators have lower growth rates than their
ancestors in nonlimiting lysine. Relative fitness compared with ancestral R←L
CT8 using monoculture growth rates.
Evolved cooperators and cheaters show a tradeoff between the ability
Waite and Shou PNAS
| November 20, 2012
| vol. 109
| no. 47
periodically measured by plating onto media containing the ap-
propriate antibiotic. As expected, the derived ancestral cocultures
dominated). Of the cocultures derived from the two ecm21 alleles
(Fig. 6, ecm21←L
4 were cheater-dominated, and 2 were indeterminate. Thus, even
after cooperators had obtained major fitness improvements in the
cooperative environment, a fraction of them were able to fend off
cheaters that were just as well-adapted.
→Aand ecm21←L), 11 were cooperator-dominated,
To uncover novel mechanisms that allow cooperation based on
the production of common goods to survive cheating, we examined
an engineered cooperative system that bypasses known mecha-
nisms of cheater control. After adding cheaters, we expected slow
but deterministic cheater takeover. Instead, identically initiated
cocultures showed rapid population divergence: Whereas cheaters
rapidly destroyed cooperation in some cocultures, cooperators
rapidly displaced cheaters in other cocultures. We determined that
this process was an adaptive race between cheaters and coopera-
tors that was driven by strong selection for improved growth in the
novel, nutrient-limited cooperative environment.
The race between cooperators and cheaters to adapt to a new
environment can eventually favor cooperation, despite the initial
fitness advantage of cheaters. As has been shown theoretically, if
the frequency of recombination is negligible and if the fitness
gain of the most adaptive mutation in cooperators exceeds that
of cheaters by at least the fitness cost of cooperation, the co-
operative trait can “hitchhike” on environmental adaptation and
rise to high frequency (54–56). Otherwise, cheaters dominate.
Thus, the probability of either type eventually dominating a co-
culture is related to its initial abundance in the population (Fig.
3B), because a larger population is more likely to sample better
mutations. However, this initial symmetry is quickly broken: If
cheaters dominate, the cooperative system will collapse (Fig. 2,
gray) because the common good necessary for survival is no
longer produced in sufficient quantities. In fact, the better the
mutation obtained by cheaters, the sooner self-extinction will
occur. On the other hand, if cooperators dominate, they con-
tinue to produce the common good and proliferate (Fig. 2, or-
ange). These cooperators have improved their ability to initiate
cooperation (Fig. 4B) and are able to defeat evolved cheaters
that are superior to the ancestral cheater (Fig. 4A). While pro-
liferating, cooperators can potentially acquire further beneficial
Two factors can synergize with the adaptive race to stabilize
cooperation. First, different types of environmental change can
provide different opportunities for adaptation. Improved coop-
erators that have survived one round of the adaptive race will
eventually generate cheating mutants. Initially, the small pop-
ulation size of these cheating mutants makes them vulnerable to
extinction by genetic drift and less likely to acquire the best
mutation during subsequent rounds of the adaptive race. Even-
tually, however, cheaters will crash the cooperative system if the
environment remains constant long enough. This is because the
magnitude of possible fitness gains from subsequent adaptations
will diminish over time (53). However, other environmental changes
in, for example, temperature or osmolarity will select for distinct
mutations that confer large fitness gains, triggering another adaptive
race. Second, if individuals can migrate between spatially separated
populations, “Simpson’s paradox” allows cooperators to dominate
globally even though cheaters grow faster in each population. This
can occur even when the population structure is periodically de-
stroyed by global mixing (58). The only requirement is sufficiently
variable cheater frequency between groups, which can be achieved
through dilution (58), or through mutations obtained during an
adaptive race. Thus, in conjunction with spatially structured envi-
ronments that change in different ways, two unavoidable realities of
nature, the adaptive race “buys time” for a cooperative system to
acquire additional mutations that may result in more reliable mech-
anisms of cheater tolerance, such as partner recognition.
Our cooperative system was engineered to cooperate in a
manner completely foreign to its ancestor; thus, it is pertinent to
consider whether theadaptive race occurs in morenaturalsystems.
We consider two examples in natural microbial cooperative com-
munities in which the adaptive race may have operated. The first
example involves the social bacterium Myxococcus xanthus, mem-
bers of which, when starved, exchange developmental signals, ag-
gregate, and differentiate to form a fruiting body. During this
process,a majorityofcellsdie,whereas theremainingcellsbecome
by disproportionately forming spores when mixed with another
strain (60). One cheater strain that could not form spores on its
own repeatedly rose to a high frequency when mixed with a co-
operative strain and drove the entire system to very low population
sizes (60). Reminiscent of our system, some of the replicate
cocultures suffered complete collapse, whereas the remaining
cocultures emerged cheater-free. If not solely due to drift, we
suspectthatthisstochasticcheater outcomemay betheresultofan
cheater created a novel environment unfamiliar to the cooperative
ancestor strain. If the emerging cooperators demonstrate an im-
proved ability to defeat ancestral cheaters, sequencing evolved
cooperators might reveal the type of novel environment imposed
by cheaters, as well as the mechanisms deployed to adapt to it.
A second example of a potential adaptive race was recently
shown in the siderophore-producing bacterium Pseudomonas
fluorescens (54). During iron limitation, P. fluorescens produces
and releases iron-chelating siderophores that can be taken up by
any nearby individual. When nonproducing cheaters, which have
a large fitness advantage over siderophore-producing cooperators,
were mixed with cooperators in an iron-limited environment,
cheaters introduced at low frequency failed to invade (61). The
authors proposed that this occurred because the numerically
cooperators and cheaters previously exposed to the cooperative environ-
ment. We replaced the adenine overproduction allele of ADE4 (PUR6) in two
Table 1) with PUR6 or WT ADE4 marked with different antibiotic resistances.
Cocultures were initiated at 1:1:1. Frequencies of cell types were periodically
determined by plating onto media containing the appropriate antibiotic. A
coculture was considered “cooperator-dominated” if the ratio of R←L
C←Lwas >100, or if the frequency of C←Lfell below 1% of its initial fre-
quency. It was considered “cheater-dominated” if the ratio of C←Lto R←L
was >100, or if the frequency of R←L
was considered “indeterminate” if none of these conditions were satisfied
after ∼2,000 h of propagation. Indeterminate cocultures were excluded from
the analysis. Because results across pairs were not significantly different from
one another, the data were combined and are presented as ecm21←L
ecm21←L. Error bars are the 95% CIs calculated as in Fig. 3B.
Stochastic cheater purging observed in cocultures composed of
→Astrains containing truncation alleles of ECM21 (CT22 and CT75;
→Afell below 1% of its initial frequency. It
| www.pnas.org/cgi/doi/10.1073/pnas.1210190109Waite and Shou
dominant cooperators had a greater chance of obtaining a high-
fitness mutation adaptive to iron limitation that could sweep
through the population. However, the actual mechanisms were
not determined. This implies that although P. fluorescens must
have experienced similar types of iron limitation throughout its
evolutionary history, as evidenced by the very existence of side-
rophore production, the experimentally imposed iron limitation
was still sufficiently “novel” to elicit an adaptive race. This sup-
ports our experimental finding that the environment does not
need to be completely novel for the adaptive race to occur (Fig. 6).
Similar genetic hitchhiking in a not-so-novel environment was
recently observed outside the context of cooperation. In an ex-
periment initially designed to track deterministic takeover of
sterile yeast mutants (which have greater fitness than nonsterile
cells), the researchers observed complex patterns consistent with
the rise of multiple nonsterile mutants with greater fitness than
the sterile mutants (62). Importantly, this occurred in rich media
thought to impose minimal selective pressure. These examples
illustrate that the conditions necessary to initiate the adaptive
race may be met more frequently than is currently assumed.
In summary, by using an engineered cooperating–cheating
yeast system to bypass all known cheater-control mechanisms, we
discovered that the adaptive race allowed cooperators to defeat
cheaters stochastically. This simple mechanism requires low rates
of recombination and can be triggered by any environment in
which adaptation offers fitness gains greater than the cost of
cooperation. Experimental evidence from naturally occurring
cooperative systems suggests that these conditions may be met
even when the environment is not completely novel (60, 61).
Because adaptation to a changing environment is the norm in
biology, we propose that the adaptive race between cooperators
and cheaters, especially in spatially structured populations, is
a fundamental mechanism for cooperative systems to survive the
perpetual onslaught of cheaters.
Materials and Methods
Strains and Cell Culture. All strains were derived from the BY designer de-
letion strains of the S288C background (63) and created essentially as de-
scribed by Shou et al. (40). A list of constructed strains used in this study is
provided in Table S1. To derive cooperating and cheating strains from a
parent strain, DNA fragments PCR-amplified off a cassette containing either
WT ADE4 and resistance to Hygromycin B (hph; WSB150) or an ADE4 over-
production allele (PUR6) and resistance to ClonNAT (nat; WSB151) were
transformed into the parent strain. The primers contained regions homol-
ogous to the site of integration. Transformants were selected on media
containing the appropriate antibiotic. The presence of ADE4 or PUR6 in
the lysine-auxotrophic parent strain was determined by patching a small
amount of transformant onto SD + lysine plates prespread with ade8 diploid
cells and cut into individual squares. Overproduction of adenine resulted in
adenine release and visible growth of ade8 cells into satellite colonies,
whereas strains with ADE4 did not. By replacing ECM21 after the 142nd
amino acid with DNA fragments PCR-amplified off an hph-containing cas-
sette (WSB117), ecm21Δ143strains were created.
Cells were grown at 30 °C in the rich medium YPD [1% (wt/vol) Bacto-yeast
extract, 2% (wt/vol) Bacto-peptone, 2% (wt/vol) dextrose, 2% (wt/vol) Bacto-
agar] or the minimal medium SD [0.67% (wt/vol) Difco Yeast Nitrogen Base
without amino acids, 2% (wt/vol) dextrose] with or without supplemented
amino acids, as necessary (64). To prepare for use, strains were struck from
the freezer and grown on YPD plates at 30 °C for 2–3 d. An individual colony
was picked and grown at 30 °C in liquid YPD until saturated (1–3 d). Unless
otherwise indicated, strains were diluted >1:1,000 into SD + lysine or ade-
nine and grown for ∼16 h to exponential phase (∼2 × 106–4 × 106cells/mL)
before being assayed.
Calculation of Average Doubling Times. The average doubling times of R←L
and R! were calculated using the formula
?, where t is time in hours,
Nð0Þ is the population size at time 0, and NðtÞ is the population size at time t.
Competition Assays. To measure the fitness difference between cooperators
and cheaters (Fig. 1B), exponentially growing cells were washed and resus-
pended in SD, diluted to 1.7 × 105cells/mL into SD (for death competition) or
SD + 164 μM lysine (for growth competition) in triplicate, and incubated at
30 °C on a rotator. Growing cocultures were periodically diluted to maintain
exponential growth, while keeping population size above 3 × 104cells per
strain. Ratios of the competing strains were determined by flow cytometry
periodically. The relationship between the C←L=R←L
erations of R←L
→Aratio and time in gen-
→A=AexpðrtÞ using the method of
→A(t) was fit to the formC←L
weighted nonlinear least-squares, where A is the initial ratio and r is the
difference in Malthusian parameters between C←Land R←L
growth rate of any population (R←L
nition, percent fitness advantage with respect to R←L
r=lnð2Þ. To determine fitness differences between R←L
descendants (Fig. 5B), which were all red-fluorescent, an otherwise isogenic
ference in fitness between Y←L
less than the difference in fitness between R←L
evolved strain (1.7%, CT22).
→A. Because the
→Ahere) is lnð2Þ per generation by defi-
→Awas calculated as
→Aand its evolved
→Astrain was competed against each strain. The dif-
→Aand the fastest growing
→A(0.1%) was an order of magnitude
Flow Cytometry. Population compositions were measured by flow cytometry
using a FACSCanto II (BD Biosciences) or DxP10 (Cytek) with three lasers
(405 nm, 488 nm, and 633 nm). If necessary, fluorescent beads of known
concentration were added to determine cell densities. Data analyses were
performed using FloJo (TreeStar) or custom software written in R (65) using
the Bioconductor (66) packages flowCore (67), flowMeans (68), and flow-
Coculture Growth and Assay of Population Compositions. Exponentially grow-
ing cells were washed in SD and counted using a CoulterCounter Z2 (Beckman
Coulter) or by flow cytometry. After counting, cells were diluted into SD and
mixed.Then,3-mLaliquots wereputin13-mmtubes andcultured ona rotator
at 30 °C. OD600was monitored at least once a day using a Gensys 20 (Thermo
Spectronic), and population compositions were determined by flow cytom-
etry. To estimate densities and ratios for the ecm21 strains (Fig. 6), cocultures
were sampled into a microtiter plate and serially diluted in 10-fold steps from
1- to 10−4-fold using a multichannel pipette; spotted onto agar plates con-
taining YPD, YPD + clonNAT (100 μg/mL), and YPD + Hygromycin B (200
μg/mL); and incubated at 30 °C for up to 7 d. Colony counts were used to
calculate population sizes of the entire coculture (YPD), ecm21←L
clonNAT), and ecm21←L(YPD + Hygromycin B).
Viability Assay.Exponentiallygrowingcellswere washed freeofsupplements.
serially diluted in a deep multiwell trough (∼10 mL per well) (Fig. 4B). Six or
twelve 200-μL aliquots of each prepared density were transferred into wells
of a microtiter plate using a multichannel pipette, which were then sealed
using parafilm and placed in a 30 °C incubator in moisturized Tupperware
bins for 1 mo. Visible growth by eye was considered successful initiation of
cooperation. Data were analyzed using a generalized linear model with a bi-
nomial link function. Strain and the logarithm of cell density were fixed effects.
→Aand ancestral and evolved R←L
→Astrains were mixed 1:1 and
Microcolony Growth Assay. Exponentially growing cells were washed and
resuspended in SD, and then diluted to ensure that they would remain at low
density after residual growth. After starvation in SD for 16–20 h, cell density
was estimated by OD600, diluted to ∼104cells/mL, and plated onto SD plus
4 μM or 1 μM lysine (Fig. 5A). Plates were observed using light microscopy
under a 10× objective after 24 h (4 μM lysine) or 48 h (1 μM lysine). Clusters
of five or more cells were considered microcolonies. Percentages and 95%
CIs for the ancestor and evolved data in Fig. 5A were estimated by a gen-
eralized linear mixed-effect model with a binomial link function using the
package lme4 (70). Cooperator/cheater (“type”), ancestor/evolved (“state”),
lysine concentration, and lysine concentration/state interaction were fixed ef-
fects, whereas strain and strain/experiment interaction were random effects.
Sequencing and Analysis. Genomic DNA was extracted using the Genomic-tip
20/G kit (QIAGEN). Libraries were prepared using the tagmentation reaction
through the Nextera DNA Sample Preparation Kit, the Nextera Index Kit (96
indices), and TruSeq Dual Index Sequencing Primers (Illumina). All clean-up
steps were performed with DNA Clean and Concentrator-5 (Zymo Research).
Libraries from 30 strains (including the ancestral R←L
tiplexed and run on a HiSeq2000 (Illumina) using 50-cycle paired-end read-
ing. Sequence data were analyzed using a custom Perl script that incor-
porated the bwa aligner (71) and SAMtools (72).
→Aand C←L) were mul-
Waite and ShouPNAS
| November 20, 2012
| vol. 109
| no. 47
ACKNOWLEDGMENTS. We thank Justin Burton, who initially observed sto- Download full-text
chastic cheater outcomes; Wayne Gerard for performing the experiment
used to generate Fig. 3A; Aric Capel for assistance with strain construction;
Sarah Holte for help with statistical analysis; and Andy Marty and Ryan
Basom for help with sequencing. Work in the W.S. group is supported by
the W. M. Keck Foundation, the National Institutes of Health (Grant 1 DP2
OD006498-01), and the National Science Foundation Bio/computation Evo-
lution in Action CONsortium (BEACON) Science and Technology Center.
1. Hollén LI, Radford AN (2009) The development of alarm call behaviour in mammals
and birds. Anim Behav 78:791–800.
2. West SA, Diggle SP, Buckling A, Gardner A, Griffin AS (2007) The social lives of mi-
crobes. Annu Rev Ecol Evol Syst 38:53–77.
3. Lee HH, Molla MN, Cantor CR, Collins JJ (2010) Bacterial charity work leads to pop-
ulation-wide resistance. Nature 467:82–85.
4. Hardin G (1968) The tragedy of the commons. Science 162:1243–1248.
5. Turner PE, Chao L (1999) Prisoner’s dilemma in an RNA virus. Nature 398:441–443.
6. World Health Organization (2008) The Global Burden of Disease: 2004 Update (World
Health Organization, Geneva) Available at: http://www.who.int/healthinfo/global_
burden_disease/2004_report_update/en/index.html. Accessed May 22, 2012.
7. Naumov GI, Naumova ES, Sancho ED, Korhola MP (1996) Polymeric SUC genes in
natural populations of Saccharomyces cerevisiae. FEMS Microbiol Lett 135:31–35.
8. Strassmann JE, Zhu Y, Queller DC (2000) Altruism and social cheating in the social
amoeba Dictyostelium discoideum. Nature 408:965–967.
9. Schaber JA, et al. (2004) Analysis of quorum sensing-deficient clinical isolates of
Pseudomonas aeruginosa. J Med Microbiol 53:841–853.
10. Dobata S, Tsuji K (2009) A cheater lineage in a social insect: Implications for the
evolution of cooperation in the wild. Commun Integr Biol 2:67–70.
11. Vos M, Velicer GJ (2009) Social conflict in centimeter-and global-scale populations of
the bacterium Myxococcus xanthus. Curr Biol 19:1763–1767.
12. Wilder CN, Allada G, Schuster M (2009) Instantaneous within-patient diversity of
Pseudomonas aeruginosa quorum-sensing populations from cystic fibrosis lung in-
fections. Infect Immun 77:5631–5639.
13. Jandér KC, Herre EA (2010) Host sanctions and pollinator cheating in the fig tree-fig
wasp mutualism. Proc Biol Sci 277:1481–1488.
14. Foster KR, Shaulsky G, Strassmann JE, Queller DC, Thompson CRL (2004) Pleiotropy as
a mechanism to stabilize cooperation. Nature 431:693–696.
15. Greig D, Travisano M (2004) The Prisoner’s Dilemma and polymorphism in yeast SUC
genes. Proc Biol Sci 271(Suppl 3):S25–S26.
16. Gore J, Youk H, van Oudenaarden A (2009) Snowdrift game dynamics and facultative
cheating in yeast. Nature 459:253–256.
17. Doebeli M, Hauert C (2005) Models of cooperation based on the Prisoner’s Dilemma
and the Snowdrift game. Ecol Lett 8:748–766.
18. Foster KR (2004) Diminishing returns in social evolution: The not-so-tragic commons.
J Evol Biol 17:1058–1072.
19. Archetti M, Scheuring I (2011) Coexistence of cooperation and defection in public
goods games. Evolution 65:1140–1148.
20. MaClean RC, Fuentes-Hernandez A, Greig D, Hurst LD, Gudelj I (2010) A mixture of
“cheats” and “co-operators” can enable maximal group benefit. PLoS Biol 8:
21. Kümmerli R, Jiricny N, Clarke LS, West SA, Griffin AS (2009) Phenotypic plasticity of
a cooperative behaviour in bacteria. J Evol Biol 22:589–598.
22. Kümmerli R, Brown SP (2010) Molecular and regulatory properties of a public good
shape the evolution of cooperation. Proc Natl Acad Sci USA 107:18921–18926.
23. Fletcher JA, Doebeli M (2009) A simple and general explanation for the evolution of
altruism. Proc Biol Sci 276:13–19.
24. Hamilton WD (1964) The genetical evolution of social behaviour. I. J Theor Biol 7:
25. Trivers RL (1971) The evolution of reciprocal altruism. Q Rev Biol 46:35–57.
26. Axelrod R, Hamilton WD (1981) The evolution of cooperation. Science 211:1390–1396.
27. Strassmann JE, Gilbert OM, Queller DC (2011) Kin discrimination and cooperation in
microbes. Annu Rev Microbiol 65:349–367.
28. Kiers ET, Rousseau RA, West SA, Denison RF (2003) Host sanctions and the legume-
rhizobium mutualism. Nature 425:78–81.
29. Heath KD, Tiffin P (2009) Stabilizing mechanisms in a legume-rhizobium mutualism.
30. Maynard Smith J (1964) Group selection and kin selection. Nature 201:1145–1147.
31. Chao L, Levin BR (1981) Structured habitats and the evolution of anticompetitor
toxins in bacteria. Proc Natl Acad Sci USA 78:6324–6328.
32. Nowak MA, May RM (1992) Evolutionary games and spatial chaos. Nature 359:
33. Harcombe W (2010) Novel cooperation experimentally evolved between species.
34. Hauert C, Doebeli M (2004) Spatial structure often inhibits the evolution of co-
operation in the snowdrift game. Nature 428:643–646.
35. Hansen SK, Rainey PB, Haagensen JA, Molin S (2007) Evolution of species interactions
in a biofilm community. Nature 445:533–536.
36. Verbruggen E, et al. (2012) Spatial structure and interspecific cooperation: Theory
and an empirical test using the mycorrhizal mutualism. Am Nat 179:E133–E146.
37. Queller DC (1992) Does population viscosity promote kin selection? Trends Ecol Evol
38. West SA, Pen I, Griffin AS (2002) Cooperation and competition between relatives.
39. Griffin AS, West SA, Buckling A (2004) Cooperation and competition in pathogenic
bacteria. Nature 430:1024–1027.
40. Shou W, Ram S, Vilar JMG (2007) Synthetic cooperation in engineered yeast pop-
ulations. Proc Natl Acad Sci USA 104:1877–1882.
41. Liu H, Styles CA, Fink GR (1996) Saccharomyces cerevisiae S288C has a mutation in
FLO8, a gene required for filamentous growth. Genetics 144:967–978.
42. Momeni B, Chen C-C, Hillesland KL, Waite A, Shou W (2011) Using artificial systems
to explore the ecology and evolution of symbioses. Cell Mol Life Sci 68:1353–1368.
43. Armitt S, Woods RA (1970) Purine-excreting mutants of Saccharomyces cerevisiae. I.
Isolation and genetic analysis. Genet Res 15:7–17.
44. Blickling S, Knäblein J (1997) Feedback inhibition of dihydrodipicolinate synthase
enzymes by L-lysine. Biol Chem 378:207–210.
45. Lin CH, MacGurn JA, Chu T, Stefan CJ, Emr SD (2008) Arrestin-related ubiquitin-ligase
adaptors regulate endocytosis and protein turnover at the cell surface. Cell 135:
46. Swaminathan S, Amerik AY, Hochstrasser M (1999) The Doa4 deubiquitinating en-
zyme is required for ubiquitin homeostasis in yeast. Mol Biol Cell 10:2583–2594.
47. Dupré S, Haguenauer-Tsapis R (2001) Deubiquitination step in the endocytic pathway
of yeast plasma membrane proteins: Crucial role of Doa4p ubiquitin isopeptidase.
Mol Cell Biol 21:4482–4494.
48. Springael JY, Galan JM, Haguenauer-Tsapis R, André B (1999) NH4+-induced down-
regulation of the Saccharomyces cerevisiae Gap1p permease involves its ubiquitina-
tion with lysine-63-linked chains. J Cell Sci 112:1375–1383.
49. Grenson M, Hou C, Crabeel M (1970) Multiplicity of the amino acid permeases in
Saccharomyces cerevisiae. IV. Evidence for a general amino acid permease. J Bacteriol
50. Regenberg B, Düring-Olsen L, Kielland-Brandt MC, Holmberg S (1999) Substrate
specificity and gene expression of the amino-acid permeases in Saccharomyces cer-
evisiae. Curr Genet 36:317–328.
51. Luhtala N, Odorizzi G (2004) Bro1 coordinates deubiquitination in the multivesicular
body pathway by recruiting Doa4 to endosomes. J Cell Biol 166:717–729.
52. Grenson M (1966) Multiplicity of the amino acid permeases in Saccharomyces cer-
evisiae. II. Evidence for a specific lysine-transporting system. Biochim Biophys Acta
53. Lenski RE, Travisano M (1994) Dynamics of adaptation and diversification: A 10,000-
generation experiment with bacterial populations. Proc Natl Acad Sci USA 91:
54. Morgan AD, Quigley BJZ, Brown SP, Buckling A (2012) Selection on non-social traits
limits the invasion of social cheats. Ecol Lett 15:841–846.
55. Santos M, Szathmáry E (2008) Genetic hitchhiking can promote the initial spread of
strong altruism. BMC Evol Biol 8:281.
56. Smith JM, Haigh J (1974) The hitch-hiking effect of a favourable gene. Genet Res 23:
57. Schwilk DW, Kerr B (2002) Genetic niche-hiking: An alternative explanation for the
evolution of flammability. Oikos 99:431–442.
58. Chuang JS, Rivoire O, Leibler S (2009) Simpson’s paradox in a synthetic microbial
system. Science 323:272–275.
59. Velicer GJ, Vos M (2009) Sociobiology of the myxobacteria. Annu Rev Microbiol 63:
60. Fiegna F, Velicer GJ (2003) Competitive fates of bacterial social parasites: Persistence
and self-induced extinction of Myxococcus xanthus cheaters. Proc Biol Sci 270:
61. Morgan AD, Quigley BJ, Brown SP, Buckling A (2012) Selection on non-social traits
limits the invasion of social cheats. Ecol Lett 15:841–846.
62. Lang GI, Botstein D, Desai MM (2011) Genetic variation and the fate of beneficial
mutations in asexual populations. Genetics 188:647–661.
63. Brachmann CB, et al. (1998) Designer deletion strains derived from Saccharomyces
cerevisiae S288C: A useful set of strains and plasmids for PCR-mediated gene dis-
ruption and other applications. Yeast 14:115–132.
64. Guthrie C, Fink GR, eds (1991) Guide to Yeast Genetics and Molecular Biology (Aca-
demic Press, San Diego).
65. R Core Team (2011) R: A Language and Environment for Statistical Computing
(R Foundation for Statistical Computing, Vienna). Available at http://www.R-project.
org. Accessed September 26, 2012.
66. Gentleman RC, et al. (2004) Bioconductor: Open software development for compu-
tational biology and bioinformatics. Genome Biol 5:R80.
67. Hahne F, et al. (2009) flowCore: A Bioconductor package for high throughput flow
cytometry. BMC Bioinformatics 10:106.
68. Aghaeepour N, Nikolic R, Hoos HH, Brinkman RR (2011) Rapid cell population identi-
fication in flow cytometry data. Cytometry A 79(1):6–13.
69. Sarkar D, Le Meur N, Gentleman R (2008) Using flowViz to visualize flow cytometry
data. Bioinformatics 24:878–879.
70. Bates D, Maechler M, Bolker B (2011) lme4: Linear mixed-effects models using S4
classes. Available at http://CRAN.R-project.org/package=lme4. Accessed September
71. Li H, Durbin R (2009) Fast and accurate short read alignment with Burrows-Wheeler
transform. Bioinformatics 25:1754–1760.
72. Li H, et al.1000 Genome Project Data Processing Subgroup (2009) The Sequence
Alignment/Map format and SAMtools. Bioinformatics 25:2078–2079.
| www.pnas.org/cgi/doi/10.1073/pnas.1210190109 Waite and Shou