Experiments on the role of deleterious mutations as PNAS PLUS stepping stones in adaptive evolution

Arthur W. Covert IIIa,b,c,d, Richard E. Lenskia,c,e,1, Claus O. Wilkec,d, and Charles Ofriaa,b,c,1

aProgram in Ecology, Evolutionary Biology and Behavior, Departments of bComputer Science and Engineering and eMicrobiology and Molecular Genetics, and cBEACON Center for the Study of Evolution in Action, Michigan State University, East Lansing, MI 48824; and dCenter for Computational Biology, Institute for Cellular and Molecular Biology, and Department of Integrative Biology, University of Texas at Austin, Austin, TX 78712

Contributed by Richard E. Lenski, July 16, 2013 (sent for review April 22, 2013) Many evolutionary studies assume that deleterious mutations spread unless they are tightly linked (5, 6). In an asexual pop- necessarily impede adaptive evolution. However, a later mutation ulation, however, the fortuitous combination, once formed, will that is conditionally beneficial may interact with a deleterious be stably inherited, thereby providing a simple way to traverse a predecessor before it is eliminated, thereby providing access to fitness valley (5–10). adaptations that might otherwise be inaccessible. It is unknown Previous research on a variety of systems, both biological and whether such sign-epistatic recoveries are inconsequential events computational, has documented many examples of epistatic inter- or an important factor in evolution, owing to the difficulty of actions between mutations (3, 6, 11, 14–24), including a few cases monitoring the effects and fates of all mutations during experi- of mutational pairs that are individually deleterious but jointly ments with biological organisms. Here, we used digital organisms beneficial (11, 12). Nonetheless, it remains unclear, even in digital to compare the extent of adaptive evolution in populations when systems, whether these cases have a significant impact on the rate deleterious mutations were disallowed with control populations of adaptive evolution or, alternatively, are mere curiosities of little in which such mutations were allowed. Significantly higher fitness overall importance. Even theoretical studies are often limited to levels were achieved over the long term in the control populations simple two-mutation, two-allele models of evolution, which because some of the deleterious mutations served as stepping may not capture the more complex adaptive dynamics on a high- stones across otherwise impassable fitness valleys. As a conse- dimensional fitness landscape. quence, initially deleterious mutations facilitated the evolution of Here, we use the Avida platform (version 2.6; http://devolab. complex, beneficial functions. We also examined the effects of msu.edu) to study the importance of deleterious mutations in disallowing neutral mutations, of varying the mutation rate, and adaptive evolution (11, 25). In Avida, self-replicating computer of sexual recombination. Populations evolving without neutral programs called digital organisms compete for energy that allows mutations were able to leverage deleterious and compensatory them to execute the behaviors encoded by their genomes. Each mutation pairs to overcome, at least partially, the absence of single-instruction processing unit, or SIP, of energy allows a digital neutral mutations. Substantially raising or lowering the mutation organism to execute one instruction in its genome. As popu- rate reduced or eliminated the long-term benefit of deleterious lations of digital organisms evolve, they may become more ef- EVOLUTION mutations, but introducing recombination did not. Our work dem- ficient, requiring fewer SIPs to replicate, and they may acquire onstrates that deleterious mutations can play an important role in new computational functions that allow them to obtain more adaptive evolution under at least some conditions. SIPs. As in nature, selection acts on the phenotype rather than directly on the genome, and many or most mutations have del- epistasis | experimental evolution | genetic drift eterious effects, whereas beneficial mutations are less common (19). Mutations in digital organisms often interact epistatically volutionary biologists usually assume that increased perfor- rather than multiplicatively, once again as observed in many Emance, including the origin of new traits, arises via a hill- climbing process, such that a population evolves toward a local Significance fitness peak. Mutational paths through fitness valleys are often viewed as inaccessible (1, 2), and even nonlethal deleterious It might seem obvious that deleterious mutations must impede mutations are seen as dead ends that can be ignored when evolution. However, a later mutation may interact with a del- considering the rate of improvement and the distribution of eterious predecessor, facilitating otherwise inaccessible adap- adaptive outcomes (3). Wright’s shifting-balance theory offers tations. Although such interactions have been reported before, one possible dynamic for a sexually reproducing metapopulation it is unclear whether they are rare and inconsequential or, al- (comprising many demes or subpopulations) to traverse a fitness ternatively, are important for sustaining adaptation. We stud- valley, but its efficacy is disputed (4). In asexually reproducing ied digital organisms—computer programs that replicate and organisms, however, there exists a simpler mechanism for crossing evolve—to compare adaptation in populations where delete- a fitness valley. Provided that a deleterious mutation is not lethal, rious mutations were disallowed with unrestricted controls. the genome carrying it has some expected “half life” and a cor- Control populations achieved higher fitness values because responding chance of reproducing one or more times before some deleterious mutations acted as stepping stones across going extinct (2, 5–10). Occasionally, the mutant subpopulation otherwise impassable fitness valleys. Deleterious mutations can might acquire a second, hypercompensatory mutation that pro- thus sometimes play a constructive role in adaptive evolution. vides a net advantage (11, 12). Although such mutations are expected to be rare, new detrimental mutations are constantly Author contributions: A.W.C., R.E.L., and C.O. designed research; A.W.C. performed re- search; A.W.C. and C.O.W. analyzed data; and A.W.C., R.E.L., C.O.W., and C.O. wrote generated, thus providing a multitude of potential stepping stones. the paper. If the second mutation would also have been deleterious had it The authors declare no conflict of interest. fi appeared without the preceding mutation, then the bene cial Data deposition: Summary data and analysis scripts have been deposited at the Dryad combination is said to have a “sign-epistatic” interaction (3, 13)— Digital Repository, http://datadryad.org (DOI: 10.5061/dryad.4hp5n). two wrongs, in effect, make a right. In a sexual population, such 1To whom correspondence may be addressed. E-mail: [email protected] or [email protected]. fi interacting mutations will often nd themselves becoming dis- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. sociated so long as they are rare, making it difficult for them to 1073/pnas.1313424110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1313424110 PNAS | Published online August 5, 2013 | E3171–E3178 Downloaded by guest on September 26, 2021 natural systems (3, 14–18, 21–24). The digital organisms in this (RpDL). Besides eliminating deleterious mutations, these treat- study live on a toroidal grid; when an organism replicates, its ments also affected the overall mutation rate and the distribution progeny is placed at random in one of the nine neighboring cells of fitness effects (Fig. S1 B–D). All other parameters were iden- (including that occupied by the parent), replacing the previous tical across treatments, including the ancestral genotype, selective occupant. Thus, population size remains constant. In our anal- environment, population size (10,000), and duration (250,000 yses, we typically focus on the most abundant genotype in updates, typically ∼45,000 generations) (Supporting Information). a population at the end of an experimental run. We refer to this After comparing the effects of these treatments, we examine genotype as the “final dominant.” We also examine the “line of a case study in more detail, and we perform replay experiments descent” that includes every organism along the lineage from the starting from adjacent genotypes to quantify the effects of in- initial ancestor to the final dominant (11, 12, 25). dividual deleterious mutations on subsequent adaptation. Fur- Lenski et al. (11) previously found instances of deleterious ther, we compare the effects of reverting or replacing deleterious mutations acting as stepping stones while using Avida to examine mutations with the effects of similar treatments applied to neu- the origin of a complex logic function called “bitwise equals” tral mutations. Finally, we explore how mutation rate and sexual (EQU). Five of 23 replicate populations that evolved EQU had reproduction influence the effects of deleterious mutations on a deleterious mutation on the line of descent immediately before long-term adaptation. evolving the EQU phenotype. Reversion of those deleterious, penultimate mutations eliminated the EQU phenotype in three Results and Discussion of the five cases, indicating that those three mutations contrib- Effects of Reverting and Replacing Deleterious Mutations. If dele- uted to the EQU function despite the fact that they reduced terious mutations merely impede adaptive evolution, then pop- fitness on the backgrounds in which they first appeared. Cow- ulations that evolve in the treatments that prevent deleterious perthwaite et al. (12) found similar stepping stones in simulations mutations should achieve higher mean fitness than control pop- of evolving RNA sequences. Two other studies with digital ulations. However, if enough deleterious mutations serve as organisms suggested that deleterious mutations could play an evolutionary stepping stones, then the control populations should important role in adaptation on rugged fitness landscapes (26), reach higher fitness levels. even in sexual populations (27). However, none of these previous We ran three experimental treatments and one control treat- studies examined whether the stepping stones were important or ment, as illustrated in Fig. S1. In the RvD treatment (Fig. S1B), even essential for sustained adaptation or, alternatively, were each deleterious (but not lethal) mutation was reverted, as if it mere detours occasionally followed by evolving populations but had never occurred. In the RpD treatment (Fig. S1C), each of little or no consequence for long-term fitness gains. deleterious mutation was replaced with a random beneficial, To address this issue, we first examined the dynamics of ad- neutral, or lethal mutation. In the RpDL treatment (Fig. S1D), aptation in 50 populations of digital organisms that evolved each deleterious or lethal mutation was replaced with a random under four mutation treatments (described in the next para- beneficial or neutral mutation. Finally, in the control treatment, graph). We seeded populations with a hand-written ancestral all mutations were allowed to enter the population normally. organism whose genome consisted of a copy loop with instruc- Fig. 1 shows the fitness trajectories for 50 populations under tions required for self-replication and a string of initially non- all four treatments. The control populations, which experienced functional instructions that change under the influences of deleterious mutations, reached higher fitness values, on average, mutation, drift, and selection as the organisms adapt to their than the treatment populations. We performed three Mann– environment. Our experiments thus correspond to a low-fitness Whitney tests to compare the fitness values of the final most- population with extensive opportunities for adaptation to a novel abundant genotypes in the control populations with the corre- environment. Genome length was held constant at 50 instruc- sponding values for each alternative treatment; the difference tions to facilitate and automate certain analyses. Each offspring was significant in every case (all P < 0.02). Differences among had a 25% chance of having a single point mutation, which the three treatments that prevented deleterious mutations were would change the instruction at a random site to one of 25 al- more subtle, but also significant (P = 0.0283; Kruskal–Wallis ternative instructions. The fitness landscape thus contained 2650 test). The RvD treatment, in which deleterious mutations were distinct genotypes, with the fitness of each depending on the reverted (thus reducing the total mutation rate), attained lower expression of the resulting genomic program and the fit of that fitness levels than the two treatments where deleterious muta- program to the environment, including its ability to produce tions were replaced (P = 0.0098 and P = 0.0575 versus RpDL offspring, its speed of replication, and its success in performing and RpD, respectively, Mann–Whitney tests). The two replacement one or more Boolean logic operations that provide additional treatments did not differ significantly (P = 0.5510, Mann–Whitney SIPs. The small genome size and high mutation rate are roughly test). Therefore, deleterious mutations in fact accelerated adaptive comparable with some RNA viruses. This genetic landscape and evolution by acting as stepping stones that facilitated the explora- these parameters are sufficiently complex to shed light on evolu- tion of rugged fitness landscapes. tion under circumstances beyond those amenable to formal theory or conventional population-genetic simulations. Moreover, the Case Study and Evolutionary Replays. To examine the importance computational environment of digital organisms allows us to of stepping stones in greater detail, we focused on the evolution perform experiments that are impossible with biological systems. of a particular function, EQU, which is the most complex (and Throughout this study, including in all treatment and control most valuable) operation in the experimental environment and populations, we disallowed double mutations to facilitate clas- was the focal trait in the study by Lenski et al. (11). In our sifying each mutation as beneficial, neutral, deleterious, or lethal. experiments, 28 of the 50 control populations evolved the EQU Before a mutant offspring was placed into the population, its function whereas 16 or fewer of 50 did so for each treatment that fitness was evaluated in an isolated test environment, thus taking precluded deleterious mutations. We performed Fisher’s exact advantage of an opportunity that exists in a computational realm tests comparing the frequency of evolving EQU in control pop- but not in a biological one, namely, to measure the effect of ulations with the corresponding frequencies under the other a mutation before it impacts a population’s evolution (Fig. S1A). treatments, and we found significant differences in all three Our initial experiments included a control treatment, where all cases (all P < 0.05), consistent with the effects observed on mutation types were allowed, and three other treatments that overall fitness. eliminated deleterious mutations: Revert Deleterious (RvD), We next chose one deleterious mutation in one replicate Replace Deleterious (RpD), and Replace Deleterious and Lethal population as a case study for in-depth analysis. That mutation

E3172 | www.pnas.org/cgi/doi/10.1073/pnas.1313424110 Covert et al. Downloaded by guest on September 26, 2021 ABCD PNAS PLUS

Fig. 1. Fitness trajectories under four treatments affecting the entry of deleterious mutations into evolving populations. Black lines show log10-transformed population mean fitness for 50 replicate populations in each treatment; red lines show the corresponding median across replicates. (A) Each deleterious (but not lethal) mutation was reverted, as if it had never occurred (RvD). (B) Each deleterious mutation was replaced with a random beneficial, neutral, or lethal mutation (RpD). (C) Each deleterious or lethal mutation was replaced with a random beneficial or neutral mutation (RpDL). (D) Control populations, in which all mutations were allowed to enter the population normally. Treatments that prevented deleterious mutations significantly impeded long-term adaptation. See text for statistical analyses.

reduced fitness by ∼9% when it first occurred (Fig. 2A). We refer pair. The remaining population followed a more complex route to this mutation as A. The subsequent mutation on the line of involving several intermediates, each with the same fitness de-

descent, N, was neutral and played no adaptive role, as evi- fect, before acquiring a mutation that finally produced EQU. EVOLUTION denced by constructing and comparing genotypes with and Thus, this lineage drifted across multiple stepping stones, illus- without that mutation and finding no differences in fitness or any trating the added effect of a neutral network (2, 28, 29). other trait. The next mutation, B, interacted with the earlier We then expanded this replay approach to include all cases deleterious mutation A to generate the ability to perform EQU among the original control populations that satisfied the fol- for the first time in that population. However, mutation B was lowing conditions: at least one pair of mutations interacted sign- nearly lethal on its own; it reduced fitness by ∼99% in constructed epistatically, the first mutation in that pair was substantially genotypes without mutation A. This example thus presents a del- deleterious in the background in which it first arose (>1% fitness eterious mutation that clearly served as a stepping stone (Fig. 2B). loss), and that same mutation was present and had become However, it is unclear from this single outcome whether other beneficial in the most abundant genotype of the final population. paths from the previous genotype could have produced the EQU A total of 36 cases fulfilled these criteria. For each of these cases, function without this or any other deleterious mutation. we used the RpD protocol to evolve 20 populations starting from To address that issue, we replayed evolution in the test case the deleterious mutant and 20 starting from its immediate pro- starting from two adjacent genotypes on the line of descent, one genitor, and we compared them after 20,000 updates (Fig. 3). carrying the first deleterious mutation (Ab) and the other its Overall, the fitness values were significantly higher in populations immediate progenitor (ab). We ran 20 replicates starting from founded by the deleterious mutants than in those founded by their each genotype for 20,000 updates (∼3,500 generations) under progenitors (P = 0.0223, Wilcoxon signed-ranks test comparing the RpD treatment that disallowed subsequent deleterious muta- medians, across replicates, of log-transformed fitness of the final tions. A duration of 20,000 updates was sufficiently long to provide dominant in each population). We also performed Mann–Whitney insight into the adaptive potential of these genotypes, while lim- tests for each pair of genotypes, using a Bonferroni correction iting the computational resources required to perform these (30) to adjust the overall false-positive rate to α = 0.05 across all experiments to a reasonable level. All 20 replays that started from 36 tests. There were significant differences in 11 cases, and, in all Ab evolved EQU whereas none of the replays starting from ab did. of them, the populations started with the deleterious mutant Therefore, the deleterious mutation A was crucial for the evolu- achieved higher fitness; it is extremely unlikely that, by chance, tion of EQU in that lineage (P  0.0001, Fisher’sexacttest). all 11 significant contrasts would occur in the same direction We also performed 20 replays with the same progenitor ge- (P = 0.0010, binomial test). In six cases, the mutant-derived notype (ab), except using the control protocol, in which delete- populations reached much higher fitness in at least some replays rious mutations could serve as stepping stones. Thirteen of these whereas, in five cases, they achieved only slightly higher levels populations evolved EQU, a highly significant difference from but did so consistently across most or all of the replays (Table the outcome under the RpD treatment (P < 0.0001, Fisher’s S1). Thus, many of the deleterious mutations that served as exact test). Nine populations that reevolved EQU acquired the stepping stones were not merely rare events of no lasting im- same pair of sign-epistatic mutations as in the original pop- portance; instead, they opened up adaptive routes that were ulation whereas three replays evolved a different sign-epistatic otherwise inaccessible.

Covert et al. PNAS | Published online August 5, 2013 | E3173 Downloaded by guest on September 26, 2021 A system (19). Also, neutral mutations typically have fewer phe- notypic effects than deleterious ones, and having some effects— even if they are deleterious—may be important for the epistatic interactions that allow stepping-stone mutations to promote adaptation. Both the RpN and RvN treatments yielded signifi- cantly lower fitness levels than did the control (P = 0.0127 and P = 0.0015, respectively, Mann–Whitney tests), but the difference between the RpNL treatment and control was not significant (P = 0.2626). On balance, the effects of excluding neutral mutations (Fig. 4) were similar to the effects of excluding dele- terious mutations (Fig. 1). There were no significant differences between the fitness levels for the RpN and RpD treatments or the RpNL and RpDL treatments (P =0.7433andP = 0.2398, respectively, Mann–Whitney tests). However, the RvD treatment produced significantly lower fitness levels than did the RvN treatment (P = 0.0490). It is important to note, however, that the RvD treatment resulted in a lower overall mutation rate than the RvN treatment because more muta- tions are deleterious than neutral in this system (19). More B generally, the fact that excluding either neutral or deleterious mutations caused comparable reductions in final fitness levels suggests that both contribute variation that is important for sustaining long-term adaptive evolution. Perhaps some fitness peaks are more accessible by neutral mutations and others by deleterious mutations. Deleterious mutations occurred under the RvN, RpN, and RpNL treatments, and they would often have replaced neutral mutations under the RpN and RpNL treatments. Therefore, we investigated whether deleterious mutations ameliorated the ab- sence of neutral mutations in these treatments by acting as stepping stones. As before, we identified all those mutations that reduced fitness by >1%, belonged to a sign-epistatic pair, and remained in the final dominant genotype. For the 50 replicates in each treatment, we found 200 such mutations in the RvN line- ages, 247 in the RpN lineages, and 95 in the RpNL lineages, compared with only 36 in the control lineages. The RpN treatment had the most deleterious mutations among the three treatments that prevented neutral mutations, and so we examined to what extent those deleterious mutations had con- Fig. 2. Analysis of a mutational sequence in a control population, showing tributed to the realized fitness gains under this treatment. We a deleterious mutation acting as a stepping stone. (A) Fitness values (W) for four successive genotypes on the line of descent and constructions of the performed paired sets of replay experiments, as before, starting four other possible combinations of the same three mutations. The initially from genotypes either with or without the initial deleterious mu- deleterious mutation, A, was beneficial only when combined with B; simi- tation in the 247 sign-epistatic mutations (Table S2). However, larly, B was beneficial only when combined with A. The other mutation, N, these replays were conducted under the RpD treatment, such that was neutral and had no effect on fitness either alone or in combination with neutral mutations were allowed but further deleterious mutations A, B, or both. All fitness values are expressed relative to the immediate were precluded, to assess the influence of the initial deleterious progenitor (abn). (B) Fitness landscape of the four genotypes, excluding the mutation only. Twenty-two of the 247 sets of replays achieved neutral mutation. Genotypes ab, Ab, and AB all appeared on the line of significantly higher fitness in the populations seeded with the descent whereas aB was constructed. Mutations A and B display a sign-epi- deleterious mutants whereas seven of the deleterious mutants led static fitness interaction. to significantly lower fitness (all P < 0.05 after Bonferroni cor- rection). Across all 247 sets of replays, the populations that started Effects of Reverting and Replacing Neutral Mutations. The finding with the deleterious mutants evolved significantly higher fitness that deleterious mutations can act as important stepping stones than those started with the progenitor (P = 0.0019, Wilcoxon for adaptive evolution raises the question of whether neutral signed-rank test). Thus, the replacement of neutral mutations by mutations play a similar role. To address that question, we deleterious ones tended to facilitate adaptation in the RpN pop- performed additional experiments in which we reverted (RvN) ulations, evidently because the populations found mutational paths involving sign-epistatic interactions. In contrast, populations or replaced (RpN) neutral mutations using the same experi- that evolved under the RvN treatment, where neutral mutations mental design described above. As before, we also controlled for fl were reverted (but not replaced), experienced an overall lower the in ated load of lethal mutations in the RpN treatment by mutation rate and fewer deleterious mutations than those under including a treatment in which we replaced both neutral and the RpN treatment. lethal (RpNL) mutations. We performed another set of experiments in which both del- One might expect that eliminating neutral mutations would eterious and neutral mutations were reverted or replaced to slow adaptive evolution even more than eliminating deleterious assess their combined effects on long-term adaptation. We thus mutations, on the grounds that neutral mutations have a longer eliminated the opportunity for evolving populations to use half life than deleterious mutations and thus neutral mutations any stepping stones that were not immediately adaptive; there- should persist as stepping stones for longer periods. On the other fore, adaptation could proceed only by a strict hill-climbing hand, there are fewer neutral than deleterious mutations in this process. As expected, these populations achieved much lower

E3174 | www.pnas.org/cgi/doi/10.1073/pnas.1313424110 Covert et al. Downloaded by guest on September 26, 2021 PNAS PLUS

Fig. 3. Median final fitness values for populations founded by genotypes with (y axis) or without (x axis) each of 36 initially deleterious mutations. The mutations were all of large effect (>1% fitness loss), and each later became beneficial in one of the original control populations via sign-epistatic interactions with subsequent mutations. We used each mutant to found 20 new populations, and we used its immediate progenitor (lacking the deleterious mutation) to found an additional 20 populations; the new populations evolved for 20,000 updates. All subsequent deleterious mutations were replaced (RpD treatment) in every population. The median fitness values for the final dominant genotypes in the 20 replicate populations are shown. The dashed line corresponds to the null hypothesis that final fitness values would be unaffected by the initial deleterious mutation. Values below the line indicate that the mutation impeded long-term adaptation; values above the line imply that the deleterious mutation promoted long-term adaptation. The symbols indicate whether the treatments with and without the initial mutation differed significantly (P < 0.05, Mann–Whitney test with Bonferroni correction; see Table S1). Filled square, significant difference; open square, no significant difference. See text for further statistical analyses.

fitness levels than the control populations that had access to fitness. Previous theoretical studies have implicitly assumed that EVOLUTION deleterious and neutral mutations (Fig. 5, all P  0.0001, Mann– the rate of compensatory adaptation increases with the mutation Whitney test). rate (5–9). In that case, one would expect that deleterious mutations would be more important at higher mutation rate than Effect of Mutation Rate on Use of Stepping Stones. All of the at lower rates. One possible explanation for this unexpected experiments above were performed with the mutation rate sim- result is that higher mutation rates drive populations toward ilar to the rate used in a previous study (11) that motivated our flatter areas of the fitness landscape, with a greater proportion of work; specifically, organisms had a 25% chance of a single point mutations being neutral (31) and thus fewer opportunities for mutation per replication cycle in our experiments. To examine compensatory changes. Further work will be necessary to de- the effect of mutation rate on the propensity of evolving pop- termine whether this explanation is correct. ulations to use deleterious mutations as stepping stones, we ran experiments at both lower and higher mutation rates, with the Effect of Recombination on Use of Stepping Stones. We also briefly probability of mutation per replication set to 8% and 75%, re- explored the role of deleterious mutations as stepping stones in spectively; other parameters were unchanged. We ran only the sexual populations. Recombination will tend to separate dele- control and RpD treatments, but we increased the number of terious mutations from subsequent mutations that interact epis- replicates to 200 for each treatment to improve statistical power. tatically with them, and so we expected recombination to reduce At the lower mutation rate, the median log-transformed fit- or eliminate the utility of deleterious mutations as stepping ness at the end of the runs was slightly lower for the RpD stones in adaptive evolution. We implemented haploid re- treatment than for the controls, but the difference was not sig- combination in Avida as described elsewhere (32). In brief, nificant (P = 0.0743, Mann–Whitney test) (Fig. S2). At the sexual digital organisms copied their genomes (potentially in- higher mutation rate, the median log fitness was actually slightly troducing mutations) just as asexual organisms did. However, higher for the RpD treatment than for the controls although this before each sexual offspring was placed in the population, a ran- difference was also not significant (P = 0.1786, Mann–Whitney dom segment of its genome was replaced by a corresponding test) (Fig. S3). Thus, the extent to which evolving populations segment from another sexual offspring that was also about to be use deleterious mutations as stepping stones is sensitive to the placed in the population. In the corresponding reversion treat- mutation rate and appears to reach some maximum at an in- ment, deleterious mutations were reverted before recombination termediate rate. Evolution via stepping stones may occur less occurred. That is, for each incipient sexual offspring, we first tested often at lower mutation rates because a deleterious mutation is whether it contained a deleterious mutation relative to its own more likely to be eliminated before the appearance of a potential parent, and, if so, the mutation was reverted before recombination compensatory mutation. This interpretation is consistent with and placement of the sexual offspring into the population. results from numerical simulations (12). We were more sur- The reversion treatment significantly depressed the final fit- prised, however, that, at higher mutation rates, the use of step- ness levels relative to the control populations in the sexual pop- ping stones did not produce a measurable increase in final ulations (P < 0.0001, Mann–Whitney test). This result implies that

Covert et al. PNAS | Published online August 5, 2013 | E3175 Downloaded by guest on September 26, 2021 A BCD

Fig. 4. Fitness trajectories under four treatments affecting the entry of neutral mutations into evolving populations. Black lines show log10-transformed population mean fitness for 50 replicate populations in each treatment; red lines show the corresponding median across replicates. (A) Each neutral mutation was reverted, as if it had never occurred (RvN). (B) Each neutral mutation was replaced with a random beneficial, deleterious, or lethal mutation (RpN). (C) Each neutral or lethal mutation was replaced with a random beneficial or deleterious mutation (RpNL). (D) Control populations, in which all mutations were allowed to enter the population normally. The RvN and RpN treatments significantly impeded long-term adaptation; the RpNL treatment tended toward lower fitness than the control treatment, but the difference was not significant. See text for statistical analyses.

many sign-epistatic interactions involved mutations that were volving sequential beneficial mutations in an empirical study on sufficiently close that their linkage was not completely disrupted the evolution of antibiotic resistance in bacteria. In theoretical by recombination events. Previous work with digital organisms work, Poelwijk et al. (1) assumed that populations can cross (32) has shown that recombination led to the evolution of more fitness valleys and thereby reach higher fitness peaks only by si- modular genomes, in which interacting sites tended to be phys- multaneous mutations, recombination, or relaxed selection. Other ically close on the genome. This modularity may protect sign- studies have considered the role of random drift on neutral epistatic interactions against disruption by recombination. A networks in allowing access to otherwise inaccessible fitness related study with Avida has confirmed that some initially delete- peaks, but these studies also ignored the role of deleterious rious mutations are indeed retained in sexual populations because mutations as potential stepping stones (1–3, 13, 14, 26, 33). Of they become beneficial after the appearance of linked sign- course, if a strictly uphill trajectory is mutationally accessible, epistatic mutations (27). Not surprisingly, however, the differ- then an evolving population will be more likely to take that path ence between the reversion and control treatments was smaller than one that involves a deleterious intermediate state, all else in the sexual populations (Fig. S4) than in the asexual populations being equal. However, strictly uphill trajectories may not exist for (Fig. 1), presumably because some sign-epistatic interactions were every population; even when they exist, they may not be readily disrupted by recombination. accessible depending on the number of mutations and their effects. Recombination might also generate additional deleterious An earlier study (11) using Avida found many cases where dele- mutations—or, more precisely, expose mutations with deleteri- terious mutations were present on the line of descent and showed ous effects—if some neutral mutations become detrimental that, in some cases, these mutations played important roles in the when moved into other genetic backgrounds. Given the role of evolution of complex new functions. Following that work, several deleterious mutations as stepping stones in long-term adapta- theoretical studies (5, 8, 9, 12, 34) examined the role of delete- tion, one might then imagine that sexual populations should rious mutations in crossing fitness valleys. However, these studies evolve higher fitness than asexual ones because sexual populations did not address whether this phenomenon was a mere curiosity— experience more of these potential stepping stones. However, we something that might occasionally happen, but without substantial found no evidence of such an effect; sexual and asexual pop- or lasting effect on the process of adaptation—or, alternatively, ulations in the control treatments achieved comparable final fit- whether deleterious mutations might be important for sustaining ness values (P = 0.6124, Mann–Whitney test). Thus, any potential long-term adaptation. Here, we took advantage of the possibility benefits of sexual reproduction were offset by costs, including the in Avida (unlike organic systems) of testing the fitness effects disruption of beneficial interactions between mutations. of mutations before they are placed in a population, and then reverting or replacing some of them to preclude all deleterious Conclusions and Synthesis. At first glance, it seems entirely rea- mutations. These experiments demonstrate that deleterious muta- sonable to think that adaptation by natural selection proceeds tions can, in fact, substantially increase fitness gains over long through the appearance and fixation of beneficial mutations, and periods. However, the generality of this effect remains unclear and indeed many studies assume that adapting populations must worthy of future empirical and theoretical investigation. We show follow paths of beneficial mutations through genotype space. For that the effect is sensitive to mutation rate, with the long-term example, Weinreich et al. (3) considered only those paths in- benefit of deleterious mutations highest at an intermediate muta-

E3176 | www.pnas.org/cgi/doi/10.1073/pnas.1313424110 Covert et al. Downloaded by guest on September 26, 2021 A B C D PNAS PLUS

Fig. 5. Fitness trajectories under treatments affecting the entry of both neutral and deleterious mutations into evolving populations. Black lines show log10- transformed population mean fitness for 50 replicate populations in each treatment; red lines show the corresponding median across replicates. (A) Each neutral or deleterious (but not lethal) mutation was reverted, as if it had never occurred. (B) Each neutral or deleterious mutation was replaced with a random beneficial or lethal mutation. (C) Each neutral, deleterious or lethal mutation was replaced with a random beneficial mutation (if one was found within 100 attempts). (D) Control populations, in which all mutations were allowed to enter the population normally. The treatments that prevented both neutral and deleterious mutations significantly impeded long-term adaptation relative to treatments that prevented only neutral or deleterious mutations. See text for statistical analyses.

tion rate in this system. We also show that the effect is diminished, set to zero. This type of unstable genotype can occur when an organism’sself- replication process is damaged. For example, a point mutation might cause the but not eliminated, by recombination in this system. EVOLUTION organism to copy only part of its genome, or to copy part of its genome more Materials and Methods than once. We disallowed mutations that produced unstable genotypes in all Treatments to Exclude Deleterious Mutations. We used three treatments to treatments, including the control treatment, because our analyses were preclude the occurrence of deleterious mutations that might act as stepping predicated on single mutations occurring in genomes of constant length. stones during adaptive evolution (Fig. S1A). These treatments differed in their impact on the overall mutation rate and the distribution of fitness Measuring the Fitness of Digital Organisms. In the isolated test environment, effects (Fig. S1 B–D). and with additional mutations prevented, we evaluated each candidate The RvD treatment reverted any nonlethal deleterious mutation to its mutant’s fitness by allowing it to execute its genome. We measured two previous state. Lethal mutations were not reverted because they cannot act as aspects of its performance: the rate at which it acquired SIPs (single-instruction stepping stones. Deleterious mutations make up a large proportion of all processing units) and the number of instructions executed to replicate itself. mutations (typically about half in evolved genomes), and therefore this The ratio of these two numbers is a close approximation to the organism’s treatment reduced the overall mutation rate (Fig. S1B). To account for this absolute fitness. If digital organism A has twice the fitness of organism B, then effect on the mutation rate, the RpD treatment replaced a nonlethal dele- A will, on average, produce twice as many offspring as B in the same amount terious mutation with another mutation by randomly sampling potential of time. This expectation is not frequency-dependent; i.e., it is independent of mutations until one belonging to a permissible class—beneficial, neutral, or the relative abundance of A and B. Note also that previous work has shown lethal—was found. The RpD treatment thus maintained the overall muta- that the outcome of direct competition is quantitatively consistent with the tion rate. However, this treatment increased the load of lethal mutations calculated differences in fitness except at very high mutation rates (31), where because they generally comprised the largest proportion of the permissible differences in offspring viability may become important; this effect is minimal C classes (Fig. S1 ). The RpDL treatment addressed the increased load of lethal at the mutation rates used in our study. mutations by replacing both deleterious and lethal mutations, again main- An organism obtained additional SIPs if, in the course of executing its taining the overall mutation rate but allowing only two classes of mutations, genomic program, it performed one or more of nine distinct one- and two- neutral and beneficial (Fig. S1D). input logic operations, according to the multiplicative schedule in Table S3.If Two exceptions sometimes occurred that are not covered above. First, in a mutant could not self-replicate (within an allotted maximum time), the the RpD and RpDL treatments, we tested a maximum of 100 candidate mutation was categorized as lethal. Otherwise, we classified the mutation as replacements before stopping the search to find a mutation belonging to an beneficial, neutral, or deleterious based on whether the mutant’s fitness was allowed class. This threshold was never reached under the RpD treatment higher than, equal to, or lower than that of its immediate progenitor. because lethal alternative mutations were always abundant. However, in the RpDL treatment, the two largest classes of mutation (deleterious and lethal) were disallowed, and this threshold was occasionally reached. We examined Reversion and Replacement of Neutral Mutations. The RvN, RpN, and RpNL 1,092,337 attempted replacements at 200 uniformly spaced updates across treatments followed the same procedures as the RvD, RpD, and RpDL the 50 RpDL populations, and only 281 (<0.03%) failed to identify an ac- treatments, respectively, except that neutral mutations were reverted or ceptable replacement. In these rare cases, the original mutant was elimi- replaced rather than deleterious mutations. For the RvN and RpN treatments, nated to prevent any deleterious mutations from entering the population. mutations that changed fitness by <1% were considered to be neutral The second exception occurred in all treatments, including the controls. We whereas, for the RpNL treatment, mutations were treated as neutral only if occasionally encountered a mutant that, when tested in isolation, produced they did not affect fitness at all. Other experiments suggested that the exact an offspring that was not an exact copy of itself, even if the mutation rate was definition of neutrality was immaterial to our general conclusions.

Covert et al. PNAS | Published online August 5, 2013 | E3177 Downloaded by guest on September 26, 2021 Reversion and Replacement in Sexual Populations. Sexual recombination in Data Analyses and Statistics. Analyses and statistics were performed in Py- digital organisms was implemented as described elsewhere (27, 32). In brief, thon, using the libraries numpy, scipy, and matplotlib. Wilcoxon signed-rank an organism first produced an asexual copy of its genome. Then, two copies tests were performed in R. All statistical tests were two-tailed. To correct for from different parents exchanged a section of their genomes to produce multiple tests in the replay experiments, we adjusted P values using the two offspring; we used two randomly chosen crossover points to define Dunn–Sidak method of the Bonferroni correction (30). that section. Potential mutations were tested for deleterious effects, and reversions ACKNOWLEDGMENTS. We thank C. Adami, J. Barrick, D. Bryson, F. Howes, M. Rupp, C. Strelioff, B. Walker, D. Weinreich, and M. Wiser for valuable were performed after genomes were copied but before recombination discussions. This research was supported, in part, by the Defense Advanced events. Therefore, mutations were judged as deleterious or not only against Research Projects Agency “FunBio” program (HR0011-09-1-0055), National Sci- the background of the parent in which they arose. We did not perform ence Foundation (NSF) Grant CCF-0643952, and the BEACON Center for the replacement (rather than reversion) treatments in sexual populations. Study of Evolution in Action (NSF Cooperative Agreement DBI-0939454).

1. Poelwijk FJ, Kiviet DJ, Weinreich DM, Tans SJ (2007) Empirical fitness landscapes reveal 18. Schrag SJ, Perrot V, Levin BR (1997) Adaptation to the fitness costs of antibiotic accessible evolutionary paths. Nature 445(7126):383–386. resistance in Escherichia coli. Proc Biol Sci 264(1386):1287–1291. 2. van Nimwegen E, Crutchfield JP (2000) Metastable evolutionary dynamics: Crossing 19. Lenski RE, Ofria C, Collier TC, Adami C (1999) Genome complexity, robustness and fitness barriers or escaping via neutral paths? Bull Math Biol 62(5):799–848. genetic interactions in digital organisms. Nature 400(6745):661–664. 3. Weinreich DM, Delaney NF, Depristo MA, Hartl DL (2006) Darwinian evolution can 20. Elena SF, Wilke CO, Ofria C, Lenski RE (2007) Effects of population size and mutation follow only very few mutational paths to fitter proteins. Science 312(5770):111–114. rate on the evolution of mutational robustness. Evolution 61(3):666–674. 4. Coyne JA, Barton NH, Turelli M (1997) Perspective: A critique of Sewall Wright’s 21. Burch CL, Chao L (1999) Evolution by small steps and rugged landscapes in the RNA shifting balance theory of evolution. Evolution 51:643–671. virus φ6. Genetics 151(3):921–927. 5. Iwasa Y, Michor F, Nowak MA (2004) Stochastic tunnels in evolutionary dynamics. 22. Woods RJ, et al. (2011) Second-order selection for evolvability in a large Escherichia Genetics 166(3):1571–1579. coli population. Science 331(6023):1433–1436. 6. Carter AJR, Wagner GP (2002) Evolution of functionally conserved enhancers can be 23. Chou H-H, Chiu H-C, Delaney NF, Segrè D, Marx CJ (2011) Diminishing returns accelerated in large populations: A population-genetic model. Proc Biol Sci 269(1494): epistasis among beneficial mutations decelerates adaptation. Science 332(6034): 953–960. 1190–1192. 7. Weinreich DM, Chao L (2005) Rapid evolutionary escape by large populations from 24. Khan AI, Dinh DM, Schneider D, Lenski RE, Cooper TF (2011) Negative epistasis local fitness peaks is likely in nature. Evolution 59(6):1175–1182. between beneficial mutations in an evolving bacterial population. Science 332(6034): 8. Weissman DB, Desai MM, Fisher DS, Feldman MW (2009) The rate at which asexual 1193–1196. populations cross fitness valleys. Theor Popul Biol 75(4):286–300. 25. Ofria C, Wilke CO (2004) Avida: A software platform for research in computational 9. Gokhale CS, Iwasa Y, Nowak MA, Traulsen A (2009) The pace of evolution across evolutionary biology. Artif Life 10(2):191–229. fitness valleys. J Theor Biol 259(3):613–620. 26. Covert AW III, Carlson-Stevermer J, Derryberry DZ, Wilke CO (2012) The role of del- 10. Campos PR, Wahl LM (2009) The effects of population bottlenecks on clonal eterious mutations in the adaptation to a novel environment. Artificial Life 13, eds interference, and the adaptation effective population size. Evolution 63(4):950–958. Adami C, Bryson DM, Ofria C, Pennock RT (MIT Press, Cambridge, MA), pp 27-31. 11. Lenski RE, Ofria C, Pennock RT, Adami C (2003) The evolutionary origin of complex 27. Covert AW, Smith L, Derryberry DZ, Wilke CO (2012) What does sex have to do with features. Nature 423(6936):139–144. it: Tracking the fate of deleterious mutations in sexual populations. Artificial Life 12. Cowperthwaite MC, Bull JJ, Meyers LA (2006) From bad to good: Fitness reversals and 13,edsAdamiC,BrysonDM,OfriaC,PennockRT(MITPress,Cambridge,MA),pp the ascent of deleterious mutations. PLOS Comput Biol 2(10):e141. 32–36. 13. Weinreich DM, Watson RA, Chao L (2005) Perspective: Sign epistasis and genetic 28. van Nimwegen E, Crutchfield JP, Huynen M (1999) Neutral evolution of mutational constraint on evolutionary trajectories. Evolution 59(6):1165–1174. robustness. Proc Natl Acad Sci USA 96(17):9716–9720. 14. Bridgham JT, Carroll SM, Thornton JW (2006) Evolution of hormone-receptor 29. Wilke CO (2001) Adaptive evolution on neutral networks. Bull Math Biol 63(4): complexity by molecular exploitation. Science 312(5770):97–101. 715–730. 15. McKenzie JA, Whitten MJ, Adena MA (1982) The effect of genetic background on the 30. Holm S (1979) A simple sequentially rejective multiple test procedure. Scandanavian fitness of diazinon resistance genotypes of the Australian sheep blowfly, Lucilia Journal of Statistics 6:65–70. cuprina. Heredity 49:1–9. 31. Wilke CO, Wang JL, Ofria C, Lenski RE, Adami C (2001) Evolution of digital organisms 16. Maisnier-Patin S, Berg OG, Liljas L, Andersson DI (2002) Compensatory adaptation to at high mutation rates leads to survival of the flattest. Nature 412(6844):331–333. the deleterious effect of antibiotic resistance in Salmonella typhimurium. Mol Microbiol 32. Misevic D, Ofria CA, Lenski RE (2006) Sexual reproduction reshapes the genetic 46(2):355–366. architecture of digital organisms. Proc Biol Sci 273(1585):457–464. 17. Maisnier-Patin S, Andersson DI (2004) Adaptation to the deleterious effects of 33. Lunzer M, Miller SP, Felsheim R, Dean AM (2005) The biochemical architecture of an antimicrobial drug resistance mutations by compensatory evolution. Res Microbiol ancient adaptive landscape. Science 310(5747):499–501. 155(5):360–369. 34. McCandlish DM (2013) On the findability of genotypes. Evolution, 10.1111/evo.12128.

E3178 | www.pnas.org/cgi/doi/10.1073/pnas.1313424110 Covert et al. Downloaded by guest on September 26, 2021