<<

Downloaded by guest on October 3, 2021 www.pnas.org/cgi/doi/10.1073/pnas.1702994114 a lift segregation Wittmann J. via Meike loci many at maintain can selection fluctuating Seasonally eprlheterogeneity temporal genetic maintaining to empiri- in contribution populations. lift its natural segregation quantify in for variation to test and formally data direction to important cal is An detect. work to future hard exhibit for and can small be loci may lift, fluc- parameters tuations often segregation in but fluctuations, differences Under -frequency seasonal to seasons. conspicuous robust and is loci and prob- across load, by genetic affected of not is lems , works diminishing-returns lift Segregation under fitness. best for traits naturally of in importance arise changes relative seasonal may the and pleiotropy that seasonal antagonistic with phenomenon requires situations a in lift,” dominance, “segregation in call changes general we This which dominant. possible sufficiently is mechanism, are loci many favored at currently epis- polymorphism simula- if stable and individual-based that selection show and fit- we of analysis nonlinear tions, strength mathematical a the Using for uses tasis. account step to second function homozygous The ness and dominance. accounts heterozygous and is, of loci loci—that across contributions additive relative in is the fitness mapping to for first mapped mod- diploids. The is of in genotype steps. multilocus two class selection the general fluctuating models, these more seasonally In a multilocus explore with we els Here, polymorphism multilocus restrictive. for are conditions the that previous However, suggests populations. theory main- in to polymorphism contribution large genetic in a taining make changes potentially could seasonal 2017) selection fluctuat- 8, by ing Seasonally March availability. review resource affected for or (received are rainfall, 2017 temperature, 3, populations October approved natural and CA, Most Irvine, California, of University Clegg, T. M. by Edited 19104-6313 PA Philadelphia, Pennsylvania, of University Universit ntesrnt n ieto fslcin oho phenotypes on both selection, of direction lim- and of fluctuations strength be temporal the widespread to in despite believed (10), is more importance heterogeneity. contrast, ited considered environmental only by mechanism spatial heterogeneity, at A is Temporal acting (7–9). powerful neu- perhaps species overdomi- and (nearly) per part, common with loci that minor variation, of is relatively tens genetic a view most playing majority nance cause the mutations although tral 6), (5, tled (4). pressures selection in spa- variability temporal and of bal- or selection, tial meaning of frequency-dependent (over- the form negative advantage substan- over dominance), some heterozygote controversy example, a by (3)]—for some that maintained “substantial” [with posits is scale selection genomic variation school ancing a of “balance” but on fraction the loci, rare tial contrast, exceedingly some are By at loci (2). frequency such may intermediate selection that that argues at admits alleles view drift, neoclassical maintain genetic muta- The mutation, deleterious between selection. weakly equilibrium and or an neutral at of to present bulk tions due 3). the is (2, that variation claim schools school scientific genetic over “(neo)classical” two Dispute the to (1). of rise Proponents populations gave natural reasons in underlying the abundance its by zled E overdominance eateto ilg,Safr nvriy tnod A94305; CA Stanford, University, Stanford Biology, of Department it er ae,tedbt a o encnlsvl set- conclusively been not has debate the later, years Fifty e ic ilgsswr rtal odtc population detect puz- to been have able they level, first molecular the were at variation biologists genetic since ver tBeeed 31 ilfl,Germany; Bielefeld, 33615 Bielefeld, at ¨ | aacn selection balancing a,b,c,1 | ylclselection cyclical lnO Bergland O. Alan , | eei diversity genetic d eateto ilg,Uiest fVrii,Caltevle A294 and 22904; VA Charlottesville, Virginia, of University Biology, of Department a,d acsW Feldman W. Marcus , | marginal b Fakult tf at ¨ rMteai,Universit Mathematik, ur ¨ 5142262 1073/pnas.1702994114/-/DCSupplemental at online information supporting contains article This been has manuscript this at 1 in (available analyses repository figshare the the underlying in code deposited Source deposition: Data the under Published Submission. Direct PNAS a is article D.A.P. This interest. and M.W.F., of conflict A.O.B., no M.J.W, declare authors and The paper; the on paper. input aspects the gave all wrote P.S.S. on input analyses; gave D.A.P. the and M.W.F., of A.O.B., simulations; and analyses performed oiac”,atog naypriua eeain n fthe 20). 19, of (16, one fittest generation, be over- might particular (“marginal homozygotes 18). any fitness heterozy- and in mean if although 17 geometric stable dominance”), refs. highest is see locus the single but have fitness a gotes 16, at mean (ref. polymorphism geometric fixation diploids, higher In to cannot a go generally have eventually alleles will and two one haploids, because In coexist (16). fit- time mean over geometric their nesses on depend selection fluctuating porally number large a at selection loci. should of organisms fluctuating many seasonally whereas polygenic, experience usually reproduction, such resource are Since survival. in traits abundant in life-history investment investment for with select sea- for may seasons seasons across stressful select example, trade-offs might For life-history supply 15). are ani- (14, there for sons availability tree Often, food between (13). in and changes mals in seasonal within fruiting to synchronized leading and often species, flowering is example, forests com- experi- For tropical predators, usually seasonality. populations of some tropical abundance ence Even the parasites. or in petitors, temporal or of availability, rainfall, type temperature, resource in particular multiple example, with a for seasonality, organisms experience heterogeneity: most year fact, per In generations (12). genotypes and (11) uhrcnrbtos ...... n ...dsge eerh M.J.W. research; designed D.A.P. and P.S.S., M.W.F., A.O.B., M.J.W., contributions: Author owo orsodnemyb drse.Eal [email protected] Email: addressed. be may correspondence whom To [email protected]. or se o ayseisadde o ufrfo rbesof problems from suffer not does models. previous and sat- species be many plausibly for may season- isfied condition by This polymorphism selection. fluctuating multilocus condition ally of general more maintenance a the obtain We for at genome. the polymorphism in maintain loci many simultaneously fluctu- seasonally to empiri- of selection recent potential ating by the reevaluate Inspired rarely. we findings, occurs cal this research that previous argued However, has polymorphism. mecha- maintain powerful to potentially a nism sea- year, is per selection generations fluctuating multiple their sonally have in they seasonality if popula- and, some maintains natural habitats, experience What in organisms observed is: Many variation tions? biology genetic evolutionary abundant in the question key A Significance ihdsrt eeain,teftso eoye ne tem- under genotypes of fates the generations, discrete With a alS Schmidt S. Paul , ). NSlicense. PNAS tWe,19 in Austria; Wien, 1090 Wien, at ¨ e n mtiA Petrov A. Dmitri and , . https://doi.org/10.6084/m9.figshare. www.pnas.org/lookup/suppl/doi:10. e NSEryEdition Early PNAS eateto Biology, of Department c Fakult a,1 tf at ¨ rBiologie, ur ¨ | f10 of 1

EVOLUTION PNAS PLUS Extending these results to the multilocus case is nontrivial, alone is not sufficient to reconcile evidence from population and, so far, only two cases are well-understood: (i) multiplicative genomics and quantitative genetics (38). Thus, we need to recon- selection across loci and (ii) temporally fluctuating selection on sider the potential of temporally fluctuating selection to maintain a fully additive trait. Under multiplicative selection in an infinite multilocus polymorphism. population with free recombination, the allele-frequency dynam- As explained above, the conditions for multilocus polymor- ics at a focal locus are independent of those at other loci. Thus, phism under seasonally fluctuating selection have been examined polymorphism is stable if heterozygotes have the highest geo- mostly in two narrow cases. Here, we examine a more general metric mean fitness, as in the single-locus case. However, devi- class of seasonal selection models with various forms of domi- ations from multiplicative selection appear to be the rule. In nance and epistasis. Using deterministic mathematical analysis particular, beneficial mutations often exhibit diminishing-returns and stochastic simulations, we show that multilocus polymor- epistasis (21–23). Additionally, there is the potential problem of phism is possible if the currently favored allele at any time is genetic load. Genetic load is commonly defined as the differ- sufficiently dominant, with dominance measured by using a scale ence between the population’s average fitness and the fitness of on which contributions across loci are additive. This mechanism, the fittest possible genotype. Lewontin and Hubby (1) noticed which we call “segregation lift,” can maintain polymorphism at that this value can become unsustainably high if there is strong a large number of loci across the genome, is robust to many at many loci. This was a conundrum for model perturbations, and does not require single individuals to the neoclassical school, which was worried that with high genetic have too many offspring. Depending on the parameter values, load, single individuals would have to produce an astronomically allele-frequency fluctuations can be large and readily detectable, large number of offspring. Others have dismissed this concern, or subtle and hard to discern. arguing, for example, that selection does not generally act on all loci independently or that only relative fitness differences within Basic Model the population are relevant, not fitness relative to some optimum We consider a diploid, randomly mating population in a season- genotype that might not even exist (24–27). However, debate ally fluctuating environment. While asymmetry in various model continues over whether genetic load should be an important con- parameters will be explored later, we start with a fully symmetric sideration (28, 29). model having a yearly cycle with g generations of winter followed The second previously studied scenario is seasonally fluctuat- by g generations of summer. The genome consists of L unlinked ing selection on a trait to which loci contribute additively (30, loci with two alleles each: one summer-favored and one winter- 31). These models generally assume additivity also within loci, favored allele. For a given multilocus genotype, let ns and nw such that the contribution of heterozygotes is exactly intermedi- be the number of loci homozygous for the summer and winter ate between the contributions of the two homozygotes. Tempo- allele, respectively, and nhet the number of heterozygous loci, rally fluctuating selection can then cause intermediate trait val- with ns + nw + nhet = L. ues to be best in the long run (32), i.e., select against variance In the basic model, loci are interchangeable in their effects in fitness. Effectively, this is on the temporal (see Stochastic Simulations for a more general model), and the average. As such, it can generally maintain polymorphism at only fitness of a multilocus genotype can be computed as a function one locus (33, 34), or two loci if their effect sizes are sufficiently of ns , nw and nhet . In the simplest case, fitness depends only on different (35) or if they are closely linked (30). The reason is that ns + 0.5 · nhet in summer and nw + 0.5 · nhet in winter, i.e., half with multiple loci and additivity within and between loci, there the number of currently favored allele copies. To allow for domi- are multiple genotypes with intermediate phenotypes. For two nance effects, we generalize this simple scenario and assume that loci, for example, there is the double heterozygote (“heterozy- fitness in summer depends on the summer score gous intermediate”) and the genotype homozygous at both loci, but for alleles with opposite effects (“homozygous intermedi- zs := ns + ds · nhet [1] ate”). These genotypes may all have the same high fitness. How- ever, matings between heterozygous intermediates produce a and fitness in winter depends on the winter score range of different genotypes, some of which are less fit than their z := n + d · n . [2] parents. By contrast, matings between homozygous interme- w w w het diates only produce new homozygous intermediates. Homozy- The parameters ds and dw quantify the dominance of the cur- gote intermediates can therefore go to fixation and eliminate all rently favored allele in summer and winter, not with respect to polymorphism. fitness, but with respect to the seasonal scores zs and zw . Because In summary, multiplicative seasonal selection is a power- we are interested in whether temporally fluctuating selection can ful mechanism to maintain multilocus polymorphism, but the maintain polymorphism in the absence of other stabilizing mech- assumed independence across loci and the associated load call anisms, we only consider values of ds and dw between 0 and 1, into doubt its plausibility. On the other hand, selection on addi- and do not allow values >1, which would correspond to standard tive traits can maintain polymorphism at only a few loci. So heterozygote advantage. far, there has been little need for further exploration because The relationship between the seasonal score z (z = zs in sum- there were no clear empirical examples to challenge the view mer and z = zw in winter), and fitness, w, is given by a mono- that temporal heterogeneity rarely maintains variation. This is tonically increasing fitness function w(z). This function specifies now changing, however, as advances in sequencing technology the strength of selection and accounts for epistasis. With discrete allow detailed studies of genetic variation across time and space. generations, the allele-frequency dynamics at a focal locus are For instance, by sampling the same temperate population of driven by the relative fitnesses of the three possible genotypes at Drosophila melanogaster at several time points, Bergland et al. that locus—for example, the ratio of the fitness of homozygotes (36) detected seasonal allele-frequency fluctuations at hundreds and heterozygotes. We say that there is no epistasis if these ratios of sites in the genome. Many of the SNPs are also shared with and thus the strength of selection are independent of the number African populations of D. melanogaster and some even with the of other loci and their contributions to z. This is the case when sister species Drosophila simulans, indicating that some of them fitness is multiplicative across loci: may be ancient balanced polymorphisms. More generally, recent L L population genomic data appear to suggest that balancing selec- Y X tion contributes more to maintaining genetic variation than pre- w = wi ⇔ ln(w) = ln(wi ), [3] viously assumed (37) and that mutation-selection-drift balance i=1 i=1

2 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1702994114 Wittmann et al. Downloaded by guest on October 3, 2021 Downloaded by guest on October 3, 2021 with still selection. allele per-locus stronger favored overall additional an with having increases of advantage increasing with selective negative more all for becomes negative is scale epistasis that increasing logarithmic reveals with the linearly to than transformation faster increases ness htrvra fdmnnewudas ept ananpoly- maintain to we that selection, help fluctuating assume also temporally multilocus would under dominance morphism of Hypothesizing (42). reversal selection that fluctuating dominance” single-locus temporally bene- in of for respective polymorphism “reversal models of the Such maintenance locus 41). the a facilitates (40, also by dominant affected is polymor- allele trait maintain ficial each to for likely if most suggest phism is form studies a pleiotropy theoretical versa, antagonistic vice Previous that and pleiotropy. score winter antagonistic low of a but score summer high 2). (step epistasis and strength dominance selection disentangles from process 1) two-step (step This the 1). of (Fig. fitness counter to (Eqs. alleles generalized favored a of score number essentially seasonal additively, a onto contribute genotype multilocus the q with parameter positive a with form the of is first The 1). (Fig. functions fitness to of loci classes other two on of contribution increasing with weaker (v epistasis returns contribution to increasing loci with other strength the in of increases locus focal with a linearly at than epistasis). faster of increases definition fitness similar of (v a epistasis synergistic for or 39 positive ref. Under (see fitness epistasis of of logarithm sure the of derivative second by the characterized is model plicative urnl ifvrdall.With allele. disfavored currently and heterozygous, allele, favored currently with setting fulfilled by achieved is oaihi scale. logarithmic In parameters. various 1. Fig. where itane al. et Wittmann , w(z) A > h eodcaso tesfntosi fteform the of is functions fitness of class second The normdl eoye ihmn umrallshv a have alleles summer many with genotypes model, our In maps first The steps. two in computed is fitness summary, In

0 5 10 15 20 1 q q 0246810 eg,mgnalnsi i.1). Fig. in lines magenta (e.g., w 1 = ≥ xmlsfrfins ucin eeae yEq. by generated functions fitness for Examples i hsfnto eue otemlilctv model multiplicative the to reduces function This 1. stefins au tlocus at value fitness the is ca ie nFg )adhspstv psai with epistasis positive has and 1) Fig. in lines (cyan d s w y [5] i and [4] o all for q exp(1) = [5] =1.5 z 00 w y tesi hw nalna cl,adin and scale, linear a on shown is fitness A, ycnrs,udrngtv rdiminishing- or negative under contrast, By . (z =4 q i d [4] =1 x()=1 = exp(0) = [4] [4] ) w w y w < w z y y =0.5 nEq. in w =1 =2 (z (z hs larger Thus, . (z ,slcina oa ou becomes locus focal a at selection 0), i y Seasonal score, 1+ (1 = ) exp(z = ) flocus if exp(z = ) exp(d = lhuhfor Although . 1 1 B and v

3 6 and v

1 10 10 (z fi shmzgu o the for homozygous is it if 00 s z 0246810 i ) ,adtescn maps second the and 2), q (z ) ) y =ln(w := ) ) y z or 00 shmzgu o the for homozygous is 2 .Eitss(v Epistasis 1B). (Fig. eteeoeuse therefore We 0. = ) eas hnEq. then because y (z y v i aetesm value, same the take but , normdl this model, our In . (z auscrepn to correspond values w ) z i y n hsselection thus and , = ) > exp(d = > ,telogarithm the 0), (z v v z 1 0 00 h multi- the )), y 4 owihloci which to (z nEq. in n hsthe thus and z z rEq. or · ) efocus We . Fg 1A), (Fig. n1+ ln(1 w samea- a as ) fi is it if na on B, 5 fit- 4, 3 with [5] [4] z 00 is z ) ) ue oiac o ttesaeo tesbta h cl of scale the for S1 necessary at Fig. nor but sufficient neither fitness is of ness scale score, because the seasonal Interestingly, at the fitness. not and dominance score sures higher a score for have same Finally, the fitness. have and intermediates homozygous and For erozygous intermediates. homozygous than seasons score, both seasonal lower a fully have For are but 2). alleles, winter (Fig. and homozygous summer of also number which same intermediates, the have homozygous heterozy- with some least compared at loci, and gous alleles winter and summer same of the number with genotypes multilocus intermediates, heterozygous of value the cur- Importantly, the If and dominant. (41), always score dominance” is allele for of favored whereas reversal rently recessive, “beneficial cur- always have the is and we (41), allele dominance” favored of rently reversal “deleterious have we Simulations Stochastic d line). solid (black intermediate homozygous a and line) (blue mediate parameter, 2. Fig. cospeorpcefcscnaiei i rnhdezm path- (44). enzyme feedbacks branched or via saturation dominance with is reces- in ways arise is changes can and such effects which fecundity pleiotropic in smaller across (blue) way dominant to specific is allele One leads and there. it winter sive tolerance whereas the starvation trait, a example, this higher of for For produces effects 3 traits. pleiotropic Fig. two the in the to on respect requires reversal; with locus scenario a dominance This in necessarily seasons. S2A) not between constant (though changes remain changes traits traits the two of the year, on the pleiotropy. effects throughout starvation antagonistic allelic is example, the there traits—for Although fecundity—and additive) and (also the tolerance 3, two Fig. of in average scenario in example score, the changes seasonal additive In seasonal without possible. map, even are But genotype–phenotype dominance expression. the in in in of changes changes reversal deleterious medi- seasonal beneficial be and by a could ated expect dominance season in thus changes one might Alternatively, dominance. we during that season, favored such other fluctuating the is is allele selection pathways If each (43). metabolic recessive multistep generally are affecting that mutations suggests theory deleterious control metabolic example, For dominance. hc en htdmnnesice ewe esn (see seasons between switches dominance that means which , utpemcaim ol neleasaoa eeslof reversal seasonal a underlie could mechanisms Multiple z ). auso h esnlscore, seasonal the of Values sadtv,ntjs ewe oi u lowti loci. within also but loci, between just not additive, is d o w xml orlcsgntps eeoyosinter- heterozygous a genotypes: four-locus example two for , z eeca eeslo oiac o fit- for dominance of reversal beneficial , d o oegnrlmdl.For model). general more a for z > d d sacmoiepeoye weighted a phenotype, composite a is , d lodtrie h eaiefins of fitness relative the determines also 0.5 > < z n hrfr oe tesin fitness lower a therefore and , eeoyosintermediates heterozygous 0.5, eas h eaieimportance relative the because eeoyosintermediates heterozygous 0.5, z safnto ftedominance the of function a as , NSEryEdition Early PNAS d d > 0 = Fig. Appendix, SI 0.5 h seasonal the .5, IAppendix (SI d het- 0.5, = d d | d > < f10 of 3 mea- 0.5, 0.5, ,

EVOLUTION PNAS PLUS detail below, appears to be the only way in which seasonally fluc- tuating selection can maintain polymorphism under additivity (d = 0.5). Next, we explore whether deviations from additivity (d =6 0.5) can facilitate multilocus polymorphism. Under multiplica- tive selection, i.e., without epistasis, the conditions for polymor- phism at one locus are not affected by the dynamics at other loci. Thus, given the fitness values for individual loci (exp(d) for het- erozygotes, exp(1) and exp(0) for currently favored and disfa- vored homozygotes, respectively, see text below Eq. 3), we can conclude that polymorphism is possible if exp(d)2 > exp(1) · exp(0) ⇔ d > 0.5. [6]

Fig. 3. Potential mechanistic underpinning for beneficial reversal of dom- That is, there must be a reversal of dominance with respect to z, inance. There is antagonistic pleiotropy for two traits, and the seasonal such that at any time the currently favored allele is dominant. scores for winter and summer are computed as weighted averages of traits 1 Now, we explore whether such a beneficial reversal of domi- and 2, with the relative importance of the two traits switching between sea- nance can also maintain polymorphism in the presence of epis- sons. The dashed line indicates the average of the two homozygote traits. If tasis. In each case, a necessary condition for polymorphism is the heterozygotes are closer to the fitter homozygote with respect to both that a population fixed for the fittest possible fully homozygous traits 1 and 2, there is a beneficial reversal of dominance at the level of the genotype can be invaded by mutants. As we have seen above, seasonal score, z. See SI Appendix, Fig. S2 for alternative scenarios. with synergistic epistasis (v 00 > 0), there are two fully homozy- gous genotypes with maximum fitness, the one with the summer allele at all loci and the one with the winter allele at all loci. In Deterministic Analysis both cases, the resident type has score L in one season and score In this section, we assume that population size is so large that 0 in the other season, whereas mutants differing in one posi- does not play a role. We also assume that muta- tion have scores L − 1 + d and d. For mutants to invade, we tions are rare enough that the allele-frequency dynamics will thus need equilibrate before a new mutation arises at one of the L loci. This v(d) + v(L − 1 + d) > v(L) + v(0). [7] simple deterministic framework allows us to develop an intuitive understanding of the conditions for stable polymorphisms for Using our example class of fitness functions with positive epista- various genotype-to-fitness maps. The intuitions developed here sis, Eq. 5 with q > 1, we thus obtain the condition will then be checked and extended with stochastic simulations in d q + (L − 1 + d)q > Lq . [8] the next section. We will first confirm that the conditions under which season- The critical value of d, dcrit , at which extreme types become inva- ally fluctuating selection can maintain polymorphism are restric- sible, satisfies tive when contributions to the seasonal score z are additive  q  q dcrit dcrit − 1 within loci (d = ds = dw = 0.5 in Eqs. 1 and 2). Then, zs +zw = L + 1 + = 1. [9] ∗ L L for all possible genotypes, and the mean z over time is z = L/2. The long-term success of a genotype depends on its geometric For q = 2, dcrit takes values 0.707, 0.954, and 0.995, with 1, 10, mean fitness, or, equivalently, on the arithmetic mean of the and 100 loci, respectively. For q > 1 in general, dcrit approaches logarithm of fitness, v(z). Jensen’s inequality or simple geomet- 1 as the number of loci increases. To see this, note first that the ric considerations (Fig. 4) tells us that the arithmetic mean of condition in Eq. 9 is always fulfilled for d = 1 and thus dcrit ≤ 1. v(zs ) and v(zw ) for a given genotype will be smaller than or Thus, as the number of loci, L, goes to infinity, (dcrit − 1)/L equal to v(z ∗) if v 00 < 0 everywhere (Fig. 4A), equal to v(z ∗) becomes small, and we can approximate the second term on if v 00 = 0 (Fig. 4B), and larger than or equal to v(z ∗) if v 00 > 0 the left-hand side of Eq. 9 by a Taylor expansion around 1 (Fig. 4C). to obtain The interannual allele-frequency dynamics (e.g., from summer to summer or from winter to winter) with multiplicative fitness (v 00 = 0) and d = 0.5 are thus neutral. No balancing selection 00 ) z

emerges. With positive epistasis (v > 0), extreme types with ( A B C either only summer or only winter alleles have the highest geo- v metric mean fitness. Therefore, the population ends up in a state where all loci are fixed for the summer allele or all for the win- ter allele. With negative epistasis (v 00 < 0), the genotypes with Summer the highest geometric mean fitness are those with the same num- Winter ber of summer and winter alleles and thus zs = zw . There are Average always some genotypes heterozygous at one or more loci that fulfill this condition (heterozygous intermediates; Fig. 2). For an 0 L/2 L 0 L/2 L 0 L/2 L Logarithm of fitness, Logarithm of fitness, even number of loci, zs = zw is also true for genotypes homozy- Summer score, zs = L − zw gous for the summer allele at half of the loci and homozygous for the winter allele at the other half (homozygous intermediates). Fig. 4. The logarithm of summer fitness (red) and winter fitness (blue) and When one of the homozygous intermediates fixes in the popu- the average logarithm of fitness (gray) as a function of a genotype’s summer lation, it cannot be invaded by any mutant starting at small fre- score, zs. Assuming d = 0.5, the winter score is zw = L − zs, leading to the mirror symmetry around L/2. If the fitness function is concave on the quency (under the assumptions of the deterministic model; see SI logarithmic scale, intermediate types have the highest average log-fitness Appendix, section S1 for a detailed proof), and all polymorphism (A); if the fitness function is log-linear, then all types have the same average is eliminated. For an odd number of loci, homozygous interme- log-fitness (B); and if the fitness function is convex on a logarithmic scale, diates do not exist, and some polymorphism may be maintained, extreme types have the highest average log-fitness and thus the highest at least at one locus. This case, which we will examine in more geometric mean fitness (C).

4 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1702994114 Wittmann et al. Downloaded by guest on October 3, 2021 Downloaded by guest on October 3, 2021 etta ti oecnuiet utlcsplmrhs than parame- polymorphism dominance ter, multilocus critical to the sug- Specifically, conducive arguments epistasis. more synergistic theoretical is above it the that gest because (e.g., and plausible and 21–23) common refs. more be to appears epistasis Eq. returns type of diminishing- on functions focus fitness We returns fitnesses. their as to sampled proportion are in individuals parents two independently, generation a in ual simulations (see forward individual-based type Wright–Fisher use We now Simulations we Stochastic questions, these address simulations. To stochastic problem. to whether turn a and segrega- is mutations efficient load recurrent how pop- and genetic finite unclear drift in genetic still polymorphism with is multilocus ulations maintaining it at Also, is one them. lift just tion or of loci, us few all tell a seg- at not of or maintained does aspect be analysis will negative segregat- polymorphism preceding the alleles whether to the two opposed However, of as load. suf- aspect locus, regation lift” same is positive the “segregation a allele at on mechanism ing favored based this is respective it call the because We seasons dominant. both ficiently in if polymor- phism maintain can selection fluctuating seasonally functions, in it. change overcome smaller to a sufficient and is weaker, dominance becomes type intermediate the words, other increasing In with epistasis. without those approach polymorphism with type intermediate increasing with scores have position one in + score differing L/2 has mutants type whereas resident The seasons can mutants. loci type by the condition intermediate invaded of homozygous necessary be this half a that other Thus, is the loci). polymorphism at at of for allele number allele winter even summer the an and (assuming the loci carries the of genotype half homozygous fully sible large requiring dominant, dominance. completely in almost changes seasonal be allele to favored respective have the but would loci, main- many at principle, polymorphism in tain can, Eq. selection in fluctuating type seasonally the of epistasis, functions fitness for Thus, large. that conclude by sides both Multiplying itane al. et Wittmann fitness of logarithm the of tive as Eq. From satisfies invasible become Eq. coefficient form for dominance the fulfilled of always is functions condition this Again, is polymorphism for vn o u oe,adapstv solution positive rel- a not and is model, which our solution, for negative evant a has equation quadratic This u eut ofrsgetta o ra ls ffitness of class broad a for that suggest far so results Our (v epistasis diminishing-returns With L  d eto S2 section Appendix, SI + 1 ost nnt.Teitiinhr sta h eodderiva- second the that is here intuition The infinity. to goes crit d d and o iiihn-eun psai nEq. in epistasis diminishing-returns for ,  crit L 2 13, w d + crit L + (L/2 L/2 = d d d crit crit 2 1  crit z eeto gis eprlvrainaround variation temporal against selection L, − q hrfr,frlarge for Therefore, .

  + 1 + 1 + erae as decreases −1 sapoiaey1i h ubro oiis loci of number the if 1 approximately is d ) + 1 d d − · z crit hs h eutn eesr condition necessary resulting the Thus, . w L = q (L/2 L L 2 2+ (2 + · o eal) hti,freeyindivid- every for is, That details). for 4 twihhmzgu intermediates homozygous which at L/2 n letting and − d ihayexponent any with crit v + 1 − L 00 L swa,adtecniin for conditions the and weak, is (z − + 1 nrae n prahs0.5 approaches and increases L) d = ) 4 1 crit s ohbcuediminishing- because both d + ) + 1  00 L psai rudthe around epistasis L, −y O > = < ot nnt,w can we infinity, to go  1+ (1 w 2+ (2 d  L ,tefits pos- fittest the 0), (L/2) 1 + 1 2 o fitness For 1. = 1  z L) 13 5 y ) 1 = L −2 2 h critical the , ihpositive with 2 2 L/2 . sgenerally is  ! . decreases 2 . . nboth in [10] [13] [11] [12] aaee Explanation explored ranges the and parameters Parameter model of Overview 1. Table rwne lee o large For allele. winter or parameters, d dominance of different three case for trajectories the frequency For loci. of 6). (Fig. only is rate and mutation loci by of influenced number weakly the with decreases which value, parameter, critical Methods) dominance and the if (Materials emerges selection balancing inde- effective is expected, behavior exponent, This the 5). zero of (Fig. pendent approaches below finally from and neutrality) value, (effective absolute numbers, in odd overall balanc- for of even decreases a negative strength becomes effective at eventually the loci selection S4). increases, ing Fig. loci two of Appendix, or number (SI one the frequency As only intermediate but at strength positive, fluctuate effective time is the become loci, selection of to balancing numbers tend of odd alleles small rare For rarer. that even indicating sim- both the negative, from where is the estimated ulations situations 2), selection balancing (Fig. i.e., of exist strength loci, effective intermediates of heterozygous numbers and even homozygous For 5). within allele (Fig. contributions rare from (d additive a expected arguments, fast loci As theoretical cycle. how yearly above and full the a whether over us frequency in tells increases which S2), Methods section and (Materials selection” balancing of individuals still between that value number population. possible offspring the smallest in in the differences three, for of simu- allows cap supplementary a role run the with also understand lations to better we To set capping, again. is gametes. offspring-number drawn fitness individu- be 10 of its cannot of most it number, parent that at that so a reached contribute 0 has as i.e., individual times generation, an 10 next Once indi- most the each at in model, this drawn als In be load. can genetic of vidual relevance the assess to computa- by muta- approximated larger rates. with well tion populations smaller be manageable may more values tionally rates on the mutation only than depend small smaller processes product genetic rates population the mutation many and here, used larger often rate are [w mutation effects explored. an the mutation dominance than ranges to gives and larger the relative selection 1 strong and scenarios, Table are parameters simulated size. most model simulations population In the supplementary in of run changes overview also seasonal but with constant, size size population lation the and ation probability mutation symmetric model. multiplicative the for simulations Eq. some in run epistasis we synergistic for one the than smaller L N g d µ y o small For . rmnwo,w ilfcso cnro ihlrenumbers large with scenarios on focus will we on, now From strength “effective an estimate we output, simulation the From model “capped” a design we model, basic the to addition In the are simulations stochastic the in parameters Additional 0 = r o odcv omliou polymorphism multilocus to conducive not are .5) ubro loci of Number 1–20 size Population 2g has year a season; per generations of Number e generation per uainpoaiiyprall oy10 parameter Dominance copy allele per probability Mutation xoeto h tesfnto Eq. function fitness the of Exponent d N ahlcsi lotfie ihrfrtesummer the for either fixed almost is locus each , µ eg,rf 5.Tu,lreppltoswith populations large Thus, 45). ref. (e.g., generations 100 y ftefins ucinEq. function fitness the of , d l oiflcut tintermediate at fluctuate loci all , oi i.7soseapeallele- example shows 7 Fig. loci, .Atog aua populations natural Although µ]. N µ egnrlyke popu- keep generally We . 0 e leecp e gener- per copy allele per (z d ) slre hnacertain a than larger is , NSEryEdition Early PNAS and d 4 and − 0.5–4 ag explored Range 1–500 100–10,000 0–1 naddition, In 9. 0.5 IAppendix, SI −6 lo as Also, 4. r much are − | 10 f10 of 5 −4

EVOLUTION PNAS PLUS the fittest individual in the population was on average 1.2 times fitter than the least fit individual and 4.9 times fitter for y = 4 (see SI Appendix, section S4 for a supporting heuristic analysis). When the offspring-number cap is set to three, substantial quan- titative differences between uncapped and capped simulations are seen (SI Appendix, Fig. S9). However, 0.5 remains the crit- ical dominance parameter. This result also holds for the multi- plicative model (SI Appendix, Fig. S10), but with otherwise larger differences between capped and uncapped model versions, even with an offspring-number cap of 10. Also, the retardation fac- tor for the multiplicative model is often below one, even when the effective strength of balancing selection is positive. The rea- son appears to be that there is larger variance in fitness under the multiplicative model (SI Appendix, Fig. S11) and that fluctuations are sometimes so large that alleles go to fixation (SI Appendix, Fig. S12). Both of these effects are weakened by offspring num- ber capping, so that the capped multiplicative model behaves more similarly to the diminishing-returns model (SI Appendix,

Fig. 5. Effective strength of balancing selection (be in Eq. 14 in Materi- Fig. S10). als and Methods) in the additive case (d = 0.5) as a function of the num- Additional simulations for the diminishing-returns model sug- ber of loci. Solid lines indicate means and dashed lines indicate means ± gest that the finding of stable multilocus polymorphism for two standard errors. Simulations are always run for successive odd and even d > dcrit ≈ 0.5 still holds under various forms of asymmetry, numbers. N = 1,000, g = 15, µ = 0.0001. e.g., when one season has more generations than the other (SI Appendix, Fig. S13); when the exponent of the fitness function, y, differs between summer and winter (SI Appendix, Fig. S14); and frequency. The critical dominance parameter, dcrit , with 100 loci when there are seasonal changes in population size (SI Appendix, is close to 0.5, independently of the exponent of the fitness func- Fig. S15). To explore the effects of asymmetry in the dominance tion, y (Fig. 8A). For d < 0.5, i.e., if the currently favored allele is parameters, we fixed the winter dominance parameter, dw , at 0.4 recessive, the effective strength of balancing selection is negative and varied the summer dominance parameter, ds . Stable poly- and polymorphism is unstable (Fig. 8A). As d increases beyond morphism then arises for ds > 0.6 (SI Appendix, Fig. S16), sug- 0.5, i.e., as the currently favored allele becomes more dominant, gesting that stable polymorphism requires an arithmetic mean effective balancing selection becomes stronger (Fig. 8A). Both dominance parameter >0.5. Compared with the diminishing- the stabilizing and destabilizing effects increase with increasing returns model, the multiplicative model seems less robust to exponent y (Fig. 8A). asymmetry (SI Appendix, Fig. S17). A tendency for rare alleles to increase in frequency does not With an increasing number of loci under the diminishing- guarantee that the average lifetime of a polymorphism is larger returns model and with d > 0.5, the strength of balancing selec- than under neutrality (42, 46). This is particularly interesting for tion, the retardation factor, and the magnitude and predictabil- fluctuating selection regimes with positive autocorrelation where ity of fluctuations all decrease (Fig. 9). Population size hardly alleles regularly go through periods of low frequency (42). We influences effective strength of balancing selection and effective therefore compute the so-called retardation factor (46), the aver- age lifetime of a polymorphism in the selection scenario relative to the average lifetime under neutrality (see SI Appendix, section S2 for detailed methodology). The results for 100 loci are consis- tent with those for the effective strength of balancing selection: For d > 0.5, polymorphism under segregation lift is lost more slowly than under neutrality (Fig. 8B). To quantify seasonal fluctuations, we compute an effective selection coefficient (Materials and Methods and SI Appendix, section S2). We also compute the predictability of fluctuations as the proportion of seasons over which the changes in the expected direction, e.g., where the summer- favored allele increases over a summer season. Both the effective selection coefficient and the predictability of fluctuations have a maximum at intermediate values of d and increase with increas- ing exponent y of the fitness function Eq. 4 (Fig. 8 C and D). For even higher values of d, fluctuations are not as strong, presum- ably because heterozygotes are fitter, and therefore more copies of the currently disfavored allele enter the next generation. Also, effective strength of balancing selection, effective selection coef- ficient, and predictability of fluctuations increase with the num- ber of generations per season (SI Appendix, Fig. S6). With an offspring-number cap of 10, the results for the capped model generally match the results for the uncapped model in all Fig. 6. Critical value of the dominance parameter, dcrit , such that the effec- d > 0.5 tive strength of balancing selection (be in Eq. 14 in Materials and Methods), respects, especially for (Fig. 8). It appears that the cap- is positive (stable polymorphism) if d > d , and negative (unstable poly- ping mechanism only rarely takes effect because of relatively nar- crit morphism) if d < dcrit . Symbols represent means across replicates, and lines row distributions of the seasonal score, z, within a generation (SI represent averages ± two standard errors. Since the pattern for odd num- Appendix, Fig. S7), and consequent relatively low variance in fit- bers of loci is more complex (Fig. 5), only even values for the number of loci ness (SI Appendix, Fig. S8). For example, for d = 0.7 and y = 0.5, are included here. N = 1,000, y = 2, g = 15.

6 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1702994114 Wittmann et al. Downloaded by guest on October 3, 2021 Downloaded by guest on October 3, 2021 hssaemr aacdi hi umradwne fetsizes effect (Fig. winter and stable summer (two-sample their only polymor- in balanced stable are more are polymorphisms, phisms that unstable with polymorphisms 10D). Compared than 10E). (Fig. win- sizes and size effect summer population larger ter have with to increases tend total polymorphisms and Detectable intermediate loci 10 an at (Fig. of highest seasons number is the polymorphisms of 5% detectable least half of at least by direction at fluctua- expected the clas- allele-frequency in in detectable polymorphisms changes as exhibit the defined also tions, of stable proportion size. as population small sified on a weakly only only of depends number However, The stable 10B). as (Fig. classified size loci plausible biologically of allele; ulations rare poly- a stable of of change see hundreds frequency loci, expected of (positive With number morphisms 10A). total (Fig. high sufficiently allele trajectories their a various with on population, same fluctuating the frequencies in maintained be can eters use We model. effects. basic winter and summer y between trade-off a sizes exhibit effect com- all Because are analogously. contributions puted Winter homozygotes. summer–summer homozygotes, winter contribution the where on h ubro oihmzgu o h urnl favored we currently First, the for steps. homozygous two loci in of fitness number to the count genotype that multilocus selection fluctuating the seasonally for maps model simple a study We Discussion sea- across them, parameter dominance of is sons average all the almost not, for or detectable but parameters, dominance asymmetric also see as computed then are scores Seasonal [0,1]. on tion oec fte.Sme n itrdmnneparameters, dominance winter and function Summer exponential corre- them. the and d 1, apply of SD then each 0, and to mean 0.9, with coefficient from distribution parameters lation of normal pair a bivariate draw a we this, For distribution. normal Inde- seasons. size effect between locus asymmetric each be for may pendently, and loci across vary under than faster slightly lost 9B). (Fig. be neutrality even can popula- small polymorphism In pre- tions, 9D). more average (Fig. have fluctuations and allele-frequency on 9B) dictable (Fig. longer based for polymorphism are maintain 9 which (Fig. changes measures allele-frequency coefficient, selection run simulation the of middle the in generations) 301–305). (150 (years years 5 for shown are (C L 7. Fig. itane al. et Wittmann s =

.Ol 0rnol eetdlc soni ifrn oos u f10loci 100 of out colors) different in (shown loci selected randomly 10 Only ). Frequency of winter allele 4 = ,l h eut niaeta oyopim ihdfeetparam- different with polymorphisms that indicate results The ial,w osdragnrlzdmdlweeparameters where model generalized a consider we Finally,

100, 0.0 0.4 0.8 eto S2 section Appendix, SI and A Fall eebcuei e otems tbeplmrhs nthe in polymorphism stable most the to led it because here . Fg 10F (Fig. >0.5 g Spring .Mn tbeplmrhsshave polymorphisms stable Many S18). Fig. Appendix, SI he xmlso leefeunytaetre for trajectories allele-frequency of examples Three d = Fall w ∆ 15, Spring ,l r rw neednl rmauiomdistribu- uniform a from independently drawn are , s Fall t ,l y ts on -test Spring = n itrefc size effect winter and Fall

4, Spring

µ Fall

= Spring ). | d

10 Fall ln(∆ s c ,l B −4 l

∆ Fall l flocus of and , o eal)cnb anandi pop- in maintained be can details) for eda orprmtr:Summer parameters: four draw we , s

s Spring ,l ,l Fall /∆

o eeoyoe,and heterozygotes, for Spring d A Fall = w and

,l Spring 5(A), 0.15 l

)|, Fall nsme s0frwinter– for 0 is summer in ∆ Spring p ,btlrepopulations large but C), w Fall C ,l < Spring ∆ r rw rmalog- a from drawn are and

d Fall 2. = r oiie l loci all positive, are C

,and (B), 0.5 Fall · .Tenumber The D).

10 Spring Fall −16

z Spring

= Fall i.10E; Fig. ; Spring N P ∆

d Fall = s l L =

,l Spring =1 1,000, Fall 0.65 for

c Spring l

, Fall osgntpscnbte dutt ohsme n winter polymorphism, and maintain summer both to heterozy- adjust more better where environments. can inter- plasticity, be phenotypic genotypes also of gous may type lift a segregation as polymorphism cases, preted multilocus some types and In population, homozygous maintained. the is fully in Unlike fix models, loci. cannot additive homozygous thus studied more previously but of number alleles, in same winter the with and individuals higher summer than have sufficiently seasons then both winter—is loci in in heterozygous scores allele many with winter summer Individuals the allele—the large. and favored summer currently in dom- average the allele the of that parameter hundreds requires lift or inance Segregation tens loci. at unlinked polymorphism of fluctu- maintain seasonally can which selection a by identify ating lift, We model. segregation fully our mechanism, a of general on cases special selection previously are and phenotype The selection additive epistasis. multiplicative and of cases selection then studied which of by is function strength increasing weighted score for monotonically seasonal loci accounts a via resulting heterozygous fitness to The of mapped parameter. number dominance the a add and allele vrg tlatsihl oiatwt epc oteseasonal the on to is respect allele with favored dominant currently slightly the Mechanism least that at a such average as time over Lift changes Segregation Variation. of Maintain to Plausibility and Robustness increases. loci of number the as 0.5 diminishing- approaches With common, loci. most appears multiple that d epistasis are of there type the for when epistasis, returns i.e., one to epistasis, close Without is loci. of number selection, multiplicative the on and epistasis i.8. Fig. n rqec eedneo esnlall rqec hne.Teverti- at The are changes. lines frequency gray allele cal seasonal of dependence frequency and replicates. all See from jointly in obtained and Lines overlapping), errors. (often in standard variant lines model dashed capped vs. vs. solid uncapped repli- the across averages for indicate cates Symbols (D). (C fluctuations of fluctuations predictability and of magnitude (B), tor ( selection balancing CD AB crit h rtclvleo h oiac aaee eurdto required parameter dominance the of value critical The i.S5 Fig. Appendix , SI ssbtnilylre hn05wt e oi u quickly but loci, few with 0.5 than larger substantially is , nuneo h oiac parameter dominance the of Influence d A; = b .Bl e. aacn eeto;ep,expected. exp., selection; balancing sel., Bal. 0.5. e o oedtie nomto ntedistribution the on information detailed more for A, Eq. , B ergto itrqie htdominance that requires lift Segregation and C, ipycnetmxmmlklho estimates maximum-likelihood connect simply d crit d 14, crit aeil n Methods and Materials s05 ihsnritceitss it epistasis, synergistic With 0.5. is D eed otyo h yeof type the on mostly depends , niaetersetv means respective the indicate ; N s e = Eq. , 1,000, 15, NSEryEdition Early PNAS d L aeil n Methods), and Materials = nefciesrnt of strength effective on 100, ,rtrainfac- retardation ), g = 15, µ | = ± f10 of 7 10 two −4 .

EVOLUTION PNAS PLUS AB between epistasis and strength of selection, future work should also consider more general fitness functions allowing for various combinations of epistasis and selection strength. Whenever there is balancing selection at a large number of loci, genetic load is a potential concern. In the case of segre- gation lift with diminishing-returns epistasis, however, genetic load does not appear to play an important role. The results for our capped model closely match the results for the origi- nal, uncapped model. Apparently, independent segregation at a large number of unlinked loci leads to relatively small vari- ance in seasonal scores within the population and, together with CD the diminishing-returns fitness function, to relatively small vari- ance in fitness. With a much smaller offspring number cap or with multiplicative selection, differences between the capped and uncapped model are more substantial, but even then, balancing

AB

Fig. 9. Influence of population size and the number of seasonally selected loci on effective strength of balancing selection (A; be in Eq. 14, Materials and Methods), retardation factor (B), magnitude of fluctuations (C; se in Eq. 15, Materials and Methods), and predictability of fluctuations (D). Symbols indicate averages across replicates and lines in A, C, and D indicate means ± two standard errors (in C and D, standard errors are too small to be visible). Lines in B simply connect maximum-likelihood estimates obtained jointly from all replicates. Note that in B, some points are missing because the rate of loss of polymorphism was too small to be quantified. d = 0.7, y = 4, g = CD 15, µ = 10−4. Bal. sel., balancing selection; exp., expected.

score. As discussed above, there are several potential mecha- nisms that can plausibly produce such changes in dominance. Moreover, the required changes are small. Unfortunately, there have been only few relevant empirical studies so far. For instance, in the copepod Eurytemora affinis, there appears to be beneficial reversal of dominance for fitness across salinity con- ditions (47). In experimental Drosophila populations, changes in dominance for gene expression across environments appear to be common (48). More empirical and theoretical work is required to find out how common changes in dominance are, in particu- EF lar on the relevant scale of the seasonal score. However, even if the required changes in dominance are rare on a per-site basis and the vast majority of polymorphisms are lost under fluctu- ating selection, there may still be many sites in the genome with appropriate reversal of dominance, and, as we show, those are then the ones that we should see as seasonally fluctuating polymorphisms. In the focal diminishing-returns scenario, the conditions for stable polymorphism via segregation lift are surprisingly robust to changes in the mutation rate (Fig. 6) and to asymmetries in number of generations, strength of selection, or population size between summer and winter (SI Appendix, Figs. S13–S15), appar- Fig. 10. Stability of polymorphism and detectability of allele-frequency ently more so than under the multiplicative model (SI Appendix, fluctuations when parameters vary across loci and seasons. (A) Snapshot Fig. S17). When the dominance parameter differs between sum- of allele-frequency trajectories for stable polymorphisms in one simulation mer and winter, polymorphism is generally stable at those loci run. (B) Average number of stable polymorphisms as a function of the total whose average dominance parameter across seasons is >0.5 (Fig. number of loci for different population sizes. (C and D) As in A and B, 10F and SI Appendix, Fig. S16). Segregation lift is also robust to but only for polymorphisms that are also detectable. (E) Winter effect size, variation in effect sizes and dominance parameters across loci ∆l,w , vs. summer effect size, ∆l,s, for stable and detectable and only sta- (Fig. 10). In reality, the strength of seasonality likely varies in ble polymorphisms. The plot shows pooled results over ten simulation runs with independently drawn parameters. Oval isoclines indicate the shape of space and time, which could make the maintenance of polymor- the original sampling distribution, with 75% of the sampling probability phism by segregation lift even more robust (SI Appendix, Fig. mass inside the outermost isocline. (F) Corresponding dominance parame- S2B and ref. 40). Future work needs to explore whether segrega- ters (see also SI Appendix, Fig. S19). The dominance parameters were orig- tion lift is robust also to linkage between selected loci. Since our inally drawn from a uniform distribution on the unit square. Parameters: diminishing-returns fitness function has a particular relationship y = 4, g = 10, µ = 10−4 and in A, C, E, and FN = 10,000, L = 100.

8 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1702994114 Wittmann et al. Downloaded by guest on October 3, 2021 Downloaded by guest on October 3, 2021 ihrwne-o umrfvrdallsdmnt.However, the dominate. of alleles case where the summer-favored subpopulations in or other winter- from 49), immigration either (31, recurrent fluctuations by seasonal with the or mutation, for recurrent only by responsible induced selection simply but maintained, bly need alternatives. will possible work and Future model lift. this test segregation empirically by et to there- explained Bergland we by are knowledge, observations (36) empirical current al. the our that on claim Based 10 cannot 7C). fore (Fig. Fig. subtle lift see more segregation but often D, our are by fluctuations explained where model, be can allele- of fluctuations magnitude observed frequency it the Finally, whether unknown. clear still completely are not is loci dominance selected of seasonally chromo- dynamics the and at few distribution effects the only Second, some sites. and between the linkage of SNPs substantial be seasonal necessarily will of there somes, reasons. hundreds explain several for potentially with warranted could First, is caution lift stable. but segregation long-term observations, results, be these to our appear on polymorphisms Based the of many and by temperate fluctuated a in of sites of population hundreds seasonal strong at detected fluctuations (36) Tests allele-frequency al. et Potential Bergland by study and Lift. sequencing Hypotheses, Segregation Alternative for Evidence, Empirical sites. individual col- at patterns their on on than based rather perhaps behavior loci, lective many allele- seasonal at subtle loci fluctuations such detecting selected frequency of ways individual new will detect explore research to Future to need fluctuations. easy allele-frequency their necessarily on of not based number is large it a at loci, contributes polymorphism lift segregation maintaining if to even substantially selection because Thus, locus. average is each weaker at This pressures to scores, neutrality. leads seasonal effective epistasis overall pre- to higher diminishing-returns i.e., is goes to zero, locus lead loci loci each to of more at go number selection the to of As dicted strength size. effective infinite fluctuations the of of infinity, rela- magnitude population the and a explore loci in to of argument number mathematical between tionship heuristic an a the at use Thus, In we maximized 10). selection. (Fig. be loci under of however, may number allele loci intermediate polymorphisms fluctuations, detectable of in of these number number fluctuations the of with seasonal magnitude decreases predictable pro- The also and frequencies. can lift strong segregation Fluctuations. duce polymorphism, Allele-Frequency stable to of addition Detectability and Magnitude polymor- multilocus lift. stable segregation for via required phism not are numbers spring for emerges selection itane al. infer et Wittmann to possible not is function, it fitness not unknown, the itself generally of is also shape and is the fitness, Since and score, measurable. genotype seasonal directly the multilocus to parameter, between relative dominance ates but pivotal fitness to the tive that be will diploids. polymorphism in maintain can loci they coexis- multiple pro- whether at species unclear other more for is some it are mechanisms and as mechanisms by tence, ecology these or in However, act, studied 54). commonly life-history not 53, (51, long-lasting does state a selection tected by which buffered where on effect” be stage storage can “temporal variation so-called a genetic (ii) densities and population 50–52) and (17, both concentrations responses resource to differential fluctuating (i) lead to are could stability long-term that and mechanisms fluctuations Alternative (36). unlikely n lentv sta eei aito sntral sta- really not is variation genetic that is alternative One nftr miia et o ergto it anchallenge main a lift, segregation for tests empirical future In 0 vrasnl esnof season single a over ∼10% tmn ie,all frequencies allele sites, many At melanogaster. D. Drosophila smnindaoe eetpooled- recent a above, mentioned As d > hs nelsial ag off- large unrealistically Thus, 0.5 bevtos hswsconsidered was this observations, eto S3, section Appendix, SI 0generations, ∼10 z z d hc under which , hc medi- which , sntrela- not is , d rmfit- from C and w In , h ert-erall-rqec yais o hs efi tnadbal- standard form a the fit of (55) we model this, selection selection, polymor- For ancing balancing of dynamics. of stability allele-frequency strength assess year-to-year effective we the an model, estimating capped the by and phism model basic the For of puzzle Methods and the Materials solving on progress seg- make for variation. modern thus genetic test use and to can lift technologies we regation sequencing how and is biology work An taxa. molecular future many for of populations question natural contribu- important in substantial variation a genetic environmental to make tion of could ubiquity lift unre- segregation the have fluctuations, Given to offspring. individuals and many heterozygous selection alistically highly require stabilizing not with under does associated polymorphism problems of reasonable the biologically maintenance circumvents of lift populations Segregation in size. loci hun- unlinked at of polymorphism dreds maintain sea- can which selection by mechanism fluctuating general sonally a as lift segregation identify We Conclusions models. possible multiple acces- between more distinguish empirically to patterns these patterns sible use regions, and and neutral spectra, disequilibrium, linked site-frequency linkage levels, of in diversity selection at predictions of look make e.g., footprint to genetic is approach the alternative for An arising study. direction research this at important from get an empirically thus is to parameters score way nance seasonal feasible the Com- and of on. scale meaningful focus the to a loci logistic with which the priori up a ing from know not apart do outdoor However, we challenges, of seasons. track number multiple and large over genotype, multilocus fitness a different stock a with to each pro- mesocosms, be that One could itself. loci in direction challenging prac- contributing ductive is In fitness of interactions. multiplicative measuring epistatic however, set and several tice, lift a of segregation with exhibit existence could might each the approaches components, account statistical fitness Such into lift. take segregation or also whether is effect assess there function, thereby fitness and not learn- parameters the machine dominance of as and parameters sizes, such estimate methods jointly statistical to different ing use at could genotypes we multilocus mea- times, different fitness with many situation, for ideal surements an com- In a background. in genetic genotypes mon single-locus different of measurements ness org/10.6084/m9.figshare.5142262. the for loss of of loss times of rate the loci the From the (see for reached. polymorphism of estimators is one maximum-likelihood years obtain at 500 we lost of replicates, is time polymorphism maximum until a allele replicates or mean 100 its the run changes report we we locus factors, cases, a three all fluctu- which replicates In of in over direction. predictability expected seasons the the of for in proportion frequency same the the do i.e., We ations, replicate. each for dently strength selection, effective balancing calculate and of combination parameter every for replicates coefficient season, selection one over effective changes an with (55) we model this, selection For directional negative seasons. we standard individual Second, a whereas over rare. fit run, fluctuations more of even long become magnitude the to the tend quantify in alleles rare common that more indicate values become cycle, to yearly tend one alleles over S2 frequency section allele Appendix, in changes average to ooti esr o ttsia netit norrsls ern10 run we results, our in uncertainty statistical for measure a obtain To + iuaincd n uprigRsrpsaeaalbeat available are scripts R supporting and code simulation C++ ± w tnaderr ftema.T banretardation obtain To mean. the of errors standard two eto S2 section Appendix, SI o eal) oiievle of values Positive details). for b ∆ e ∆ n fetv eeto coefficient, selection effective and , y x s x ∆ = eto S2). section Appendix, (SI s x b = e x (1 z s e x n siaeterlvn domi- relevant the estimate and − (1 x − )(1 s o details). for e x oaeaeallele-frequency average to , − ), 2x NSEryEdition Early PNAS ) b e niaeta rare that indicate s ∆ e https://doi. indepen- , y | b x e (see f10 of 9 from , [14] [15] SI

EVOLUTION PNAS PLUS ACKNOWLEDGMENTS. For helpful discussion and/or comments on the were performed on Stanford’s FarmShare Cluster and on the Vienna Scien- manuscript, we thank Michael Desai, Joachim Hermisson, Oren Kolodny, tific Cluster. M.J.W. was supported by fellowships from the Stanford Center Mike McLaren, Richard Nichols, Pleuni Pennings, Jitka Polechova,´ and mem- for Computational Evolutionary and Human Genomics and from the Aus- bers of D.A.P.’s laboratory, as well as two anonymous reviewers. Simulations trian Science Fund (M 1839-B29).

1. Lewontin RC, Hubby JL (1966) A molecular approach to the study of genic heterozy- 29. Charlesworth B (2013) Why we are not dead one hundred times over. Evolution gosity in natural populations. II. Amount of variation and degree of heterozygosity 67:3354–3361. in natural populations of Drosophila pseudoobscura. Genetics 54:595–609. 30. Korol AB, Kirzhner VM, Ronin YI, Nevo E (1996) Cyclical environmental changes as a 2. Lewontin RC (1974) The Genetic Basis of Evolutionary Change (Columbia Univ Press, factor maintaining genetic polymorphism. 2. Diploid selection for an additive trait. New York). Evolution 50:1432–1441. 3. Beatty J (1987) Weighing the risks: Stalemate in the classical/balance controversy. 31. Burger¨ R, Gimelfarb A (2002) Fluctuating environments and the role of mutation in J Hist Biol 20:289–319. maintaining quantitative genetic variation. Genet Res 80:31–46. 4. Hedrick PW (2007) Balancing selection. Curr Biol 17:R230–R231. 32. Lande R (2008) Adaptive topography of fluctuating selection in a Mendelian popula- 5. Barton N, Keightley P (2002) Understanding quantitative genetic variation. Nat Rev tion. J Evol Biol 21:1096–1105. Genet 3:11–21. 33. Wright S (1935) Evolution in populations in approximate equilibrium. J Genet 30: 6. Turelli M, Barton NH (2004) Polygenic variation maintained by balancing selection: 257–266. Pleiotropy, sex-dependent allelic effects and G x E interactions. Genetics 166:1053– 34. Burger¨ R, Gimelfarb A (1999) Genetic variation maintained in multilocus models of 1079. additive quantitative traits under stabilizing selection. Genetics 152:807–820. 7. Asthana S, Schmidt S, Sunyaev S (2005) A limited role for balancing selection. Trends 35. Nagylaki T (1989) The maintenance of genetic variability in 2-locus models of stabi- Genet 21:30–32. lizing selection. Genetics 122:235–248. 8. Hedrick PW (2012) What is the evidence for heterozygote advantage selection? 36. Bergland AO, Behrman EL, O’Brien KR, Schmidt PS, Petrov DA (2014) Genomic Trends Ecol Evol 27:698–704. evidence of rapid and stable adaptive oscillations over seasonal time scales in 9. Croze M, Zivkoviˇ c´ D, Stephan W, Hutter S (2016) Balancing selection on immunity Drosophila. PLoS Genet 10:e1004775. : Review of the current literature and new analysis in Drosophila melanogaster. 37. Gloss AD, Whiteman NK (2016) Balancing selection: Walking a tightrope. Curr Biol Zoology 119:322–329. 26:R73–R76. 10. Hedrick PW, Ginevan ME, Ewing EP (1976) Genetic polymorphism in heterogeneous 38. Charlesworth B (2015) Causes of natural variation in fitness: Evidence from studies of environments. Annu Rev Ecol Syst 7:1–32. Drosophila populations. Proc Natl Acad Sci USA 112:1662–1669. 11. Siepielski AM, DiBattista JD, Carlson SM (2009) It’s about time: The temporal dynamics 39. Desai MM, Weissman D, Feldman MW (2007) Evolution can favor antagonistic epista- of phenotypic selection in the wild. Ecol Lett 12:1261–1276. sis. Genetics 177:1001–1010. 12. Thurman TJ, Barrett RD (2016) The genetic consequences of selection in natural pop- 40. Rose MR (1982) Antagonistic pleiotropy, dominance, and genetic variation. ulations. Mol Ecol 25:1429–1448. 48:63–78. 13. van Schaik CP, Terborgh JW, Wright SJ (1993) The phenology of tropical forests– 41. Curtsinger JW, Service PM, Prout T (1994) Antagonistic pleiotropy, reversal of domi- Adaptive significance and consequences for primary consumers. Annu Rev Ecol Syst nance, and genetic polymorphism. Am Nat 144:210–228. 24:353–377. 42. Hedrick PW (1976) Genetic variation in a heterogeneous environment. II. Temporal 14. Schmidt PS, Conde DR (2006) Environmental heterogeneity and the maintenance of heterogeneity and . Genetics 84:145–157. genetic variation for reproductive diapause in Drosophila melanogaster. Evolution 43. Kacser H, Burns JA (1981) The molecular basis of dominance. Genetics 97:639–666. 60:1602–1611. 44. Keightley PD, Kacser H (1987) Dominance, pleiotropy and metabolic structure. Genet- 15. Behrman EL, Watson SS, O’Brien KR, Heschel MS, Schmidt PS (2015) Seasonal variation ics 117:319–329. in life history traits in two Drosophila species. J Evol Biol 28:1691–1704. 45. Gillespie JH (1998) Population Genetics—A Concise Guide (The John Hopkins Univ 16. Haldane JBS, Jayakar SD (1963) Polymorphism due to selection of varying direction. Press, Baltimore). J Genet 58:237–242. 46. Robertson A (1962) Selection for heterozygotes in small populations. Genetics 17. Dean AM, Lehman C, Yi X (2017) in the Moran. Genetics 47:1291–1300. 205:1271–1283. 47. Posavi M, Gelembiuk GW, Larget B, Lee CE (2014) Testing for beneficial rever- 18. Novak R, Barton NH (2017) When does frequency-independent selection maintain sal of dominance during salinity shifts in the invasive copepod Eurytemora affi- genetic variation? Genetics 207:653–668. nis, and implications for the maintenance of genetic variation. Evolution 68:3166– 19. Dempster ER (1955) Maintenance of genetic heterogeneity. Cold Spring Harb Symp 3183. Quant Biol 20:25–32. 48. Chen J, Nolte V, Schlotterer¨ C (2015) Temperature stress mediates decanalization 20. Gillespie JH (1973) Polymorphism in random environments. Theor Popul Biol 4:193– and dominance of gene expression in Drosophila melanogaster. PLoS Genet 11: 195. e1004883. 21. Chou HH, Chiu HC, Delaney NF, Segre` D, Marx CJ (2011) Diminishing returns epistasis 49. Kondrashov A, Yampolsky L (1996) High genetic variability under the balance among beneficial mutations decelerates adaptation. Science 332:1190–1192. between symmetric mutation and fluctuating stabilizing selection. Genet Res 68: 22. Khan AI, Dinh DM, Schneider D, Lenski RE, Cooper TF (2011) Negative epista- 157–164. sis between beneficial mutations in an evolving bacterial population. Science 332: 50. Armstrong RA, McGehee R (1976) Coexistence of species competing for shared 1193–1196. resources. Theor Popul Biol 9:317–328. 23. Kryazhimskiy S, Rice DP, Jerison ER, Desai MM (2014) Global epistasis makes adapta- 51. Chesson P (2000) Mechanisms of maintenance of species diversity. Annu Rev Ecol Syst tion predictable despite sequence-level stochasticity. Science 344:1519–1522. 31:343–366. 24. Feller W (1967) On fitness and the cost of . Genet Res 9:1–15. 52. Yi X, Dean AM (2013) Bounded population sizes, fluctuating selection and the tempo 25. Milkman RD (1967) Heterosis as a major cause of heterozygosity in nature. Genetics and mode of coexistence. Proc Natl Acad Sci USA 110:16945–16950. 55:493–495. 53. Svardal H, Rueffler C, Hermisson J (2015) A general condition for adaptive genetic 26. Sved JA, Reed TE, Bodmer WF (1967) The number of balanced polymorphisms that polymorphism in temporally and spatially heterogeneous environments. Theor Popul can be maintained in a natural population. Genetics 55:469–481. Biol 99:76–97. 27. Turner JRG, Williamson MH (1968) Population size, natural selection and the genetic 54. Gulisija D, Kim Y, Plotkin JB (2016) Phenotypic plasticity promotes balanced polymor- load. Nature 218:700. phism in periodic environments by a genomic storage effect. Genetics 202:1437–1448. 28. Agrawal AF, Whitlock MC (2012) Mutation load: The fitness of individuals in popula- 55. Durrett R (2008) Probability Models for DNA Sequence Evolution (Springer, tions where deleterious alleles are abundant. Annu Rev Ecol Evol Syst 43:115–135. New York).

10 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1702994114 Wittmann et al. Downloaded by guest on October 3, 2021