ebook img

Unifying the Theories of Inclusive Fitness and Reciprocal Altruism PDF

12 Pages·2016·0.41 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Unifying the Theories of Inclusive Fitness and Reciprocal Altruism

PPoorrttllaanndd SSttaattee UUnniivveerrssiittyy PPDDXXSScchhoollaarr Systems Science Faculty Publications and Systems Science Presentations 7-14-2006 UUnniiffyyiinngg tthhee TThheeoorriieess ooff IInncclluussiivvee FFiittnneessss aanndd RReecciipprrooccaall AAllttrruuiissmm Jeffrey Alan Fletcher Portland State University, [email protected] Martin Zwick Portland State University, [email protected] Follow this and additional works at: https://pdxscholar.library.pdx.edu/sysc_fac Part of the Social and Behavioral Sciences Commons Let us know how access to this document benefits you. CCiittaattiioonn DDeettaaiillss Fletcher, J.A., and Zwick, M., 2006. Unifying the Theories of Inclusive Fitness and Reciprocal Altruism. The American Naturalist 168: 252-262. This Article is brought to you for free and open access. It has been accepted for inclusion in Systems Science Faculty Publications and Presentations by an authorized administrator of PDXScholar. Please contact us if we can make this document more accessible: [email protected]. vol. 168, no. 2 the american naturalist august 2006 (cid:1) Unifying the Theories of Inclusive Fitness and Reciprocal Altruism Jeffrey A. Fletcher1,2,* and Martin Zwick2,† 1.DepartmentofZoology,UniversityofBritishColumbia,#2370- More than 40 years ago, Hamilton developed an expla- 6270UniversityBoulevard,Vancouver,BritishColumbiaV6T1Z4, nationfortheevolutionofaltruismamongrelativesbased Canada; ontheideaofinclusivefitness(Hamilton1963,1964,1970, 2.SystemsSciencePhDProgram,PortlandStateUniversity, 1972,1975).Hismostfamousresult,knownasHamilton’s Portland,Oregon97207 rule (HR), is usually interpreted as specifying the condi- SubmittedDecember19,2005;AcceptedMay31,2006; tions under which the indirect fitness of altruists (due to ElectronicallypublishedJuly14,2006 helping relatives have more offspring) sufficiently coun- terbalancestheimmediateself-sacrificeofaltruists.Inthis Onlineenhancement:appendix. way, the altruistic trait can increase overall. This mecha- nismisalsoknownaskinselection(MaynardSmith1964). Twenty-five years ago, Axelrod and Hamilton (1981) launched a still vigorous areaofresearch(forreviews,see Dugatkin 1997; Sachs et al. 2004; Doebeli and Hauert abstract: Inclusive fitness and reciprocal altruism are widely thought to be distinct explanationsforhowaltruismevolves.Here 2005) in which computer-based models of the iterated we show that they rely on the same underlying mechanism. We prisoner’s dilemma (IPD) areusedtostudytheevolution demonstratethiscommonalitybyapplyingHamilton’srule,normally of cooperation via reciprocal altruism (Trivers 1971). In associatedwithinclusivefitness,totwosimplemodelsofreciprocal their article, Axelrod and Hamilton suggestthatthereare altruism:one,aniteratedprisoner’sdilemmamodelwithconditional thesetwoalternativeexplanationsfortheevolutionofco- behavior;theother,amutualisticsymbiosismodelwheretwointer- operative traits: when the benefits of altruism fall to rel- acting species differ in conditional behaviors, fitness benefits, and atives, cooperation can evolve by inclusive fitness, and costs.WeemployQueller’sgeneralizationofHamilton’srulebecause when benefits fall to nonrelatives, cooperationcanevolve the traditional version of this rule does not apply when genotype and phenotype frequencies differ or when fitness effects are non- by reciprocal altruism.1 A third theory for the evolution additive, both of which are true in classic models of reciprocalal- ofaltruism,basedonmultilevel(orgroup)selection(Wil- truism. Queller’s equation is more general in that it applies to all son1975;Wade1978),isnotaddresseddirectlybyAxelrod situations covered by earlier versions of Hamilton’s rule but also and Hamilton (or in this article), but several researchers handlesnonadditivity,conditionalbehavior,andlackofgeneticsim- havedemonstratedtheunderlyingunitybetweeninclusive ilarity between altruists andrecipients.Ourresultssuggestchanges fitness and multilevel selection theories (Wade 1980;Bre- tostandardinterpretationsofHamilton’srulethatfocusonkinship den 1990; Queller 1992b; Frank 1998; Sober and Wilson andindirectfitness.Despitebeingmorethan20yearsold,Queller’s 1998). For reasons discussed below, reciprocal altruismis generalizationofHamilton’sruleisnotsufficientlyappreciated,es- peciallyitsimplicationsfortheunificationofthetheoriesofinclusive often left out of these unification efforts. Here we focus fitnessandreciprocalaltruism. on this missing piece: the unification of the theories of inclusive fitness and reciprocal altruism. Keywords:conditionalbehavior,Hamilton’srule,iteratedprisoner’s It is understandable thatAxelrodandHamilton(1981) dilemma,kinselection,mutualism,synergy. donotsuggestthatHRcouldapplytoreciprocalaltruism andtheirIPDmodels.Inadditiontoassumingthatplayers are unrelated, these models involve conditional strategies (genotype/phenotype differences) and nonadditive fitness * E-mail:fl[email protected]. † E-mail:[email protected]. 1 Weusecooperationandaltruismsynonymouslybecausecooperationinour Am.Nat.2006.Vol.168,pp.252–262.(cid:1)2006byTheUniversityofChicago. modelsinvolvesanimmediatealtruisticsacrificeinfitnessthatprovidesrel- 0003-0147/2006/16802-41512$15.00Allrightsreserved. ativefitnessbenefitstorecipients. Unifying Theories of Altruism 253 functionsthatHRcouldnotaccommodate.Yetjustafew they incur for their own helping behaviors. The required yearslater,Queller(1985)developedaversionofHRthat combination of genotype-phenotype assortment, benefit handles both of these issues, and in his article, Queller to cost ratio, and any nonadditive effects is given by suggeststhathisversioncouldapplytoreciprocalaltruism. Queller’s generalization of HR. This rule applies whether More recently, Sober and Wilson (1998) suggest a unifi- the source of positive assortment is interactions among cationofinclusivefitnessandreciprocalaltruismtheories. relatives (the original application), conditional behaviors They show how additive versions of the prisoner’s di- among nonrelatives, or reciprocalinteractionsacrossspe- lemma (PD) correspond to fitness functions used in in- cies(mutualisticsymbiosis).Thissinglerequirementgov- clusive fitness models, but they do not address the two erned by Queller’s version of HR brings unitytothesep- critical issues mentioned above: genotype/phenotype dif- arate theories of inclusive fitness and reciprocal altruism. ferenceswhenbehaviorsareconditionalandnonadditivity. We begin by briefly reviewing HR, Queller’s contribu- Frank (1994, 1998) notes that regression coefficients be- tions, and the original IPD experiments (Axelrod and tween species in his model of mutualism measure reci- Hamilton1981;Axelrod1984).WethenuseQueller’sver- procity and are similar to coefficients of relatedness in sionofHRforgenotype/phenotypedifferencestoanalyze inclusivefitnessmodels,buthealsodoesnotaddressthese an experimentinvolving an additiveIPD.UsingQueller’s two issues. The emphasis of his analysis (Frank 1998) is nonadditive version of HR, we also show how additive onpartitioningselectionandtransmissionandonunifying behavior that is iterated within generations gives fitness quantitative genetic and population genetic approaches. consequences similar to those of synergistic behavior Despite these suggestions that inclusive fitness and re- whereapairingisasingleinteractionpergeneration.(Syn- ciprocal altruism theories are related, unification of these ergy is defined below as positive nonadditivity.) Finally, two theories requires that HR be effectively applied to we extend this model so that there are conditional co- reciprocal altruism. However, until now, there has been operator and defector types in each of two mutualistic no directandsuccessfuldemonstrationofusingQueller’s speciesthatinteract.Heretheparticularconditionalstrat- moregeneralversionofHRinreciprocalaltruismmodels. egy, benefit level provided, cost paid for cooperative be- In fact, in an expansion of his original results, Queller haviors, and any nonadditive effects can be different in (1992a, 1992b) drops any mention of its applicability to each of the species. For each species, we use Queller’s reciprocal altruism. version of HR to accurately predict whether the cooper- HerewedemonstratethatQueller’sequationsdoindeed ative trait will increase in that species. Finally, we discuss provideafoundationfortheunificationofinclusivefitness the implications of these results for understanding and and reciprocal altruism theories. Our approach differs unifying the theories of inclusive fitness and reciprocal from Nee’s (1989) application of a version of HR to an altruism. IPDmodelofreciprocalaltruism.InNee’swork,thegen- erality of Queller’s equation was not utilized; instead, The Progressive Generalization of Hamilton’s Rule Queller’sversionwasbroughtbacktothesharedgenotype level by adding an additional term that related the phe- Hamilton’srule(Hamilton1963,1964)givesthecondition notype of others to their common genotype with focal necessaryforanaltruistictraittoincreaseinapopulation altruists. Here we use Queller’s equation in its full gen- in the next generation and is deceptively simple: eralitybysimplytakingitatfacevalue:itisthephenotype of others that is crucial, not their genotype. This more rb1c, (1) encompassing viewpoint allows us to include heterospe- cific interactions in mutualistic symbiosis, where altruists wherebisusuallyinterpretedastheaveragefitnessbenefit and recipients are clearly genetically unrelated and toarecipientofthealtruisticbehaviorandcistheaverage nonidentical. cost to an altruist for this behavior. Complications arise Ourmodelsandanalysissuggestthat,ratherthanbeing inthemeaningoftherterm,whichhasbeenprogressively fundamentallydifferentmechanisms,inclusivefitnessand generalizedovertheyears.Originallythoughtofasasim- reciprocal altruism are alternative ways to satisfy a com- ple measure of relatedness via descent (Hamilton 1963, mon single requirement for self-sacrificing traits to in- 1964), Hamilton (after interacting with Price [1970]) crease in a population. This requirement can be statedas broadened the meaning of r to be a measure of the as- follows: there must be sufficient positive assortment be- sortment of genetic types regardless of relatedness by de- tween individuals with the altruisticgenotypeinquestion scent (Hamilton 1970, 1972, 1975): and the helping phenotypes of others they interact with, suchthatonaveragethosewiththefocalgenotypebenefit Cov(G ,G ) A O b1c, (2) from the helping behaviors of others more than thecosts Var(G ) A 254 The American Naturalist Table1:ProgressivegeneralizationofHamilton’sruleillustratedbysituationsforwhichdifferentversions (eqq. [1]–[4])apply Kin Nonkin genetic Genotype-phenotype Nonadditivefitness Equation interactions similarity differences(G(P) functions(d(0) (1) Yes … … … (2) Yes Yes … … (3) Yes Yes Yes … (4) Yes Yes Yes Yes whereG isthegenotype(orbreedingvalue)withrespect interpretedasthecontributiontothedirectfitnessofthose A tothealtruistictraitforeachpotentialactor(subscriptA) with the altruistic genotype from the behavior of others. and G is the average genotype value of others(subscript Intheindirectfitnessconcept,bisseenasthecontribution O O) that interact with each potential actor. Queller (1985) by the actor to the fitness of others. This is a possible further generalized Hamilton’s r term by explicitly in- interpretationofHRwhenthesetwobenefitsarethesame, cludingtheconsequencesofthephenotype(behaviors)of but when the amount given and the amount received by actors and others on selection for a genetic trait rather focal altruists differ, as in our model of symbiosis below, than focusing on the effect of genotypes directly. This only the interpretation suggested by Queller’s version yields works correctly. Whenthereisadeviation(d)fromfitnessadditivityfor Cov(G ,P ) mutual cooperation (as there is in many IPD models), A O b1c, (3) Cov(G ,P) thenanadditionaltermisneededthatspecifiesthedegree A A to which the focal genotype covaries with mutual coop- whereP isthephenotypeoftheactorandP istheaverage eration,2scaledbytheamountofdeviation(Queller1985): A O phenotype of others interacting with each actor. In the appendixintheonlineeditionoftheAmericanNaturalist, weprovidemoredetailsaboutthegeneralizationofHam- Cov(G ,P )b(cid:1)Cov(G ,PP )d1Cov(G ,P)c. (4) A O A A O A A ilton’sruleaswellasthemathematicaldetailsofourmod- els and analysis. Thisdeviationvaluecanbepositive(representingsynergy), These covariance ratio expressions (eqq. [2], [3]) may negative (representing diminishing returns), or 0 (repre- be convenient in comparing different versions of r, but sentingadditivity).Notethatdividingbothsidesofequa- crossmultiplyingresultsinamoreeasilyinterpretedform tion (4) by Cov(G , P ) results in the same r term as in A A of Queller’s equation. For example, equation (3) can be equation (3), plus an additional covariance ratio related written as to the deviation from additivity. There are other versions of HR (for reviews, see Pepper 2000; West et al. 2002), Cov(G ,P )b1Cov(G ,P)c. (3a) butequations(1)–(4)representsignificantstepsinapro- A O A A gressivegeneralizationofHRthataresummarizedintable ThissaysthatthegenotyperepresentedbyG willincrease 1. A in the population if the covariance between its presence Queller’s versions (eqq. [3], [4]) apply to all situations in each potential actor and the helping behaviors (phe- covered by Hamilton’s versions (eqq. [1], [2]), plus they notypes) of others, scaled by the benefit of this help, is handleadditionalsituationsthatmaynotallowarecursive more than the covariance between its presence and the analysis (Grafen 1985), such as when the frequency of helping behaviors of actors themselves, scaled by thecost cooperative behavior depends on both genotype and en- of these behaviors. Simply put, the genotypewillincrease vironmental factors (e.g., the behaviors of others). To see if on average individuals carrying it receive more fitness thateachequationaboveismoregeneralthantheprevious benefits than they pay out. Note that it is the phenotype ones, note that if fitness functions are additive (dp0), or behaviors of others (P ) that is critical, not their ge- thenequation(4)becomesequation(3);ifphenotypefre- O notype; there is no G term in this equation. This has quencies equal genotype frequencies (G pP and O O O consequences for the usual indirect fitness interpretation G pP), then equation (3) becomes (2); andif thesim- A A of HR. As Frank (1998, p. 68) points out, “The direc- tionalityofQueller’srelatednesscoefficient…isopposite 2 TheproductPP representsmutualcooperation,wherecooperate(C)be- A O to the directionality of Hamilton’s inclusive fitness coef- haviorshaveaphenotypeof1anddefect(D)behaviorsaphenotypeof0. ficient.” The benefit (b) term in Queller’s version is best Thisisexplainedfurtherinthefollowingsections. Unifying Theories of Altruism 255 ilarityingenotypebetweenactorsandothersissolelydue Table 2: Typical prisoner’s dilemma fitness values for an to interactions within kin groups, then equation (2) can actor,givenitsbehavior(phenotypeP )anditsopponent’s A become equation (1), where r represents relatedness by behavior(phenotypePO) descent. Queller’s version (eq. [4]) is the most general in Opponent’sbehavior that it works without these restrictions or assumptions. C (P p1) D (P p0) O O Actor’s behavior contributesb contributes0 C (P p1) sacrificesc 3 0 The Iterated Prisoner’s Dilemma Model A w (cid:1)b(cid:2)c(cid:1)d w (cid:2)c of Reciprocal Altruism 0 0 D (P p0) sacrifices0 5 1 A The PD captures a fundamental problem of social life: w (cid:1)b w 0 0 individually rational behavior may lead to a collectively Note:Thesefitnessvaluescanberepresentedastheresultofadditive irrationalanddeficientoutcome.Inn-playerversions,this benefit contributions (b) fromits partner, itsownsacrificeorcost dilemma is also known as a “tragedy of the commons” (c),thebasefitnessvalueuncorrelatedwithCandDbehaviors(w), 0 (Hardin 1968) or a freeloader (free rider) problem (Mc- andthedeviationfromadditivityformutualcooperation(d).Forthe shownfitnesspayoffvalues,bp4,cp1,w p1,anddp(cid:2)1. Millan 1979; Avile´s 2002). Typical two-player PD fitness 0 values for the actor, given its own and its opponent’sbe- was also one of the simplest. Called “tit for tat” (TFT), haviors, are shown in table 2. Here behaviors are either this strategy always cooperates with an opponent in the cooperate(C)ordefect(D).Table2alsoshowsparameters first interaction (PD game) and, in all subsequent inter- thatdecomposethesefitnesspayoffsintermsofthebenefit actions, simply plays whatever the opponent did in the (b)providedtoanopponentbyaCbehavior,thecost(c) lastgame.ThisconditionalbehaviorallowedTFTtomin- paid by the cooperator, the base fitness (w) that is in- 0 imizeexploitationbydefectingopponents,suchas“always dependentofcooperation,andthedeviation(d)fromad- defect” (ALLD), while taking advantage of mutual coop- ditivitywhencooperationismutual.InthePD,eachplayer eration when it met other “nice” strategies. Since these has a dominant strategy to defect (D), but if they both original experiments more than 25 years ago, much re- cooperate, both can receive more (in this case, three in- search has been done on the IPD (Dugatkin 1997; Sachs stead of one) than if theybothdefect.Itispresumedthat et al. 2004; Doebeli and Hauert 2005). players exhibit their behaviors simultaneously and that From the perspective of Queller’s version of HR (eq. there is no knowledge or guarantee about what the other [3]), the combination of iterated games and conditional player will do. The dilemma is that cooperation makes a play can create positive assortment among the helping player vulnerable to exploitation; in the case of mixed behaviors (phenotypes) of others and the conditionally behaviors,thedefectorgetsthehighestpayoff(five),while cooperative genotype (e.g., TFT), even when there is no the cooperator gets the lowest (zero). positive assortment among genotypes. That is, if one cal- Note that while a two-player fitnessmatrixcanberep- culatesHamilton’srusingonlygenotypes(eq.[2]),itwill resented in terms of these four parameters, other param- be 0 in the case of random binomial pairing. Therefore, eterizations are also possible. Also note that this typical the traditional version of HR cannot be satisfied, and it PD matrix, which is the one used by Axelrod (1984), is appearsasiftheincreaseincooperation(e.g.,theincrease nonadditive: it cannot be achieved without a nonzero d in TFT types) observed in IPD models of reciprocal al- term.Inthiscase,therearediminishingreturnsformutual truism is not due to inclusive fitness as measuredbyHR. cooperation (dp(cid:2)1), but synergistic (d10) matrices Thisisnotinfactthecase;aswewillsee,Queller’sversion that still define a PD are also possible. While Axelrod of HR predicts exactly when conditional cooperationwill purposely chose a nonadditive PD to ensure that tour- increase in these models. nament results did not depend on additivity (R.Axelrod, personal communication, 2005), the nonadditive nature of these common PD fitness values is not generally ap- Hamilton’s Rule Applied to a Classic preciated.Becauseofnonadditivity,thisPDcannotbean- Reciprocal Altruism Model alyzed with a pre-Queller HR. Confirming Queller’s Version AlthoughinaPDsituationitisindividuallyrationalto defectineachsingleplayofthegame,AxelrodandHam- Here we offer a simple example where Queller’s version ilton(1981)providedearlysupportforreciprocalaltruism of HR (eq. [3]) is applied to reciprocal altruism using a theory (Trivers 1971) by showing that conditional coop- population consisting of the two classic types mentioned erativestrategiescandowellwheninteractions(PDgames) above, TFT and ALLD. This kind of population hasbeen areiterated.ThemostsuccessfulstrategyinAxelrod’stour- used previously to apply a modified HR to an IPD (as naments (submitted by social scientist Anatol Rapoport) mentioned above; Nee 1989), to classify typesofaltruism 256 The American Naturalist production as a function of the benefit value (b) for dif- ferentvaluesoftheinitialTFTfrequency(Q)andnumber of iterations (i). For convenience, other parameters are held constant at cp1 and w p1. Notice that, all else 0 being equal, more iterations or higher starting Q make it easier for DQ to increase. Arrows in figure 1 indicate the predicted equilibrium points (DQp0) using Queller’s (eq. [3]) version of HR and the same parameters used in figure 1 (appendix). If instead of equation (3), which in- volves P , we use the more restrictedequation(2),which O involvesG ,rp0forthisbinomialpopulationstructure, O and we do not correctly predict the increase in the pro- portion of TFT (Q). In contrast, Queller’sversionexactly predictsthese“tippingpoints,”thatis,thevalueofbenefit, b, beyond which TFT increases in each population. Figure 1: Change in the proportion of TFT players (DQ) after one generationasafunctionofbenefit(b)levelinapopulationofTFTand ALLDplayerswithbinomialpairing.Resultsshownforvariousstarting Iterations and Synergy TFT proportions (Q) and numbers of iterated games (i), where other parameters for each game are held constant at cp1, w p1, and 0 Queller’s version of HR (eq. [4]) suggests two ways to dp0. Arrows indicate where DQp0. These equilibrium points are enhance the evolution of cooperation, given random predictedbyQueller’sversionofHR(seetableA2intheonlineedition oftheAmericanNaturalist). grouping: if cooperative behaviors toward altruists are more frequent than the frequency with which they are groupedwithotheraltruistsorifthereisnonadditivesyn- (KerrandGodfrey-Smith2002),andtostudytheevolution ergyformutualcooperation.Hereweshowthatthesetwo of altruism in finite populations (Nowak etal.2004).Be- effects can have equivalent fitness consequences. Coop- cause one of the types (TFT) uses conditional behaviors, eration evolves in the populations plotted in figure 1 be- we must measure genotype and phenotype frequencies cause conditional behavior positively assorts cooperation separately.Tocalculatethecovariancesneededinequation withTFTgenotypes,butanalternativeanalysisispossible. (3), we take TFT as our focal genotype (G) and assign it Assuming additive PD parameter values of bp4, cp a value of 1 and ALLD a value of 0. For phenotype (P), 1, w p1, and dp0 and iterations of ip10, the cu- thecooperate(C)behaviorhasavalueof1,andthedefect 0 mulative intergenerational fitness consequences of differ- (D) behavior a value of 0. In a population of these two entpairingsareshownintable3asiftheyweretheresult types,therewillbethreepossibleparings(TFT-TFT,TFT- of a single interaction between the players. This resulting ALLD,ALLD-ALLD),eachwithpredictablevaluesforG , A game matrix, no longer a PD, can be decomposed into P ,andP ,giventhenumberofiteratedgames(i)within O A ourfourparameters.Weuseprimestodistinguishparam- generations(tableA1intheonlineeditionoftheAmerican etersoftheresultinggamefromthoseoftheoriginal.The Naturalist).3 Again, subscript A indicates the focal actor, andsubscriptOmeansothers(inthiscase,thefocalactor’s opponent). We can now calculate the change in TFT fre- Table3:Cumulativefitnessvaluesforpairings quency in the population (DQ), assuming fitness payoffs of tit-for-tat(TFT)andalwaysdefect(ALLD) are proportional to offspring representation. Given the players thatlast for ip10 iteratedgames population averages forthefrequencyofCbehaviorsand Opponent’sbehavior(s) theinitialfrequencyofTFTtypes(Q),wecanalsocalculate Actor’s behavior TFT ALLD whether Queller’s version of HR is satisfied for any given 40 9 parameter settings (see appendix). TFT w(cid:1)(cid:1)b(cid:1)(cid:2)c(cid:1)(cid:1)d(cid:1) w(cid:1)(cid:2)c(cid:1) We begin with an additive PD game (dp0). Figure 1 0 0 14 10 showsthechangeintheproportionoftheTFTtype(DQ) ALLD w(cid:1)(cid:1)b(cid:1) w(cid:1) 0 0 after one generation of random pairing and asexual re- Note: For each game, bp4, cp1, w p1, and 0 dp0.Theshownfitnesspayoffvaluesinterpretedas 3 For mathematical convenience, i represents a fixed number of games in theresultofonlyasingleinteractioncanbedecomposed eachinteraction.Similarresultsholdforgamesofaveragelengthi,andthe withb(cid:1)p4,c(cid:1)p1,w(cid:1)p10,andd(cid:1)p27.Thisgame simpleplayersinthismodelarenotcapableofusingknowledgeofthenumber isnotaprisoner’sdilem0 ma;inthegametheorylitera- ofgamesforbackwardinduction. ture,itiscalled“assurance”or“staghunt.” Unifying Theories of Altruism 257 parameters for table 3 are b(cid:1)p4, c(cid:1)p1, w(cid:1)p10, 0 d(cid:1)p27 (appendix). Now suppose that we do not know the fitness conse- quences of each social interaction (game) or how often interactions(iterations)occurwithineachgeneration.In- stead, we see only who is paired with whom and the re- sulting fitness consequences to each type at the end of eachgeneration.Thefitnessconsequencesarethesameas in the original situation, but from this “black box” per- spective, there are no iterations and no difference in ge- Figure 2: Fitness relationships between two interactingspecies(1and notype versus phenotype frequencies. There are just bi- 2). The origin of each arrow correspondsto acooperatebehavior(C) nomial single-interaction (ip1) pairings but a strong bythespeciesatitsorigin,andtheterminationofeacharrowindicates fitness synergy when TFT meets TFT. Analyzing thissyn- whichspecies’fitnessisdirectlyaffectedbythisbehavior.Forinstance, ergistic (d(cid:1)p27) situation with equation (4) and no ge- aCbehaviorexhibitedbyamemberofspecies1hastwodefiniteeffects— decrementofc initsownfitnessandanincrementofb toitsspecies notype/phenotype differences gives the exact same in- 1 1 2partner’sfitness—aswellasonepotential“interaction”effect(indicted equality as assuming additivity (dp0) and genotype/ bydashedlines),achangeofd initsownfitnessonlyifthereisalsoa phenotype differences due to iterations (ip10) and simultaneous C behavior by its1species 2 partner. We label benefitsby conditional play (appendix). Thisperspectiveprovidesan theirsourcetoemphasizethathelpisgivenheterospecifically. alternativeexplanationforhowconditionalstrategiessuch as TFT evolve. From this point of view, itis not somuch species1exhibitingaCbehaviorareb (benefittospecies 1 that iterations and conditional play “solve” the PD itself 2) and c (its own cost) and similarly for species 2. We 1 (in which defection is favored) but that they effectively labelthe benefits bytheirsourcetohighlightthefactthat change the game into one in which mutual cooperation the benefit received by a member of one species comes (in table 3 labeled TFT) has the highest fitness payoff. fromacompletelyunrelatedmemberoftheotherspecies. Note,however,thatthe“assurance”gamethatthePDhas Nonadditive effects, which have their source in bothspe- been converted into is not itself dilemma free: ALLD still cies, are subscripted with the species whose fitness is di- receivesmorethanTFTinallheterogeneouspairings(table rectly affected. The proportion of the cooperative typein 3), and therefore a maximin strategy results in a Nash eachspeciesisgivenbyQ andQ,respectively.Ingeneral, 1 2 equilibrium of mutual defection, which is Pareto whether the focal genotype (e.g., TFT) of species 1 in- nonoptimal. creases in the next generation is predicted by Queller’s version of HR (eq. [4]), where A (actor) is a member of species 1 and O (other) is a member of species 2: Hamilton’s Rule Applied to Cooperation across Species Cov(G,P)b (cid:1)Cov(G,PP)d 1Cov(G,P)c . (5) Mutualism Model 1 2 2 1 1 2 1 1 1 1 Here we use Queller’s version of HR to analyze a simple A symmetric equation (where all subscripts areswitched) model of mutualistic symbiosis in which there are two predicts whether the focal type in species 2 increases or interacting species (labeled 1 and 2), each with two dif- not. We will refer to these two instances of Hamilton’s ferenttypes:ALLDand(usually)aconditionalcooperator rule as HR and HR, respectively (appendix). 1 2 type such as TFT. For convenience, the cooperative be- This model of symbiosis has some similarities to one haviors of interest take place only heterospecifically. For developed by Frank (1994), but his model assumes that instance,thebehaviorsbetweencleanerfishandtheirhosts phenotypeandgenotypefrequenciesarethesameandthat (Bshary and Grutter 2002) are strictly heterospecific: fitnessfunctionsareadditive.Whilethissimplemodeldoes cleaner fish do not clean conspecifics, and hosts are not not capture all types of mutualisms (Bronstein 2001), as cleanedbyotherhosts.Wealsoassumerandom(binomial) far as we know, the analysis presented here is the first pairings, but conditional behavior will provide the asym- example of the use of Queller’s version of HR to analyze metry in benefits within species necessary for mutualists cooperation across species in which behavior canbecon- to increase (Ferriere et al. 2001). Note that these inter- ditional (G(P). actions, unlike our within-species cooperation examples above, can be asymmetric between species in terms of Dynamic Simulations costs,benefits,anddeviationfromadditivity(Frank1994; Sachsetal.2004).Figure2illustratesthesefitnessparam- Figure 3 illustrates the coupled dynamics in our model, eters between the two species. The benefit and cost for where cooperation can reach saturation in both species 258 The American Naturalist Figure3:Dynamicsbetweentwospecies,eachwithacooperativetypeandanALLDtypeforvariousparametersettings.A,Cooperationevolves even though in species 1 cooperation costs more than the benefit it produces. Parameters areip4,b p1.5,c p2,d p0, initialQ p0.2, 1 1 1 1 b p5,c p0.1,d p0,andinitialQ p0.1;inbothspecies,thecooperativetypeisTFTandw p2.B,Allparametersarethesameforboth 2 2 2 2 0 species,exceptthatthecooperative type inspecies1 isTF2Tand in species2isPavlov.Parametersareip80,b pb p4,c pc p1,d p 1 2 1 2 1 d p0,initialQ pQ p0.1,andw p1.C,Cooperativetypeisunconditional(ALLC)inspecies1andTFTinspecies2.Otherparametersare 2 1 2 0 ip100,b p2,c p1,d p0,initialQ p0.01,b p2.2,c p0.1,d p1.3,initialQ p0.5,andinbothspeciesw p1.Thebarsaboveeach 1 1 1 1 2 2 2 2 0 graphindicateatwhichgenerationsQueller’sversionofHR(e.g.,eq.[5])issatisfied(solidbar)foreachspeciesrespectively(HR andHR).Note 1 2 thatsatisfyingHRcorrespondstowhenthecooperativetypeineachspeciesincreases. evenundersomesurprisingconditions.Infigure3A,mu- an increase in altruism is not that the benefit given by tual symbiotic cooperation evolves even though cooper- altruists sufficiently exceed their costs but rather that, on ation costs members of species 1 more than the benefit average, the benefit received by those with the altruistic they provide (c 1b ). This inefficient formof altruismis genotype exceed their costs. 1 1 not generally thought to evolve, but it can evolve here In the run shown in figure 3B, all parameters are the because of the high benefits of cooperation provided by same for both species and additive, but the cooperative the other species; that is, b is sufficiently greater than c strategies differ: in species 1, it is tit for two tats (TF2T), 2 1 (eq. [5]). In this case, fitness is additive,d pd p0.In which plays C unless the previous two plays by its op- 1 2 these dynamic numerical simulations, HR and HR are ponentwerebothD,andinspecies2,itisPavlov(Nowak 1 2 calculatedeachgeneration,andhorizontalbarsaboveeach and Sigmund 1993), which initially cooperates but graphinfigure3indicatewhentheserespectiveinequalities switchesitsbehaviorifitdidnotgetoneofthetwohighest are satisfied in each species. In all cases, Queller’s version payoffs in the last game. The terms Q and Q give the 1 2 of HR accurately predicts when the cooperative type will fraction of the more cooperative type in each species, re- increase. For instance, at the start of the run depicted in spectively. Initially, when both species are dominated by figure 3A, HR (eq. [5]) is not satisfied and the TFT type ALLD types (Q pQ p0.1), Pavlov looses ground in 1 1 2 decreasesinspecies1,butbecausethefractionoftheTFT species 2 as it alternates C and D behaviors when paired type in species 2 (Q) is simultaneously increasing, Q is with the ALLD type in species 1; it never gets one of the 2 1 eventually pulled up in this coupled system. two highest payoffs and therefore keeps switching. The Notethatifthebenefitprovidedbycooperativebehav- TF2T type in species 1 fares better because it cooperates iors in species 1 (b) is used in equation (5) instead of b onlyinthefirsttwogameswhenitmeetsanALLD.Even- 1 2 for comparison with c, HR does not accurately predict tually, the fact that the TF2T type in species 1 increases 1 1 the fate of TFT in this species. To work, the b term must providesmoreopportunitiesforthePavlovtypeinspecies be the benefit received, not the benefit provided, by car- 2 to end up in a mutually cooperative interaction, and it riers of the focal genotype (e.g., TFT). The criterion for too eventually increases. Unifying Theories of Altruism 259 Finally, in figure 3C, the cooperative type in species 1 procity (even if nonconditional): on average, carriers of is unconditional always cooperate (ALLC) and in species the altruistic genotype must receive directbenefits.While 2 is TFT. In this case, even having conditional behaviors this reciprocated benefit can be asymmetric, on average, in only one species can be enough for cooperation to it must sufficiently exceed focal carriers’ costs, where the evolve in both. Here species 1 faces a PD in each inter- meaning of “sufficiently” is captured by Queller’sversion action, experiences no synergy for mutual cooperation of HR. Inclusive fitness is also broadenedbeyondtheno- (d p0),andstartswithonly1%ofitspopulationbeing tion of genes helping other copies of themselvesinrecip- 1 the ALLC type (Q p0.01). We would expect this naive ients (Williams 1966; Dawkins 1976). This “selfish gene” 1 cooperator to be selected out of species 1 under random interpretation of HR holds only in the special case where pairing,butinsteaditsteadilyincreasestosaturation.Here helping behaviors are predicted by the common alleles species 2 starts with an even mixture of TFT and ALLD betweendonorandrecipients.Abroadernotionofinclu- types and experiences synergy for mutual cooperation sive fitness, as Queller (1985) argued for, is fitness aug- (d p1.3). mented by help from others regardless of their genotype. 2 Of course, many other parameter settings are possible, Therearetwoimportantandrelatedideasherethatare including where cooperation goes extinct. Here we just reflected in the most general form of HR (eq. [4]; table illustrate that conditional behavior with iterationsand/or 1), which can encompass not only kin selection but also nonadditivity can allow a cooperative symbiotic relation- reciprocal altruism and symbiosis. For its most general shiptoevolveunderrandomgroupingevenifonespecies application, first, the direct cost of behaving altruistically is inefficient (c1b) in its help (fig. 3A), less effective in shouldbecomparedwiththedirectbenefitgaintocarriers avoiding exploitation by ALLD (fig. 3B), or even uncon- of the altruistic genotype from others (rather than to an ditionallycooperative(fig.3C).Ourdifferentstrategytypes indirectbenefitviatheenhancedfitnessofothercarriers), are not meant to represent any particular symbiosis ex- and second, the direct fitness benefit depends on the be- amples in nature but to show that the evolution of mu- haviors (phenotypes) of others, not their genotypes. The tualismsneedsnotdependonparticularstrategiesandthat firstpointisplainlyevidentinoursymbiosisresults,which strategiescanvarybetweenmutualisticpartners.Notethat show that, in the generalcasewherebenefitprovideddif- in this simple model with its obligatoryheterospecificin- fers from benefit received, only an interpretation of HR teractions,thefatesofthecooperativetypesineachspecies basedondirectbenefitreceivedbycarriersgivesacorrect are ultimately tied together, and the system reaches an result. Rather than employing multiple interpretations of equilibriumofeitherallcooperation(Q pQ p1.0)or HR—one for relatives, one for nonrelatives having com- 1 2 all defection (Q pQ p0.0). At each generation along mon alleles, and one for different alleles (including in 1 2 the way, Queller’s version of HR accurately predicts the different species)—it is more parsimonious for a theory direction of selection for the (conditionally) cooperative of altruism to be based simply on the most general in- type in each species. terpretation of HR. From this perspective, the equation (1) version (for relatives) of HR is just a special case of the equation (2) version, which does not depend on re- Discussion latedness by descent, and the genotype of others (G ) in O A recent review article (Sachs et al. 2004), with the same the equation (2) version is just a stand-in for the phe- titleasAxelrodandHamilton’s(1981)seminalarticleand notypeofothers(P )inthemostgeneralformoftherule, O Axelrod’s (1984) book, echoes the traditional view that that is, Queller’s equations (3) and (4). It is also more inclusivefitnessandreciprocalaltruismarefundamentally parsimonious to see HR as measuring whetherthefitness distinct explanations for the evolution of altruism. For gains to carriers are sufficiently greater than their costs instance, these authors state that these theories differ be- rather than in terms of indirect fitness. cause reciprocal altruism can operate “between nonrela- Note that explaining how, on average, benefits to car- tives and between species” (Sachs et al. 2004, p. 139) and riersofthealtruisticgenotypeendupexceedingtheircosts that inclusive fitness is unique because “the cooperative doesnotaffectthedefinitionofaltruismattheindividual individualneednotbenefitfromitsact”(Sachsetal.2004, level (Kerr et al. 2004). The conventional perspective in p. 143). which an individual altruist incurs cost and gives benefit Inthisarticle,wedemonstratethat,onthecontrary,the remainsessentialtodefiningaltruism.Giventhatindivid- distinction between inclusive fitness and reciprocal altru- uals have no guarantees abouttheirpartner’spresentand ismisnotsharp.WeshowthatHamilton’sinclusivefitness future behaviors, cooperation (C) in any given PD inter- rule (in Queller’s generalized form) applies to reciprocal action (game) is altruistic because, comparedwiththeal- altruism. Analysis of Hamilton’s rule also reveals thatthe ternativebehavior(D),acooperatorgivesbenefittoothers evolution of altruism by inclusive fitness involves reci- ata cost toitself.Evensummingoveriterations,TFTcan 260 The American Naturalist beseenasaltruisticfromarelativefitnessperspective(Wil- At the same time, cooperative interaction may create son 2004) in that TFT never does better than its paired new opportunities for defection (Michod and Nedelcu opponent (Rapoport 1991; Sober and Wilson 1998) and 2003), for example, the free-riding hunter or swimming its opponent does better than if it had been paired with spermthatexpendslessenergythanaveragebutstillreaps ALLD. The fact that pairs of TFT do better than pairs of the benefit of others’ cooperation. The relationship be- ALLD for particular parameters helps explain how altru- tween synergy and exploitation by defectors is difficultto ism evolves overall, and this is captured in Queller’s ver- appreciate in the paired interactions modeled here. This sion of HR and in the multilevel selection framework is because when there is one C behavior, there is no syn- where groups are of size 2 (Sober and Wilson 1998). ergy, and when there are two C behaviors, there are no Wehaveintentionallyusedsimplemodelsofreciprocal defectorstodotheexploiting.Moregenerally,wheresyn- altruisminordertoillustrateourpointaboutunification, ergistic benefits are an increasing function of thepropor- but Queller’s version of HR can be applied to a much tion of cooperators in a group (and benefits are shared broader array of circumstances. This includesvaluesofb, among all group members), there are necessarily more c, and d that fall outside the definition of a PD, group cooperators insituationswiththehighestsynergisticpay- sizes greater than two, diploid genetics, other population offs,whiledefectorsareatarelativeadvantagewithineach structures besides binomial random grouping, degrees of group because they do not pay the cost. cooperation rather than just all C or D, and other forms Synergy may also be important in addressing recent of conditional behavior beyond those based on just the argumentsthatreciprocalaltruismrarelyoccursinnature past behavior of others. (Hammerstein2003).Someoftheseassessmentsarebased on the low frequency of repeated interactions, but as we have shown, fewer iterations are required in the presence The Role of Synergy of nonadditive synergy. Elsewhere, we showed that mul- Models of the evolution of altruism typicallyassumevar- tiple generations within groups similarly result in non- iouscombinationsofrandominteractions,additivefitness additivity(FletcherandZwick2004).Thoughnotexplored functions, and a one-to-one correspondence between ge- here, nonadditivity may also be negative, as inthetypical notypeandphenotypefrequencies.Droppingoneormore PD(table2)orothercasesofdiminishingreturns(Foster of these assumptions can make the evolution of cooper- 2004; Hauert et al. 2006). ationmorelikely.Traditionalinclusivefitnessmodelsfocus on nonrandom interactions due to kinship while leaving A General Theory with Many Specific Mechanisms the other assumptions in place. Traditional reciprocal al- truism models assume random encounters but use con- In a review on the evolution of mutualism, Herre et al. ditional behavior to break the correspondence between (1999, p. 52) lament that “there is no general theory of genotype and phenotype. mutualism that approaches the explanatory power that What has been less explored but is explicitlyaddressed ‘Hamilton’s rule’ appears to hold for the understanding by Queller’s version of HR (eq. [4]) is the role of non- of within-species interactions.” In this article, we have additive synergy, something quite different from the po- shownthatinfactHRitself,inQueller’sgeneralizedform, tentially additive benefit of mutual cooperation (see provides a general theoretical basis for understandingthe Hauert et al. 2006). Its significance will of course depend evolution of cooperation across species. Fundamentally, on fitness consequences in particular interactions. In the the evolution of altruism (within or between species) de- accompanying commentary on Queller’s (1985) original pends on sufficient positive assortment between individ- article, Grafen (1985, p. 311) argued that “for genes of ualswiththealtruisticgenotypeofinterestandthehelping small effect, additivity is restored and the correctness of behaviors(i.e.,phenotypes)ofothersand/orsufficientsyn- Hamilton’s rule is restored with it.” While HR in its ad- ergistic effects of mutual cooperation (eq. [4]). ditive form may be a good approximation when fitness Sufficient association between cooperators and coop- effects are small, cooperative traits may have strong syn- eration from others or synergistic effects can be created ergisticeffects.Fromcooperativehuntersthatbringhome inavarietyofways.Theseincludeinteractionsinspatially spoilsgreaterthantheycouldgetalone(PackerandRuttan structured populations among kin (Hamilton 1964) or 1988), to cooperatively swimming sperm that reach the across species (Doebeli andKnowlton1998),iteratedand egg faster than individual swimmers (Moore et al. 2002), conditional behavior based on the pastbehaviors(Trivers to the potential synergistic benefits of mutualisms(Herre 1971;AxelrodandHamilton1981;Axelrod1984;Dugatkin etal.1999;Bronstein2001;Ferriereetal.2001),thenatural 1997) or reputations (Nowak and Sigmund 1998, 2005; worldisfullofpotentiallysuperadditivesituations(Wright Panchanathan and Boyd 2003) of others, policing (Frank 2000; Michod and Nedelcu 2003). 1995, 2003), punishmentofnonaltruists(BoydandRich-

Description:
on the idea of inclusive fitness (Hamilton 1963, 1964, 1970,. 1972, 1975). His most Frank (1994, 1998) notes that regression coefficients be- . (Hardin 1968) or a freeloader (free rider) problem (Mc- Bronstein, J. L. 2001.
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.