Robert Trivers and colleagues on Nowak, Tarnita, and Wilson’s “The evolution of eusociality”

One of the most difficult things about being the only full-time biologist on the Pratt Institute campus is that I do not have the opportunity to discuss serious science in my field with colleagues or guest speakers. To help alleviate this problem, I have my friends who are at serious research institutions on the lookout for opportunities to engage in this kind of exchange. My friend Ari Novy, a graduate student at Rutgers University, came through for me big-time by alerting me to a special seminar given by Robert Trivers regarding a recent Nature paper authored by Martin A. Nowak, Corina E. Tarnita, and Edward O. Wilson (henceforth referred to collectively as “NTW”).

The paper, boldly entitled “The evolution of eusociality” (Nowak et al. 2010), seeks to question the validity of kin selection as an explanation of eusociality. Eusocial organisms live in colonies that appear to act in a unified fashion. By analogy they seem very much like sexually-reproducing multicellular organisms in that they are composed of workers, who generally take on somatic roles, and reproducers, who form the germ line and disperse the genes of the colony. Of course this analogy breaks down in several ways, as each individual member of a eusocial colony is in fact a multicellular organism, and can thus operate independently of the colony. This is the brilliant adaptive feature of eusocial colonies: they have a form of “detachable soma” that can act on behalf of the colony while traveling great distances to gather resources. The millions of Atta ant species in each colony disperse through their tropical forest homes to methodically defoliate the tree species which provide their means of sustenance (after the leaves are digested by a mutualist fungus, of course). Bees produce great amounts of honey by the concerted pollen and/or nectar gathering effort of their workers. While some individuals are out gathering food, other workers tend to a nursery that pumps out offspring at a staggering rate. To protect these offspring and the colony as a whole from a host of competitors, predators, and even parasites, some individuals stand guard over the nest, and are often willing to die defending their communal home. Cumulatively these are feats of effort that could only be accomplished by a vast colony composed of specialized individuals, each performing their appointed task.

What is surprising about eusocial colonies from an evolutionary standpoint is the way they reproduce. Although the pattern varies somewhat from species to species, the general trend is to maintain specialization in not only foraging, care of offspring, and guarding, but also in reproduction. A “queen” who founds the colony often does all of the reproducing, giving birth to workers who are sisters. These workers raise their sisters and brothers who will disperse from the colony and be the sole means of passing on the genes of the colony. This arrangement has served as a challenge to evolutionary biologists since Darwin tackled the issue in The Origin of Species, as it is difficult to understand why workers would forgo reproduction in favor of helping their mother raise massive numbers of her own offspring.

In the 1960’s William D. Hamilton appeared to have solved this problem with the publication of his paper “The Genetical Evolution of Social Behaviour” (Hamilton 1964). Formalizing an argument first suggested by J.B.S. Haldane, Hamilton inferred that reproductive fitness must include not only the genes passed on through the offspring but also the genes passed on through other relatives. He called this more complete measure “inclusive fitness”, and derived a simple inequality (now known as Hamilton’s Rule) that sought to answer the question “when should I provide help to relatives”?:

r B > C

In this equation, C and B are the costs and benefits in terms of reproductive fitness of an altruistic behavior and r is the coefficient of relatedness. Hamilton meant for his equation to be applied to a particular gene that would code for altruistic behavior, implicitly assuming that the gene was the unit on which selection acted.

One of the interpretations of Hamilton’s work that emerged strongly was the idea that the unique genetic system of the hymenoptera (ants, bees, and wasps) might account for their tendency to form eusocial colonies. This conclusion followed from the observation that in a haplodiploid system of sex determination, sisters from a once-mated mother are all related to each other with a coefficient of relatedness of r = 0.75. This means that sisters are actually more closely related to each other than they are to their own offspring, with which their coefficient of relatedness is r = 0.5. This odd pattern of relatedness provided a really parsimonious answer for why hymenopteran workers forgo reproduction in favor of raising their sisters. Of course parsimony is not the same as proof, and no one has ever been able to show that haplodiploid relatedness patterns account for the behavior of eusocial worker drones. Nonetheless, the argument by inclusive fitness spread like wildfire, and was used to argue against previously-articulated hypotheses that suggested that altruism among eusocial colonies was the product of selection acting at the level of the colony (in other words, group selection). George C. Williams was probably the most successful carrier of this banner, and he continued to make the argument in even his later works (Williams 1996) that kin selection was the best explanation for the origin of eusociality.

As much as Hamilton’s Rule makes intuitive sense, how to actually use it remains a bit of a mystery. Measuring r has gotten quite a bit easier with the advent of cheap and easy DNA sequencing, but measuring B and C is considerably more challenging. How exactly would you measure the exact benefits and costs associated with helping behavior in eusocial species? While it might be possible to sum up the genetic benefits of helping behavior, how would you measure the costs, all of which are opportunity costs to eusocial workers? And what exactly is the unit of behavioral effort that you are measuring? Is it a lifetime of helping behaviors? If so, how do I measure these costs? What share of the reproductive benefits should be assigned to each individual? If I die in defense of my colony, is that it for my inclusive fitness benefits, or do I continue to accrue these benefits for the lifetime of the colony?

I think that it is also really important to understand what Hamilton’s Rule represents. It is an optimality model — perhaps the most successful optimality model in all of evolutionary biology — and not a mechanistic model explaining how optimal behavior might evolve. Optimality models can be quite valuable, but they need to be used within the domain where they are valid. When we observe a behavior or suite of behaviors, it makes sense to ask whether the individual displaying the behavior(s) is acting in an optimal fashion. If we discover, for instance, that the organism behaves in a manner that is consistent with the predictions of inclusive fitness theory, we might feel comfortable saying that kin selection is the mechanism by which the behavior evolves. In practice, discovering such an optimum requires that we go well beyond the simplistic comparison of coefficients of relatedness, and understanding the actual optimum may require more mechanistic modeling (for example, look at the way that relatedness is incorporated in Archetti 2009). Even after employing a more mechanistic approach to discovering optimal behavior, we need to consider by what evolutionary process a population might reach that optimum, and simply knowing the relatedness is insufficient.

What NTW contend in their paper is that the evolution of the kind of cooperation that typifies eusocial species cannot be explained by kin selection theory. Instead, they propose that a form of group selection is necessary to create the behaviors that best define eusociality: the retention of worker-caste offspring in the nest. They offer a simple, five-step model of the evolutionary process by which solitary species become eusocial. This model — which in its conceptual form is not actually new (see Hölldobler and Wilson 2008) — relies not on relatedness but on a multilevel selection process wherein between-colony selection is stronger than within-colony selection. What is important to note in their formulation is that parent-offspring conflict (which is an extension of kin selection) selects against eusociality, as female offspring may enjoy better reproductive fitness if they disperse. It is only a higher-level selection process favoring those colonies where “worker mutiny” can be suppressed that can produce the beginning stages of eusociality. Once the genetics coding for eusocial “worker compliance” are fixed, this system becomes robust to invasion by solitary alleles. Only then does relatedness go up as a byproduct of being eusocial breeders. Although NTW do not explicitly make these points, I would add two more observations: 1) Such a “genetic revolution” would certainly open up the doors for pre-zygotic isolating mechanisms that might lead to speciation; and 2) Once the eusocial lifestyle has been established, it is still prone to invasion by selfish “cheating” workers who may wish to advance their interests by reproducing themselves.

The novelty of the NTW paper is that it provides several mechanistic models that explore the conditions under which eusociality might evolve by the process conceptualized by Hölldobler and Wilson (2008). Unfortunately, due to the manner in which Nature is published, almost all of the novel content provided in this paper is found in the supplementary material. What is valuable about their model is that it has the potential to provide testable predictions about when we should see the evolution of eusociality. After all, this ought to be the primary focus of research in this area: to explain why some species are eusocial and some species are solitary. NTW prominently point out that patterns of relatedness alone (haplodiploidy versus diplodiploidy) do not account for the prevalence of eusociality. Clearly ecological conditions influence whether eusociality will evolve, and so we need a mechanistic model to explore when eusociality will be favored. If this model incorporates the most relevant ecological variables, it should predict the pattern of evolved eusociality that we observe in nature.

The NTW paper does not get as far as predicting the pattern of eusociality, but it at least provides a starting point for this sort of analysis. The one thing that NTW mention that hints at explaining the observed pattern of evolved eusociality is that in their model crossing the threshold over to eusociality can only occur under very specific conditions but once this threshold has been crossed the system is robust to the invasion of solitary genotypes. This is held up as consistent with the rare-but-robust pattern of eusociality seen in nature. Pretty weak as far as predictive power goes, but perhaps a start. What we would really like is a model that makes stronger predictions about the kinds of ecological conditions that allow eusociality to evolve (potentially problematic, as it would require pretty accurate reconstruction of ancient environments) and persist (potentially possible, if we can measure the relevant ecological variables in current environments).

So what alternative does kin selection offer to the NTW model? The answer is “none”, and that is why NTW want to toss out the idea of inclusive fitness. I can see where they are coming from: forty-six years later and no one can formulate a population dynamic or agent-based model that explains the evolution of eusociality by relatedness alone? Of course this assessment of inclusive fitness theory is a little unfair, as it asks that an optimality model serve as a population genetic model. We should not pick on Hamilton because he provided us with a different tool than we now need to answer the question “how did eusociality evolve?”. What is remarkable is that no one has successfully picked up Hamilton’s idea and incorporated it into a mechanistic model of genetic evolution that demonstrates how and when eusocial species reach the optima his rule predicts. The potential to measure ecological conditions in terms of costs and benefits are there, but we are still busy talking about coefficients of relationship. How do we turn inclusive fitness theory into a predictive science?

I apologize for trying to be the reconciliator who throws water on the fire of a good scientific gang fight, but it is also possible that a hybrid approach might produce the model that best predicts the ecological conditions under which eusociality will evolve. After all, the preponderance of eusocial species do have a haplodiploid genetic system, so relatedness might tip the scale towards eusociality in many species. The conceptual form of this model is pretty simple to verbalize. If you imagine a eusocial system that has evolved by the mechanisms proposed by NTW, the big question after primitive eusociality evolves is how that eusocial pattern will develop. It could become further specialized, with genetic interests being further aggregated by more extensive role differentiation and the residence of multiple generations of kin within the colony. It could also be destabilized by worker mutinies. I would suggest that here is where kin selection theory could be valuably considered through the model. There is no doubt that if the model proposed by NWT is correct, eusociality depends on the optimal functioning of the colony as a whole, as it must compete against the results of other colonies. But the ability to function optimally will be constrained by individual-level selection within each colony. Perhaps relatedness can temper the effects of this within-colony competition, allowing a slightly larger domain of ecological conditions to foster eusociality. In the end, ecological conditions will dictate which is a stronger force of selection — competition between colonies or competition within colonies — but without considering the effects of inclusive fitness, we might inaccurately predict the dividing line between solitary and eusocial species.

After reading over and partially digesting the NTW article, I was excited to see what Trivers and other folks at Rutgers would have to say about it. I did not know what to expect from Trivers’ seminar presentation, but understanding his previous work I had some sense that he might have objections. First of all, he has published an important book entitled “Genes in Conflict: the Biology of Selfish Genetic Elements” (Burt and Trivers 2008) that shows that many genomes are littered with cheating fragments that may reduce the overall fitness of the organism that carries them. Trivers himself has published numerous papers on B chromosomes, extra genetic elements which have been shown to “cheat” in the meiotic process so as to be over-represented in offspring. He published the seminal paper on parent-offspring conflict (Trivers 1974) demonstrating that asymmetrical genetic interests can cause mothers and their young to fail to cooperate. He is also the originator of reciprocal altruism (Trivers 1971), a key mechanism by which cooperation can evolve, so he has played an important role in this field.

Every professor has his own style, and Trivers’ style is definitely pretty didactic. Once we had all arrived and seated ourselves in the mini-auditorium of Rutger’s Marine and Coastal Sciences building, Trivers proceeded to erupt into a pretty massive attack on the paper. From the beginning he warned us that some of this critique would be personal and some would be scientific, and we got what we were promised. Although Trivers did pause to answer a few questions, anyone with the illusion that this would be a “discussion” had that idea squashed within the first few minutes.

Trivers made a number of really valid criticisms of the paper, although these were not the majority of his comments. He rightly pointed out how the paper arrogantly ignores the work of many ethologists and behavioral ecologists who have dedicated their lives to testing various explanations of altruistic behavior, many of whom have made great discoveries using kin selection as their guiding principle. In a related point, he also appropriately criticized the authors for suggesting that explaining eusociality was in some way the central pursuit of all sociobiologists. He railed against the final paragraph, which makes the claim that the arguments of the paper can be applied to humans.

A central focus of the NTW paper is a criticism of the haplodiploid hypothesis as a means of explaining the origin of eusociality. Watching Trivers tackle this one was fascinating. On the one hand, he made the valid point that while the haplodiploid hypothesis is an extension of inclusive fitness theory, it certainly does not define the value of kin selection. If it fails to explain eusociality, that does not invalidate the use of inclusive fitness theory to explain other phenomena. But then he launched into a pretty generic discussion of how haplodiploidy leads to odd patterns of relatedness between male and female offspring of the foundress queen. Perhaps this was meant to be a reminder to those in the audience who might be unfamiliar with the nature of haplodiploid relatedness, but it came off as a big tangent to the main point made by NTW: that haplodiploidy fails to be predictive of eusociality. Trivers went on to discuss some valid parallels between the behavioral patterns found in some eusocial insects (such as the 3:1 skewing of offspring sex ratios by female workers) and the predictions of kin selection theory.

But here is the problem: NTW are not talking about what happens once the species becomes eusocial, they are talking about how the species becomes eusocial. This is a critical difference that seems to be missed by many. In his criticism of the paper, Jerry Coyne appears to be making the same mistake when he quotes a recent PNAS paper (Strassmann and Queller 2007):

Although I am not sure that Strassman and Queller meant it to serve this function, this quote actually does a wonderful job of distinguishing what kin selection can and cannot claim to have explained about eusociality. Their metaphor of the eusocial colony as a clock is perfect, as it explains the two levels of selection that are likely acting on the system. There is selection acting on the whole colony, rewarding those colonies that operate as the most efficient clock. And there is also selection acting on individuals, which may cause them to throw sand into the clockworks. These two forces are in opposition, and if enough sand is thrown into the gears the whole clock stops working. If you buy the argument of NTW, group selection is required to select for the clocklike design, as it is the only force that opposes the frictional, destructive forces of individual interest. And ironically, it is the very success of the whole colony that later fosters high relatedness, leading to the kinds of conflicts between the queen and her daughters that have been well-characterized in the literature on eusocial species.

You can tell when people are wholly committed to the idea of gene-level selection, because they insist that all arguments rest on the tenet that the gene is the sole unit of selection. Trivers showed himself to be of this ilk, emphasizing that relatedness is not a measure of the fraction of genes shared by relatives but instead should be considered as the “conditional probability” of a particular gene being shared by a relative. Strictly speaking this is a correct interpretation of Hamilton, which means that it is “wrong” to be thinking about the fraction of genes that an altruistic (donor) individual may share with a relative (recipient) individual (for a good recent review of this “mistake” being made in related fields, see Park 2007). But the only thing “wrong” about thinking of relatedness as a measure of the fraction of genes I share with relatives is that it does not take the gene-centered approach assumed by Hamilton and his many intellectual offspring. For kin selection to work I need to perform altruistic acts for other individuals — whole individuals — with whom I will share some fraction of my genes. For this reason the gene-centered approach may not make much sense. Ironically, the best argument I have read against thinking of genes as the unit of selection comes from Trivers’ own Ph.D. advisor, Ernst Mayr, who argues strongly in his book What Evolution Is (Mayr 2001) that the individual — not the gene — is the fundamental unit on which selection acts.

The issue of how to measure relatedness is tricky. In his presentation Trivers seemed happy enough to accept Hamilton’s original formulation as valid, which is to just look at the proportion of genes being transferred from parent to offspring and then infer the relatedness based on the probability that a particular gene that one individual possesses will be found in a relative. But that is not the way the NTW define relatedness. Their measure of relatedness is:

R=(Q-q)/(1-q)

Where Q is the average relatedness between donor and recipient individuals and q is the average relatedness of the population. The logic behind this formulation is that if the whole population is already highly related (in other words inbred, as is found in a lot of eusocial species), the degree to which my shared inheritance with relatives matters is decreased. This makes sense, as at some level of inbreeding my relatedness to any random individual is going to become nearly comparable to my relatedness to my closest relatives. This also means that the strength of kin selection should go down rather than up when colonies of organisms are inbred, which directly contradicts the claim that high levels of inbreeding in organisms like mole rats helps to explain their eusocial lifestyle.

So why isn’t this formulation of relatedness being used by folks like Trivers? To that question I do not have an answer. Which is the correct way of measuring relatedness? My instinct is that the “relative relatedness” used by NTW makes more sense, but I am not certain. Even if we agree that relatedness ought to be measured in a particular way, the actual act of measuring it in the field — even with the advent of cheap DNA sequencing — is not trivial. All of this does call into question the usefulness of kin selection as a primary means of explaining altruistic behavior. I am not denying that it works sometimes, but it should not be considered the default explanation.

Beyond this confusion over what relatedness is and on what unit selection acts, the ugliest part of Trivers’ monologue was his frequent use of character assassination to make his points. Perhaps the “informal” nature of this seminar on his home campus made him feel like he was among “family”, but in my eyes speaking in public is public speaking. He criticized both Nowak and E.O. Wilson for being overly-ambitious, relating their appointment at Harvard to an exceptional level of ego (Trivers ought to know, having earned his entire higher education from Harvard). Making sure to go after all the members of the small band that still even allows group selection to enter into their hypothesis list, Trivers also suggested that David Sloan Wilson “came out of the womb advocating group selection”. Too bad that he could not stay on topic, as I would have appreciated twice as many scientific arguments in exchange for not hearing the political and personal analysis.

One thing that interested me was the audience reaction. There were few questions, and few people seemed to want to foster a dialogue (I gave up after a few questions meant to open up discussion). Most in attendance were graduate students, and I understand the reluctance to question an icon of the field on his home turf. I guess what maybe upset me most about Trivers’ presentation is that I cannot imagine that it engendered in the audience much excitement about the field of evolutionary biology: through Trivers’ eyes, those trying to explain how cooperation evolves seem like a fractious, Balkanized, unproductive bunch of scientists. Perhaps this is true, but I doubt that focusing on this aspect of the field is going to do much to change it.

I think that one of the most unfortunate things about the NTW paper is that it distracts readers from its more valuable ideas because it overreaches on a couple of issues. The first issue is that of taxonomic extent, and the overreach appears in the final paragraph of paper. In this paragraph, the authors mention humans as a potentially eusocial species but then fail to elaborate. This comes off as a pretty irresponsible “oh and by the way all this could apply to humans but we don’t have time to explain why” sort of statement. I agree with Trivers’ suggestion that this statement would have been better left out, but I have the ability to ignore this poor decision and focus instead on the remainder of the paper. But tactically this overreach into the human realm was pretty bad, because a lot of people are going to trip up on this extension and allow it to obscure their view of material presented in the rest of the paper.

The second and much more egregious overreach is to disparage kin selection theory simply because it might fail to explain the evolution of eusociality. Going hand-in-hand with this exaggerated claim is the unwarranted elevation of the study of eusociality to the pinnacle of all social evolution studies. It simply is not the case that all areas in which social evolution is studied would be better off without kin selection theory. And as Trivers rightly pointed out, some aspects of eusociality are still best explained by applying the thinking underlying kin selection. Even if the origin of eusociality lies not in kin selection but in group selection, we need to acknowledge that different theoretical constructs will be useful to explain different evolutionary phenomena. Ironically, one of my favorite recent papers is “Five Rules for the Evolution of Cooperation” by Martin Nowak (2006), which includes kin selection as one of the five mechanisms by which cooperation can evolve. Now that he is convinced that kin selection is not the mechanism by which eusocial species evolve, is Nowak ready to shorten his list of mechanisms to four? What is very painful about all this is that it was not necessary to try to wholly supplant kin selection theory in order to write a paper that — as David Sloan Wilson has been quoted as saying — “knocks inclusive fitness off its perch”. I believe that kin selection as a mechanism is probably overused and more importantly applied in a sloppy manner, but this does not mean that it ought to be scrapped entirely. Ironically, it appears that some advocates of group selective processes are bent on doing the same thing to kin selection that was done to group selection in the 1960’s: throwing out the baby (a theory that makes sense in many contexts) with the bathwater (sloppy or unsubstantiated use of that theory).

Many who oppose the NTW paper have suggested that it should have been better reviewed in order to prevent it from being published. Ironically, I concur with the idea that it should have been better reviewed, even as I am not convinced that it should have been denied publication. Had the two overreaches outlined above been removed through a more thoughtful review process, the article might have generated more light and less heat. The valuable content provided in the article (much of which is buried in the supplementary material — another downside of publishing in journals like Science and Nature) is overshadowed by the inappropriately grandiose claims it makes.

Out there somewhere are two pools of potentially-guilty parties: 1) the referees for this paper who failed to point out some of its fundamental flaws; and/or 2) the editors who either ignored the flaws that were pointed out by referees or who failed to get referees adequate to the task. Add me to the list of those frustrated by the priorities that top journals like Nature use to decide on what to publish, although I guess something is better than nothing for those of us who just do not buy the “relatedness explains everything” argument.

I am curious to see whether the furor caused by NTW leads to any clarification of evolutionary theory in relation to eusocial species. But even if this never happens, I believe that the paper has already — even before formal responses have been published objecting to it — exposed some very disturbing things about the sociology of evolutionary biology. This publication has really angered a great number of people, and this anger has revealed the kind of tribal orthodoxy that exists in the mainstream of evolutionary biology. A great number of evolutionary biologists will become enraged whenever someone in the field espouses a hypothesis that contradicts the gene-centered approach to explaining evolutionary processes. Even in this case — where a provocational paper was published in a revered journal — the anger often seems excessive in relation to the affront. After all, if an idea is clearly dumb and can easily be dispatched by theoretical or empirical demonstration, why get upset about it? We need more publication of relevant results, not more rhetoric.

Over the summer I had the pleasure of hearing a really great talk by current Society for the Study of Evolution President H. Allen Orr. In his talk, he suggested that data is the ultimate arbiter of truth in science, and when we see scientists engaged in emotional arguments about competing theoretical visions, this probably results from the inability to bring data to bear on the controversies. Clearly in this situation we need to find ways of creating more robust tests of predictions made by opposing theories. Otherwise, we are left with only ugly and/or directionless rhetoric.

Beyond Robert Trivers’ unholy mixture of rhetoric and scientific argument described above, the web offers some examples of other prominent evolutionary biologists departing from the scientific issues at hand in order to level personal attacks on the authors of this paper. Most egregious of these are comments made by Jerry Coyne, the publisher of the book and blog called “Why Evolution is True”. He compares scientists in favor of group selection to creationists in a recent blog comment, and also allows others to attack E.O. Wilson by suggesting that his involvement in this paper stems from “senescence”. These sort of comments demonstrate that scientific issues have become emotional and political, and while I do not deny the humanity of scientists (being human means being both emotional and political), we do nothing to move forward scientific understanding by engaging in name-calling. Want to comment on the issues on this blog? Go ahead, but expect me to edit out anything that amounts to character assassination.

To an anthropologist visiting this strange island tribe, we must look like a funny culture:

The sad part about this kind of knee-jerk rejection of dissent and diversity of perspective amongst evolutionary biologists is that it opens us up for criticism from a myriad of cultural actors who would like to contend that we are “just another culture”. What makes science unique is that it is not a wholly constructed culture: it has a few cultural ideals — the foremost of which is bringing objective data to bear on hypothesized answers to critical questions — which subject our own culture to being shaped by the outside world and its realities. We ought to embrace a diversity of hypothesized answers, as only through the generation of this diversity can our culture be advanced by the force of natural selection exerted by scientific inquiry.

Although I am focusing on the negative because it is so prominent in this case, there are also critiques of the NTW paper provided by Richard Dawkins, David Sloan Wilson, John Hawks, Ben Allen, and “Johnny”, all of which do not resort to character assassination in order to make their points. These are complemented by news articles published in Nature, Wired, and The New York Times. I do not agree with all the ideas expressed in all of these resources, but they are worth checking out for more perspective.