Finally, we may seal the case by citing Grantham's important argument (1995, p. 301) that “species selection does not require emergent traits because higher-level selection acting on aggregate traits can oppose lower-level selection.” Vrba herself has argued (1989, p. 80) that “the acid test of a higher level selection process is whether it can in principle oppose selection at the next lower level.” Surely such an opposition can arise “in principle” (and probably in actuality) in this case — for planktotrophy could be positively selected at the organismic level, but may, through its strong effect on population structure, and the resulting consequences for rates of speciation, enjoin negative selection at the species level.
To summarize, we all agree that an independent theory of macroevolution must identify higher-level causal processes that are not reducible to (or simple effects of) causes operating at conventional lower levels of gene and organism. [Page 662] This premise defines the theoretical salience of the debate about species selection — for if such a process exists, and can also be validated as both common in evolution and irreducible in principle, then macroevolutionary theory has been achieved. For this reason, evolutionary biologists, who usually eschew academic philosophy (as the mildly philistinistic culture of science generally dictates), have joined in such classical philosophical debates as the meaning of reduction and emergence.
Vrba's criterion of emergent characters establishes an obvious case for irreducibility because the trait that causes species selection can claim neither existence nor representation at the conventional organismic level. Grantham writes (1995, p. 308): “When a component of species-level fitness is correlated with an emergent trait, this correlation cannot be reduced because the trait cannot be represented at the lower level.” But Lloyd's broader criterion of emergent fitness also establishes irreducibility, even if the trait involved in the correlation between trait and fitness is reducible under the effect hypothesis. In Lloyd's case, the fitness is irreducible (as shown practically in the previous example of gastropod lineages, where higher-level fitness based on speciation rate opposes lower-level fitness based on the same trait of larval adaptation). The technical point may be summarized in the following way: selection is defined by the correlation between a species-level trait and species-level fitness; therefore, the irreducibility of either component of the correlation establishes irreducibility for the selection process. Grantham notes (1995, p. 308): “Emergent traits are not, however, necessary for species selection. If an aggregate trait affects a component of species-level fitness (e.g. rate of speciation) and this component of fitness is irreducible, then the trait-fitness correlation will be irreducible.”
Vrba's emergent character approach embodies one great strength, but two disarming weaknesses. This criterion does identify the most irrefutable, and in many ways the most interesting, subset of cases for species selection — examples based on genuine species adaptations (for an emergent character that evolved as a consequence of its value in fitness is, ipso facto, an adaptation); whereas nonemergent characters that contribute to species fitness via the effect hypothesis are exaptations (Gould and Vrba, 1982; Gould and Lloyd, 1999), at the species level (and adaptations at the lower level of their origin).
But the emergent character criterion suffers from two problems that would render the theory of species selection, if framed exclusively in its light, eternally contentious and, perhaps, relatively unimportant as well. First, by including only the “hardest-line” cases within the concept, we may be unduly limiting species selection to an unfairly small compass. (For example, and as an analogy, we wouldn't want to restrict the concept of “adaptation” only to the small subset of true biomechanical optima — for most adaptations only hold the status of “better than,” not ne plus ultra). Second, emergence can often be extremely difficult to document for characters — so, in practice, the concept may be untestable in most circumstances. To differentiate between a truly emergent species character and an effect of a lower-level character, one often needs a great density of narrative information about the actual history [Page 663] of the lineage in question — information only rarely available in the fossil record, not to mention our spotty archives for living species.
By contrast, the emergent fitness approach enjoys the great virtue of fully general applicability. For, when one only has to consider current circumstances (the trait-fitness correlation), and need not reconstruct prior history (as the designation of emergence for a species-level character so often requires), then we can study any present reality that offers enough information for a resolution. We certainly use this most broadly applicable, nonhistorical approach in traditional studies of natural selection at the organismic level — that is, we identify current selective value whether the feature conferring differential reproductive success arose as an adaptation for its current contribution to fitness, or got coopted for its present role from some other origin or utility. (In other words, both preadaptations and spandrels — features that arose as adaptations for something else, or for no adaptive purpose at all — can function just as well in a regime of current selection as true adaptations forged by the current regime.) The historical origin of characters, and their later shifts in utility, constitute a central and fascinating question in evolutionary theory — and provide a main theme for Chapter 11 of this book. But we define the process of selection ahistorically — as differential reproductive success based on current interaction between traits of evolutionary individuals and their environments — that is, the concept of selection remains agnostic with respect to the historical origin of the traits involved.
The emergent fitness approach presents four favorable features that establish species selection as a central, fully operational, and vitally important subject in evolutionary biology — thereby validating both the necessity and the distinctness of macroevolutionary theory.
1. Rather than depending upon a documentation of prior history in the narrative mode (often untestable for lack of information), we move to a fully general mathematical model that can, in principle, identify components of higher-level selection in any case where we can obtain sufficient data on the current operation of a selection process. Arnold and Fristrup (1982) expanded Price's (1970, 1972) covariance formulae to encompass a set of nested levels, and devised an approach closely allied to analysis of covariance, considering selection at one level as a “treatment effect” upon selection at an adjacent level. Damuth and Heisler (1988) developed a similar method, also based on covariances (or regression of fitness values on characters); this procedure has been expanded by Lloyd (1988; Lloyd and Gould, 1993). As Lloyd and Gould (1993, p. 596) describe the method: “This is done by describing interactors at the lower level first. If a higher-level interactor exists, the higher-level correlation of fitness and trait will appear as a residual fitness contribution at the lower level; we must then go to the higher level in order to represent the correlation between higher-level trait and higher-level fitness.”
Lest this method seem to fall into the very reductionistic trap that species selection strives to overcome — because we begin at the lowest level and only move higher if we find a residual fitness — I point out that we use this procedure only as a convenient and operational research method, and decidedly [Page 664] not with the reductionistic hope that no residuals will appear, and that the lowest level will therefore suffice for a full explanation. We may be stuck with the technical term “residual” as a common statistical usage in such circumstances — but there is nothing conceptually residual about higher-level selection. Selection at lower levels cannot be designated as more true or basic, with higher levels then superadded if necessary. The statistical “residual” of our procedure exists as a separate but equal natural reality in our fascinating world of hierarchical selection.
2. The emergent fitness approach establishes a large and general realm for the operation of species selection. Any evolutionary trend that must
be described, at least in part, as a result of species sorting automatically becomes subject to the analytical apparatus here proposed, and therefore a candidate for identification of species selection. (And I can hardly imagine that any important trend unfolds without a major — I would say almost always predominant (see Chapter 9) — component of species sorting, for extensive anagenesis rarely occurs in single lineages, and none can persist very long without branching in any case.)
3. The emergent fitness approach allows us to use a single, familiar, and minimalist definition of selection in the same manner at each level — differential proliferation of evolutionary individuals based on interactions of their traits with the environment. We therefore achieve a unified theory of selection at all scales of nature. The availability of a fully operational analytical apparatus, connected with this definition, greatly enhances the scientific utility of emergent fitness as a definition of species selection.
4. As an admittedly more subjective and personal point, the emergent fitness approach allows us to encompass under the rubric of species selection several attributes of populations that many participants in this debate have intuitively wished to include within the causal compass of species acting as evolutionary individuals, but which the more restrictive emergent character approach rules out. Many of us have felt that two distinct kinds of species properties should figure in species selection because, for different reasons, such features cannot function at the lower and traditional level of organismic selection. In the first category, emergent characters of species obviously can't operate at the organismic level because they don't exist for organisms. These features clearly serve as criteria of species selection in either the emergent character or the emergent fitness approach.
In a second category, some important aggregate characters of species can't function in selection at the organismic level, not because they have no expression at this lower level (for they clearly exist as organismic properties, at least in the form of traits that aggregate additively to a different expression at the species level), but because such properties do not vary among organisms, and therefore supply no raw material for selection's necessary fuel. I speak here of a common phenomenon recognized by different jargons in various sub-disciplines of our field — autapomorphies for cladists, or invariant Bauplan characters for structuralists. Suppose that each species in a clade has evolved a unique state of a homologous character — and that, within each species, [Page 665] all organisms develop the same state of the character, without meaningful variation. In this situation, all variation for the homologous character occurs among species, and none at all within species. If a trend now develops within the clade when some species live and proliferate because they possess their unique state of the character, while others die because their equally distinct and unvarying state has become maladaptive in a changed environment, should we call such a result species selection — for each species manifests a single attribute, and all variation occurs among species? Interestingly, de Vries originally coined the term species selection (see pages 448–451) for precisely this situation, where no relevant variation exists within species, and all variation occurs among species.
To summarize: in the first situation, the character doesn't exist at the organismal level, and each species develops only one state of the (emergent) character because the character belongs to the species as a whole. Therefore, selection for this character can only occur among species. In the second situation, the character doesn't vary at the organismal level, and each species in a clade has evolved a unique and different state of the character. Again, selection can only occur among species. In either situation, each species manifests one different and unvarying state of a feature that cannot operate in organismic selection — so selection for this feature can only occur among species.
The emergent status of the character leads us to designate the first situation as species selection without any ambiguity or alternative. But we balk at designating the second situation as species selection because the relevant species-level character (lack of variation) represents an aggregate, not an emergent, feature. The emergent fitness criterion rescues us from this dilemma, and forges an intuitive union between the two situations by designating both as species selection. Lack of variation — the aggregate species character — interacts with the environment to influence differential rates of proliferation among species. This character imparts an emergent fitness to the species, and therefore becomes an agent of species selection. (After all, the species doesn't die because organism A, or B, or C, possesses a trait that has become maladaptive; the species dies because none of its parts (organisms) can develop any other form of the trait — and this lack of variation characterizes the species, not any of its individual organisms.)
I believe that such “species selection on variability” — the title that Lloyd and I gave to our 1993 paper — will prove to be a potent style of selection at this level. (When I was struggling with the issue of whether such an aggregate character as variability could count as a property of species, I asked Egbert Leigh, a brilliant evolutionist and the leading late 20th century disciple of R. A. Fisher, whether he thought that variability could operate as a character in species selection — and he replied: “if variability isn't clearly a character of a species, then I don't know what is.”)
To cite just one hypothetical example that I have often used to illustrate this issue and to argue for species selection on variability: Suppose that a wondrously optimal fish, a marvel of hydrodynamic perfection, lives in a pond. This species has been honed by millennia of conventional Darwinian [Page 666] selection, based on fierce competition, to this optimal organismic state. The gills work in an exemplary fashion, but do not vary among individual organisms for any option other than breathing in well-aerated, flowing water. Another species of fish — the middling species — ekes out a marginal existence in the same pond. The gills don't work as well, but their structure varies greatly among organisms. In particular, a few members of the species can breathe in quite stagnant and muddy waters.
Organismic selection favors the optimal fish, a proud creature that has lorded it over all brethren, especially the middling fish, for ages untold. But now the pond dries up, and only a few shallow, muddy pools remain. The optimal fish becomes extinct. The middling species persists because a few of its members can survive in the muddy residua. (Next decade, the deep, well-aerated waters may return, but the optimal fish no longer exists to reestablish its domination.)
Can we explain the persistence of the middling species, and the death of the optimal form, only by organismic selection? I don't think so. The middling species survives, in large part, as a result of the greater variability that allowed some members to hunker down in the muddy pools. (We may even argue that the optimal fish always prevailed against most members of the middling species, even at the worst time, so that most middlings died quickly when the pond dried, while the optimals hung on longer, but eventually succumbed.) The middling species survived qua species because the gills varied among its parts (organisms), not because all its members gained advantage when the environment changed. (For most middling organisms continued to fare worse than the optimal fishes.) We may represent this story at the organismal level by discussing the gills of the few middling fishes that carried the species through the crisis. But the middling species prevailed by species selection on variability — for this greater variability imparted an emergent fitness to the interaction of the species with the changed environment.
Species selection on variability also possesses the salutary property of uniting the two major themes of this book, the concepts that I regard as the most important revisions now needed to mend and strengthen the two main legs of the essential Darwinian tripod: the hierarchical theory of natural selection as a vibrant expansion of Darwin's focus on the organismal level, and the centrality of constraint as a channeler of evolutionary direction in concert with natural selection (wh
ich can no longer maintain the exclusivity that strict Darwinians wished to impart). An important component for explaining the patterning of life's history lies in limitations and channels imposed and retained by developmental architecture — and these constraints do much of their work at higher levels, in large part by influencing “species selection on variability.”
I close this discussion with three points that validate the status of species selection as an irreducible macroevolutionary force, and place the proposed criteria of emergent characters and emergent fitnesses under a common rubric. [Page 667]
THE FALLACY OF “NECKER CUBING” The philosophical doctrine of conventionalism, as expressed by Dawkins (1982) in his Necker Cube metaphor (see pages 640–641), presents an important challenge to claims for an independent macroevolutionary theory based on higher-level selection. For if all cases of higher-level selection, however cogently defended, represent only one legitimate way to describe a process that can always be causally expressed in terms of selection at conventional lower levels as well, then why bother (except for the fun of it, or for the psychological insight thus provided) with the alternative higher level, when the traditional Darwinian locus invariably works just as well?
I do not doubt that some evolutionary events can be alternatively expressed (and I shall mention one category under my second point below), but Necker cubing will not apply to genuine cases of irreducible species selection because the nature of the world (not the conventions of our language) regulates the locus of causality. Two reasons debar the Necker cube for true cases of species selection. First, for Vrba's “hardest” category of species selection based on emergent characters, no expression at conventional lower levels can be formulated because the relevant species character does not exist at the usual Darwinian locus of organisms. Second, for Lloyd's broader category of species selection based on the emergent fitness associated with aggregate species characters, the “Necker cubers” commit a basic error in logic. They correctly note that the aggregate character can be represented at the organismic level — so they invoke the conventionalism of alternative and equally valid expression. But, as discussed on page 659, the species-level fitness imparted by the aggregate character, not the character itself, denotes the irreducible feature that defines species selection on this criterion.
The Structure of Evolutionary Theory Page 106