Finally, let us say that, in general, the surviving (and ultimately speciating) deme lies at or near the smaller-bodied end of the random distribution (around the parental mean) of average body size. Let us also posit that the smaller average body size of new species arises as a consequence of a deme-level property conveying differential success in interdemic selection among the ten peripheral isolates initially spun off from each species. An obvious (and not implausible) reason might be found in a strong correlation between small bodies and larger N in any population (the “more ants than elephants” principle, albeit in a more restricted range). The surviving deme might owe its success to generally larger population size in a tough peripheral environment.
The cladal trend to smaller body size among species would then arise by a drive of directional speciation (new species biased to originate with smaller-bodied organisms than those of their ancestors, in a situation where no species selection can occur). The cause of the trend, in this hypothetical case, will be interdemic selection — for the ten peripheral isolates arise as demes of the parental species. Selection among these demes favors those with smaller average body size, based on correlation with the causally controlling deme-level property of larger population size. This deme-level property confers an irreducible fitness upon demes in their interaction with the environment. (In an extreme, albeit improbable, case, interdemic selection based on larger population size could even outweigh negative organismic selection against small bodies.) Again, a drive at the species level arises by selection among lower-level parts, in this case demes rather than organisms.
Finally, an irreducible species-level character may cause a drive at the species level. Suppose that each species spins off only one peripherally isolated population and that, invariably, the parental population dies while the peripheral isolate becomes a new species. Suppose that the single peripheral deme, generated by each species, generally features organisms with a smaller average body size than the organisms of the parental population. Suppose that this directional bias arises as a result of a species-level trait in the parental population. Perhaps, for example, the social structuring or territorial system of the parental population preferentially excludes smaller organisms of both sexes, and that these smaller organisms therefore tend to migrate to the species border, where they aggregate to form the isolated population that will generate a new species. (Again, the case merely requires conceivability for purposes of illustration, not plausibility.) In this situation, no selection can occur at the species level because each parental species produces one daughter species and then dies. The cladal trend to species with smaller average body size arises by the driving process of directional speciation — and the cause lies in a species-level trait of the parental population. As stated above, we here encounter a case of irreducible macroevolution not based on species selection. Examples of this kind illustrate that the domain of independent [Page 728] macroevolutionary theory extends well beyond the phenomenology of Darwinian analogs in species selection.
As an additional argument for the importance of directional speciation as a driving force in evolution — and as an example of interesting complexity engendered by the hierarchical model (and of differences in the character of explanation between this hierarchical reformulation, and the traditional one-level world of Darwinian evolution) — note what often happens when causes at one level correlate with emergent properties involved in causes at a higher level; for we then encounter the fascinating situation of disparate theoretical meanings for inexorably linked phenomena at two levels. (We have already discussed one common example in the causation of higher-level drives by lower-level selection.) Another important example, potentially encompassing one of the dominant phenomena of macroevolution, translates the results of ordinary selection at the organismal level into strong constraints acting as causes of directional speciation at the species level. In this sense, when considered at the appropriate higher level, macroevolutionary pattern results much more from immediate constraint, and less from the traditional selectionist mode, than we have generally been willing to allow — thus suggesting another potentially important reform and expansion of Darwinian thinking (see Chapter 10 for a fuller discussion).
Consider two cases of cladal trends produced by the driving cause of directional speciation. Figure 8-6 depicts the common pattern in both examples. At a starting point, the clade contains two kinds of species in equal numbers — those bearing trait A, and those bearing trait B. Every reproducing species generates two daughters and no variation exists for differences in species birth rates among those species that have offspring — so no species selection can occur. Evolution proceeds rapidly by directional speciation because A-Species can only produce A-Daughters, while B-Species produce 50 percent
8-6. A cladal trend produced entirely by directional speciation with no species selection. A species can only produce A daughters, while B species produce 50% A daughters and 50% B daughters. Under these conditions of strongly directional speciation, a powerful trend towards A leading to quick disappearance of B from the clade, will arise, even under a regime of random mortality among species.
[Page 729]
A-Daughters and 50 percent B-Daughters. (If we posit random mortality of a given percentage of species before they split into their daughters, as in Fig. 8-6, then B-Species will eventually be entirely eliminated, and A-Species will become fixed in the clade.)
A first kind of example would not disturb the tranquility of any committed adaptationist, for a functionalist theme translates well across the levels. Suppose that Cope's Rule were true in the classical sense — it is not, by the way (Stanley, 1973; Jablonski, 1987, 1997; McShea, 1994; Gould, 1988b, 1997b, and pp. 902–905 of this book) — and that organismal selection always favored size increase because big organisms prevail in competition. A-Species are large and B-Species are small; A's only give rise to other A's, while B's give rise either to A's (given the pervasive advantage of increasing size), or to B's at equal frequency (for small size may still be favored in some habitats of the clade). The strong Cope's-Rule trend in the clade occurs by directional speciation. The adaptationist theme prevails at both levels. Average organisms in the clade become larger because bigger is better; and species increase in average body size because their parts (organisms) do better at larger size. (No species-level trait exists in regulating this trend, and the entire phenomenon arises by conventional organismal selection based on advantages of increased body size.)
But a second kind of example — undoubtedly quite common in evolution — would perturb a strict adaptationist by translating selection at the organismic level to regulation of the cladal trend by constraint. Suppose now — and such an explanation has been urged as an alternative to species selection for the increase of nonplanktotrophic species within Tertiary clades of gastropods (Strathmann, 1978, 1988) — that a molluscan clade begins with an equal number of species of nonplanktotrophs (A-Species) and plankotrophs (B-Species). Planktotrophic larvae stay aloft through the motion of complex ciliary bands that beat in concert. Selection pressures for nonplanktotrophy lead to loss of these bands, and consequent benthic development of a maternally protected brood. Plankotrophs can always, in principle, convert to nonplanktotrophy because the bands can be lost; but the transition cannot proceed in the other direction because ciliary bands can't be reconstituted once they have disappeared in evolution (see Gould, 1970b, on the proper meaning of Dollo's Law of irreversibility in evolution).
The origin of each species may be governed entirely by the conventional route of adaptation based on natural selection of organisms. But a structural limitation in possible directions of change produces the cladal trend by directional speciation towards increasing frequency of nonplanktotrophic species — for a planktotrophic parent species can generate either planktotrophic or nonplanktotrophic daughters, while a nonplanktotrophic parent can only produce nonplanktotrophic daughters. The numerical situ
ation corresponds exactly with Figure 8-6 and the previous example based on Cope's Rule (with A-Species now read as nonplanktotrophs, and B-Species as planktotrophs), but the explanation at the cladal level differs crucially — for the trend arises by structural constraint upon possible directions of change, not from [Page 730] any general or global advantage for nonplanktotrophic organisms. (In fact, planktotrophic species might hold a small advantage in species selection for longevity, and the trend to nonplanktotrophy might still arise by directional speciation under this potent constraint.)
I strongly suspect that trends driven by structural constraints within large systems, and not by adaptational advantages to organisms, pervade evolution, but have been missed because we focus on means or extremes in a distribution and not on the full range of variation as a more telling “reality” (see Gould, 1996a, for an entire book on this subject, written for popular readers; and Gould, 1988b, for a technical account). The vaunted trend to increasing complexity in the history of life, for example, only records the small and extending tail of an increasingly right-skewed distribution through time — but with a strong and persistent bacterial mode that has never altered during life's entire 3.5 billion year history, leaving this planet now, as always, in the Age of Bacteria (see pp. 897–901 for a further development of this example). This extending right tail may record little more than the constraint of life's origin right next to the lower bound of preservable complexity in the fossil record. Only one direction — towards greater complexity — remained open to “invasion,” and a small number of species dribble in that direction through time, thus extending the right tail of the skewed distribution.* But no evidence now exists to support an argument that higher complexity should be construed as a “good thing in general” (in adaptive terms, or otherwise), either at the organismal or species level. In fact, the few studies based on patterns of speciation in clades where founding members lie far from any upper or lower structural boundary, and therefore impose no constraint upon either decreasing or increasing complexity, show no trend at all towards increasing complexity. Approximately equal numbers of species arise with less complex and with more complex phenotypes than their ancestor (see McShea, 1993, [Page 731] on mammalian vertebral columns; McShea, Hall, Grimsson and Gingerich, 1995, on mammalian teeth; and Boyajian and Lutz, 1992, on ammonite sutures).
Species selection, Wright’s Rule, and the power of interaction
with directional speciation
I have long regarded species selection as the most challenging and interesting of macroevolutionary phenomena, and the most promising centerpiece for macroevolutionary theory. While I continue to espouse this view, my rethinking for this chapter has led me to appreciate the significant power of two other species-level processes: drives of directional speciation as just discussed (see also Gould, 1982c), and species drift, the higher-level analog of genetic drift. I would now argue that the interaction of these three processes sets the distinctive character of macroevolution.
As for natural selection at the organismic level, the two major modes of species selection operate by differential rates of generating daughter species (the analog of birth biases in natural selection) and differential geological longevity before extinction (the analog of death biases in natural selection). At the species level, however, the difference between these two modes does not rest upon the same basis that distinguishes their analogs at the organismic level.
At the organismal level, natural selection by birth bias works mainly upon such “internal” traits as reproductive rate and brood size, and often doesn't increase adaptation in the conventional sense of phenotypic molding to better biomechanical design for local environments. For example, an organism gains a large selective advantage merely by breeding a bit earlier, though nothing else about the phenotype need alter (Gould and Lewontin, 1979, referred to this mode as “selection without adaptation”). But natural selection by death bias among organisms usually yields phenotypic adaptation for better fit to the ambient environment.
At the species level, however, our main concern moves to an interesting difference in causal locus. Most cases of selection by differential speciation operate by the interaction of an irreducible species-level character — some feature of population structure — with the environment, and therefore represent genuine species selection. After all, and as stated before, organisms don't speciate; only populations do. But for selection by differential extinction, a higher frequency of cases can probably be explained as the simple summation of organismal deaths, and may therefore be causally rendered at this conventional lower level — for both organisms and species die. Thus, students of species selection have rightly focussed on differential speciation as their most promising category (see Gilinsky, 1981, for both theoretical arguments and empirical examples).
However, the most interesting of all differences between organismal and species selection may lie not in the phenomena themselves, but rather in the character of their interaction with the two other primary modes of evolutionary change: drives, and drift (I shall discuss drift in the next section). Our [Page 732] sense of the commanding potency of organismal selection rests upon the conformity of Mendelian genetics to one of the cardinal prerequisites of Darwinian systems (see Chapter 2): that the variation serving as raw material for natural selection be “random” (with an operational meaning of “undirected towards adaptive states,” not “equally likely in all directions”) — so that selection, rather than biases inherent in variation, can become the “creative” force in evolutionary change (see p. 144 for further discussion in a related context). This crucial condition can be validated at the organismic level — not because mutations (and other sources of genetic variation) are truly random in the mathematical sense, but because mutation represents a process so different from natural selection, and operating on material (the structure of DNA) so disparate from the bodies of organisms (integrated tissues and organs), that we cannot postulate a reason why favored directions of mutation should correspond in any way to the needs of organisms.
But no comparable argument exists for any a priori expectation that the analogous variation (among species within a clade) made available for species selection should also be random with respect to the direction of a trend. Species do not discourage drives among their parts (organisms), while organisms usually do suppress directional variation at lower levels (because the proliferative “interests” of individual genes and cell lineages generally run counter to the adaptive needs of organisms). Moreover, the adaptive features of organisms often confer benefits upon their species as well — as when species live longer because their well-designed organisms prevail in competition. Therefore, we cannot defend an a priori basis for asserting randomness in the variation that serves as raw material for species selection.
This situation creates both a problem and a challenge for the analog of Darwinism at the species level — for maximal efficiency of species selection does demand undirected variability, and by the same classical argument originally devised for the organismic level. The randomness of species-level variation with respect to the direction of a trend therefore becomes a matter for empirical testing, rather than a claim predictably flowing from the nature of materials and processes. Such a test should also receive high priority for anyone interested in discovering the frequency and strength of species selection in the explanation of evolutionary trends.
For these reasons, Gould and El dredge (1977) formulated such a test under the name of “Wright's Rule.” We took our cue from a prescient statement by Sewall Wright (1967) that the direction of speciation might be random with respect to the origin of higher taxa, just as we consider mutation to be random relative to the direction of natural selection. Wright's Rule, in our formulation, therefore asserts either that drives of directional speciation do not exist at all in a given situation (the strong version), or at least that any existing directional bias not occur along the vector o
f an established trend (a weaker version, but fully adequate for assertions of species selection). If Wright's Rule holds, then trends must be attributed to differential proliferation of certain kinds of species (by selection or drift), and not to any drives from within based on directional variation arising from [Page 733] lower-level processes.
Wright's Rule represents a strong test for putative species selection, but I now realize that its failure does not eliminate species selection from consideration. When Wright's Rule holds, a trend must be attributed to species sorting, for no directional component exists at the lower level of variation among units of sorting. But if Wright's Rule fails in any particular case, then species selection cannot forge the trend exclusively — although species selection may still operate as one contribution in a hierarchical system. A speciational drive may act synergistically with species selection to intensify a trend. (Since drives tend to be more potent than selection, a powerful drive, with strong violation of Wright's Rule, will probably relegate species selection to an insignificant role. But small departures from Wright's Rule permit a substantial intensification of the trend by species selection. In any case, and in situations of unusually complete paleontological data, we should be able to measure the relative strengths of drive and sorting when the two modes act synergistically.)
Wright's Rule has been tested in some cases, but not often enough — and the subject remains ripe for future research, including several Ph.D. theses! Gould and Eldredge (1977) found Wright's Rule validated for Gingerich's data on early Tertiary Hyopsodus. MacFadden (1986) failed to confirm Wright's Rule in the evolution of horses, where a directional bias exists for descendant species to arise at larger body sizes than their ancestors. Arnold, Kelly, and Parker (1995) validated Wright's Rule for a remarkably complete data set of 342 ancestral-descendant pairs in Cenozoic planktonic foraminifera. An equal number of species arose at larger and at smaller sizes than ancestors; see Figure 8-7. In a pioneering study, notable for completeness and density of data (and a consequent capacity to distinguish among all the various modes of evolutionary change), Wagner (1996) documented three general and speciational trends in the evolution of gastropods during the lower Paleozoic (Cambrian through Silurian): towards higher spires, more inclined apertures and narrower sinuses. For 276 ancestor-descendant pairs over the entire clade, Wagner confirmed Wright's Rule for spire height and inclination, where as many species differed from ancestors in a direction away from the general trend, as along the ultimately favored route. But data for sinus width, where a statistically significant bias exists for speciation in the direction of
The Structure of Evolutionary Theory Page 117