Barraco no PNAS: Vocês são darwinistas heréticos! Nós não somos darwinistas heréticos!

1. Vocês são darwinistas heréticos!

Morphogenera, monophyly, and macroevolution

Jablonski and Finarelli (1) suggest that morphogenera, even when they are nonmonophyletic, serve as good representatives for large-scale evolutionary studies. We feel there are two issues that warrant further discussion. First, the test used to evaluate the effect of using nonmonophyletic groups for macroevolutionary studies was not conservative and thus does not provide strong evidence about the impact of nonmonophyly on evolutionary studies. Their test examined whether the median trait value for the species in a nonmonophyletic genus correlated with the median trait value for those same species plus the additional species needed to make the set monophyletic. Correlation of the median value of a set of measurements with the median value of a superset of those measurements is to be expected (as is brief ly mentioned in the methods section of ref. 1). To demonstrate this, we correlated the median values of ln body mass from random sets of 3, 7, and 15 mammal species and 0–30 additional species for 45 simulated genera, using data from ref. 2 (Fig. 1A). Even when the median of 3 species is correlated with those 3 plus 15 additional species, the correlation is significant. Moreover, slightly better than random assignment of species to genera increases the expected correlation. To show this, we compare the results of figure 3A from ref. 1 with simulations where 45 genera of 3 species each were created from species chosen randomly from (i) the same family (Fig. 1B) and (ii) the same order (Fig. 1C) (data from refs. 2 and 3). Correlations between morphogenus median and the smallest clade containing the morphogenus species were significant (P
0.0001). Thus, the observed correlations presented in ref. 1 are not surprising. Second, ref. 1 also appears agnostic as to whether the morphogenera should be used together with a phylogeny when conducting large-scale evolutionary analyses or whether the morphogenera would themselves be independent data points (as in some of the papers it cites). As discussed in ref. 4, accounting for the bias of shared ancestry in analyses of biological data is essential. We illustrate the importance of using phylogenies by examining the correlation of body size and gestation in mammals (data from refs. 2 and 3). First, we conducted a generalized least-squares analysis (taking into account phylogeny) of gestation on body size by using specieslevel data [slope, 0.083 (0.07–0.096, 95% confidence interval); intercept, 0.561 (0.306–0.815, 95% confidence interval); Fig. 1D]. Then, we examined the same relationship by using the phylogenetically corrected generic-level data (slope, 0.119; intercept, 0.282) and phylogenetically uncorrected generic level data (slope, 0.244; intercept, 0.851). Both the phylogenetically corrected and uncorrected generic-level analyses result in intercepts and slopes outside the 95% confidence interval of the species level dataset. This demonstrates that generic-level data may mislead large-scale evolutionary studies, and so using morphogenera rather than species should be done with caution. More importantly, the worse results achieved by the generic-level analysis when not accounting for phylogeny (a slope 3 times greater than the slope estimated from all the data) suggest that, even if morphogenera are used, correction based on phylogeny must be considered.

Stephen A. Smith and Brian C. O’Meara1
National Evolutionary Synthesis Center, 2024 West Main Street
A200, Durham, NC 27705

PDF gratuito do artigo aqui.

+++++

2. Nós não somos darwinistas heréticos!

Reply to Smith and O’Meara:

The utility of morphogenera

We do not disagree with Smith and O’Meara’s points (1), which do not affect our results (2). P values must be viewed with caution when the x variable is a subset of the y variable, and we could have noted this point more prominently in our paper. From a purely pragmatic standpoint, however, most analyses employing morphogenera (e.g., analyzing fossil taxa) will lack prior knowledge of their phyletic status with respect to molecular cladograms. Therefore, the key consideration is the amount of variance explained by genera of unknown status compared to known monophyletic taxa. Furthermore, their analysis of mammalian body masses (1) suggests we should exclusively expect significant correlations, but when we re-create their analyses with the same body-mass data, varying the number of additional species required to create monophyletic groups across a realistic range, a substantial number of replicates yield insignificant P values (Table 1). Thus, significant correlations are not a foregone conclusion. This approach can also assess morphogenera as species level proxies, although this issue was quite secondary to our main focus. Selecting body masses randomly from all Mammalia, we created 23 paraphyletic genera (as in ref. 2), assigning 13 congeneric species and a single outgroup responsible for the paraphyly (as in Cercopithecus), and 22 polyphyletic genera (2) with 16 congeners and 2 outgroups (as in Gazella). This configuration of congeners and outgroups represents a conservative assessment because many morphogenera had more outgroup taxa and therefore a greater ability to alter median mass. The reported Spearman rank correlation for the mammalian body masses (2) was higher than all 10,000 replicates (Fig. 1). With respect to phylogenetic correction at the genus level, we explicitly cautioned that the robustness of morphogenera as species proxies may not hold for all variables. However, consider encephalization in the mammalian order Carnivora, where data are readily available. Phylogenetically corrected ordinary leastsquares (OLS) regressions for ln brain volume on ln body mass are statistically indistinguishable between species- and genuslevel analyses (Table 2; for comparative purposes, we use their cited phylogeny, adding brain and body data from ref. 3). Our intention was to encourage such cross-level tests, particularly for features that might be preserved in fossil taxa, although we do note that genus-level data are often interesting in themselves, not just as proxies for species (4). We were surprised that Smith and O’Meara mistook agnosticism for apostasy on the point of phylogenetic comparative methods. We agree that phylogeny should be taken into account whenever possible (e.g., ref. 5), indeed our analysis of encephalization supports that view (compare corrected and uncorrected genus-level values in Table 2). Species-level molecular phylogenies are the most desirable framework for macroevolutionary analyses, but this situation will never obtain for all but a handful of the extinct 99% of species that have lived on Earth (and will be a long time coming for a large fraction of the extant 1%, owing to the unavailability of suitable samples). Our analysis showed that lack of a molecular phylogeny does not necessarily preclude quantitative macroevolutionary analysis for morphospecies and morphogenera.

David Jablonski a,1 and John A. Finarelli b

a Department of Geophysical Sciences, University of Chicago, 5734
South Ellis Avenue, Chicago, IL 60637; and

b Department of Geological Sciences, University of Michigan, Ann Arbor, MI 48109

PDF gratuito do artigo aqui.