Technical Comments

Comment on “Nocturnality in Dinosaurs Inferred from Scleral Ring and Orbit Morphology”

Science  23 Dec 2011:
Vol. 334, Issue 6063, pp. 1641
DOI: 10.1126/science.1208442

Abstract

Schmitz and Motani (Reports, 6 May 2011, p. 705) claimed to definitively reconstruct activity patterns of Mesozoic archosaurs using the anatomy of the orbit and scleral ring. However, we find serious flaws in the data, methods, and interpretations of this study. Accordingly, it is not yet possible to reconstruct the activity patterns of most fossil archosaurs with a high degree of confidence.

Schmitz and Motani (13) described a method for using orbit and scleral ring morphology to discriminate living species by activity pattern, which they subsequently used to classify 31 species of Mesozoic archosaurs (1) as photopic (diurnal), scotopic (nocturnal), or mesopic (cathemeral/crepuscular). If correct, this analysis would represent a major advance in amniote paleobiology. However, we have major concerns that call these conclusions into question.

First, the statistical approach employed by Schmitz and Motani (1) does not provide a basis for such definitive interpretations. Their work (13) uses discriminant function analysis (DFA), which tests whether groups that are defined a priori differ significantly from one another based on a set of predictor variables (4). In DFA, group membership for known species is assigned before analysis, and the model predicts group membership of unassigned species based on a set of independent variables [here, orbit length and internal and external scleral ring diameters (13)]. DFA is useful because the group identity of a fossil species can be inferred based on how closely it plots to a group centroid, which is calculated from the known species.

Given these points, we find key problems with how Schmitz and Motani interpret their DFA results. Although their whole-model results exhibited significant P values, substantial overlap exists between their three groups even among extant taxa. In particular, mesopic taxa nearly completely overlap the other two groups, a fact evident in Schmitz and Motani’s figure S1 (1). These results are consistent with previous research showing that orbit and scleral ring morphologies of diurnal and nocturnal birds and lizards overlap to a degree that precludes confident reconstruction of activity pattern in fossil taxa (5, 6). Furthermore, although DFA tests for differences among all groups, it does not test for differences between any two particular groups (4). Classification statistics reveal the ability of the variables to correctly predict species activity patterns by describing the proportion of species that the model correctly assigns to their a priori groups and the percentage chance that each animal will belong to each group. Currently, the only software available to conduct a phylogenetically corrected DFA is the R script created by Schmitz and Motani (13), which does not provide classification statistics. Therefore, we used their data and the same prior probabilities to conduct a linear DFA and found that ~21% of taxa were considered “misclassified” by the model (table S1). Because Schmitz and Motani reported a very weak phylogenetic effect (Pagel’s lambda = 0.08) (1), similar results can be expected in both analyses (7). Importantly, our model showed that 80% of the 30 species with a mesopic a priori assignment were incorrectly classified. For example, the great blue heron is mesopic, but our model predicts it is only 30.5% likely to be considered mesopic using the Schmitz and Motani data (3). These results suggest that the underlying anatomical features are unreliable predictors of activity pattern in extant species. We question whether a method that does not confidently discriminate between living species of known activity pattern should be used to infer the activity patterns of extinct taxa.

Our second concern is that the DFA prior probabilities rely on flawed data and assumptions. Schmitz and Motani used observed proportions of activity patterns among extant amniotes (1) but obtained most of their data from the Animal Diversity Web, a nonrefereed resource. Data that we gathered from primary literature and more authoritative sources [e.g., (8)] reveal that 26% of the mammals classified as “photopic” in Schmitz and Motani’s table S4 (1) are either cathemeral or nocturnal. Regardless, no theoretical or empirical justifications are provided for the assertion that Mesozoic amniote activity patterns should conform to those of extant amniotes. Indeed, the Mesozoic archosaurs sampled lived across some 160 million years and did not constitute a single, coeval fauna. These fossils were chosen because they preserve relevant osteological structures and were neither sampled randomly through time nor selected to capture the range of known Mesozoic ecomorphologies. Therefore, they cannot be construed to represent a typical Mesozoic world and should not be apportioned based on modern taxa.

Our third concern is that nearly half of the fossil archosaurs analyzed do not overlap with extant taxa on discriminant axis 2 in Schmitz and Motani’s figure S1 (1). Interpretations beyond the bounds of the originally sampled data are statistically dubious (9). Most fossils overlapping with extant animals are either relatively small-sized fliers or closely related to birds. The outlier fossils are not closely related to each other, and although most are medium- to large-sized species, only a few are outside the size range of the extant sample. These taxa occupy a morphospace that may have no living analog, suggesting that definitive activity pattern assignments may not be possible.

Schmitz and Motani (13) argue that they can interpret activity pattern from the orbit and scleral ring because these variables are functionally informative for vision. To interpret the visual capabilities of an eye, two measurements are minimally required: corneal diameter, correlated with how much light can enter the eye, and axial length, correlated with focal length (10). We agree that the inner diameter of the scleral ring surrounds the cornea and is a good proxy for corneal diameter (13, 5, 6, 11). Similarly, because mammalian eyes are essentially spherical, their transverse diameter is nearly the same as axial length (1214). Accordingly, orbit diameter, an osteological correlate of transverse eye diameter, can be used to approximate axial eye diameter in mammals, although the correlation deteriorates with increasing body size (12, 13, 15). However, many living diapsid eyes are nonspherical, with a transverse eye diameter that is substantially greater than axial length (14). For many diapsids, orbit diameter is thus a poor proxy for axial eye diameter (5, 6). None of the fossil archosaurs included preserved osteological features that permit reliable inference of eye axial diameter.

In sum, the model presented by Schmitz and Motani is much less powerful than they suggest. Definitive interpretation of fossil activity patterns based on these osteological characteristics may not be possible except in cases of morphological extremes. Substantial improvements are likely to come only from concentrated efforts to analyze eye shapes and identify new osteological correlates in extant animals.

Supporting Online Material

www.sciencemag.org/cgi/content/full/334/6063/1641-b/DC1

Table S1

References

View Abstract

Subjects

Navigate This Article