Adaptive evolution toward larger size in mammals

Abstract

The notion that large body size confers some intrinsic advantage to biological species has been debated for centuries. Using a phylogenetic statistical approach that allows the rate of body size evolution to vary across a phylogeny, we find a long-term directional bias toward increasing size in the mammals. This pattern holds separately in 10 of 11 orders for which sufficient data are available and arises from a tendency for accelerated rates of evolution to produce increases, but not decreases, in size. On a branch-by-branch basis, increases in body size have been more than twice as likely as decreases, yielding what amounts to millions and millions of years of rapid and repeated increases in size away from the small ancestral mammal. These results are the first evidence, to our knowledge, from extant species that are compatible with Cope's rule: the pattern of body size increase through time observed in the mammalian fossil record. We show that this pattern is unlikely to be explained by several nonadaptive mechanisms for increasing size and most likely represents repeated responses to new selective circumstances. By demonstrating that it is possible to uncover ancient evolutionary trends from a combination of a phylogeny and appropriate statistical models, we illustrate how data from extant species can complement paleontological accounts of evolutionary history, opening up new avenues of investigation for both.

Keywords: Cope’s rule; adaptive evolution; ancestral state reconstruction; evolutionary trends; macroevolution.

Fig. 1.

Faster path-wise rates have led to larger body size in mammals. (A) Relationship across all mammals is plotted, and data points are colored by order (n = 3,321). The black line is the fitted phylogenetic slope of the relationship between body size and path-wise rates (Methods) across all mammals. (B) Fitted phylogenetic slopes of the relationship within each of the 11 mammalian orders investigated here. Orders that contain aquatic groups are indicated by an asterisk; for these orders, only the terrestrial members are plotted. Aquatic groups are plotted separately (Cetacea, pinnipeds, and Sirenia).

Fig. 2.

Comparisons between the cumulative distribution of observed mammalian body sizes (n = 3,321, black lines) and simulated data (n = 1,000, colored lines). The real data are compared with simulations generated from our separate-slopes regression model (blue lines) and a conventional homogeneous Brownian motion model (red lines, Inset).

Fig. 3.

PAD comparisons and reconstructed ancestral sizes. (A) Projection of ancestral state reconstructions into a phylomorphospace (n = 5,234, including all tips and internal nodes). Points are connected by phylogeny, and each internal node of the tree has been reconstructed using the parameters of our separate-slopes regression model. Our estimate for the therian root (24.5 g) falls within the ranges given by the paleontological data (20–25 g, midpoint indicated by the pale blue square). This estimate is in contrast to the estimate made by a conventional homogeneous Brownian motion model, which is more than an order of magnitude too large (pale pink square, 610.7 g). Orders that contain aquatic groups are indicated by an asterisk; for these orders, only the terrestrial members are plotted. Aquatic groups are plotted separately (Cetacea, pinnipeds, and Sirenia). (B–D) PAD changes (Δlog₁₀ body size) across every branch of the mammalian phylogeny (n = 5,233). The red dashed line indicates no change in size. (B) Frequency (f) distribution of Δlog₁₀ body size across individual branches. There is a significant bias toward body size increase (exact binomial test, P < 0.001). (C) Plot of the inferred rate of evolution along individual branches (Methods) against Δlog₁₀ body size. The regression line is significantly positive (β = 0.015, P < 0.0001). (D) Ancestral body size plotted against body size change across individual branches. The gray bars represent the SD of Δlog₁₀ body size calculated from the variance associated with each data point (σ²_{Δlog10 body size}; Methods). The regression line and the SDs in D have been corrected for the regression to the mean artifact (Methods and SI Text). The slope of the relationship between ancestral size and Δlog₁₀ body size is significantly positive (β = 0.020, P = 0.0006). Highlighted by a red square on each of these plots is the branch leading to modern bats.

Fig. 4.

Comparisons of reconstructed body sizes with fossil estimates. The solid colored lines in both plots are the predicted phylogenetic slopes from a regression model of fossil sizes as given in the paleontological literature against reconstructed values (n = 65). The dashed black lines indicate a one-to-one relationship, which is the expected slope if models are predicting body sizes accurately. (A) Predicted body sizes from a homogeneous Brownian motion model compared with fossil estimates. (B) Predicted body sizes from our separate-slopes model in comparison to the fossil record.