Associations between physiological traits and the performance of whole-organisms are the very stuff of evolutionary physiology. Recently, this approach has been taken several steps further by linking cellular level biochemistry with whole animal performance in a variety of organisms. Some aspects of the molecules to macroecology approach have proven controversial, such as the metabolic theory of ecology, but overall the field has started to reveal surprisingly strong connections among levels in the ecological hierarchy, that are rivalled in novelty perhaps only by those of community genetics (see Whitham et al., 2006, Nature Reviews Genetics 7, 510-523).

Here, Ruf and colleagues provide an elegant demonstration that variation in membrane composition (specifically n-6 polyunsaturated fatty acids) is strongly related to maximum running speed of mammals. Moreover, the strength of this relationship is almost as great as that of body mass, a more typical correlate of speed. The authors provide a convincing, though as yet not fully tested, mechanism for this relationship, and go on to argue that natural selection could easily operate thereon, so accounting for intra- and interspecific differences in performance.

This work raises several significant issues, most prominently among these the importance of biochemical variation for a trait typically not examined from this perspective (see review in Lovegrove, 2006, Physiological and Biochemical Zoology 79, 224-236), and the relative unimportance of size alone as a correlate (or determinant) of performance. Moreover, the study reveals avenues for research, such as repeatability of PUFA-related performance within species and taxonomically broader investigations of its underlying mechanisms, that are required to more fully justify its conclusions.

One concern with the present work is the inclusion, in 14 of the 36 species, of less than two individuals per sample. As studies elsewhere on scaling have shown (see Farrell-Gray & Gotelli, 2005, Ecology 86, 2083-2097), small sample sizes can substantially bias regression results. The bootstrap approach does somewhat compensate for the problem, given that sign and strength of association, rather than value of the slope, are of interest. However, likely bias should nonetheless be borne in mind, especially when calculations of the likely increase in the dependent variable for a given step size of the independent variable are being undertaken.