ASPM and mammalian brain evolution: a case study in the difficulty in making macroevolutionary inferences about gene–phenotype associations

Identifying the genetic basis of adaptive phenotypes can be a significant step towards understanding how that phenotype evolved. With the increased availability of interspecific molecular data, an approach to uncover such genes has been to search for signatures of adaptive evolution at the molecular level. Many analyses have adopted a candidate gene approach, focusing on genes with important developmental roles. One such candidate gene is ASPM, which is involved in neurogenesis and associated with major neurological disorders [1]. The molecular evolution of ASPM has been investigated for a decade (see electronic supplementary material, table S1), under the hypothesis that it contributes to primate brain evolution. A recent study by Xu et al. [2] extends the taxonomic scope by demonstrating that ASPM evolved adaptively in cetaceans. However, descriptive studies of patterns of selection are now being supplanted by those that explicitly test for gene–phenotype associations. Using such an approach, we find that Xu et al.'s conclusion that ASPM is linked to increases in cetacean encephalization quotient (EQ), a measure of relative brain size, is not supported. We highlight developments in the analysis of molecular data and phylogenetic methods that are capable of resolving major issues in functional gene–phenotype coevolution, which we hope will provoke discussion and aid future studies.

One approach for making gene–phenotype associations is to test for shifts in selection pressure acting on a gene in taxa that display the phenotype of interest. This frequently involves comparing estimates of dN/dS, a measure of the strength of selection acting on a protein-coding gene, using a range of tests implemented in software such as PAML (see electronic supplementary material, table S2) [3]. The results of these tests can be influenced by the nature of the data and, in particular, require sufficient evolutionary variation to make reliable estimates. Data with few substitutions or from a restricted number of taxa can lead to spurious results. These effects may be evident in Xu et al.'s analysis [2]. First, they suggest that a high proportion of branches in the cetacean phylogeny have an elevated dN/dS, which they interpret as evidence of increased positive selection, but do not perform explicit tests of this hypothesis. Further analysis (see the electronic supplementary material) suggests that none of these values are significantly greater than one, the threshold for accepting adaptive evolution. The apparent elevation in dN/dS is likely to be influenced by the low number of substitutions on short branches. This problem is particularly strong for cetaceans, which have low substitution rates [4]. Second, it is suggested that positive selection is limited to mammalian orders with high EQs. However, this result is likely to be influenced by sampling bias, and inclusion of further taxa provides evidence for positive selection across mammals (see the electronic supplementary material). Identifying robust shifts in selection pressure clearly requires both adequate and even sampling, and sufficient numbers of substitutions.

A related method involves testing for shifts in selection acting on a gene and changes in the associated phenotype along a subset of branches in a phylogeny. This method is particularly useful when applied to novel or discrete traits but has also been applied to continuously variable, quantitative traits. This can lead to two problems; first, identifying the branches that show high rates of phenotypic evolution, and, second, applying models of molecular evolution that assume episodic positive selection in the presence of pervasive positive selection. A previous study on ASPM suggested an association between episodic positive selection and branches showing major increases in cortical volume in primates, identified using parsimony-based ancestral state reconstructions [5]. However, closer analysis revealed that this result was not robust, as positive selection was not episodic but pervasive, and the identification of key branches was not supported by alternative methods [6]. Xu et al. [2] suggest an association between high rates of evolution and major increases in cetacean relative brain size, but do not explicitly test for phenotypic shifts. Instead, they rely on previous assumptions about cetacean evolution to highlight key branches. Recent comparative analyses unfortunately suggest that these assumptions are not valid [7]. Furthermore, their results demonstrate that positive selection was again pervasive and not limited to a subset of branches. Hence, although this approach may be valid for some phenotypes in some taxa, care is needed on both the phenotypic and molecular side of the analyses. Methods are available that explicitly identify phenotypic rate shifts [8], and, combined with tests for episodic versus pervasive positive selection, robust tests for gene–phenotype association could be performed in some situations in this way.

If positive selection acting on a locus was pervasive and the phenotype did not evolve in a punctuated manner, a potentially more relevant approach is to test for correlated rates of gene and phenotypic evolution across the whole phylogeny. Several methods have now been proposed to perform such analyses [6,9,10], and a handful of studies have found evidence for macroevolutionary gene–phenotype associations. For example, a method that has been applied to ASPM is to test for a significant regression between the selection pressure acting on a gene during the descent of each species (measured by root-to-tip dN/dS) and alternative phenotypes along branches of the phylogeny [6]. Using this approach, selection on ASPM has been linked to absolute brain mass, and in particular neonatal brain mass, in anthropoid primates [6]. This result is supported by a significant association being found in two largely independent datasets representing both increases and decreases in brain mass [6,11], and is consistent with the hypothesis that selection on ASPM may contribute to the evolution of neurogenic output.

Explicit hypothesis testing is challenging, but clearly desirable when arguing for a gene–phenotype association at a macroevolutionary level where comparative functional tests may not be forthcoming. Careful planning is required to ensure maximum statistical power in such analyses (e.g. by targeting the collection of genetic data according to the availability of phenotypic data when the latter represent a restrictive commodity). This is clearly an issue with brain volume data. The overlap between Xu et al.'s genetic data and cetacean brain size data is incomplete; nevertheless, one can still test hypotheses while acknowledging this caveat. When the available data are used to test for a macroevolutionary association between selection on ASPM and either EQ or absolute brain size, no significant association is found (EQ: t₉ = 0.445, p = 0.667; brain mass: t₉ = −0.741, p = 0.478; see electronic supplementary material). Although better data could overturn this result, we currently find no support for an association between ASPM and cetacean brain size either based on the patterns of positive selection within cetaceans or across mammals, or through explicit hypothesis testing.

This absence of evidence does not rule out the possibility that ASPM does indeed play some role in cetacean brain evolution. Xu et al. clearly demonstrate that ASPM evolved adaptively in cetaceans, and patterns of evolution in primates are suggestive of a link between ASPM and brain mass, raising the possibility that ASPM has a conserved role in mammalian brain evolution. Explicit tests using comparative methods, combined with functional data, are necessary to assess this hypothesis.

The methodology for such tests is in its infancy and further developments are required. In addition to poor overlap between genetic and phenotypic datasets, one can envisage several other limitations. For example, if selection is restricted to a subset of sites or domains, then the signal of a gene–phenotype association could be lost when using gene-wide dN/dS values. Should we then perform association tests on functional domains or is a sliding-window analysis across a locus desirable? If phenotypic reversals are common, then the signal could again be lost as dN/dS may increase during both increases and decreases of a phenotypic trait [6]; is it possible to account for such effects? For polygenic traits, how do we detect real associations with genes that are only targeted by selection intermittently? Beyond candidate genes, do we have sufficient power to perform genome-wide scans for macroevolutionary phenotypic associations? And beyond protein-coding genes, what tests can be applied to promoter regions or levels of gene expression? The development of new methods should begin to offer answers to these questions [9–10,12].

Xu et al.'s study of the evolution of ASPM in cetaceans [2] is a welcome addition to a field frequently mired by a narrow focus on the singular case of human brain evolution. Furthermore, it raises important questions about the genetic basis of complex and convergent phenotypes. However, the issues discussed above limit the conclusions derived regarding the phenotypic relevance of selection on ASPM in cetaceans. These problems are frequently found in similar studies, and we highlight them here only because they need to be addressed if we are to move beyond the descriptive phase of comparative adaptive genetics to one capable of applying powerful statistical tests to gene–phenotype associations.

Funding statement

S.H.M. is supported by a Research Fellowship from the Royal Commission for the Exhibition of 1851.