Genomic basis of neotropical primate adaptations

Neotropical primates(infraorder Platyrrhini) are among the most species-rich primate lineages. They arrived in South America ∼40 Ma, probably by transoceanic dispersal, when the continent had no connections with other landmasses. Living platyrrhines comprise five evolutionary offshoots (families) with diverse morphologies and ecologies: the seed-eating sakis and uakaris and the titi monkeys (Pitheciidae), the large-sized Atelidae with their prehensile tails, the tiny and brightly colored marmosets and tamarins (Callithrichidae), the nocturnal owl monkey (Aotidae), and the Cebidae (squirrel monkeys and capuchins) (1). Cebids present one of the largest encephalization quotients found in primates. Capuchins are characterized by advanced cognitive abilities and sensorimotor intelligence, which is particularly true of the robust capuchin monkeys (genus Sapajus), who frequently exhibit imaginative use of tools. Although research on capuchin monkey behavior is extensive, the genetic basis underpinning the development of advanced cognition of cebids remains unknown. Up to now, two of the three genera of extant cebids have publicly available genomes: the squirrel monkey Saimiri boliviensis and the gracile capuchin Cebus imitator. In PNAS, Byrne et al. (2) sequenced a high-quality genome of the robust capuchin Sapajus apella. They performed a comparative study of representative species from all three cebid genera, unveiling the evolutionary history of genetic modifications that allowed the rise of the distinctive neurobiology of capuchins and squirrel monkeys.

Byrne et al. (2) devised a strategy to interrogate a total of 9,216 genes in 10 primate species to search for signatures of molecular adaptation during the diversification of the three cebid genera. By investigating each branch of the cebid phylogeny at a time, they were able to identify adaptive evolutionary changes that were shared by all cebids (ancestral Cebidae), those shared by capuchin monkeys (the ancestor of Cebus and Sapajus), and those exclusive to each cebid genus (refer to figure 1 of ref. 2). They found that, early in the evolution of cebids, from 20 to 15 Ma, genes related to neurobiological traits underwent adaptive molecular evolution. This trend was maintained in the Saimiri and Sapajus lineages and the ancestor of capuchins, indicating sustained selective pressures for genes related to the central nervous system. Moreover, the split between Cebus and Sapajus was marked by adaptive changes in genes associated with face and skeletal system development in each species. Interestingly, both species are distinguished by their contrasting gracile and robust morphologies.

Several functional categories listed by the authors have been previously reported as possible targets of adaptive evolution in other primate lineages, especially the marmoset Callithrix jacchus (3, 4) and the gracile capuchin C. imitator (5). However, compared to the marmoset genome, there is an overrepresentation of genes related to neurobiological traits in most branches of the cebid phylogeny, suggesting their importance for cebid evolution (figure 2 of ref. 2). Additionally, they found correspondence between the morphological and ecological features associated with each branch and the functional categories of the genes inferred to be subjected to adaptation.

Challenges in Studying Adaptations Using Phylogenies

Byrne et al. (2) coped with several methodological complications when analyzing their data. The advent of genomics has brought new challenges to studying molecular adaptation. The genetic basis of adaptations is frequently investigated by comparing closely related lineages, such as populations or recently diverged sister species. In both cases, methods based on differences in allele frequencies (F_ST-like) and linkage disequilibrium along the genome are generally employed. However, such approaches are inadequate when studying species that diversified deeper in the evolutionary past, such as cebids. Alternatively, estimating the nonsynonymous (amino acid–altering) to synonymous rate ratio (d_N/d_S or ω) has been used to investigate the action of natural selection on coding sequences. Because synonymous changes in genes are thought to be nearly invisible to selection, d_S is used as an approximation of the neutral rate of molecular evolution. When d_N/d_S < 1, alleles with nonsynonymous changes tend to be eliminated from the population by negative purifying selection, whereas d_N/d_S = 1 indicates neutral evolution. If d_N/d_S > 1, there is evidence of positive selection, meaning that protein-changing alleles were kept in populations or were fixed with a probability greater than the neutral expectation, possibly because they conferred an increase in fitness to individuals carrying them. This has been widely used as evidence of adaptation at the molecular level.

When initially proposed in the 1980s, d_N/d_S methods were based on pairwise nucleotide sequence comparisons (6, 7). Analysis of episodes of molecular adaptation in the deeper evolutionary time relied on the statistical reconstruction of ancestral sequences (8). In the 1990s, maximum likelihood methods for estimating the ω parameter from multiple sequence alignments given a phylogenetic tree were developed (9, 10). This has significantly expanded the scope of molecular adaptation analysis, allowing the comparison of distantly related species. Furthermore, it was theoretically possible to identify episodes of positive selection on specific codon sites and branches of a phylogenetic tree, elucidating adaptive evolutionary changes in ancestral species (11). Byrne et al. (2) used such methods in their analysis of cebid evolution.

These attractive methods have some shortcomings, nevertheless. Most obviously, the analysis is restricted to coding regions of the genome, although many examples of molecular adaptation involve noncoding regions (12). Additionally, because of genetic recombination associated with population-level phenomena, the evolutionary history of the gene may differ from the species phylogeny (13). In the absence of natural selection, this gene tree–species tree topological discrepancy is mathematically described by the multispecies coalescent model (14). However, this model has mainly been used to develop new phylogenetic reconstruction methods and has not been fully incorporated into phylogeny-based d_N/d_S analysis. As a result, to have a one-to-one mapping of the changes along the branches of the gene tree onto the species tree, researchers have either to eliminate genes with trees that are discordant from the species tree or, as Byrne et al. (2) did, to enforce the consensus species tree topology onto gene trees. The impact of the latter approach has not been investigated thoroughly.

Furthermore, most phylogeny-based d_N/d_S estimation methods were developed before the advent of high-throughput genome sequencing technologies, when only a few genes were analyzed. Now, with thousands of genes investigated, some correction for multiple testing is required. Researchers have relied on standard statistical procedures to accomplish this task, such as calculating the false discovery rate. However, when interrogating a single gene, because its associated tree has several branches, multiple tests are frequently conducted, adding another layer of complexity to false discovery rate control. For each gene investigated, Byrne et al. (2) tested 11 models. The species sampling was also adjusted for each gene to increase the number of loci investigated (4,636 genes were available for all species analyzed). The authors employed the Benjamini–Hochberg correction to control for false positives but found this strategy too conservative because it reduced the set of candidate genes to figures never reported in similar studies with primates. Indeed, the performance of multiple testing corrections in genome-wide scans for positive selection has not been comprehensively evaluated.

Avoiding Molecular Spandrels

When novel vertebrate genomes are communicated, it is customary to scan for positively selected genes and comment on their relevance to species biology and evolution, highlighting putative adaptive processes. However, the function of several coding regions generally relies on the knowledge gathered from studies on model organisms, particularly humans (15). As Byrne et al. (2) point out, analyzing the physiological consequences of these adaptive changes requires that the function of those genes remains constant along the evolutionary process. Therefore, the possibility of enacting spurious causal relationships should not be ignored. After Gould and Lewontin’s famous criticism of the adaptationist program (16), Barrett and Hoekstra (17) deemed similar instances as molecular spandrels in the context of genome analysis.

For closely related lineages, where gene function is more easily evaluated, two methodological approaches are recognized when studying the genetics of adaptation, the forward and the reverse approaches (17, 18). In the forward approach, researchers first identify the adaptive phenotype and later ask what genes underlie the trait. In the reverse approach, after scanning genomes employing some metric, loci with significant statistical evidence for selection are identified and recorded as putatively involved in the (supposedly) adaptive process generating the phenotypes. Ideally, functional follow-up studies should be carried out to validate the results from the reverse strategy. Studies of adaptive genomic changes on distantly related species using d_N/d_S calculation consist of the reverse approach. Functional validation, in this case, is hardly possible.

Despite methodological hurdles and the possibility of inaccurate interpretation of the physiology of genes inferred to have undergone adaptive changes along the cebid history, the study of Byrne et al. (2) enables a broader understanding of the genetic basis of the evolution of advanced cognition in capuchins. Using genomic data, they backed up the hypothesis that neurobiological traits played a crucial role in the diversification of cebids. The authors also compared the phenotypic changes associated with each branch with the results from the enrichment analysis of adaptively evolving genes, adding an extra filter to avoid speculative adaptive histories. Although experimental confirmation of ancestral gene function is difficult, their research provides a candidate set of genes that merit evaluation by future studies.