Abstract
Programmed cell death, or apoptosis, plays an essential role in mammalian development, especially the development of the nervous system. Here, we systematically examine the molecular evolution of the mammalian intrinsic apoptosis program. We divided the program into its several constituent pathways and examined the evolution of each pathway in diverse mammalian taxa spanning primates, rodents and carnivores. We observed that genes involved in the caspase-dependent apoptosis pathway stood out in several ways. First, these genes display an accelerated rate of protein sequence evolution in primates relative to rodents or carnivores. Secondly, this acceleration is most pronounced along the lineage leading to humans, and it is associated with signatures of positive selection. Finally, several genes in this pathway, including APAF1 , CASP9 and CASP3 , have been shown to be associated with dramatic defects in neuronal cell number and brain size when mutated in mice. These observations suggest the possibility that evolutionary changes in the caspase-dependent apoptosis pathway may have contributed to brain evolution in primates and humans. Our results also lend further support to the hypothesis that genes regulating brain size during development might have played a particularly important role in transforming brain size during evolution.
INTRODUCTION
Apoptosis is one of the most extensively studied cellular processes, having been characterized in several model organisms including nematodes, flies and mammals ( 1 ). Although apoptosis is often thought of as a mechanism for removing damaged cells, it also plays a critical role in normal development ( 2 ). The developmental functions of apoptosis are many, from the removal of unneeded structures and shaping of existing structures to the control of cell number and organ size ( 3 ). There are two mechanisms of apoptotic action: the intrinsic program, which is characterized by mitochondrial depolarization, and the extrinsic program, which is characterized by the direct action of death receptor-induced signaling cascade on caspases. As schematized in Figure 1 , the intrinsic apoptosis program can be subdivided into three components: (i) the caspase-dependent pathway downstream of mitochondrial depolarization, which executes apoptosis, (ii) the caspase-independent pathway leading to apoptosis and (iii) the BCL2 protein family that regulates mitochondrial depolarization, which can be further divided into pro-apoptotic members of the BCL2 family and anti-apoptotic members of the family.
Schematization of the mammalian intrinsic apoptosis program.
Schematization of the mammalian intrinsic apoptosis program.
In mammals, the intrinsic apoptosis program has been shown to play a particularly important role in the formation of the central nervous system ( 4–9 ). During mammalian brain development, both neurons and glia are produced in overabundance, with programmed cell death ultimately eliminating as many as 50% of the cells. Studies in mice have demonstrated unequivocally that apoptosis is critically involved in brain development. This work has been further extended to primates and humans, which also supports the role of apoptosis in brain development, especially the development of the neocortex ( 10 , 11 ).
Mouse knockout studies have been particularly informative in elucidating the role of the intrinsic apoptosis program in brain development. Knockouts of many apoptosis genes show changes in neuronal cell number or neuronal sensitivity to apoptotic stimuli ( 4–9 ). Nowhere is this more pronounced than in the knockouts of Apaf1 , Casp9 and Casp3 , three genes that form the core components of the caspase-dependent apoptosis pathway. Mice carrying Apaf1 mutations display a failure of neural tube closure as well as an overgrowth of the forebrain, both the result of reduced apoptosis in neural cells ( 12–14 ). Casp9 knockout animals also show brain overgrowth, and indeed, the overgrowth is so dramatic that it leads to the appearance of cortical folds ( 15 , 16 ). Likewise, Casp3 knockout mice display neuronal overgrowth and disorganization ( 17 ).
Given the critical function of the intrinsic apoptosis program—especially the caspase-dependent apoptosis pathway—in the regulation of neuronal number during development, it is of great interest to examine the evolution of the program in the primate and human lineages where the brain has expanded dramatically. In this study, we carry out a comprehensive and detailed analysis on the molecular evolution of intrinsic apoptosis genes in primates, rodents and carnivores. We show that genes in the caspase-dependent apoptosis pathway have evolved significantly faster in primates than in rodents and carnivores and that this acceleration is most dramatic in the lineage leading to humans. We hypothesize that adaptive evolution of these genes may have contributed to the emergence of primate- or human-specific phenotypes of the brain.
RESULTS
Evolutionary comparisons among mammalian orders
Through literature survey, we identified 34 mammalian genes involved in the intrinsic apoptosis program. Eight of these belong to the BIRC family, which comprises a set of closely related homologs that have undergone varying degrees of amplification in various lineages. The complex relationship among these homologs leads to uncertainties in the assignment of gene orthology as well as difficulties in interpreting evolutionary data due to the occurrence of neofunctionalization and subfunctionalization during gene family evolution ( 18 , 19 ). We therefore excluded the BIRC family from our analysis.
For each of the genes, the ratio of non-synonymous to synonymous substitution rate ( Ka / Ks ) was calculated for three separate mammalian orders: primates (for which the calculation was based on human–macaque sequence comparison), rodents (based on mouse–rat comparison) and carnivores (based on cat–dog comparison). The Ka / Ks ratio is a commonly used metric for gauging the rate of protein sequence evolution as scaled to neutral mutation rate ( 20 ). Measurements for all the genes are summarized in Supplementary Material, Table S1.
In most cases, comparisons involving individual genes do not reach high statistical significance due to the relatively small number of nucleotide substitutions in each gene, especially for primates. Nevertheless, the majority of the genes was found to have higher Ka / Ks in primates relative to rodents or carnivores, suggesting a bias in these genes toward higher rate of protein sequence change in primates. To examine this possibility more closely, we considered the average Ka / Ks of all the intrinsic apoptosis genes as a whole. We found that primate Ka / Ks is indeed significantly higher than rodents and carnivores ( P <0.005) (Fig. 2 A). We next investigated whether this elevated Ka / Ks in primates is due to accelerated evolution of particular components of the intrinsic apoptosis program. As noted earlier, intrinsic apoptosis genes can be subdivided into four categories, including caspase-dependent apoptosis genes, caspase-independent apoptosis genes, pro-apoptotic members of the BCL2 gene family and anti-apoptotic members of the BCL2 family (see Supplementary Material, Table S1 for exact breakdown). For the caspase-dependent apoptosis subgroup, Ka / Ks is substantially and significantly higher in primates relative to rodents and carnivores ( P <0.0001), whereas the other subgroups of genes do not show significant differences among the three mammalian orders (Fig. 2 A). Indeed, upon removal of the caspase-dependent apoptosis subgroup from the entire list of intrinsic apoptosis genes, Ka / Ks values become statistically indistinguishable among the three orders. Thus, the caspase-dependent apoptosis subgroup stands out as a clear outlier in terms of its elevated rate of evolution in the primate order.
Comparing evolutionary rates of intrinsic apoptosis genes in diverse mammals. ( A ) Comparison among primates, rodents and carnivores. Intrinsic apoptosis genes are considered as a whole as well as subdivided into several constituent groups. ( B ) Comparison between the human lineage and the macaque lineage since their divergence from their common ancestor.
Comparing evolutionary rates of intrinsic apoptosis genes in diverse mammals. ( A ) Comparison among primates, rodents and carnivores. Intrinsic apoptosis genes are considered as a whole as well as subdivided into several constituent groups. ( B ) Comparison between the human lineage and the macaque lineage since their divergence from their common ancestor.
Evolutionary comparisons within primates
To further investigate the evolution of intrinsic apoptosis genes within primates, we compared Ka / Ks values between the human lineage and the macaque lineage. Gene sequences for the last human–macaque common ancestor were generated by inference from multiple-species alignments. This allowed the calculation of branch-specific Ka / Ks for either the human or the macaque lineage since their divergence from their common ancestor (Supplementary Material, Table S1). For intrinsic apoptosis genes as a whole, Ka / Ks is higher in the human lineage than in the macaque lineage, but this does not reach significance ( P <0.08) (Fig. 2 B). When each subgroup of apoptosis genes is considered separately, the caspase-dependent apoptosis subgroup shows substantially and significantly higher average Ka / Ks in the human lineage when compared with the macaque lineage ( P <0.01), whereas the other subgroups do not show statistically significant disparities between the two lineages (Fig. 2 B). Upon removal of the caspase-dependent subgroup from the entire list of apoptosis genes, the average Ka / Ks of the remaining genes become virtually identical between human and macaque branches. Thus, similar to comparisons involving different mammalian orders, the caspase-depended subgroup stands out as an outlier in showing significant rate disparities between human and macaque lineages.
The elevated Ka / Ks of caspase-dependent genes in the human lineage could be due to functional relaxation or positive selection. One way to test for positive selection is the McDonald–Kreitman test ( 21 ). The test examines whether the non-synonymous-to-synonymous ratio for divergence between species is significantly higher than the ratio for polymorphism within a species. If so, it is taken as evidence that positive selection has elevated the rate of non-synonymous changes over evolutionary time. Using resequencing-based human polymorphism data as previously reported ( 22 ), we applied a modified version of the McDonald–Kreitman test to each of the four subgroups. For each test, we combined all the genes in the category to determine category-wide totals of synonymous and non-synonymous changes. This showed that the caspase-dependent subgroup of genes indeed exhibit a significant excess of non-synonymous substitutions in the human lineage ( P <0.05), suggesting the action of positive selection. In contrast, the test did not yield any significant results for the other subgroups of apoptosis genes.
We next compared the evolutionary rates of intrinsic apoptosis genes between human and chimpanzee terminal branches. Using macaque sequences as the outgroup, we calculated Ka / Ks for either the human or the chimpanzee terminal branch since their divergence from a common ancestor. This showed, qualitatively, the same evolutionary trends as those seen in the comparison between human and macaque branches, i.e. there is higher rate of evolution in the human branch, especially for the caspase-dependent subgroup. However, owing to the small number of nucleotide substitutions between human and chimp, there is no statistical power to test the robustness of the trend.
Evolution of the APAF1 – CASP9 – CASP3 signaling cascade
Three genes, APAF1 , CASP9 (encoding caspase-9) and CASP3 (encoding caspase-3), form a signaling cascade that lies at the core of the intrinsic apoptosis program ( 23 ) (Fig. 1 ). A key event in apoptosis is mitochondrial depolarization. Following this event, cytochrome c is released from mitochondria and induces the oligomerization of APAF1, which in turn recruits and cleaves procaspase-9 into its active form. This multi-protein structure, known as the apoptosome, then binds and cleaves procaspase-3 into its active form. The activation of caspase-3, known as the executioner, leads to further cleavage events and ultimately DNA fragmentation and cell death. Besides caspase-3, other caspases can also serve as executioners, including caspase-6 and caspase-7; but caspase-3 is the primary executioner and is also the most abundant.
As noted earlier, although many apoptosis genes are associated with neural phenotypes when knocked out in mice ( 4 , 5 , 7–9 ), mutations in Apaf1 , Casp9 and Casp3 produce the most pronounced defects in the nervous system. Knockout of any one of these genes leads to a dramatic overgrowth of the brain, implicating these genes in the regulation of neuronal cell number and brain size during embryonic development ( 12–17 ). Casp3 , in particular, has been suggested to regulate neuronal cell number by mediating apoptosis in both neural stem cells and post-mitotic neurons ( 24 ). In light of the emerging concept that genes regulating brain size during development might be particularly relevant to brain evolution ( 25–36 ), we sought to more closely scrutinize the molecular evolution of these three genes.
For APAF1 , the primate Ka / Ks is significantly higher than the rodent rate ( P <0.002); it is also higher than the carnivore rate but only near significance ( P <0.07) (Supplementary Material, Table S1). Moreover, within primates, the branch from human–macaque ancestor to human has significantly higher Ka / Ks than the corresponding macaque branch ( P <0.0005). The branch undergoing the most rapid evolution is the one leading from human–macaque ancestor to human–chimpanzee ancestor. Here, the Ka / Ks value is 1.73, suggesting an episode of positive selection occurring during this time period. We next performed a sliding-window analysis of Ka / Ks to identify regions of the gene that have undergone particularly rapid evolution. Comparison of the sliding-window Ka / Ks profile in primates with that in rodents and carnivores identified a region within the WD40 repeat domains of APAF1 , which has an extraordinarily elevated Ka / Ks in primates (Fig. 3 ). Here, the Ka / Ks value peaks at close to 6, which is much greater than 1. This peak, combined with the overall heterogeneity of the distribution of non-synonymous mutations, is unexpected by neutrality and is statistically highly significant ( P <0.0001), indicating positive selection rather than relaxation of constraint in this region of the gene. The function of the WD40 domains is to bind the cytochrome c that is released following mitochondrial depolarization ( 37 ). It is this interaction that allows APAF1 to oligomerize and activate procaspase-9 ( 38–40 ). It is therefore intriguing to hypothesize that the amino acid changes in the WD40 domains of primate APAF1 might alter their binding affinity and thus change the regulation of apoptosome formation.
Sliding-window Ka / Ks analysis of APAF1 . Protein domains are schematized on the top of the graph. The pattern of non-synonymous substitutions is highly unlikely to be attributable to chance ( P <0.0001, see Material and Methods).
Sliding-window Ka / Ks analysis of APAF1 . Protein domains are schematized on the top of the graph. The pattern of non-synonymous substitutions is highly unlikely to be attributable to chance ( P <0.0001, see Material and Methods).
CASP3 also shows strong evolutionary rate disparities when compared among primates, rodents and carnivores. Its Ka / Ks in primates is significantly higher than either rodents ( P <0.003) or carnivores ( P <0.009) (Supplementary Material, Table S1). Within primates, however, the limited number of nucleotide substitutions precludes statistically meaningful analysis of rate disparities among the various branches. In contrast to APAF1 and CASP3 , the molecular evolution of CASP9 is unremarkable. It does not show statistically distinct Ka / Ks rates when compared among primates, rodents and carnivores. The lineage from human–macaque ancestor to human is also not distinct from the corresponding macaque lineage. Removal of APAF1 and CASP3 from the caspase-dependent apoptosis gene list does not abolish the trend of elevated Ka / Ks in primates, indicating that they are not entirely responsible for the trend.
DISCUSSION
It has long been recognized that apoptosis plays a critical role in regulating mammalian development, especially the morphogenesis of the nervous system. It is then perhaps not unreasonable to surmise that the molecular evolution of apoptosis genes might have contributed to the morphological evolution of primates in general and humans in particular. Indeed, previous studies have produced suggestive evidence that apoptosis genes may have experienced positive selection in the primate and human lineages. Human–chimpanzee genome comparisons as well as human polymorphism studies showed that apoptosis genes have an elevated rate of protein sequence divergence relative to the genome average and that they are enriched for putatively positively selected genes ( 22 , 41 , 42 ). In this study, we carried out a systematic analysis on the molecular evolution of the intrinsic apoptosis program in mammals. We divide the apoptosis program into its several constituent pathways and investigated the evolution of each pathway in diverse mammalian species. We found that genes in the caspase-dependent apoptosis pathway downstream of mitochondrial depolarization exhibit an accelerated rate of protein sequence evolution in primates relative to other mammalian orders. This acceleration is most pronounced along the lineage leading to humans.
By itself, an acceleration of protein sequence evolution in a lineage is not necessarily indicative of positive selection. Indeed, it has been shown that variations in effective population size may have an effect on rates of non-synonymous fixation ( 43 ). To adequately take this into account, we compared the rate of acceleration in primates relative to rodents and carnivores between several categories of apoptosis genes. Differences in effective population size should have a genomic, rather than genic, effect. The statistically significant differences in protein evolution rate observed between categories of genes are thus reflective of category-specific processes rather than a genome-wide phenomenon.
It is also the case that a category-specific acceleration of protein sequence evolution in primates is, by itself, not necessarily reflective of positive selection. A relaxation of constraint specific to a category of genes in primates would produce the same pattern. However, we disfavor this scenario because these genes have been shown to be of crucial importance to normal human development. Further, additional lines of evidence indicate that positive selection may be playing a role in the evolution of these genes. The McDonald–Kreitman test, as well as specific examples of lineages with Ka / Ks ratios greater than one, shows that the evolution of at least some of these genes is consistent with positive selection.
Interestingly, several genes in the caspase-dependent apoptosis pathway are linked to the regulation of neuronal cell number and brain size during embryonic development. For three of these genes, APAF1 , CASP9 and CASP3 , which form a signaling cascade at the core of the intrinsic apoptosis program, loss-of-function mutations result in dramatic brain overgrowth in mice. APAF1 and CASP3 both show clear evidence of accelerated evolution in the primate and human lineages. These observations suggest—though they do not yet definitively prove—the possibility that the caspase-dependent apoptosis pathway in general and the APAF1 and CASP3 genes in particular might have contributed to brain evolution in primates and humans.
One point worthy of some discussion is how to reconcile the pleiotropism of the apoptotic program across many cell types and the fact that disruption of certain key apoptosis genes such as APAF1 , CASP9 and CASP3 predominantly affects neural development. Indeed, although Apaf1 , Casp9 and Casp3 are expressed in many tissues and the list of peptide substrates of CASP3 is extensive ( 44 ), knockouts of these genes show overgrowth specifically of neural tissues ( 12–17 ). One clue is the finding that most substrates of CASP3 are appropriately cleaved even in Casp3 knockout mice, suggesting functional redundancy ( 45 ). Such redundancy is also consistent with the phylogenetic history of apoptosis genes ( 46 ). Thus, APAF1 , CASP9 and CASP3 may have been relieved of their pleiotropic function by other members of the apoptosis program and have become specific modulators of cell death in mammalian neurogenesis. Such specificity might have been key in facilitating the rapid evolution of certain apoptosis genes in primates.
It has been hypothesized that genes involved in regulating brain size during development might have also been involved in transforming brain size during evolution ( 25 ). A number of studies are indeed consistent with this hypothesis. Several genes implicated in the developmental regulation of brain size, such as ASPM , microcephalin ( MCPH1 ), CDK5RAP2 , CENPJ , Cernunnos and Sonic Hedgehog ( SHH ), show accelerated evolution in the primate and human lineages, and the acceleration is often associated with robust signatures of positive selection ( 26–36 ). Our results lend further support to the hypothesis. However, given the many functions of apoptosis in developmental and physiological processes outside of the nervous system, it remains an open possibility that adaptive evolution of the apoptosis program could have contributed to the phenotypic evolution of non-neural systems. This notwithstanding, it is clear that the caspase-dependent apoptosis pathway has experienced distinct selective regimes in primates. Our results thus provide an impetus for additional studies to further examine the role of apoptosis in primate and human evolution.
MATERIALS AND METHODS
Gene sequences were obtained from the current ENSEMBL builds of the human ( Homo sapiens ; NCBI 35), chimpanzee ( Pan troglodytes ; PanTro 1.0), rhesus macaque ( Macaca mulatta ; Mmul 0.1), mouse ( Mus musculus ; NCBI m34), rat ( Rattus norvegicus ; RGSC 3.4) and dog ( Canis familiaris ; CanFam 1.0) genomes. In some cases, when ENSEMBL genes appeared incomplete, RefSeq genes were used. Additional sequences were obtained for cat ( Felis catus ) from trace archives available through NCBI. A complete list of accession numbers for all the genes used in our analysis is given in Supplementary Material, Table S2. Orthologs were determined from a combination of reciprocal best BLAST hits, synteny and prior description in the literature. For statistical outlier genes, PCR and sequencing primers were designed to resequence them within the laboratory, which in all cases confirmed the sequences from the databases.
Genes were aligned in frame using the Pileup and Frame Align programs as implemented in the Wisconsin Package v10.2 (Accelrys Inc., San Diego, CA, USA). Evolutionary parameters, including numbers of synonymous and non-synonymous mutations corrected for multiple hits as well as the Ka / Ks ratio, were calculated using the Li method ( 47 ). Heuristic methods such as the Li method have been shown to be more reliable than codon-based likelihood methods such as those used by PAML in estimating Ka , Ks and Ka / Ks for sequences less than 300 codons and for genes with transition/transversion ratios less than 3 ( 48 ). Ancestral sequences were determined using maximum parsimony, though the maximum likelihood method gave the same results.
Summary statistics and standard errors were calculated as described previously ( 31 ). The statistical significance of evolutionary rate disparities between lineages was calculated by two-tailed Fisher's exact test, as reported previously ( 27 ). A Bonferroni correction for multiple testing was applied (corrected significance was determined by multiplying the Fisher's P -value by the number of tests performed, in this case the number of genes). Sliding-window analysis of Ka / Ks follows previously published procedure ( 49 ) and has a window size of 33 codons and a step size of 10 codons. Computer simulations, as described previously, were employed to test the statistical significance that the spatial distribution of non-synonymous substitutions along a gene departs from the neutral expectation with respect to maximum peak height and the proportion of windows with values greater than 1 ( 27 , 30 ).
SUPPLEMENTARY MATERIAL
Supplementary Material is available at HMG Online.
ACKNOWLEDGEMENTS
We are indebted to the macaque sequencing consortium for allowing us to use their unpublished macaque genome sequence data. We wish to thank S. Dorus, P.D. Evans and N. Mekel-Bobrov for helpful discussions. We also wish to thank W.H. Li, U. Schmidt-Ott and W.B. Dobyns for their helpful criticisms of the study. This work was supported in part by a University of Chicago William Rainey Harper Dissertation Fellowship to E.J.V.
Conflict of Interest statement . None declared.
