Complete COX5A nucleotide coding region and deduced amino acid sequences encoding the mature peptide were analysed in 26 vertebrate species, including newly generated data from 14 primate species: common chimpanzee (Pan troglodytes), bonobo (Pan paniscus), gorilla (Gorilla gorilla), orangutan (Pongo pygmaeus), red-cheeked gibbon (Nomascus gabriellae), siamang (Symphalangus syndactylus), mantled guereza (Colobus guereza), olive baboon (Papio anubis), pygmy marmoset (Callithrix pygmaea), white-lipped tamarin (Saguinus labiatus), white-eared titi monkey (Callicebus donacophilus), slow loris (Nycticebus coucang), brown greater galago (Otolemur crassicaudatus) and brown lemur (Eulemur fulvus); and one marsupial, the parma wallaby (Macropus parma). Partial sequences were obtained for three additional taxa (cat, rabbit, and chicken). Apart from a single amino acid difference in the elephant sequence, the protein sequence alignment showed no amino acid changes from the last common ancestor of placental mammals to present-day non-anthropoids, in a sampling that includes representatives from every major placental mammalian clade except Xenarthra (Fig. ). Interestingly, the elephant is similar to anthropoids in that both share the aerobically demanding feature of lengthy gestation. Within anthropoids, 13 amino acid replacements were inferred by all ancestral state reconstruction methods, and eight of these occurred on the human lineage from the last common ancestor (LCA) of haplorhines (tarsier and anthropoids) (Fig. ).
Figure 1 Phylogenetic tree depicting the omega ratios and (number of nonsynonymous and synonymous substitutions) for COX5A in each branch of the tree, as estimated by PAML under the free ratio (M1) model. "0" and "*" indicate, respectively, lineages on which the (more ...)
Figure 2 Phylogenetic tree depicting amino acid replacements inferred in COX5Ap within the anthropoid clade determined via maximum parsimony (ACCTRAN and DELTRAN) as implemented in PAUP* , and the codon substitution model  implemented in PAML 3.15. Non-prosimian (more ...)
Evidence for adaptive evolution
Results from the model-based codeml analyses confirm that COX5A omega ratios vary among lineages in vertebrates. Specifically, the statistical model assuming one omega value for all branches (i.e., the one ratio model) had a likelihood value of -1431.66, with an omega value of 0.048. The model that assumes an independent omega value for each branch (i.e., the free ratio model) had a likelihood value of -1388. According to the likelihood ratio test, the free ratio model fit the data significantly better than the one ratio model (p < 0.01). The inferred omega values and the numbers of nonsynonymous and synonymous substitutions under the free ratio model are presented in Fig. . In addition, the model that segregates anthropoids from non-anthropoid placental mammals fit the data significantly better than the model that assigned the same omega value to all placental mammals (p < 10-5), indicating that the ratio of nonsynonymous/synonymous substitution rates differs in this clade; in particular, the omega value inferred for anthropoids is 52 times greater than that inferred for non-anthropoid placental mammals (0.431 vs. 0.008; Table ).
Parameter estimates and likelihood values under branch and branch-site models.
Despite this elevated ratio in anthropoids, we note that the overall ratio is still less than one. This result, however, is based on the assumption that all codons are under the same selective pressure and thus share the same underlying omega value. Because this assumption is "grossly unrealistic" [20
], and because important changes in function can result from a few key amino acid changes that do not elevate the overall omega ratio above one [21
], we conducted a branch-site test of adaptive evolution. Here, the Anthropoidea as a total group was fixed as the foreground lineage, i.e., the branches on which there was permitted a class of sites with an omega > 1. Results from this test provide evidence of positive selection on particular sites during anthropoid evolution: the model allowing a class of sites with an omega higher than one (i.e., positive selection) fit the data significantly better than an alternative model that fixed omega at one for this same site class (p < 0.002). Specifically, five sites were identified as positively selected, four of which had a posterior probability > 0.95 (Table ). Furthermore, we note that despite the many amino acid replacements inferred to have occurred throughout the Anthropoidea (see below), the five positively selected sites show changes occurring predominantly on stem lineages (anthropoid, ape, and New World monkey) and on the human terminal branch. The localization of these particular sites to these three stem lineages, combined with the M1 analysis (Fig. ) depicting a classic sequence [22
] of markedly increased dN relative to dS, followed by descendant lineages with a markedly decreased dN relative to dS, presents a pattern strongly indicative of positive selection.
Ancestral state reconstruction methods confirm and extend results obtained by the codeml analysis. Among the five sites identified as positively selected by the branch site test, four were inferred to have changed on more than one anthropoid lineage, and three were inferred to have changed on the anthropoid stem (Fig. ). In addition, all three ancestral state reconstruction methods inferred a fourth amino acid change on the anthropoid stem that was not identified by the branch site test: a methionine to isoleucine at residue 33. It is tempting to speculate that this replacement may have acted as a "permissive" change that then allowed the selection of the other three amino acid changes occurring on this and perhaps other descendant anthropoid branch(es). This sequence of events was recently demonstrated in an elegant study by Ortlund et al. [23
] in their investigation of cortisol-sensitive glucocorticoid receptor (GR) evolution in vertebrates. Using a combination of structural, phylogenetic and functional analyses of ancestral corticosteroid receptors, the authors identified a small set of permissive amino acid changes that stabilized specific elements of the protein. In conjunction with additional changes that diminished the receptor's sensitivity to aldosterone and increased its sensitivity to cortisol, these permissive changes allowed an additional and subsequent set of function-switching amino acid changes that conferred a fully GR-like phenotype that is sensitive only to cortisol. With respect to COX5Ap, verification of this hypothesis in future studies would require additional, fine-tuned functional data regarding the effects of each of the observed replacements alone and in concert on the activity of the reconstructed protein.
Expression and localization of COX5A
Quantitative RT-PCR results showed that relative COX5A
gene expression levels in the anterior cingulate cortex vary among human, chimpanzee, gorilla, and macaque (Fig. ). Specifically, human and macaque showed similar gene expression levels, as did chimpanzee and gorilla, and the first pair of species showed higher levels of expression than the second pair (Fig. ). This pattern confirms the trend previously suggested by microarray analysis [24
], and represents a more robust dataset from which to make such observations: unlike the previous experiment, which involved hybridizing non-human samples to a human-specific microarray, these qRT-PCR results are not confounded by sequence mismatches between sample and microarray chip. Specifically, the COX5A
expression results reported by Uddin et al. [24
] were based on an Affymetrix probe set (203663_s_at) that shows 0.8% difference with the published chimpanzee genome sequence [25
] and 7.3% difference with the published rhesus macaque sequence [26
]. Thus, the present results suggest that COX5A
has undergone not only coding region sequence evolution, but also regulatory evolution in anthropoid primates. This hypothesis is supported by the observation of primate COX5A
expression patterns in other tissues that differ from those reported here: in microarray experiments performed on fibroblast cell lineages, humans showed a lower COX5A
expression level than did bonobos or gorillas, and comparison of the human and bonobo expression levels showed a significant difference [27
]. Nevertheless, without additional samples from additional chimpanzees and gorillas, the possibility that our results are due to interindividual differences within these species and/or age-related differences in gene expression profiles cannot formally be ruled out.
Figure 3 COX5A Expression levels as determined by quantitative RT-PCR (solid bars) and microarray signal values (hatched bars). qRT-PCR expression levels are expressed in relative terms, with samples compared to a calibrator amplified from human reference total (more ...)
Among macaques, gorillas, chimpanzees, and humans, the localization of COX5Ap as revealed by immunohistochemistry showed a punctate staining pattern that is consistent with the morphology of mitochondria (Fig. , data from gorilla not shown). Similar staining patterns were obtained in all species with both monoclonal and polyclonal antibodies. As evident by colocalization with microtubule-associated protein 2 (MAP2) and neuron-specific nuclear protein (NeuN), COX5Ap was particularly enriched in the cytoplasm of the soma and proximal apical dendrite of neurons, especially large pyramidal cells. We also observed diffuse COX5Ap-immunoreactive puncta throughout the neuropil, which corresponds to the space occupied by glia, dendritic processes, axons, and synapses. Double immunostaining against glial fibrillary acidic protein (GFAP) demonstrated that astrocytes do not show high levels of COX5Ap. Indeed, 95.3% of cells that are enriched with COX5Ap also contain the neuron-specific protein NeuN in macaques, as estimated by optical disector counting. The white matter did not display a high density of COX5Ap staining.
Figure 4 Immunohistochemical staining of COX5A protein in the dorsolateral prefrontal cortex. The distribution of staining using a monoclonal antibody against COX5Ap in macaque monkey (A), chimpanzee (B), and human (C). Panels D-L show double label immunostaining (more ...)
These results suggest that COX5Ap is most abundant in the mitochondria of large-size projection neurons of the neocortex. It is noteworthy that neocortical enlargement in primates is associated with increasing numbers of large pyramidal neurons that are intensely neurofilament H protein-immunoreactive, have long-range projecting axons, and are presumably metabolically costly [28
]. To understand more completely the functional role of COX5A
in neocortical circuits and the possible consequences of positive selection on this gene, it will be necessary to define further the phenotype and distribution of cells in the brain that show high levels of expression for COX5Ap using strategies such as laser microcapture dissection combined with qRT-PCR or mass spectroscopy proteomics in future studies.
Role of accelerated COX evolution
As we have demonstrated, changes in COX5Ap are rare among placental mammals outside the primate clade. Among four positively selected sites identified by the branch site test with posterior probabilities > 0.95, two showed changes in their physicochemical properties: residue 5, which changed from positively charged to neutral on the ape and New World monkey stems and on the baboon terminal (Fig. ); and residue 64, which changed from nonpolar to polar in the anthropoid stem and from polar to nonpolar on the human terminal. Interestingly, the charge neutralizations are parallel to those observed at the binding site for cytochrome c
on COX in anthropoids, where the binding site changes reduced the electrostatic interaction between the docked molecules [15
]. Of note, one amino acid change on the human terminal at position 70 (Ala to Val) is a reversal of a change on the anthropoid stem (Fig. ).
Although the role of the inferred amino acid changes in COX5Ap is currently unknown, it is noteworthy that the changes in this protein are located in residues that are in proximity to other, physically close, COX subunits (Table ). Of the five positively selected sites identified by the branch site test in COX5Ap, seven replacements within 10 Å of these sites have occurred in COX2p and COX4p during anthropoid evolution. Examination of the nature of the changes suggests there has been an electrostatic change such that the interaction between position 5 of COX5Ap and residues 52 and 57 of COX2p has been neutralized; in addition, residues 55 and 56 of this subunit have changed from polar to nonpolar. Furthermore, the potentially interacting, altered residues of COX4p (22 and 63) have changed from neutral to positive charge (Fig. and Table ), forming a positive patch. Of particular interest is that COX4p residue 22, which has changed from Tyr to His, is adjacent to the proposed ATP binding site that includes COX4p 20 [29
Inferred amino acid replacements in COX2p and COX4p that occur in proximity to positively selected sites in COX5Ap among anthropoid primate stem lineages.
An additional consequence of the location of COX5Ap in close contact to the nucleotide-binding domain of COX4p is potential regulation by 3,5-diiodothyronine (T2
), a thyroid hormone produced when the iodothyronine deiodinase D3 acts over thyroxine (T4
has biological activity stimulating mitochondrial respiration by nuclear independent pathways [30
] and is known to bind to COX5Ap and abolish the allosteric inhibition of respiration at high ATP/ADP ratios [32
]. More generally, triiodothyronine (T3
) thyroid hormone is considered to be a major regulator of mitochondrial activity [33
] and has been shown to regulate the expression of a number of COX subunits, inducing functional increases in COX enzyme activity [34
amino acid evolution may thus represent the product of selective pressures acting on the regulation of metabolism. In the context of this hypothesis, we note that thyroid hormones are known to play a critical role in early brain development [35
] and that humans show evidence for a greater affinity of transthyretin and/or thyroxine-binding globulin for thyroid hormone when compared to chimpanzees [38
]. Thus, it is plausible that the interaction of thyroid hormones with COX5Ap and other COX subunits may affect both the regulation of metabolism and the growth and development of important organ systems.