We prepared cloned cDNA libraries from 11 M. mulatta tissues derived from nine separate animals. In addition, the liver was independently sampled from one animal each of the M. mulatta, M. nemestrina, and M. fascicularis species. cDNA libraries were prepared by directional lambda-based cloning into Escherichia coli and sequenced using standard fluorescent dye-terminator chemistry. Sequencing was performed from the vector-insert junction distal to the polyadenylate sequence.

A preliminary dataset of 48,642 independent clone sequences were collected as described in Table 1. We screened and analyzed these data as described in Materials and methods. Sequence data quality was assessed using the phred algorithm 9, with a mean of 539 high-quality base-pairs per read over the entire dataset. High-quality sequence bases are defined as those with a computed phred quality value of 20 or greater (Q ≥ 20) and an expected error rate of less than 1%. Of the cloned sequences, 9,219 contain a mammalian polyadenylation consensus sequence followed by a polyadenosine tail 10. Data meeting minimum quality criteria (n = 36,921) have been submitted to GenBank and contribute to all subsequent analyses. Project data and associated information are also publicly available on the project website 11.

We compared each macaque sequence to the mRNA RefSeq 12 component of GenBank using the MEGABLAST algorithm 13. The most similar human sequence was identified as that reference sequence with the most significant match by bit score. In some cases, this method will identify matches between macaque and human sequences that are not orthologs, and so should be interpreted with caution. For all subsequent analyses, those macaque sequences with equally probable matches to more than one distinct human UniGene cluster have been excluded 14. The entire dataset taken together provides a sampling of the putative macaque orthologs for 6,216 human genes (unique human LocusLink IDs), representing approximately 25% of the human gene content by recent estimate 15.

Although libraries were constructed from poly(dT)-primed cDNAs, the dataset includes a significant amount of coding sequence. Of the 6,216 unique human LocusLink IDs that were sampled in macaque, 69.3% include coding sequence (mean aligned coding length = 602 bp), whereas 30.7% include only 5' or 3' untranslated region (UTR) sequence (mean aligned UTR length = 485 bp). Of those 69.3% of genes with sampled coding sequence, the average extent of coding sequence coverage in the macaque database is 49.9% (data not shown).

We used the initial alignment information from the above data to define a subset of sequences whose alignment with their best human match extended 150 bp in each direction around a well defined stop codon. This dataset was used to compute the distribution of sequence similarity between macaque and human as represented by the histograms in Figure 1. The use of this constrained dataset permitted a direct comparison between the distributions for coding and noncoding sequence in the vicinity of the stop codon. Data for 1,180 macaque-human alignments are included in this analysis. Sequence-similarity distributions are not normal, with a modest tail toward lower values. The average degree of similarity for coding sequence is 97.79 ± 1.78% and 95.10 ± 4.15% for the 3' UTR. This analysis excludes data where the macaque stop codon was either mutated or in a different location relative to the human reference sequence. This analysis uses the 3' UTR proximal to the stop codon as a surrogate for all untranslated sequences. However, human-chimp comparative analysis suggests that the 5' UTR may be more divergent between species than other gene regions 16. We did not have a sufficiently sized dataset to locate and independently test conservation of the 5' UTR.

In order to determine if local regions of poor data quality contribute to biases in the computed degree of sequence similarity, we recomputed the histogram using alignments composed of only high-quality (Q ≥ 20) sequence. Constraining the dataset to include only high-quality bases (n = 633 sequences) did not result in significant differences in either the shape or the mean of the distributions (Figure 1).

To provide a reference dataset with which to evaluate the current results, we computed the degree of sequence similarity between human and Pan troglodytes (chimpanzee) using the same method as above. This analysis was performed using chimpanzee expressed sequence tag (EST) and cDNA sequences, as most currently available chimpanzee reference sequences are computationally predicted and therefore lack data from the 3' UTR. However, our chimpanzee-human analysis was hampered by the relative paucity of chimpanzee full-length cDNA and EST sequence in the public databases. There are currently only 209 full-length chimpanzee cDNA sequences and 6,930 EST sequences of varying quality in GenBank.

These data together provide a sampling of the 150 bp proximal and distal to the stop codon for only 134 human genes. On the basis of this small dataset, the degree of nucleotide identity between human and chimpanzee for coding and 3' UTR sequences is 98.3 ± 3.0% and 97.65 ± 3.2% respectively (Additional data file 1). As expected, the distribution of sequence similarity is strongly biased toward larger values, with 59.0% of sampled chimpanzee coding sequences and 46.3% of 3' UTR sequences identical to their best human match over the 150-bp window. The distribution of sequence identity between human and chimpanzee is presented in Additional data file 2.

We expect that most observed nucleotide substitutions between macaque and human within coding sequence will be conservative. To evaluate the degree of similarity between human and macaque at the amino-acid level, we analyzed macaque sequences that overlapped with their best-matching human reference sequence by at least the terminal 450 bp proximal to the stop codon. Data from the terminal 450 bases were favored for this analysis in order to include more of the overall dataset and to be directly comparable to our previous nucleotide-based analysis. We also constrained the dataset to again include only high-quality bases. The distribution of amino-acid similarity was as expected, given the distribution of nucleotide similarity, with a bias toward higher values (Figure 2). The mean similarity between macaque and human protein sequences over the aligned window is 96.83 ± 4.95%. A relaxation of data quality constraints resulted in a broadening of the distribution toward lower values (data not shown).

We identified 21 high-quality macaque sequences with very weak amino-acid similarity (< 90%) to their best-matching human reference sequence (Table 2). Of these, 15 are either highly expressed in placenta or immune tissue (peripheral blood mononuclear cells (PBMCs) or spleen mononuclear lymphocytes) and/or are associated with pregnancy or the immune response. The observation of poor sequence identity for immune genes is not surprising, as increased divergence and evidence for positive selection have previously been reported for members of this group 1718. The most interesting example of divergence from our study is APOBEC3C, a member of the cytidine deaminase family. Rhesus macaque APOBEC3C is only approximately 85% identical to its putative human ortholog. Members of the APOBEC family are important mediators of lentivirus infection 19, and accelerated evolution has been reported for several members of this gene family 20.

We also identified ten placentally expressed pregnancy-related transcripts with very weak similarity to their putative human ortholog. Prominent among these are the pregnancy-specific glycoproteins (PSG5 and PSG11). For example, the best macaque match to human PSG11 shows only 68% identity and is not better matched to any other member of the human PSG family. Other placentally expressed weak orthologs include the growth mediators angiogenin (ANG) and growth hormone 1 and 2 (GH1 and GH2). Episodic accelerated evolution has previously been reported for both angiogenin and the growth hormones, although its biological and developmental implications are not well understood 2122.

We compiled amino-acid similarity data into gene functional groupings using the 'biological process' classifications from the Gene Ontology (GO) Consortium 23 (Table 3). Data are shown for only those classes containing three or more entries. The data reveal a wide degree of variation in class-specific values of sequence similarity between human and macaque. Highly conserved classes include those involved in intracellular signaling, small GTPase-mediated signal transduction, translation initiation, and protein biosynthesis and folding. Poorly conserved biological process groups include pregnancy and immune and inflammatory response. We note that the small size of the dataset is reflected in large standard deviations for several classes of genes.

These data share similarity with recent comparative analyses between human and chimpanzee 424. For example in chimpanzee, a high degree of sequence conservation and low rates of nonsynonymous substitution were found for several biological classes, including protein transport, small GTPase-mediated signal transduction, regulation of DNA-dependent transcription, intracellular signaling, and glycolysis. However, not all biological functional groups demonstrate consistent conservation among the three species. For example, the signal transduction biological class is highly conserved between chimpanzee and human, whereas its conservation between macaque and human does not significantly deviate from the mean over all classes.

Our dataset includes sequence data from nine M. mulatta, one M. fascicularis, and one M. nemestrina. The breadth of the dataset provides an opportunity to conduct a preliminary analysis of the polymorphism frequency within M. mulatta and the degree of nucleotide divergence between macaque species. We estimated the polymorphism frequency within M. mulatta by assembling sequencing reads from multiple animals for the same gene using phrap 9. Polymorphisms were identified by a modified version of phred that calls two alleles at each base in the assembly and assigns each allele a quality score based on combined phred quality values (C.M., unpublished work). High-scoring polymorphisms were manually verified and are presented in Table 4 for a sample of 24 genes. This analysis includes both coding and noncoding transcribed sequences. The average nucleotide diversity (π) for this gene set in M. mulatta is 15.8 ± 12.5 × 10^-4 25. A large standard deviation in nucleotide diversity across genes is consistent with reports from other primate species 262728. The animals included in this analysis were primarily bred from wild-caught parents of Indian origin. A more comprehensive determination of nucleotide diversity will require sequence data from a greater number of genes and animals from multiple geographic locations.

We were also able to evaluate the degree of nucleotide sequence divergence between the three macaque species for a sample of 21 genes in this dataset (Table 5). Phred and phrap were again used to assemble overlapping sequences from multiple species and to identify species-specific variants that were then manually confirmed. Given the high degree of nucleotide similarity among the species and the small sample size, the three species did not differ beyond the measured standard deviations. However, M. mulatta and M. fascicularis appear more closely related to each other than either is to M. nemestrina, with an average sequence divergence between the two of 0.380 ± 0.380%. The degree of sequence divergence between M. mulatta and M. nemestrina is 0.588 ± 0.438% and 0.522 ± 0.419% between M. fascicularis and M. nemestrina. However, the dataset is not large enough for any of these pairwise differences to reach statistical significance.

Analysis of the entire dataset revealed a small number of transcribed macaque sequences that had little or no sequence similarity to any human cDNA or genomic sequence (Table 6). We speculate that some of these macaque sequences are without orthologs in the human genome. The observation of species-specific transcribed sequences among the primates is consistent with recent comparative analysis between human and chimpanzee 429. Although an absolute determination of species specificity will require a complete macaque genome sequence, we conducted preliminary computational and PCR-based analyses to test the presence or absence of these sequences in the human and other primate genomes.

As above, we used MEGABLAST to test each macaque nucleotide sequence for one or more significant hits to the human EST or genome databases. The absence of an orthologous human sequence was defined as either no significant MEGABLAST hit in the human subset of GenBank or hits with sequence identity less than three standard deviations below the mean as measured over the entire dataset (Figure 1). Because the data were not normally distributed, the identity cutoff (approximately 92.2%) was computed using the geometric mean, which relies on a logarithmic transformation of the data. All sequences meeting this cutoff definition were also outliers based on Tukey's test 30.

We selected eight of the resulting macaque sequences for PCR-based analysis using a number of primate and human genomes (Table 6, Figure 2). The purpose of this analysis was simply to verify the presence or absence of the observed sequences in a panel of primate genomes. Selected primers had an average computed annealing temperature of 59.6 ± 0.9°C with an average amplified length of 108 ± 12 bp (Materials and methods). For each primer pair, PCR analysis was conducted at several annealing temperatures between 55 and 60°C. Genomic DNA was selected from independent M. nemestrina and M. mulatta animals in order to confirm the presence of these sequences in multiple independent genomes. Of the eight tested primer pairs, two resulted in amplification of consistent bands in both human and macaque genomic DNA, two were indeterminate in human but present in the macaques, and four, while obviously present in the macaque genomes, resulted in no consistent human-specific product under any cycling conditions.

The eight tested sequences fall generally into three categories: those with weak sequence similarity to the human genome or human-derived ESTs (class I), those with weak sequence similarity only to genes and proteins from nonhuman species (class II), and those with no significant amino-acid or nucleotide sequence similarity to any GenBank nucleic acid or protein sequence (class III).

Those with weak similarity to human sequences (class I) include CX078602, a 657-bp cDNA sequence derived from macaque liver with 79-87% nucleotide sequence identity to CYP2C18 from several mammalian species. Its closest matches to human are two regions of 86-93% identity to human chromosome 10, one of which contains four cytochrome P450 2C genes. PCR-based analysis failed to amplify a consistent band from any primate species other than M. nemestrina, M. mulatta, and Lagothrix lagotricha (woolly monkey) (Figure 3a).

Likewise, CX078592 from brain demonstrated 88-90% nucleotide similarity to the IL15RA gene and other immune-derived transcripts, as well as to a region of human chromosome 10 containing IL15RA. PCR primers derived from this sequence amplified multiple specific products from macaque, human, and other primates (data not shown). Similarly, CX078596 from placenta, although having no significant match to any human EST, demonstrated significant similarity to a region of human chromosome 22. CX078596 contained a clear mammalian polyadenylation signal and poly(A) tail, and primers derived from this sequence amplified an appropriately sized product from macaque. Alignment of this sequence with human chromosome 22 revealed a 284-bp insertion in human relative to macaque, which was reflected by amplification of a proportionately larger product in two human genomic DNA samples (data not shown). Finally, although CB552301 from spleen demonstrated significant sequence identity to regions of human chromosomes 4 and 15 and multiple ESTs from UniGene cluster Hs.459311, we failed to amplify a specific product from any primate species using primers derived from this sequence (data not shown).

The second class of sequences (class II) in Table 6 had no identified human match, while demonstrating weak sequence identity to nucleic acid or protein sequences from other species. For example, CX078598, a 670-bp transcript from PBMCs, demonstrated weak amino-acid identity (67%) to the endogenous retrovirus (ERV)-BabFc^env envelop polyprotein, a member of the ERV-F/H family of primate retroviruses 31. PCR with primers derived from CX078598 under a variety of thermal cycling conditions resulted in the consistent amplification of a product of expected size from only M. mulatta and M. nemestrina (Figure 2b). Similarly, CX078591 from macaque brain demonstrated weak amino-acid identity (20-45%) to ariadne homolog 2 (ARIH2/TRIAD1) from rodents and to two unnamed proteins from the puffer fish Tetraodon nigroviridis. Primers derived from this sequence amplified the appropriately sized product only from macaque genomic DNA (data not shown).

The last class of sequences (class III) in Table 6 demonstrated no significant similarity to any protein or nucleotide sequence in GenBank (represented by CB555845 and CB552531). Both showed evidence of a mammalian polyadenylation consensus sequence near their 3' terminus, with CB552531 additionally demonstrating a clear poly(A) tail. CB555845, a 485-bp sequence from spleen, amplified expected products from both M. nemestrina and M. mulatta. However, this clone was ultimately scored as indeterminate because of its consistently weak amplification of a discrete product from all hominids including human (Figure 2c). CB552531 amplified products of the expected size from macaque species and from Ateles geoffroyi and Lemur catta, but not from human (data not shown).

It is important to note that PCR-based analysis of divergent sequences is subject to a variety of influences and may result in different conclusions under different conditions. Furthermore, we cannot rule out the possibility that one or more of the sequences in Table 6 are alternatively spliced relative to human, pseudogenes, or genomic DNA contamination. However, each clone sequence in Table 6 demonstrated similarity to known expressed sequences or a polyadenylation consensus sequence and poly(A) tail at their 3' terminus upon complete sequencing of the clones.

Genome-based technologies such as DNA microarrays are now commonplace in human biomedical research. Similarly, species-specific arrays exist for model organisms such as the mouse and rat, for which a considerable amount of genome information is available. In contrast, researchers wishing to carry out gene-expression analyses on nonhuman primate cells or tissues are currently forced to use human DNA microarrays. As part of our effort to bring genome-based technologies to researchers using nonhuman primates, we have used ESTs generated by this project to construct a rhesus macaque-specific oligonucleotide microarray.

Oligonucleotides were designed as described in Materials and methods and arrayed onto glass slides by Agilent Technologies. Briefly, macaque cDNA sequences were assembled into 9,344 distinct clusters using The Institute for Genome Research (TIGR) clustering tools 32. From these, 7,973 macaque-specific oligonucleotide probes were identified for inclusion on the array. These probes represent the putative macaque equivalent of 3,519 unique human UniGene clusters 14 and 3,045 unique human RefSeqs 12. To quality control the microarray, we measured tissue-specific differences in gene expression as a means of evaluating whether the oligonucleotides were successfully binding target sequences. For these experiments, we hybridized the microarray with probes derived from RNA isolated from various rhesus macaque tissues. Probes were paired in different combinations and two dye-flipped technical replicates were performed for each pair of samples. Of the 7,973 rhesus macaque oligonucleotides present on the microarray, 6,215 showed differential expression (equal or greater than twofold; P ≤ 0.01) in at least one of the three experiments.

Plots of the log-transformed ratios for genes in each experiment that showed an equal to or greater than twofold difference in expression between two tissues are shown in Figure 4. In each plot, points are colored according to the source library of the sequence used to derive the corresponding oligonucleotide. From this analysis, it is apparent that the majority of genes that were more highly expressed in the spleen correspond to sequences derived from the spleen cDNA library. Similarly, the majority of genes that were more highly expressed in the brain correspond to sequences derived from the brain cDNA library. These results show that a majority of the oligonucleotides were successfully binding target sequence. In addition, it is likely that many of the oligonucleotides that did not measure differential gene expression in these experiments are also successfully binding target sequences, as not all genes would be expected to be expressed in all tissues or to show differential levels of expression between the tissues analyzed.