Introduction

Cancer is a dramatic consequence of somatic evolution within the body, posing fundamental challenges to clinical interventions and therapy (Nowell 1976). A cancer cell accumulates mutations, many of which can give the cell a growth advantage over neighboring noncancerous cells. Cancer development is a multistage process that involves the activation of oncogenes and deactivation of tumor suppressor genes (TSGs). To date, there have been about 120 genes found to induce tumorigenesis when mutated (Vogelstein et al. 2013). Typically somatic cells need to accumulate enough mutations to cause cancer, and about 95 % of mutations in cancer-related genes are single-base substitutions (Vogelstein et al. 2013).

Although the idea of cancer as a Darwinian process within an organism is by no means new, relatively little attention has been paid to the evolution of cancer-related genes between species, primarily due to the unavailability of genome sequences from non-model organisms in the pre-genomics era. Since tumors decrease fitness, selection for anticancer adaptations that prevent or delay their occurrence is expected in populations (Burt and Trivers 2006). Indeed, as humans are about 3,000 times larger and live 20–30 times longer than mice, we would not make it out of the womb alive, let alone reach puberty, did human cells have the same probability of becoming cancerous as mouse cells (Peto et al. 1975). Similarly, blue whales (Balaenoptera musculatus) must be capable of suppressing cancer 1,000 times better than humans, given the body size difference (Caulin and Maley 2011; Lichtenstein 2005).

Comparative genomics has already begun to provide examples of divergence in anticancer genetic mechanisms between mammalian species. One of the most striking includes mole rats from genera Heterocephalus and Spalax, the longest living rodents (>20 years), remarkably resistant to both congenital and experimentally induced cancerogenesis (Edrey et al. 2011; Manov et al. 2013). The naked mole rat (Heterocephalus glaber) has undergone an expansion of seven cancer-related gene families relative to rat (Yang et al. 2013), as well as accumulated unique non-synonymous mutations in hyaluronan synthase 2 (HAS2) (Tian et al. 2013) and several DNA repair genes (Kim et al. 2011). Intriguingly, the cancer resistance in subterranean blind mole rats (Spalax) occurs in spite of an amino acid substitution in TP53 that normally induces tumorigenesis in mice and humans (Avivi et al. 2007). While in rodents and primates TP53 is present in one copy only, the gene seems to have multiple paralogs in elephants (Loxodonta africana) (Caulin and Maley 2011).

Thomas et al. 2003 analyzed the relative substitution rate Ka/Ks for 311 human disease genes between human and mouse, and demonstrated a significant purifying selection on cancer genes compared to other disease genes, a result later confirmed on a human-macaque comparison (Blekhman et al. 2008). However, a more accurate and complete view of cancer, as well as the evolution of oncogenes and TSGs is expected to emerge from the new genomes (including human cancer genomes) being added to the public domain. To examine a possible selection pressure operating on cancer-related genes at a finer evolutionary time scale, we compared genomic sequences of 120 oncogene and TSG orthologs among seven hominoid species, including two extinct hominids, Neanderthal (Green et al. 2010) and Denisovan (Meyer et al. 2012). Using the number of non-synonymous substitutions per non-synonymous site (Ka), the number of synonymous substitutions per synonymous site (Ks), and their ratio (Ka/Ks), we found opposing selection pressures exerted on oncogenes and TSGs. Notably, as many as 108 substitutions (SNPs) present in humans only as somatic cancerous mutations occurred in other hominoid species as prevalent germline genotypes different from human.

Materials and Methods

Oncogenes and TSGs

A total of 120 genes classified as either oncogenes or TSGs were used. Genes were classified as oncogenes if they had an oncogene score > 20 % and classified as a TSG if the TSG Score was >20 %, using the 20/20 rule (Vogelstein, et al. 2013). The oncogene score was defined as the number of clustered mutations (i.e., missense mutations at the same amino acid or identical in-frame insertions or deletions) divided by the total number of mutations, while the TSG score was defined as the number of truncating mutations (e.g., nonsense mutations, insertions, deletions, and splice-site mutations that alter reading frame or the protein product length) divided by the total number of mutations (Vogelstein et al. 2013).

Genome Sources and Tools

Genome sequences and annotation files for chimpanzee (CHIMP2.1.4.74), human (GRCh37.74), gibbon (Nleu1.0.74), gorilla (gorGor3.1.74), and orangutan (PPYG2.74) were downloaded from Ensembl (www.ensembl.org), and transcripts with the longest CDS regions were chosen to represent corresponding genes for each species. Alignment results for Neanderthal and Denisovan against the human (hg19) genome were downloaded from Neanderthal genome project (www.eva.mpg.de/neandertal/) and Denisovan genome project (www.eva.mpg.de/denisova). Samtools (Li et al. 2009) was used to generate pileup outputs for Neanderthal and Denisovan, with Phred scores lower than 20 filtered out, and the major allele set to be the genotype for each nucleotide position. Positions with coverage depth lower than five were excluded and marked as missing to avoid an impact on Ka/Ks calculations.

Ka/Ks Ratio Calculations

Sequences of same ortholog sets for all species were grouped together, and Clustal-omega (Sievers et al. 2011) was used to conduct the multiple sequence alignments. Nucleotide sequences were parsed to amino acid sequences before carrying multiple-sequence alignments to avoid a possible frame-shift, and the amino acid sequences were changed back to nucleotide sequences for Ka/Ks calculations. PAML (version 4.7) (Yang 2007) was used to calculate the Ka/Ks ratio (ω) values, setting the model = 0 in the control file of codeml, and applying a single tree to all Ka/Ks estimates.

Results and Discussion

A total of 53 oncogenes and 67 TSGs were used, along with their oncogene/TSG scores (see “Materials and Methods” section, Table S1). The average Ka (0.019) and Ks (0.107) for all genes were lower than those from a comparison of 153 orthologs between human and mouse (0.065 and 0.830, respectively) (Nekrutenko et al. 2002), reflecting closer phylogenetic distances between species in our study. No significant difference in Ka and Ks was found between oncogenes and TSGs (t test, P value = 0.388 and 0.849 for Ka and Ks, respectively, Table 1). However, when we compared average Ka/Ks between oncogenes and TSGs, we found a higher Ka/Ks in TSGs (0.212) than in oncogenes (0.151), an indication that TSGs were presumably under weaker purifying selection compared to oncogenes (t test, P value = 0.039). BRCA1, an important tumor suppressor in breast and ovarian cancers, had the second highest Ka/Ks value (0.670). We then tested for a correlation between Ka/Ks and the TSG score or oncogene score (Vogelstein et al. 2013). Interestingly, genes with higher TSG scores tended to have higher Ka/Ks values (Spearman’s r = 0.223, P value = 0.014), while genes with higher oncogene scores tended to have lower Ka/Ks values (Spearman’s r = −0.269, P value = 0.003), a possible hallmark of opposing selection pressures operating on oncogenes and TSGs.

Table 1 Summary of Ka/Ks analysis
Full size table

Why would selection pressures differ between oncogenes and TSGs? To answer this question, we note that the main distinction between oncogenes and TSGs refers to dominant effects of the former and recessive effects of the latter. The concept of a TSG emerged from a statistical analysis of retinoblastoma incidence in children, leading to the 2-hit hypothesis that inactivating mutations (hits) in both alleles of the retinoblastoma gene are required for this particular cancer to develop (Knudson 1971). Unlike TGSs, oncogenes can contribute to tumorigenesis even only one allele is mutated, amplified, or inappropriately expressed (Vogelstein and Kinzler 2002). As oncogenes confer an immediate disadvantage to the individual even when in heterozygosity, purifying selection will be more efficient in purging mutations across the population. TSG mutations will only start being discerned by purifying selection when in homozygosity, which means that the mutation has to reach considerable frequency in the population. The purging process can be further slowed down or even reversed if the first hit within a TSG has selectively advantageous effects (Iwasa et al. 2005).

We then asked a question whether any substitution different between humans and the other six species in the 120 cancer-related genes have been observed in humans as a cancer-specific somatic mutation, similar to the TP53 genotype that occurs as wild-type variant in Spalax mole rats, but induces tumorigenesis when present in mice and humans (Avivi et al. 2007). We mapped interspecies substitutions against the Catalogue of Somatic Mutations in Cancer (COSMICS, v68), and to minimize chances of the substitutions being rare human alleles, we further excluded all alleles that matched dbSNP database (Build 139) with allelic frequency greater than 0.01. Out of 7,600 non-human genomic substitutions (SNPs), we found a total of 108 substitutions that in humans are present only as somatic cancerous mutations, but in other species can be a germline wild-type (Fig. 1 and Table S2). Among them, 62 were non-synonymous substitutions, out of which 38 were shared by multiple species as wild-type, presumably as ancestral genotype. For example, a human stomach cancer mutation in APC (3512G > A) occurs as a wild genotype among four ape species (chimpanzee, gibbon, gorilla, and orangutan, Fig. 2). It thus seems that a common ancestor of human and all apes carried the “cancer” APC genotype “A”, while the common ancestor of modern humans, Neanderthals, and Denisovans acquired “G” that mutates back to “A” only in cancer. A mutation in TP53 (245C > T or P82L, Fig. 1) found in many types of human tumors (e.g., haematopoietic, intestine, salivary, and upper aerodigestive tract), and shown to have a higher transcriptional activity than wild-type TP53 in adenocarcinoma cell lines (Bisio et al. 2013), was present in gibbons as a wild-type genotype (Figs. 1 and 2). Among the 108 differential SNPs, 59 of them were found in gibbon, 36 in orangutan, 25 in gorilla, 13 in chimpanzee, and only one in Neanderthal and one in Denisovan (Fig. 1 and Table S2), consistent with the evolutionary distances among the hominoid species. Neanderthal had a BRAF variant (1781A > C) typical of many human cancers (skin, stomach, ovary, lung, and intestine), whereas Denisovan had a variant TNFAIP3 (322A > G) found in humans only as a somatic mutation in tumor samples from haematopoietic and lymphoid tissues (Zhang et al. 2013). These results show that there may be alternative genetic routes to cancer and cancer suppression even among relatively closely related species of primates. As a cautionary note, however, we should add that although the substitutions are cancer specific, they do not necessarily need to be cancer causative, or driver mutations.

Fig. 1
figure 1

Numbers (in parentheses next to species names) and examples of non-human genomic variants matching cancer somatic mutations in humans in relation to phylogeny of hominoids. Distance between branches approximately based on estimated divergence time (Chatterjee et al. 2009). Asterisk indicates a substitution present in multiple types of cancer

Full size image
Fig. 2
figure 2

Two examples of interspecies genomic substitutions with the non-human variant present in humans only as somatic cancer mutations. a Gibbon shows a wild-type T at chr17: 7,579,442 (TP53:245), while other species show a wild-type C, with TP53:245C > T being a somatic mutation found in tumor samples from haematopoietic, intestine, salivary, and other tumors; b Chimpanzee, gorilla, orangutan, and gibbon all have a wild-type A at chr5:112,174,803 (APC:3512) while Denisovan, Neanderthal and human have G as wild-type; a human somatic mutation in stomach tumor occurs in same position (APC:3512G > A)

Full size image

The challenge of suppressing cancer increases with larger bodies and longer lifespan, as each proliferating cell within the body is at risk of becoming malignant. For example, osteosarcomas inflict large dog breeds 200 times more frequently than small- and medium-sized breeds (Dobson 2013). However, this principle does not hold above the species level, since large animals do not seem to suffer from cancer more often than small animals, a phenomenon known as Peto’s paradox (Caulin and Maley 2011). If natural selection operating on cancer-related genes is a solution to Peto’s paradox, one could predict that (1) there is substantial genetic divergence between species in the genes and (2) unlike oncogenes, TSGs exhibit signatures of either positive or at least relaxed purifying selection. We showed that both of the predictions were valid with respect to hominoids, despite relatively close phylogenetic distances and small body size differences in the primate group.