Phylooncogenomics: Examining the cancer genome in the context of vertebrate evolution.

Currently, human cancer genomics is making great progress, and many mutations of new cancer driver genes have been detected at an unprecedented rate in a variety of human cancers. Many details of the genetic alterations in cancer cell genomes have been revealed by the massively parallel sequencing. Long-lasting aneuploidy caused large-scale somatic copy number alterations remains a difficulty as there are too many genes located on such big chromosomal fragments, and this cannot simply be solved by increasing sequencing depth and tumor sample numbers. Comparative oncogenomics may provide us with a solution to this problem. Here, we review some of the common animal cancer models and propose to analyze cancer cell genomics in vertebrate phylogenetic backgrounds. Thus phylooncogenomics may provide us with a unique perspective on he nature of cancer biology unattainable by single species studies.

Introduction

Cancer is essentially an aging-related disease, with most of adult cancers found in the latter half of the lifespan. With the increase of the expected average lifespan of human beings, cancer will continue to be one of the major health threats in the future due to its worldwide prevalence and the lack of effective treatments (Yancik, 2005). In the past several decades, substantial research endeavors have been made, because the effectiveness of controlling cancer depends on our knowledge of the nature of the disease. For example, upon learning that cancer cells usually divide more rapidly compared to normal cells, chemotherapy targeting cellular proliferation was created and it remains the most common treatment regime of cancer afterwards (Varmus, 2006). With advances in molecular and cellular mechanisms of tumorigenesis, targeting therapies using less toxic agents such as hormones, antibodies, and enzyme inhibitors were invented. The best example is imatinib, a kinase inhibitor which specifically inhibits the chimera protein, ABL–BCR, in some forms of chronic myeloid leukemia (O'Brien et al., 2003). More importantly, this specific targeting approach provides great hope in conquering this notorious disease.

One of the prerequisites of cancer targeting therapy is knowing the specific alterations that are only or mainly present in cancer cells, thus we can specifically target cancer cells in effective ways. Towards this rationale, many levels of alterations have been explored, such as histology, biochemistry, metabolism, and genetics (Pierce et al., 1978). With the advances of nucleic acid sequencing and microarrays, genetic changes in the cancer genome have been discovered at an unprecedented rate. These changes include: single nucleotide point mutations; frame reading shifting mutations caused by both insertions and deletions; chromosomal rearrangements including translocations, inversions, and copy number changes of protein coding and regulatory regions and may occur simultaneously in a single tumor (Stratton et al., 2009). Additionally, the mutation repertoires can be different among cancer cells in a single tumor; this is even evident by the different chromosome numbers. It has been estimated that a tumor can generate billions of mutations (Klein, 2006). Clearly, not all genetic alterations equally contribute to cancer progression. Mutations that are positively selected for and are advantageous in growth, tissue invasion and metastasis are defined as “drivers.” Mutations that are byproducts of genomic instability, are not selected for, and do not confer cancer development are named “passengers.” Very recently, a “mut-driver” definition has been proposed to precisely describe the genes whose mutations could cause cancers (Vogelstein et al., 2013).

Identifying cancer drivers is one of the central goals of current cancer research (Stratton et al., 2009), as not only are they essential for understanding the molecular mechanisms of cancer biology, but they may also serve as potential therapy targets and markers for diagnosis and prognosis. The current list of known “cancer genes” is about 488, according to the Cancer Gene Consensus (Futreal et al., 2004, Santarius et al., 2010). This list appears far from complete, as many new cancer driver genes are constantly being discovered with the completion of more and more cancer genomes. The International Cancer Genome Consortium (ICGC) network of cancer genome projects was initiated to target the mutational repertoire in 50 of the most common human cancer types (Hudson et al., 2010). This was only possible because of the completion of the Human Genome Project and advances in massive parallel sequencing technologies. To date, many important discoveries have been made or confirmed using these high-throughput technologies. However, even with these new technologies, identifying cancer driver genes still remains challenging due to the heterogeneities and mutational hierarchies.

Heterogeneity in cancer genome and somatic evolution

Tumor development is thought to occur as a somatic evolutionary process in which mutations are accumulated in a sequential manner (Merlo et al., 2006, Nowell, 1976). Like evolution on the whole organism level, the mutation process in cancer is stochastic. Very recently, evidence of Darwinian evolution has been confirmed in human pancreatic cancer and leukemia using sequencing and microarray (Campbell et al., 2008, Campbell et al., 2010, Notta et al., 2011, Sisman and Geyikoglu, 2008, Yachida et al., 2010). Selections from the micro- and macro-environments of the cells determine which mutation(s) are retained and which ones are eliminated (Gillies et al., 2012, Merlo et al., 2006). The mutations giving cell growth advantages over surrounding cells are generally selected, and thus they are likely to be cancer drivers that give rise to the tumor cells' hallmarks (Hanahan and Weinberg, 2011, Stratton, 2011, Stratton et al., 2009). Conversely, many other bystander gene mutations cannot be eliminated rapidly enough and thus they stay in the cancer genome as passengers. Therefore, multiple levels of genetic heterogeneity (intra- and inter-tumoral, inter- and intra-metastatic, and inter-patient heterogeneities) were frequently revealed by traditional cytogenetics and recent genomic sequencing analysis (Almendro et al., 2013, Heppner, 1984, Marusyk et al., 2012). For example, it has long been recognized from earlier cytogenetic studies that there is almost no consistent karyotype in different cancer cells within the same solid tumor (Wolman, 1986). Similarly, the genes and genomes have recently been noticed in a similar situation. If tumorigenesis really is an evolutionary process, the evolutionary biological approaches, such as phylogenetic analysis, should be able to be applied to trace the natural history of cancer cells. Indeed, recently genomic sequencing and copy number analysis methods successfully tracked the cancer cell development process and the relationships between the original tumor and subsequent metastatic tumors (Campbell et al., 2008, Gerlinger et al., 2012, Navin et al., 2010, Tsao et al., 2000). This kind of information on cancer cell nature history not only is very important for us to understand the dynamic processes of tumor formation, but also might serve as guidance for therapeutic strategies.

Animal models of human cancers

Animal models play a very important role in our understanding of cancer biology, such as in the identification of novel cancer drivers, validating potential oncogenes and tumor suppressor genes, investigation of molecular mechanisms, and testing new cancer therapy strategies. Currently, there are several popular cancer models in the cancer research community.

The mouse model has a long history in cancer research as the most extensively used model system due to its mature genetic manipulations, relative short breeding time and the availability of inbred strains. For example, inducible and tissue-specific gene manipulation can be achieved using mouse embryonic stem cell and advanced cre-recombinase mediated knockout and knockin technologies (Cheon and Orsulic, 2011, Frese and Tuveson, 2007). In addition, N-ethyl-N-Nitrosourea (ENU) chemical, murine retrovirus-mediated (murine leukemia virus and mouse mammary tumor virus), and transposon-based mutagenesis (Sleeping Beauty and PiggyBac) have also been utilized in the mouse to identify novel cancer driver genes (Ding et al., 2005, Dupuy et al., 2005, Hrabe de Angelis et al., 2000, Kool and Berns, 2009, Nolan et al., 2000). Moreover, chromosomal engineering strategies have been successfully applied to the mouse model in order to mimic bigger regions of chromosomal deletion and duplication that frequently occur in human cancer genomes (Yu and Bradley, 2001, Yu et al., 2006). Since the mouse is a different species, it is not surprising that many differences between human and mouse tumor biology have been reported in the literature. For example, the common laboratory mouse (Mus musculus) possesses more active telomerase, and thus the tumors in genetically engineered mice generally possess fewer genetic alterations, including aneuploidy, when compared to corresponding tumors in humans. In order to make the mouse tumor cells aneuploid, like those of human tumors, more genes have to be manipulated in multiple pathways (Maser et al., 2007, Moens, 2008). Differences in human and mouse tumor biology are also evident with regard to the tumor type spectrums within the same orthologous cancer gene mutations. For example, mouse mutations to p53 result in multiple sarcomas and lymphomas, while human p53 mutations result predominantly in carcinomas and some sarcomas (Jacks et al., 1994). In regard to the number of essential genetic alterations to convert normal fibroblasts to tumorigenic cells, a minimum of six alterations is needed for human cells; only two are sufficient for mouse cell transformation (Rangarajan et al., 2004). Interestingly, this fits the Peto's paradox hypothesis, which suggests that there are stronger tumor repressor mechanisms in larger longer-lived animals than in smaller sized animals with shorter lifespans due to natural selection (Peto et al., 1975). Recent comparative genomics revealed that the mouse and human share about 15,213 genes. The mouse has 2785 unique genes that do not have homologous genes in human; conversely, human has 3189 genes that the mouse does not possess (Howe et al., in press). Though greater knowledge has been achieved using the mouse model, clearly human cancers cannot be completely recapitulated using this model. Thus, caution should be made when general conclusions are extrapolated from single species data.

The zebrafish is rapidly becoming a popular model organism for studying cancer and a number of tumor models have been made by the transgenic expression of oncogenes or via the mutation of tumor suppressor genes (Liu and Leach, 2011, Mione and Trede, 2010). The evidence that fish can mimic human cancer comes from multiple sources. First, human oncogenes and tumor suppressor genes can induce tumors in zebrafish. Human oncogene active mutants (e.g. c-myc, b-raf, k-ras) have been successfully used for creating zebrafish cancer models (Dovey et al., 2009, Langenau et al., 2003, Patton et al., 2005). Likewise, loss of function in the tumor suppressor gene, tp53, resulted in the formation of cancer in the zebrafish although it gave rise to a different tumor spectrum (Berghmans et al., 2005, Parant et al., 2010). This phenomenon that hereditary mutations in the same orthologous cancer gene may give rise to different tumors spectrums in different species is supported in the use of the mouse model. For example, certain MET gene mutations lead to hereditary renal carcinoma in human, while in the mouse they give rise to various sarcomas depending on the types of mutations (Graveel et al., 2005). This may be caused by the varied tissue specificity of the cancer genes in different animal models. Second, zebrafish tumors have been shown to have similar gene expression signatures to human tumors. For example, the chemically induced fish liver tumor, and conditional k-ras induced melanomas and rhabdomyosarcoma have been reported to share similar transcriptomes with human corresponding tumors, respectively (Dovey et al., 2009, Lam et al., 2006, Langenau et al., 2007). Third, the fish tumors are cytogenetically aneuploidy like humans. Very recently, after strictly examining zebrafish malignant peripheral nerve sheath tumors (MPNST) that were induced by ribosomal protein (rp) and tp53 mutations, we revealed that these tumors are highly aneuploid, and that the aneuploid chromosomes show preferential presentations within a tumor and among fish tumors. For example, within a single fish tumor there is often preferential overrepresentation of chromosome 25 with simultaneous underrepresentation of chromosome 15 (Zhang et al., 2010). This finding is very similar with the aneuploidy chromosomes in human solid tumors (Wolman, 1986). Moreover, there are also recurrent focal copy number alterations (CNA) as well as human solid tumors. We succeeded in finding cancer driver genes within recurrent focal CNAs using array-comparative genomic hybridization (CGH) and Illumina deep sequencing technology (Zhang et al., 2010). Similar to the mouse cancer model, genetic manipulation such as forward genetic screening with ENU, retroviruses, and targeted mutagenesis have also been successful in zebrafish (Amsterdam et al., 1999, Mullins et al., 1994). It is not surprising that many developmental processes and disease mechanisms were found to be similar to humans. Zebrafish share 12,897 strict homologous genes with humans. This gene number is just slightly less compared with the orthologous genes between mouse and human (Howe et al., in press). Thus, it is reasonable to expect that certain properties of human cancer biology can be recapitulated using zebrafish cancer models.

While the mouse and zebrafish compose the two major cancer model organisms, there exist other vertebrate cancer models. Among them, the dog is the most prominent one. Compared to tumors of mouse and zebrafish, most dog tumors are spontaneous, like the ones in the human population (Rowell et al., 2011). Similar to humans, the cancer rate in dogs has increased in recent years due, in part, to longer life expectancy resulting from better nutrition and veterinary care. Many human cancer gene mutations, like TP53 and KIT, have subsequently been identified in canine cancers (Paoloni and Khanna, 2008). Genetic variations within each dog breed have been greatly reduced compared to humans and this makes them an ideal model system for genetic association studies of disease mapping and pharmacogenomics. One disadvantage of the canine models is the few genetic manipulation approaches available to test causal effects of candidate genes. Human, mouse, and dog genomes share about 13,816 genes with a 1:1:1 orthology (Lindblad-Toh et al., 2005). Thus, similar to the mouse and zebrafish, human cancer biology cannot be completely recapitulated in the dog.

Again, there is no single species disease model that completely mimics human disease pathogenesis, as every species has a unique ecological and evolutionary niche. Employing models from multiple species and proper comparative analysis are essential for understanding the mechanisms of human diseases including cancer.

Comparative oncogenomics

Currently, the method predominately used for finding cancer drivers is limited to finding statistical significance among certain number of tumors, usually within a single species. The approach is greatly affected by the very small signal-to-noise ratio as well as the choice of sampling sites within a given tumor. As such, this method requires relatively large sample sizes (e.g. 250 samples per tumor in human cancer genome projects) (Hudson et al., 2010) and it is very inefficient when dealing with large-scale genetic alterations. This is especially true for large and whole chromosomal copy number changes, which are mainly caused by aneuploid chromosomes in cancer cells. Aneuploidy is one of the most common genetic changes in human tumors, especially solid tumors (Mitelman et al., 2013). The discovery of the chromosome changes can be traced back to a century ago. Theodor Boveri and David von Hanseman proposed the first hypothetical cancer origin based on chromosome number changes (Hardy and Zacharias, 2005). However, the biological meaning of this aneuploidy remains unclear. Whether aneuploidy is causative or simply a byproduct of genomic instability has been actively debated as the etiology of human cancers (Duesberg, 2005). Accumulating evidence shows that these chromosomal abnormalities are indeed not random and suggests that they play certain roles during tumorigenesis. In terms of cytogenetics, aneuploidy is synonymous with whole chromosome or arm-level copy number changes that are used for modern genomic level studies, usually with array comparative genomic hybridization (aCGH). Recent large-scale investigations showed that there were indeed common chromosome changes in human solid tumors and some alteration patterns are even shared by many solid tumors (Baudis, 2007, Beroukhim et al., 2010). In most of the published CNA analysis using a single animal model species, only the small-sized focal CNAs (or minimum common regions) are informative, as they contain manageable numbers of genes for candidate gene functional validations. In contrast, identification of cancer drivers on the large-sized CNAs, usually chromosome or chromosomal arm lengths, remains a challenge. These regions often contain hundreds to thousands of genes, each requiring functional studies — well beyond the ability of most average-size laboratories. It has been proposed that high-throughput screening using RNAi and ORF libraries could be combined for this purpose in order to narrow down the scope (Boehm et al., 2007, Hahn et al., 2009). Others have shown that integrating the gene expression data with CNAs could lead to this same purpose (Akavia et al., 2010). Here, we only focus on an alternative approach, comparative genomics and genetics, for identification and biological validation of cancer driving genetic events, using copy number alterations as an example.

Animal models, especially the mouse cancer model, have been extensively used for many years to study many aspects of human cancers, from the discovery of tumor initiation genes' to testing cancer therapeutics (Frese and Tuveson, 2007). Recently, it has been shown that some mouse tumors also contain many CNAs, including both those at the whole chromosome level and more focal changes (Kim et al., 2006, Zender et al., 2006). Like humans, recurrent focal changes in the mouse genome are useful in identifying cancer drivers. More importantly, the mouse models have helped pinpoint cancer driver genes by comparing the gene contents of recurrent CNAs in both human and mouse tumors. Likewise, it has been shown effective with the focal CNAs in mouse liver tumors, melanoma, and osteosarcoma (Kim et al., 2006, Maser et al., 2007, Molyneux et al., 2010, Zender et al., 2006). However, this approach reaches potential limitations when applied to larger regions, especially at whole chromosome levels. This is because the gene locations in humans and mice are so highly conserved that the majority of the gene linkages are nearly identical. Thus it is still not efficient in eliminating the passenger genes. Other currently available mammal cancer models, such as the dog, are similar to the mouse for comparative purposes due to their close evolutionary distance.

Due to these common features among fish, human and mouse tumors and the conservation of genes' functions, it is reasonable to assume that most driver genes within the preferential whole-chromosome and focal-CNAs are also shared among vertebrates. As these and other data show that, some preferences are specific to one type of tumor while some preferences are shared by many different tumor types. Because the orthologous gene locations along the chromosomes were reshuffled during evolution in different vertebrates, phylogenetically close species share large blocks of conserved syntenies (Kasahara et al., 2007, Kirkness et al., 2003). For example, the mouse and human genomes share large regions of conserved chromosome fragments containing similar gene contents (Fig. 1A). This is also the rationale for using the mouse to model human aneuploid syndromes (e.g., Down syndrome) (Reeves et al., 1995). On the other hand, zebrafish and humans share conserved regions to a much lesser degree; instead, most of the conserved syntenies contain much smaller fragments (Fig. 1B). It is apparent from numerous phylogenetic studies that fish genes were widely dispersed among the 23 pairs of human chromosomes during evolution (Kasahara et al., 2007, Schartl et al., 2013, Woods et al., 2000). Conversely, genes that lie on a particular human chromosome are often found together on a single mouse chromosome. The small syntenies between the zebrafish and human genome make the zebrafish cancer models uniquely useful in identifying cancer drivers on large CNAs by comparative oncogenomics. This is because the overlapping of genes between fish and human preferential whole-chromosome and focal-CNAs is much less.

Fig. 1

Gene positions were reshuffled in human, zebrafish and mouse genomes. (A) Circos scatter plot for the chromosomal locations of orthologous genes between mouse (Mmu) and human (Hsa). (B) The syntenic relationship between zebrafish (Dre) and human (Hsa). The plot was generated using the published synteny database (Catchen et al., 2009). The orthologous gene counts in the shared chromosome fragments are connected with lines.

In human malignant peripheral nerve sheath tumors and many other human cancers, the most frequently over-represented chromosome is 7, followed by 8q, and 17q (Beert et al., 2011, Brekke et al., 2010). In zebrafish MPNSTs, the most frequently over-represented chromosome is 25, followed by 11, 10, 23 and 22 (Zhang et al., 2010). Human chromosome 7 alone possesses ~ 1750 genes. In contrast, there are only 68 on fish chromosome 25, 18 on chromosome 11, and 34 on chromosome 10 (Fig. 2). As the presence of unbalanced chromosomes is the result of tumor evolution selection, these common genes could be candidates that underlie the growth advantage conferred by extra copies of human chromosome 7. The same logic can be applied to genes that lie on chromosomes that are under-represented in both fish and human tumors and generating a list of genes whose under-representation favors tumor growth. In humans, the MET oncogene lies on chromosome 7. Amplification of chromosome 7 carrying a mutated form of MET underlies an inherited cancer in humans (Lee et al., 2000, Schmidt et al., 1997). It is thought that the presence of wild-type MET may explain, at least in part, the frequent over-representation of chromosome 7 in many human cancers (Grabellus et al., 2010, Jenkins et al., 1998, Zhuang et al., 1998). Strikingly, the zebrafish met gene lies on fish chromosome 25. One of other genes also presented in the overrepresented zebrafish chromosome 25 is fgf6a. We have shown that fgf over-expression accelerates the onset of MPNSTs in combination with a p53 mutation, demonstrating that fgf is probably a driver in fish MPNSTs (Zhang et al., 2010). This result indicates that there are at least two drivers on chromosomes that are over-represented in fish tumors and it is possible that there are more. This discovery is consistent with the report of multiple tumor suppressors in chromosomal region 8p22 (Solimini et al., 2012, Xue et al., 2012). Combined with genetic functional validations, the zebrafish model provides us a unique model to catalogue human cancer drivers on the recalcitrant large scale CNAs.

Fig. 2

Dot plot (generated from the synteny database, Catchen et al., 2009) shows the gene position relationships between human chromosome 7 and all zebrafish chromosomes.

Phylooncogenomics, beyond the CNAs

Although mouse–human and zebrafish–human cancer genomic comparisons are helpful to pinpoint the cancer driver genes on large CNAs in corresponding human tumors, this comparative approach should not be limited to two species only. Comparison of cancer genomics from multiple vertebrate species will be more powerful to narrow down the scope of cancer driver genes on the large scales CNAs. The early vertebrates, chondrichthyes, may be particularly interesting, as their genomes were not duplicated as teleosts (Venkatesh et al., 2007). The increased phylogenetic distance of these cartilaginous fishes to human could be useful to breakdown the chromosomally linked drivers and passengers. However, at the current time, the zebrafish is particularly useful as it is a model organism whose genomic and other biological infrastructures are much more mature compared to other non-model organisms. With advances in current sequencing technologies, we expect that other fish tumor models will play their roles in the future. For example, platyfish and lamprey genomes have recently been published (Schartl et al., 2013, Smith et al., 2013). Ultimately, multiple species tumor comparison could further reduce the passenger genes in a given tumor, thus we coined the word “phylooncogenomics” to accommodate this approach (Fig. 3). Common evolutionarily conserved cancer drivers and species-specific drivers can be identified and functionally tested in this way. For example, cancer genomes of MPNSTs from human (Beert et al., 2011, Brekke et al., 2010), mouse (Cichowski et al., 1999, Vogel et al., 1999), Tasmanian devil (Murchison et al., 2010, Murchison et al., 2012) and zebrafish (Zhang et al., 2010) DNA could be compared in the future to identify common shared cancer drivers and biological tumorigenic mechanisms. Another potential candidate for tumor phylooncogenomic studies could be melanomas from human (Gast et al., 2010, Stark and Hayward, 2007), mouse (O'Hagan et al., 2003), zebrafish (Patton et al., 2005), medaka (Schartl et al., 2010) and platyfish (Xiphophorus) (Meierjohann and Schartl, 2006). The platyfish itself is a long used melanoma model as certain hybrid species naturally develop a variety of degrees of melanoma (Meierjohann and Schartl, 2006).

Fig. 3

Phylooncogenomic approach for cancer driver identification, and understanding the nature of cancer biology.

Though we use CNAs as the basis for phylooncogenomic analysis, this approach should not be only limited to CNAs. Other genetic abnormalities, including point/frameshift mutations, indels, chromosome translocations, even gene expression and methylation and other epigenetic patterns that present in the tumors could also be investigated in a similar way. With phylooncogenomics, cancer genomes may be examined and compared in vertebrate phylogenetic background for common and distinct features. This is particularly important, as it will not only deepen our understanding of cancer biology, but will also give us hints to the discovery of novel cancer drugs. We have seen, on the cellular level, that a phylogenetic approach has been applied to intra-tumor heterogeneity and mutational hierarchies.