Genetics of cutaneous melanoma

Cutaneous melanomas have historically been classified into superficial spreading, nodular, lentigo maligna, desmoplastic, mucosal, or acral subtypes based upon anatomic site and histologic morphology [1]. The most common subtype is superficial spreading melanoma, which is thought to originate in the epidermis before entering a vertical growth phase and invading into the dermis. In contrast, nodular melanoma is thought to progress to a vertical growth phase more rapidly. Lentigo maligna melanoma presents as a slow-growing macule or patch in sun-exposed skin and typically affects elderly patients. Desmoplastic melanoma is also associated with sun exposure, is locally infiltrative, and has a lower rate of metastasis to draining lymph nodes. Mucosal melanoma affects mucous membranes while acral melanoma affects palmoplantar and subungual skin. Mucosal and acral melanomas share the characteristic that they typically have less sun exposure than other forms of cutaneous melanoma and are frequently classified together as acral lentiginous melanomas (ALM) [2, 3]. Acral melanoma accounts for less than 3% of all primary melanomas [4, 5] but is the most common subtype affecting people of African and Asian heritage [6] and has a similar incidence in all ethnicities and skin types.

The clinical utility of this classification scheme has been called into question, however, both because a significant proportion of melanomas fail to fit neatly into a class and because histologic subtype is not an independent prognostic factor [7]. Recent efforts examining the genetics of melanoma on a large scale have allowed for synthesis of genetic data with the historic classifications, allowing for a greater understanding of the biologic underpinnings of melanoma types and the role of sun exposure. In this section, we will first review the role of ultraviolet radiation in melanomagenesis, then the known genetic drivers that underlie cutaneous melanomas arising on intermittently sun-exposed (non-CSD) skin, and follow with a discussion of melanomas associated with either minimal (acral/mucosal) or chronic sun damage (CSD) [8].

Ultraviolet signature

Exposure to ultraviolet radiation has long been considered an important environmental risk factor for cutaneous melanoma. Genetic studies have supported this notion, revealing that cutaneous melanomas harbor a very high somatic mutation burden relative to other tumor types [9]. Approximately 45% of mutations in melanoma are cytosine-to-thymine or tandem CC-to-TT transitions, while approximately 10% are guanine-to-thymine transversions [10]. This is consistent with exposure to UV-B radiation, which is known to cause crosslinking of adjacent pyrimidine bases (CC, CT, TC, or TT); when recognized by DNA polymerases, two adenine bases are often inserted opposite to the dimerized pyrimidines, and during the subsequent replication cycle, the alteration is incorrectly repaired and results in a cytosine-to-thymine or a CC-to-TT transition [11]. Less commonly, guanine-to-thymine transversions have been reported [12]. Both the rates of SNV and mutations among melanomas are highest on CSD skin and lowest in acral and mucosal sites [13].

Non-CSD melanoma

MAP Kinase pathway

Greater than 80% of melanomas on non-CSD skin have been found to harbor mutations that constitutively activate the MAP Kinase/ERK signaling pathway, which regulates cell proliferation and survival [3]. Data from The Cancer Genome Atlas (TCGA) cohort revealed that non-CSD melanomas can be classified into BRAF, RAS, NF1, and triple-wild type subtypes based upon the most prevalent mutations affecting this pathway [13]. BRAF is the most commonly mutated gene and is altered in approximately 50% of non-CSD melanomas. BRAF encodes a serine–threonine kinase that phosphorylates MEK (MAP2K) when bound to active RAS-GTP [14, 15]. Up to 90% of BRAF mutations affect exon 15 and result in V600E substitutions that cause constitutive activation of BRAF [14]; less common substitutions include V600K, V600R, and K601E [13]. Approximately 25% of melanomas contain mutations in NRAS, which encodes a small G protein that binds to and activates BRAF when in the GTP-bound state [13]. Most NRAS mutations are found either within exon 1 leading to substitution of glycine at position 12 or 13 or within exon 2 leading to substitution of glutamine at position 61; in both cases, mutation prevents hydrolysis of GTP leading to constitutive activation of both NRAS and its downstream effectors [3]. In addition to NRAS, rare mutations in HRAS (G13D, G13S, and Q61K) and KRAS (G12D, G12R, and Q61R) were identified in tumors from the TCGA cohort [13]. Of note, BRAF and RAS mutations are seldom, if ever, detected in the same tumors [14]. Other perturbations in the MAP Kinase pathway include loss-of-function mutations in the genes for RAS-GAPs NF1 and RASA2 that are found in approximately 15 and 5% of non-CSD melanomas respectively [10, 16] and gain of function mutations in MEK1 (MAP2K1) and MEK2 (MAP2K2) that have been identified in 8% of non-CSD melanomas [17]. Interestingly, NF1 mutations were found to be anti-correlated with BRAF mutations but not with NRAS mutations in tumors from the TCGA cohort [13].

Despite the prevalence of BRAF and NRAS mutations in non-CSD melanomas, these mutations are insufficient to drive malignancy in isolation. It is well known that benign acquired nevi and benign congenital nevi predominantly contain BRAF and NRAS mutations, respectively [18, 19]. BRAF mutations have also been shown to induce growth arrest in vitro [20]. This suggests that constitutive activation of BRAF triggers checkpoint mechanisms designed to restrict malignant growth, a phenomenon known as oncogene-induced senescence (OIS) [20]. To overcome OIS, additional mutations or epigenetic changes, often affecting the PI3 Kinase pathway and/or CDKN2A, are necessary [21].

PI3 Kinase pathway

Up to 50% of non-CSD melanomas exhibit mutations affecting the PI3 Kinase pathway and its effectors [3]. Inactivation of PTEN, a phosphatase that dephosphorylates PIP3 and thereby inhibits the anti-apoptotic function of Akt, was identified in 10% of tumors in the TCGA cohort, while other studies have reported PTEN mutations, deletions, or epigenetic silencing in up to 30% of non-CSD melanomas [3, 22, 23]. In addition to PTEN inactivation, gain-of-function mutations or amplifications affecting either the catalytic subunit of PI3 Kinase (4%) or Akt (up to 30%) may be present [24, 25]. 4–9% of tumors contain mutations in the gene encoding RAC1, a small G protein in the Rho family that induces lamellipodia formation and contributes to cell motility downstream of PI3 Kinase and other pathways in vitro [3]. The most commonly identified mutation, a P29S substitution, is caused by a cytosine to thymine transition consistent with a UV signature; it results in increased GDP/GTP nucleotide exchange that favors the active form and is thought to enhance cell migration and proliferation [16, 26]. Of note, this mutation has also been associated with increased PD-L1 expression on melanoma cells [27]. Gain-of-function mutations in the gene encoding PREX2, a guanine nucleotide exchange factor (GEF) for RAC1, were also found in 26% of non-CSD melanomas in the TCGA cohort, although its role as a driver mutation remains unclear [28]. Finally, greater than 15% of tumors in the TCGA cohort contain mutations in components of the mTOR signaling pathway that operates downstream of Akt including MTOR, TSC1, TSC2, RICTOR, and RPTOR [13].

CDKN2A

Although known for being the most commonly altered gene in familial melanoma, somatic loss-of-function mutations in CDKN2A were found in 15% of tumors in the TCGA cohort, and losses or epigenetic downregulation have been reported in up to 70% of sporadic melanomas [8, 29]. This gene encodes both p16/INK4a, a G1-CDK inhibitor, and p14/ARF, a protein that blocks MDM2-mediated degradation of p53 [3].

TERT

Mutations in the promoter region of telomerase reverse transcriptase (TERT), a component of the telomerase holoenzyme, are found in up to 70% of melanomas [30,31,32,33]. Consistent with UV signature, cytosine-to-thymine transitions at one of four positions upstream of the transcriptional start site have been shown to produce new binding motifs for the ETS transcription factor GA-binding protein (GABP) that lead to increased expression of TERT and therefore increased telomerase activity, which may enable cells to overcome senescence [34].

MITF

The MITF locus on chromosome 3p is amplified in 10% of non-CSD melanomas [29, 35]. MITF, or microphthalmia-associated transcription factor, is a bHLH transcription factor that is responsible for melanocyte differentiation and function [35]. It is known to control expression of several cell cycle regulators and pro-proliferative proteins including CDK2, TBX2, CDKN2A, p21, Bcl2, and c-Met as well as HIF-1α, which is believed to promote tumor survival and metastasis [36]. Additionally, transcription of MITF is known to be upregulated by canonical WNT signaling through β-catenin, and gain-of-function mutations in CTNNB1 (β-catenin) are found in 5–7% of non-CSD melanomas [37].

Other genes

Analysis of the TCGA cohort has led to the implication of several other genes in melanoma tumorigenesis. 34% of tumors contain somatic mutations in GRIN2A, the gene encoding the ionotropic glutamate receptor NMDAR2A, that prevent NMDA-R complex formation, increase anchorage-independent growth, and increase migration in vitro [38]. Similarly, 23% of tumors contain somatic mutations in GRM3, the gene encoding metabotropic glutamate receptor 3, that have been shown to disrupt melanosome trafficking via dysregulation of cAMP signaling [39, 40]. Approximately 16% of non-CSD melanomas were found to harbor loss-of-function mutations in TP53, which encodes a protein that restricts cell cycle progression in response to DNA damage [10, 13, 30]. In addition, upregulation of MDM2 and MDM4, E3 ubiquitin ligases that signal p53 degradation, has also been observed in a subset of tumors [41]. 7% of tumors contain mutations in PPP6C, the gene encoding a phosphatase known to restrict G1-to-S phase transition [10, 42]. 7% contain mutations in ARID2, a gene encoding a component of the PBAF chromatin-remodeling complex and known tumor suppressor [10]. 6% of tumors contain mutations in DDX3X, a gene encoding a DEAD-box RNA helicase involved in translation initiation and stress granule assembly, resulting in globally reduced translation that may provide a survival advantage to tumor cells [43]. Less common mutations affect citric acid cycle enzyme IDH1 (5.7%), tumor suppressor RB1 (4.3%), 40S ribosomal protein RPS27 (2.6%), and mitochondrial ribosomal protein MRPS31 (1.3%) [13, 44].

Chromosomal aberrations

Prior to the advent of deep sequencing technology, cytogenetic approaches were used to study the genetics of melanoma. Commonly reported alterations found in non-CSD cutaneous melanomas include losses on chromosomes 1p, 4, 5, 6q, 8p, 9p, 10q, 11q, 12q, 14, 15, 16, 21, and 22 and gains on chromosomes 1q, 6p, 7, 8q, 18, and 20q; of note, AKT3 is located on chromosome 1q, BRAF is located on chromosome 7q, and CDKN2A is located on chromosome 9p [3, 45,46,47].

Acral, mucosal, and CSD melanoma

Chromosomal aberrations

Several studies have shown that acral and mucosal melanomas harbor significantly increased chromosomal aberration when compared to both non-CSD and CSD melanomas [8, 48, 49]. Up to 89% of acral melanomas and 85% of mucosal melanomas have been shown to contain amplifications, in contrast to infrequent amplifications found in both non-CSD and CSD melanomas [5, 8]. The most commonly identified structural variations include gains in 4q, 5p, 6p, 11q, and 22q and losses in 6q, 9p, and 10 [8, 47, 49, 50]. In particular, 10% of acral or mucosal melanomas have gain in the CDK4 gene on chromosome 12q, up to 45% demonstrate gain in the CCND1 gene on chromosome 11q, and approximately 60% demonstrate loss in the CDKN2A gene on chromosome 9p [5, 48, 51].

KIT and PDGFRA

Unlike non-CSD melanomas, amplifications of chromosome 4q containing the gene for the receptor tyrosine kinase KIT are found in approximately 30% of mucosal melanomas and 25% of acral melanomas while gain-of-function mutations are found in approximately 20% of mucosal melanomas and 10% of acral melanomas [5, 52]. 28% of CSD melanomas also harbor KIT mutations or amplifications [52]. Among ALM and CSD melanomas, KIT alterations are rarely if ever found in tumors that also contain BRAF or NRAS mutations [52]. Most KIT mutations occur in exon 11, 13, 17, or 18, and the most common mutations are the substitutions L576P on exon 11, K642E on exon 13, and V559A on exon 11 [30, 31, 53], resulting in constitutive kinase activity and stimulation of both the MAP Kinase and PI3 Kinase pathways [54]. Of note, this spectrum of mutations overlaps with that of both sporadic and familial gastrointestinal stromal tumors [37, 55]. Also located on chromosome 4q, the gene for receptor tyrosine kinase PDGFRA is commonly amplified in ALM [13, 48]. Gain-of-function mutations in the gene have also been reported in up to 6.8% of acral melanomas and 3.6% of mucosal melanomas [5, 56]. These mutations cause ligand-independent activation of the receptor leading to dysregulated melanocyte proliferation, differentiation, and migration [56, 57]. Notably, PDGF is thought to play a role in melanocyte epithelial-to-mesenchymal transition (EMT) that precedes metastasis [58]. In contrast to ALM, CSD and non-CSD melanomas rarely contain aberrations in PDGFRA [56].

BRAF and NRAS

In contrast to non-CSD melanomas, ALM and CSD melanomas less commonly contain mutations in BRAF. Although reports vary, BRAF mutations are found in approximately 3–11% of mucosal melanomas, 21–23% of acral melanomas, and 6–11% of CSD melanomas [8, 37, 48]. In contrast to BRAF mutations, NRAS mutations occur at a similar rate of approximately 10–25% among acral, mucosal, CSD melanomas, and non-CSD melanomas [59].

Other genes

Other genes that have been implicated in acral or mucosal melanoma include MITF, PREX2, ARID1A, PTEN, TP53, AURKA, and APC [10, 16, 28, 50, 60, 61]. In contrast to non-CSD melanomas, only 6–11% of ALM contain TERT promoter mutations [48, 62]. However, up to 25% of acral melanomas have been found to contain copy number gains in the TERT gene [61]. Mutations in DDX3X, RASA2, PPP6C, RAC1, or RB1 are rarely reported in ALM [48].

Genetics of ocular melanoma

Ocular melanoma is the most common primary malignancy of the eye among adults [63], although it comprises only 5% of all melanoma [64]. It arises in the uvea in 95% of cases and in the conjunctiva in the remaining 5% [63]. Information regarding the genetic alterations in conjunctival melanoma is scarce, but the available data suggest that these tumors have more in common with cutaneous melanomas than uveal melanomas. Indeed, BRAF, NRAS, and TERT mutations are commonly found in both cutaneous and conjunctival melanomas but not in uveal melanomas [3, 65]. Of note, unlike cutaneous melanoma, uveal melanoma is not considered to be related to UV exposure [66]. Below, we will review the known genetic drivers of uveal melanoma.

GNAQ/GNA11

The most commonly mutated genes in uveal melanomas include GNAQ and its paralog, GNA11, affecting 45 and 32% of primary uveal melanomas, respectively [67, 68]. Both genes encode the α subunit of a heterotrimeric Gq protein that functions to activate the protein kinase C, MAP Kinase, and YAP1 pathways via a phospholipase C-dependent mechanism [69]. 97% of the reported mutations consist of Q209L or Q209P substitutions in exon 5 while the remaining 3% consist of R183C substitution in exon 4; all of these mutations impair the intrinsic GTP-ase activity, producing an increase in downstream signaling [67, 68]. Recently, activating mutations have been identified both upstream and downstream of GNAQ/GNA11 signaling. In a subset of uveal melanomas, a L129Q substitution in the gene encoding CYSLTR2, a leukotriene-sensing G-protein-coupled receptor (GPCR) that acts as a GEF for GNAQ/GNA11, has been shown to favor the active conformation and increase GNAQ/GNA11 activity [70, 71]. PLCB3 and PLCB4, both genes that encode phospholipase C enzymes activated by GNAQ/GNA11, have recently been found to harbor K898N and D630Y substitutions, respectively, although the role of these genes in oncogenesis is unclear [70, 72].

BAP1 and other genes

Up to 40% of uveal melanomas contain inactivating (usually truncating) mutations affecting BAP1 (BRCA1 associated protein-1) [73, 74], a protein that has been shown to play a role in DNA repair [75], melanocyte differentiation [76], and epigenetic regulation of the RB1 gene [3, 64, 77]. Recurrent mutations have also been identified in EIF1AX, a gene encoding a component of the 43S preinitiation complex that aids in recruitment of the 40S ribosomal subunit to mRNA, and SF3B1, a gene encoding a component of the U2 small ribonucleoprotein (snRNP) involved in RNA splicing [74, 78]. Of note, mutations in EIF1AX and SF3B1 occur almost mutually exclusive to each other and to mutations in BAP1 [78].

Chromosomal aberrations

Commonly reported chromosomal aberrations in uveal melanoma include monosomy of chromosome 3, gain in 1q, 6p, and 8q, and loss in 1p, 6q, 8p, and 16q [3, 47, 63, 64]. Monosomy 3, found in up to 50% of cases, is associated with worse prognosis and losses in BAP1 (located on chromosome 3p), while 8q gain is also associated with worse prognosis [79].

Genetics of familial melanoma

Between 5 and 10% of cutaneous melanomas are thought to arise in the context of a family history of melanoma [3, 80]. Approximately 40% of these tumors are associated with loss-of-function mutations in CDKN2A, resulting in loss of p16/INK4A, a key negative regulator of cell cycle progression [80, 81]. Additionally, germline gain-of-function mutations in CDK4 including R24C and R24H have been reported; these mutations prevent the interaction between CDK4 and p16/INK4 resulting in increased CDK4 activity and cellular proliferation [3, 82]. Patients with familial retinoblastoma due to germline alteration of the RB1 cell cycle regulator are also at increased risk for cutaneous melanoma [83]. Other candidate genes that may explain up to 2% of familial cutaneous melanoma include BAP1 [84, 85] along with the genes encoding telomere-associated proteins POT1, ACD, TERF2IP, and TERT [86]. In addition to these highly penetrant yet rare mutations, more common single nucleotide polymorphisms (SNPs) in MITF and MC1R have been associated with intermediate risk of cutaneous melanoma [3]. A recently discovered germline mutation in MITF causing a E318K substitution has been found to impair post-translational SUMOylation of MITF leading to increased transcription of its target genes including CDK2, Bcl2, c-Met, and HIF-1α [36, 87,88,89]. MC1R encodes the melanocyte stimulating hormone receptor that controls MITF expression and the pigmentation process. Variants in this gene have been associated with synthesis of the less photoprotective pigment, pheomelanin, and the resulting red hair color phenotype [90]. These variants also inhibit the interaction between MC1R and PTEN, allowing for increased activation of the PI3 Kinase pathway [91].

1–2% of uveal melanomas are thought to be hereditary [3], although less is known about the genetics of familial uveal melanoma than familial cutaneous melanoma. Germline BAP1 mutations in uveal melanomas have been reported, and recent studies have suggested that inherited BAP1 mutations cause a tumor predisposition syndrome that predisposes patients to cutaneous melanoma, uveal melanoma, mesothelioma, renal cell carcinoma, and several other tumors [64]. In contrast to familial cutaneous melanoma, CDKN2A mutations and MC1R variants do not seem to be associated with uveal melanoma [92].

Genetics of metastasis

Metastasis is the end result of a complex, multi-step biological process known as the metastatic cascade or the “seed and soil” hypothesis [93, 94]. The genetic basis of tumorigenesis may vary between tumor types, but the cellular and molecular steps necessary for metastasis are generally similar among solid tumors [95, 96]. The step-wise process has been elucidated as follows: (1) local invasion through the surrounding extracellular matrix and stroma and breach of the basement membrane, (2) invasion of lymphatics and/or blood vessels, (3) arrest at a distant organ, (4) invasion into the parenchyma of distant tissues, and (5) survival and proliferation in a foreign microenvironment to form metastases. In this section, we will review what is known about the genetic drivers of melanoma metastasis as well as the changes associated with metastasis in breast, prostate, lung, and colorectal cancer.

Melanoma

In order to identify potential therapeutic targets for metastatic melanoma, several groups have sought to determine the genetic perturbations that drive melanoma metastasis. Somewhat surprisingly, few changes aside from those already identified in primary melanoma have been identified. In addition, there is no consensus about the role and effect of these potential metastasis-specific genetic changes. Below, we review their findings.

An array comparative genomic hybridization (aCGH) study of 25 primary cutaneous melanomas and 61 metastatic melanoma tumors [97] found that metastatic tumors harbor significantly more genomic alterations than their primary counterparts. 30 genes were found to contain both genomic alteration and differential expression in metastatic tumors when compared to both primary tumors and normal cells: ASPM, HCG2P7, BAT1, RPA3, TM4SF13, IMP-3, SCAP2, CBX3, SFRP4, KIAA0877, LSM5, PSPH, WBSCR20C, WBSCR20B, PEX1, TFPI2, SGCE, AKAP9, PEG10, ASNS, NPTX2, PILRB, VGF, SLC26A4, DNAJB9, IFRD1, MET, CALU, PRKCA, and BIRC5. Of note, increased expression of MET, a gene that encodes a receptor tyrosine kinase activated by HGF, has been previously reported in metastatic melanoma [98]. Of the 30 reported genes, 25 are located on chromosome 7, consistent with reports of chromosome 7 gains in metastatic melanoma, non-small cell lung cancer and peripheral nerve sheath tumors [99, 100]. Functional assays confirmed the pro-invasive properties of MET, ASPM, AKAP9, IMP3, PRKCA, RPA3, and SCAP2 [97, 101].

Another aCGH study comparing nonmetastatic and metastatic melanoma tumors from an inducible mouse melanoma model identified gains in chromosome 6p and resulting overexpression of NEDD9 in metastatic tumors. This gene encodes a CAS family adaptor molecule that activates focal adhesion kinase (FAK) and regulates adhesion and migration of cells. Functional assays confirmed that overexpression of NEDD9 in tumors with activating BRAF or RAS mutations increased invasiveness both in vitro and in vivo when compared to BRAF or RAS melanomas lacking NEDD9 overexpression. These findings are consistent with the fact that approximately 40% of human metastatic melanomas (in contrast to only 8% of primary melanomas) sustain gains in chromosome 6p and demonstrate increased NEDD9 expression [102].

A meta-analysis of genome-wide association studies (GWAS) published prior to 2010 identified 18 genes that were found by three or more GWAS to be altered in metastatic melanoma: CKS2, DSC3, EGFR, CDC6, CTNNBIP1, H2AFV, CXCL14, CSAG2, WNT5A, SPP1, CLIC3, PLP1, AP1S2, BCL2A1, AHNAK, S100A2, and KRT15 [103]. Interestingly, several other studies have reported EGFR amplification on chromosome 7, often due to polysomy 7, in metastatic melanoma tumors [104, 105].

Studies comparing whole genome sequencing (WGS) and whole exome sequencing (WES) data between primary tumors and their paired metastases have thus far revealed that the majority of genetic changes in metastatic tumors are shared with their primary counterparts [60, 106, 107]. One example includes a comparison of WES data from two cutaneous melanomas and their paired metastases that demonstrated 96–98% genetic similarity between primary and metastatic tumors [107]. Another study compared WGS data from an acral melanoma and its paired lymph node metastasis, demonstrating highly similar single nucleotide variants (SNVs), copy number alterations (CNAs), and loss of heterozygosity (LOH) among the paired tumors [60]. In the latter study, only two SNVs—one in the coding region of WNT1 and one in a splice site of SUPT5H—were found exclusively in the metastatic tumor [60].

In addition to these comparative studies, epidemiological data has suggested a role for several other genes in melanoma metastasis. For example, although BRAF and NRAS-mutant melanomas have similar rates of metastasis, studies suggest that they are both slightly more likely to metastasize than BRAF-and-NRAS-wild type melanomas [108, 109]. MITF amplification has been found in 20% of metastatic tumors in contrast to 10% of primary tumors [35] while TERT promoter mutations have been reported in up to 85% of metastatic melanomas in contrast to 30–40% of primary melanomas [32]. Other genetic changes that have been found more commonly in metastatic than primary tumors include loss-of-function CDKN2A mutations [108], chromosome 1 deletions causing loss of Kiss-1 [110], and gain-of-function mutations in PREX2 [28, 66].

In contrast to cutaneous melanoma, more is known about the genetic changes involved in uveal melanoma metastasis. Uveal melanomas are commonly classified into one of two prognostic groups based on their gene expression profiles [111]. Melanomas in the first group, which harbor mutations in either EIF1AX or SF3B1, are associated with low risk of metastasis [78] while melanomas in the second group, which harbor truncating mutations in BAP1, are associated with high metastatic risk [112]. Approximately 80% of metastatic uveal melanomas contain truncating mutations in BAP1 [73] and target the liver in up to 90% of cases [113]. Interestingly, BAP1 mutations usually arise early in tumor evolution and are followed by a period of neutral evolution before tumors become clinically evident, suggesting that the metastatic potential of uveal melanoma is determined early in its course [112].

Breast cancer

Breast cancer is the most prevalent cancer in women. Breast cancer metastasis begins with the loss of myoepithelial cells and subsequent local invasion of the mammary gland [114]. The secretion of IL-6 by adipocytes present in the tumor microenvironment and the proliferation of stromal CD4 + T lymphocytes are both known to stimulate tumor-associated macrophages (TAMs) and promote tumor invasion through the basement membrane. TAMs and TGF-β can subsequently enhance tumor invasion into lymphatics and blood vessels [115]. The process by which distant organs are colonized may be tumor- and/or gene-specific [116], and a number of genes whose expression facilitates metastasis of breast cancer to bone, lung, and liver have been identified.

Breast cancer that has metastasized to bone often presents with pathologic bone fractures due to upregulation of osteoclast activity [117]. By overexpressing osteoclast-inducing factors such as IL-8, IL-11, and PTHrp, cancer cells inhibit the pro-osteoblastic BMP and Wnt pathways, leading to the formation of lytic lesions [117, 118]; IL-11 in particular disrupts the normal signaling between osteoblasts and osteoclasts through modulation of RANK/RANKL signaling [119]. Breast cancer metastasis to bone also induces secretion of proteases that cleave RANKL into a more active form, thereby activating additional osteoclasts to create osteolytic lesions [120]. In addition, TGF-β and IGF are upregulated, further promoting osteoclast proliferation and function [117, 119, 121, 122].

Several genes that encode extracellular factors important for growth and survival have been shown to be upregulated in breast cancer metastases to the lung. The products of these genes include HER/ErbB receptor ligand epiregulin, chemokine CXCL1, cell-surface receptor VCAM1, cell adhesion protein ROBO1, and matrix metalloproteinases (MMPs) MMP1 and MMP2 [123]. Several intracellular effectors, such as COX2 and transcriptional regulator ID1 have also been implicated in metastasis [123, 124]. TGF-β and NF-κB have been identified as key regulators of breast cancer metastasis to lung, although the downstream mediators remain largely unknown [125, 126]. Expression of MMP2 is rare in primary tumors but increases once cancer cells reach the lung [127].

The liver is a common site of metastasis for all solid tumors and is the third most common site for breast carcinoma [128]. Colonization and proliferation within hepatic tissue involves altering the cellular junctions. Claudin-2, a component of cellular tight junctions encoded by the CLDN2 gene, has been shown to act as a mediator of breast cancer metastasis to liver. Increased CLDN2 expression by metastatic breast cancer enhances adhesion to hepatocytes via claudin-2–claudin-2 interactions [129]. Interestingly, claudin-2 enrichment is selective, and similar expression profiles are not seen in breast cancer metastases to skin [129].

Prostate cancer

Prostate adenocarcinoma is the most common cancer in American men and the second leading cause of cancer-related death in men [130]. A review of the metastatic patterns of prostate cancer at autopsy revealed that 90% of patients developed bone metastases while fewer patients developed lung and liver metastases [131]. Studies have implicated several proteins including EGFR, Wnt, Notch, Sonic hedgehog, CXCR4, and stromal cell-derived factor 1 in the metastatic progression of prostate cancer [132]. Several additional factors may play a role in the EMT [132,133,134].

As with metastatic breast cancer, metastatic prostate cancer involves disruption of RANKL-dependent bone homeostasis [117]. However, in contrast to breast cancer, metastatic prostate cancer upregulates the activity of bone-depositing osteoblasts resulting in a net increase in bone density [118]. Several studies have linked BMP and Wnt upregulation with the differentiation and proliferation of osteoblasts [118, 135]. Prostate cancer has also been shown to secrete endothelin-1 and PDGF, both osteoblast activators, as well as osteoprotegerin, a RANKL inhibitor that inhibits osteoclast activation [118]. Additionally, members of the RAF, NTRK2, and MERTK families are also upregulated in prostate cancer metastases to bone [136].

Lung cancer

Lung cancer is the second most common cancer in both men and women, and it is the leading cause of cancer-related deaths [137]. Primary lung tumors metastasize quickly [138], and lung cancer is most often detected as a metastatic stage IV carcinoma [139]. Several genes have been implicated in lung cancer metastasis. Upregulation of VEGF, for example, allows the primary tumor to mitigate the hypoxic environment created by the tumor’s own growth while EGFR mutations are thought to promote EMT through activation of TWIST1 [140]. Additionally, upregulation of cytokine CXCR4 and adhesion protein VCAM-1 are thought to help tumor cells colonize distant organs [140].

Lung cancer commonly metastasizes to bone, producing mixed osteolytic/osteoblastic lesions [141]. Recent studies have demonstrated RANK expression in primary lung carcinomas, with clonal inactivation of RANK associated with prolonged survival [142]. Additionally, both TGF-β-dependent osteoclast activation and enhanced activity of TCF4 and PRKD3 have been associated with aggressive bone colonization [143]. NF-κB, MAPK, and Wnt signaling pathways are also involved in bone metastasis through promotion of angiogenesis and upregulation of MMPs [141].

Metastasis to the brain appears to be partially dictated by the histologic type of lung cancer. Squamous cell lung carcinoma invades the thoracic wall before traveling to the brain likely due to EGFR mutations [139]. In contrast, metastasis of lung adenocarcinoma to the brain correlates with FGFR1 amplification [144]. The brain cells themselves also appear to play a role in metastasis; microglia are thought to facilitate invasion of brain tissue via a Wnt-dependent mechanism [145] while astrocytes secrete MMP-2 and MMP-9 to allow invading tumor cells to colonize brain tissue [146].

Colorectal cancer

Colorectal carcinoma (CRC) is the third most common cancer type in the U.S. [147], and at the time of diagnosis, approximately 20% of CRC patients already have metastatic disease [148]. Due to the anatomy of the portal vein system, CRC metastasis most commonly affects the liver, carrying a 5-year survival rate of 25–40% [149]; less frequently, metastatic lesions are found in the lung [124, 128, 147]. Studies have shown that the incidence of metastasis to both liver and lung is higher with left-sided colon cancer than right-sided colon cancer [150, 151].

Genetic alterations associated with CRC metastasis include loss of TP53, allelic losses on chromosomes 13q, 14q, 17q, 18q, and 22q, and decreased expression of DCC [152, 153]. In addition, P-cadherin is often overexpressed in CRC metastases to the liver relative to primary tumors. P-cadherin, a cell-adhesion molecule, is thought to both downregulate expression of E-cadherin and upregulate the expression of β-catenin leading to increased cellular motility and invasiveness [149, 154,155,156]. Finally, levels of prothrombin and fibrinogen have been found to mediate the metastatic potential and growth of CRC [157].

Non-genetic factors that may regulate metastasis

The global prevalence of cancer along with the poor prognosis associated with metastasis underscores the importance of understanding the biology of tumor outgrowth. The lack of consistency among studies comparing metastatic and primary tumors, however, has made it difficult to conclusively identify clear metastatic drivers within the genome. One explanation for this lack of consistency involves the studies’ technical limitations. Intratumor heterogeneity increases the likelihood that the clones from which metastases derive are missed when sampling primary tumors, thus leading to overestimated genetic divergence between primary and metastatic tumors [158, 159]. Conversely, comparisons based on WES or gene panels would likely underestimate genetic divergence by detecting only genetic changes existing within the coding regions of genes in the case of WES or within known driver genes in the case of gene panels [160]. Further complicating the study of metastatic melanoma genetics is the fact that normal skin harbors a high burden of UV-signature mutations within oncogenes, thus increasing the number of mutations shared between primary and metastatic tumors and leading to underestimation of their true divergence [161, 162].

In addition to these technical limitations, there may also be biological explanations for the lack of conclusive information about the genetic drivers of metastasis. First, metastasis may be driven primarily by genetic changes already present within the primary tumor, as seems to be the case with uveal melanoma. Furthermore, epigenetic changes causing altered gene expression likely play an important role in metastasis. Finally, stochastic features of the micrometastatic microenvironment may modulate tumor growth, allowing for some disseminated tumor cells to develop into clinically-apparent metastases while others remain dormant occult tumors [163].

Epigenetics of cancer metastasis

Epigenetic changes include altered DNA methylation and histone modification patterns that result in chromosomal instability, changes in chromatin compaction, and changes in gene expression [164]. Several recent studies have found that epigenetic changes may play a significant role in metastasis. For example, an analysis of aggregated transcriptomic data from melanoma patient samples found that upregulated ALDH1A1 and HSP90AB1 and downregulated KIT, KRT16, SPRR3, and TMEM45B distinguished metastatic melanomas from their primary counterparts despite the absence of mutation or CNA in these loci [165]; notably, several immunohistochemical studies have correlated reduced KIT expression with metastatic potential [166, 167]. Another study found that metastatic melanomas contain reduced expression of LFNG, which encodes an enzyme that glycosylates NOTCH and thereby increases its signaling, and demonstrated that CRISPR/Cas9 knockdown of LFNG leads to increased pulmonary metastases in a mouse model [168]. Other epigenetic changes associated with melanoma metastasis include loss of 5-hydroxymethylcytosine (5-hmC) bases in DNA likely due to downregulation of IDH2 and TET family enzymes [169], reduced expression of NM23 [170], overexpression of RhoC [171], and downregulation and nuclear localization of BRMS1 [172]. In addition to melanoma, DNA hypermethylation within tumor suppressor gene promoters including BRCA1 has been associated with metastatic progression in breast, lung, colorectal, and liver cancer [173, 174].

The tumor microenvironment

The tumor microenvironment consists of malignant cells along with endothelial cells, fibroblasts, immune cells, stromal cells, and the various cytokines that convey messages among them all [175]. Features of the microenvironment including metabolic changes, hypoxia, angiogenesis, proteoglycan expression, the presence of matrix metalloproteinases, growth factor secretion, and autophagy may influence both malignant cells’ ability to invade novel tissues and their survival in these tissues [175, 176]. Some tumors have elevated rates of metastasis to specific tumor microenvironments, as exemplified by the high rates of uveal melanoma metastasis to liver [113].

Inflammation in particular appears to play a role in tumor metastasis. Although anti-tumor immune responses attempt to kill tumor cells, they often fail to contain tumors and instead produce a chronic inflammatory state that may paradoxically promote malignant transformation [177]. Inflammatory cytokines that have been shown to enhance tumor growth and invasiveness, including TNF-α, IL-6, and IL-10 [178], along with growth factors such as TGF-β, PDGF, and VEGF, amplify the inflammatory state and lead to upregulation of MMPs and cysteine cathepsin proteases that degrade the extracellular matrix and clear a path for invading tumor cells [179].

Future directions

The statistical power afforded by large tumor cohorts such as TCGA has facilitated the discovery of much of what is now known about the genetic drivers of carcinogenesis. Our ability to draw conclusions about the genetic drivers of metastasis, however, is limited by a relative paucity of studies comparing genomic data from paired primary and metastatic tumors. Further complicating the matter is the fact that the studies that have performed such analyses have done so using relatively few tumor pairs. To address these limitations, future efforts should focus on collecting a large cohort of paired primary and metastatic tumor samples from a variety of different cancer types. Once such a cohort is established, WGS and transcriptomic studies can be used to determine which, if any, genetic changes and expression profiles are significantly associated with metastasis and whether these metastasis-specific changes differ among tumor types. Any genes or pathways identified through such studies would provide researchers and clinicians with potential therapeutic targets.

Summary

Analysis of genomic data has led to the discovery of several genetic changes thought to drive melanomagenesis. Comparisons of genetic alterations between matched primary and metastatic tumors, however, have generally failed to detect clear drivers of metastasis that are consistent across studies. Similar studies of other tumor types including breast, prostate, lung, and colorectal cancer have demonstrated that altered expression of cell adhesion molecules, growth factors, and cytokines are important to the metastatic process but have yielded few candidate metastasis driver genes. Instead, it appears that metastasis is a result of the complex interplay between genetic alterations, epigenetic changes, and the tumor microenvironment. With the use of higher-resolution genetic technologies, larger sample sizes, and comprehensive epigenomics, the future may shed more light on this elusive question. Given the unmet needs in the treatment of metastatic cancer, there is no shortage of work to be done.