To gain a comprehensive view of the genomic landscape in human melanoma tumors, we sequenced the genomes of twenty-five metastatic melanomas and peripheral blood obtained from the same patients (Supplementary Table S1
). Two tumors (ME015 and ME032) were metastases from cutaneous melanomas arising on glabrous (or hairless) skin of the extremities, representing the acral subtype. The other tumors were primarily metastases from melanomas originating on hair-bearing skin of the trunk (the most common clinical subtype). Further, ME009 represented a metastasis from a primary melanoma with a clinical history of chronic ultraviolet (UV) exposure.
We obtained 59-fold mean haploid genome coverage for tumor and 32-fold for normal DNA (Supplementary Table S2
). On average, 78,775 somatic base substitutions per tumor were identified, consistent with prior reports3,4
(Supplementary Table S3
). This corresponded to an average mutation rate of 30 per megabase (Mb). However, the mutation rate varied by nearly two orders of magnitude across the 25 tumors (). The acral melanomas showed mutation rates comparable to other solid tumor types (3 and 14 mutations per megabase)5,6
, whereas melanomas from the trunk harbored substantially more mutations, in agreement with previous studies3,7,8
. In particular, sample ME009 exhibited a striking rate of 111 somatic mutations per Mb, consistent with a history of chronic sun exposure.
Figure 1 Elevated mutation rates and spectra indicative of UV radiation damage. Top bar plot shows somatic mutation rate of 25 sequenced melanoma genomes, in decreasing order. Middle matrix indicates BRAF/NRAS somatic mutation status, with left-adjacent bar plot (more ...)
In tumors with elevated mutation rates, most nucleotide substitutions were C/G > T/A transitions consistent with UV irradiation9
. The variations in mutation rate correlated with differences in the UV mutational signature. For example, 93% of substitutions in ME009 but only 36% in acral melanoma ME015 were C>T transitions (); these tumors contained the highest and lowest base mutation rates, respectively (111 and 3 mutation per megabase). Interestingly, the acral tumor ME032 also showed a discernible enrichment of UV-associated mutations (). Thus, genome sequencing readily confirmed the contribution of sun exposure in melanoma etiology.
In agreement with prior studies7,9
, we detected an overall enrichment for dipyrimidines at C>T transitions. Analysis of intragenic C>T mutations yielded a significant bias against C>T mutations on the transcribed strand for most melanomas, consistent with transcription-coupled repair (TCR) (Suppl. Fig. 1
. Most commonly, C>T mutations occurred at the 3′ base of a pyrimidine dinucleotide (CpC or TpC) (Suppl. Fig. 2
). In contrast, the C>T mutations in sample ME009 (with hypermutation and chronic sun exposure history) more often occurred at the 5′ base of a pyrimidine dinucleotide. As expected, the acral tumor ME015 exhibited mutation patterns observed in non-UV associated tumor types11
, such as an increased mutation rate at CpG dinucleotides relative to their overall genome-wide frequency (Suppl. Fig.2
). These different mutational signatures suggest a complex mechanism of UV mutagenesis across the clinical spectrum of melanoma, likely reflecting distinct histories of environmental exposures and cutaneous biology.
We detected 9,653 missense, nonsense, or splice site mutations in 5,712 genes (out of a total of 14,680 coding mutations; Supplementary Tables S4, S5
), with an estimated specificity of 95% (Supplementary Methods
). The BRAFV600E
mutation was present in 16 of 25 tumors (64%), including the acral melanoma ME015. NRAS
was mutated in 9 of 25 tumors (36%) in a mutually exclusive fashion with BRAF
, with the exception of one non-canonical substitution (NRAST50I
) in the hypermutated sample ME009. We also identified 6 insertions and 34 deletions in protein coding exons (Supplementary Table S6
), including a 21-bp in-frame deletion involving exon 11 of the KIT
oncogene in the acral tumor ME032 (Supplementary Fig. S3
mutations occur in 15% of acral and mucosal melanomas12
, and melanoma patients with activating KIT
mutations in exon 11 have demonstrated marked responses to imatinib treatment13
We identified an average of 97 structural rearrangements per melanoma genome (range: 6-420) (Supplementary Table S7
). In addition to displaying a wide range of rearrangement frequencies, the proportion of intrachromosomal and interchromosomal rearrangements varied widely across genomes. ME029, which harbored the largest number of rearrangements (420), contained only 8 interchromosomal events (). In contrast, ME020 and ME035 contained 95 and 90 interchromosomal rearrangements, respectively (). In both cases, the vast majority of interchromosomal rearrangements were restricted to two chromosomes. This pattern is reminiscent of chromothripsis14
, a process involving catastrophic chromosome breakage that has been observed in several tumor types15,16
Figure 2 Hubs of rearrangement breakpoints affect known and putative oncogenes. (a) Circos plots representing 4 melanoma genomes with notable structural alterations. Interchromosomal and intrachromosomal rearrangements are shown in purple and green, respectively. (more ...)
106 genes harbored chromosomal rearrangements in two or more samples (Supplementary Table S8
). Many recurrently rearranged loci contain large genes or reside at known or suspected fragile sites17
; examples include FHIT
(6 tumors), MACROD2
(5 tumors), and CSMD1
(4 tumors). On the other hand, several known cancer genes were also recurrently rearranged, including the PTEN
tumor suppressor (4 tumors) and MAGI2
(3 tumors), which encodes a protein known to bind and stabilize PTEN. MAGI2
was also found disrupted in recent whole genome studies of prostate cancer18
and a melanoma cell line7
. Rearrangements involving the 5′ untranslated region of the ataxin 2-binding protein 1 gene (A2BP1
) were observed in 4 tumors. A2BP1
encodes an RNA binding protein whose genetic disruption has been linked to spinocerebellar ataxia and other neurodegenerative diseases. A2BP1
undergoes complex splicing regulation in the central nervous system and other tissues19
; in melanoma, these rearrangements may disrupt a known A2BP1
splice isoform or enable a de novo
splicing product. Together, these results suggest that chromosomal rearrangements may contribute importantly to melanoma genesis or progression.
Acral melanoma (ME032) harbored the second-largest number of total rearrangements (314; ). We employed high throughput PCR followed by massively parallel sequencing to successfully validate 177 of 182 events tested in this sample, confirming its high rate of rearrangement. The elevated frequency of genomic rearrangements in acral melanomas has been reported previously20
. In comparison, ME032 exhibited one of the lowest base pair mutation rates of the melanomas examined (21st
out of 25 samples), suggesting that different tumors might preferentially enact alternative mechanisms of genomic alteration to drive tumorigenesis.
As noted above, many rearrangements in ME032 involved multiple breakpoints within a narrow genomic interval. One such event disrupted the ETV1
locus. We previously demonstrated an oncogenic role for ETV1 in melanoma, whose dysregulated expression was associated with upregulation of microphthalmia-associated transcription factor (MITF)21
, the master melanocyte transcriptional regulator and a melanoma lineage survival oncogene22
. We validated 6 distinct rearrangements (4 interchromosomal translocations) in ME032 involving breakpoints within ETV1
introns (). These events join regions of ETV1
to distal loci on chromosomes 8, 9, 11, and 15. In support of their possible functional relevance, these rearrangements were associated with high-level ETV1
amplification in this tumor.
A second complex rearrangement involved the PREX2
encodes a phosphatidylinositol 3,4,5-trisphosphate RAC exchange factor recently shown to interact with the PTEN tumor suppressor and modulate its function2
. We validated 9 somatic rearrangements in the vicinity of PREX2
(6 interchromosomal translocations), including 5 with intronic breakpoints (, Supplementary Fig. S4
). One event joined specific intronic regions of PREX2
. Like ETV1, PREX2
is highly amplified in this tumor, as verified by FISH analysis (, Supplementary Fig. S5
). The presence of these complex structural rearrangements in addition to amplification may indicate multiple mechanisms of PREX2
dysregulation in melanoma. More generally, these findings raised the possibility that sites of complex rearrangement might denote genes of functional importance in melanoma.
Next, we calculated the mutational significance of each gene based on the number of mutations detected, gene length, and background mutation rates (, Supplementary Table S9
) (See Methods). Eleven genes were found to be significantly mutated across the 25 samples (q < 0.01). As expected, the two most significant genes were BRAF
, mutated in 16 and 9 samples, respectively. Interestingly, PREX2
scored as one of the top significant genes (). Furthermore, 4 samples harbored nonsense truncation mutations in PREX2
, more than any of the other genes identified as statistically significant in this analysis. PREX2
mutations have occasionally been reported in colon, lung, and pancreatic cancer23
, albeit at low frequencies. Here, we detected 13 non-synonymous point mutations in PREX2
— including 4 nonsense mutations — and 1 synonymous mutation, with 11 of 25 melanomas harboring at least 1 non-synonymous mutation. The mutations were distributed throughout the entire length of PREX2
(, green circles), and 13 of 14 mutations were non-synonymous, suggestive of strong positive selection. An analysis of the mutant allele frequencies and estimated tumor purities indicates that at least 2 mutations are homozygous. One melanoma, ME018, harbors 3 missense mutations, two of which (I534M and G1581R) appear to co-occur on a single allele based on their observed mutation frequencies. Notably, a PREX2
nonsense mutation was detected in ME032, in addition to the rearrangements and amplification of this locus present in this tumor (). This PREX2
mutation was truncating (E824*), removing the C-terminal region with homology to an inositol phosphatase domain. Based on the allele frequency of this mutation, we infer that it occurs on the non-amplified allele. Taken together, whole-genome sequencing of this 25-sample discovery cohort identified PREX2
as a candidate melanoma gene whose amplifications, rearrangements or mutations appeared to undergo positive selection in human melanoma genesis.
Significantly mutated genes in 25 melanoma tumors
Figure 3 Mutant PREX2 expression promotes melanoma genesis. (a) Non-synonymous sequence mutations detected from Illumina sequencing of 25 melanomas (green) or from capillary sequencing of a validation cohort of 107 additional melanomas (purple). Mutations are (more ...)
To determine the prevalence of PREX2
mutations in melanoma, we performed bidirectional capillary sequencing in an extension cohort of 107 tumor/normal pairs, comprising 45 tumors and 62 short-term cultures collected from multiple institutions and geographic regions (Supplementary Table S10
). We identified 23 somatic base pair mutations and one frame-shift insertion in PREX2
in this cohort (; Supplementary Table S11
), 15 of which represented non-synonymous changes. We therefore inferred a 14% frequency of non-synonymous PREX2
mutations in this melanoma cohort.
Discrepant non-synonymous:synonymous ratios were observed between the tumor samples and short-term cultures in the extension cohort. In line with results from the discovery cohort, 100% of PREX2 mutations detected across 45 tumor samples were non-synonymous in nature (n = 4), consistent with positive selection. In contrast, only 55% of the sequence mutations found in the 64 short-term cultures were non-synonymous (a ratio of 11:9). Conceivably, these findings may indicate that subsets of melanoma cells capable of robust growth in vitro may have experienced reduced selective pressure for PREX2 mutations. Alternatively (or in addition), the PREX2 locus may exhibit an enhanced “local” mutation rate, a by-product of which is the production of variants that undergo positive selection in vivo.
To demonstrate the functional relevance of PREX2
mutations in melanoma tumorigenesis, we ectopically expressed six representative mutations (three truncation variants and three non-synonymous point mutations predicted to carry functional impact24
) in TERT-immortalized human melanocytes engineered to express NRAS(G12D) (PMEL-NRAS*)21
. These melanocytic lines were transplanted into immunodeficient mice alongside control melanocytes expressing either wild-type PREX2 or GFP. Overexpression of all 3 truncated variants as well as a point mutant (G844D) of PREX2 significantly accelerated in vivo
tumorigenesis when compared to GFP control or WT PREX2 expressing melanocytes (, Supplementary Fig. 6
). These results therefore affirmed the aforementioned genomic data suggesting that PREX2
mutations may undergo positive selection in vivo
. Although the spectrum of PREX2
mutations in human melanoma () is reminiscent of inactivating mutations, our findings suggest that PREX2
somatic mutations generate truncated or variant proteins that gain oncogenic activity in melanoma cells.
In summary, following recent efforts to characterize whole genomes from several hematologic and solid tumors, we provide the first high-resolution view of the genomic landscape across a spectrum of metastatic melanoma tumors. The analysis reveals global genomic evidence for the role of UV mutagenesis in melanoma, and identifies several recurrently mutated and rearranged genes not previously implicated in this malignancy. In particular, we discovered that PREX2 mutations are both recurrent and functionally consequential in melanoma biology. Although its precise mechanism(s) of action remain to be elucidated in melanoma, PREX2 appears to acquire oncogenic activity through mutations that perturb or inactivate one or more of its cellular functions. This pattern of mutations may exemplify a category of cancer genes that is distinct from “classic” oncogenes (often characterized by highly recurrent gain-of-function mutations) and tumor suppressors (inactivated by simple loss-of-function alterations). Instead, (over)expression of certain cancer genes with distributed mutation patterns may promote tumorigenicity either through dominant negative effects or more subtle dysregulation of normal protein functions.
Cancer genomics has enabled the discovery and rational application of the first truly effective targeted therapy for metastatic melanoma: BRAF
mutations predict sensitivity to selective RAF inhibitors25-27
. However, the emergence of acquired resistance is rapid and often driven by other genomic events28
. Our genomic exploration of the melanoma genomes revealed a large number of complex alterations that likely impact on many other genes in addition to PREX2
. Understanding how this spectrum of genomic aberrations contributes to melanoma genesis and progression should provide new insights into tumor biology, therapeutic resistance, and developing treatment regimens aimed at durable control of this malignancy.