Melanoma Genomics: Shifting Focus from Inter- to Intrapatient Variation

Thousands of melanomas have been sequenced through The Cancer Genome Atlas (TCGA) project and other large-scale sequencing initiatives. These studies profiled, for the most part, individual tumors from each patient that were thought to be representative of the disease. However, patients who succumb to melanoma tend to have dozens of macroscopic metastases at the time of their death. Moreover, there is a bias in the types of tumors that are sampled from living patients with widely metastatic disease, as most biopsies are taken for diagnostic purposes or with palliative intent, favoring accessible lesions and/or relatively large tumors. Although it may seem as if the omic landscape of melanoma has been exhaustively covered in the post-TCGA era, intrapatient heterogeneity remains poorly understood.

To address this question, Spain and colleagues sampled 573 melanoma biopsies from autopsy material of 14 patients to study their genetic and transcriptomic alterations (1). Their work is not the first study to profile multiple metastases from individual patients (2–6), but it stands out in its scope. Several important insights came from their work.

First, this study sheds light on patterns of metastatic dissemination by comparing metastases from different organ sites (Fig. 1A). When a patient has brain metastases, they are at the highest possible disease stage. However, the authors show that brain metastases are not necessarily the fastest-growing or the genetically most advanced tumors in the body. They observed instances in which brain metastases diverged early in the phylogenetic tree, having fewer copy-number alterations than clones elsewhere in the body. These observations support the growing consensus (7) that metastatic cells begin to seed early, likely as soon as a melanoma becomes invasive, and that the cells disseminate in parallel to both regional and distant sites directly from primary tumors.

Furthermore, MYC was nominated as a driver of resistance to immune-checkpoint inhibitors (ICI). Patients with cancer are categorized as responders or nonresponders to therapeutic interventions based on RECIST criteria. However, patients can have mixed responses at the lesion level, in which some tumors shrink, whereas others continue to grow. The authors leveraged this intrapatient heterogeneity, comparing the genomic landscapes of individual tumors with radiologic response data of each tumor (Fig. 1B). They discovered that tumors with amplification of chromosomal arm 8q, encompassing the MYC oncogene, were less likely to respond to ICI treatment. In a separate study (8), Haq and colleagues found an extreme example of a heterogeneous response to ICI treatment within a single patient. The patient had diffusely metastatic melanoma and showed a complete response to pembrolizumab with the exception of a single stubborn mass. The unresponsive lesion in the Haq study had an FBXW7 mutation that was not present in other samples. FBXW7 is the E3 ubiquitin ligase responsible for MYC degradation, and thus a loss-of-function mutation in FBXW7 increases MYC protein levels. Taken together, there is compelling genetic evidence from two independent studies linking the MYC signaling axis to ICI resistance.

In addition to MYC, the authors identified mutations in B2M and JAK1, confirming previous work that these aberrations drive resistance to immune-checkpoint inhibition. They also observed widespread loss of heterozygosity in antigen presentation genes, typically driven by deletion of one allele, though more work is needed to conclusively demonstrate that antigen presentation genes are haploinsufficient. Unexpectedly, the authors did not detect a significant loss of neoantigens. However, if there had been selection against neoantigens prior to the most recent clonal sweep, then their loss would not be detectable here. Future studies that are focused on earlier stages of evolution may reveal stronger signals of immunoediting.

Next, the phylogenetic histories of the melanomas indicated the mutational factors operating at different evolutionary time points (Fig. 1C). The authors constructed clone-based trees (as opposed to sample-based trees) from the somatic point mutations. Most trees had long trunks and short branches, though two patients who had received chemotherapy and one patient with acral melanoma defied this pattern. The authors also constructed phylogenetic trees based on copy-number alterations, and in contrast to the point mutation phylogenies, the copy-number trees tended to have short trunks with long branches. Previous work elucidated the evolution of primary melanomas from benign precursors (e.g., common nevi), showing that UV radiation steadily induced point mutations until melanomas became invasive, at which point, copy-number alterations became more prevalent (9, 10). The phylogenies from Spain and colleagues generally support this view, though some therapies (e.g., chemotherapy) may alter the evolutionary trajectory, and certain subtypes of melanoma (e.g., acral melanoma) likely follow different routes.

The study mostly performed bulk-cell DNA and RNA sequencing of tumors, but single-cell DNA sequencing was carried out on 22 cells from one tumor. The single-cell DNA-sequencing data revealed clonal heterogeneity that was not apparent from the bulk-cell DNA-sequencing data. This observation serves as an important reminder that there is intrapatient heterogeneity between tumors as well as heterogeneity between individual cells within each tumor.

These examples represent only a few highlights from the study. More generally, the data described by Spain and colleagues will be an important resource for the melanoma research community, and the authors provide a blueprint on how to perform and interpret a large-scale intrapatient interrogation of metastatic disease. In closing, Spain and colleagues underscore the need, in the post-TCGA era, to better understand tumor heterogeneity. Although the challenge is daunting, there is a promise because of the growth of tissue banks, as elegantly demonstrated by Spain and colleagues, and future studies can also emphasize single-cell and spatial-profiling technologies, providing an even more granular portrait of heterogeneity.