Dissecting melanocytes to predict melanoma

Abstract

With advanced high-throughput technologies, scientists can now use transcriptional signatures to study melanocytes as they become cancer. A new study identifies transcriptional programs at single-cell resolution across platforms and species, which enables prediction of melanoma prognosis and response to immune-checkpoint inhibitor therapy.

Melanocytes are found in the epidermis and hair follicles of human skin and serve to produce pigment and protect against UV radiation. Revealing the gene-expression programs that regulate the development and maintenance of these cells is essential for gaining a better understanding of melanoma. Melanoma arises from melanocytes and is the deadliest type of skin cancer. In this issue of Nature Cell Biology, Belote et al. perform single-cell RNA sequencing (scRNA-seq) on epidermal melanocytes spanning various developmental ages and anatomical locations to compile the first human melanocyte atlas of this scale¹. Furthermore, they determine stage-specific transcriptional programs that can be used to identify cell states in melanoma and ‘fortune tell’ prognosis and treatment response (Fig. 1).

Fig. 1 ∣. Melanoma subtypes categorised by transcriptional signature.

Melanomas with a neonatal melanocyte transcriptional signature (NEO) were associated with the worst survival and immunotherapy resistance outcomes, whereas those with an adult melanocyte signature (ADT) were associated with the highest levels of immune infiltration, treatment response and survival. Those with a melanocyte stem cell signature (MSC) or a foetal melanocyte signature (FET) were associated with moderate survival, with high druggability or high immune evasion, respectively. Credit: BioRender.com.

Melanocytes develop from embryonic neural crest cells, which differentiate into melanoblasts and mature into melanin-producing dendritic melanocytes of the skin. These developmental stages have been well characterized in many model organisms, including mouse, chick and zebrafish. Despite the largely conserved nature of melanocyte development, species-specific differences exist, and the precise transcriptional programs present in developing human melanocytes is ill defined. Since its conception in 2009 (ref. ²), scRNA-seq has been used extensively to identify gene-expression signatures of individual cells in a variety of organisms and contexts. Although scRNA-seq has been previously performed on normal human skin, those studies focused on abundant cell types such as keratinocytes and fibroblasts^3,4.

To study human melanocytes in detail across developmental time points and anatomical locations, Belote et al. performed scRNA-seq on FACS-enriched melanocytes from 34 skin samples from 22 people ranging in age from 9.5 foetal weeks to 81 years¹. These samples were taken from the leg, arm, foreskin, palm and sole and were obtained from people with diverse sex and racial backgrounds. Despite the diverse background of these samples, the predominant factor that determined sample clustering was developmental age. When querying differences in donor-matched samples from unique anatomical locations, the authors found a bifurcation between genes expressed in volar (palm and sole) samples and those expressed in non-volar samples at 12–18 weeks of age. The authors identified genes with volar-specific expression and validated their expression in human skin by RNA FISH. Their data shed new light on the transcriptional profiles of volar melanocyte subpopulations that probably give rise to acral melanoma, a particularly deadly subtype of melanoma.

The authors wanted to characterise the gene-expression programs that define melanocytes at specific stages during human development. To do this, they used logistical regression modelling, creating a program called ‘DevMel’. Their modelling identified gene-expression signatures of around 50–70 genes that are expressed uniquely by melanocyte stem cells (MSCs), as well as by foetal (FET), neonatal (NEO) and adult (ADT) melanocytes. They then took these signatures and evaluated their expression in previously generated mouse datasets. The MSC signature was most closely associated with mouse melanoblasts isolated from embryonic days 15.5 and 17.5 or with postnatal CD34⁺ hair follicle melanocyte stem cells^5,6. Lef1⁺ melanocytes isolated from the embryonic or postnatal mouse are differentiated melanocytes in the mouse skin^7,8; however, these cells did not show specifically enrichment for non-MSC markers. This indicates that they are not representative of human melanocytes at a specific developmental age. In vitro differentiation of melanocytes from embryonic stem cells was also not representative of a specific developmental stage of human melanocytes⁹, which highlights the limited use of these cells in modelling mature adult melanocytes. Future studies could use the gene-expression programs identified by Belote et al.¹ to improve in vitro differentiation protocols of neural crest cells.

Identifying the similarities and differences between the melanocyte transcriptional signatures of various species is important for accurate modelling of human melanocyte biology during homeostasis and in the context of disease. Studies using mouse melanocyte and melanoma models typically investigate the biology of hair-follicle melanocytes, which represent only a portion of human melanocytes. Zebrafish melanocytes are located in the hypodermis and lack specific anatomical associations with structures such as hair follicles or scales¹⁰, which perhaps makes them more representative of interfollicular melanocytes of the human skin. It would be interesting to do a more complete cross-species comparison of the developmental stage–specific transcriptional signatures identified by Belote et al.¹ in order to characterize the direct relevance of melanocytes at different developmental stages and locations in various model organisms to those in humans. An analysis such as this would complement previous a previous study by Minnoye et al. that compared chromatin accessibility of melanoma samples across six different species¹¹.

Belote et al. were also interested in learning if the developmental melanocyte programs were heterogeneously expressed in melanoma¹. The authors ran their modelling program against previously published single-cell malignant melanoma samples^12,13. They identified melanoma cell states that were similar to MSC, FET, NEO and ADT. Most tumours queried consisted of multiple programs, in support of the idea that melanoma consists of heterogeneously de-differentiated cell states. The classification of melanoma into different cell states, such as invasive, proliferative, pigmented, neural crest–like MITF-high, and AXL-high, has been previously explored by scRNA-seq^12,14; however, few studies have directly compared the developmental signatures of human melanocytes to those of melanomas. Through this comparison, the authors were able to determine that sequential de-differentiation is prevalent in melanoma progression.

Given that melanoma contains a heterogeneous mix of de-differentiated cell states, is it possible to use the gene signatures curated from developmental melanocytes as ‘tarot cards’ to predict the stage and treatment response of melanomas? In their final analysis, Belote et al. estimated the fraction of melanoma cells in The Cancer Genome Atlas that were similar to their four developmental cell states¹. Interestingly, tumour-sample clustering was determined by developmental signature, as opposed to genetic driver or tumour site. They used these developmentally defined melanoma subclasses to evaluate patient survival and found that the most differentiated tumours were associated with the best median overall survival. Unexpectedly, melanomas with a predominant NEO signature were associated with the worst survival, relative to the more de-differentiated MSC and FET groups. This appears to be due to a lack of immune infiltration and higher immune-resistance signatures for the NEO signature. Furthermore, melanomas containing a strong NEO signature had a significantly greater incidence of resistance to immune-checkpoint inhibition. Thus, it is a bad omen for melanomas to contain cells with transcriptional states similar to those of neonatal melanomas, as they are likely to yield a worse prognosis (Fig. 1).

Although this work elegantly details the transcriptional states of cutaneous melanocytes, further studies are needed to characterize other melanoma subtypes. Mucosal melanomas are a rare melanoma subtype that arises from melanocytes located in mucous membranes and have a less favourable prognosis than that of cutaneous melanomas¹⁵. Contrasting the signatures of cutaneous developmental melanocytes to those of mucosal melanomas or exploring the signatures of melanocytes found in mucous membranes could reveal novel information about the aetiology and progression of this deadly disease. Expansion of the transcriptional profiling of melanomas freshly isolated from patients could lead to improved predictive prognostic tools.

In summary, the work by Belote et al. provides a novel way of characterizing human melanocytes and melanoma¹. Their unique scRNA-seq studies of primary melanocytes provide detailed insight into transcriptional programs specific to a range of developmental time points and anatomical locations. Notably, their findings elucidate signatures that can be used to predict patient outcomes such as survival and resistance to immune-checkpoint inhibition.