Skin and hair pigmentation is the product of a remarkably complex communication between two histologically-distinct cell types; the pigment-producing melanocytes of neural crest origin and the pigment-accepting keratinocytes of epithelial origin. The interaction of one melanocyte with a highly stable number of keratinocytes generates a range of functional pigment or melanin units with unique features, which contribute to the homeodynamic balance of the epidermis and hair follicle. Melanocytes constitute a minor cell population (~3%) in the adult human interfollicular epidermis (see Figure 1 ). Their most apparent trait is the production of a light-absorbing indole biopolymer called melanin, via a phylogenetically-ancient biochemical pathway called ‘melanogenesis’ (14, 15). Apart from this primary role, melanocytes play important regulatory functions in skin, as potential contributors to skin immune response, stress sensors or as neuroendocrine cells, through the production of neuropeptides, neurohormones and neurotransmitters (16–18). For example, as a near-permanent cell in our epidermis, melanocytes act as ‘sentinel’ cells, through stress sensing and immune response modulation, especially in the context of UVR. While cutaneous melanocytes of the interfollicular epidermis are essentially post-mitotic cells, even with (UVR) stimulation, their cousins in the cycling hair follicle (see below) exhibit considerable plasticity, including in their proliferative capacity.

Melanin synthesis is a product of a series of enzymatic reactions, commencing with the hydroxylation of phenylalanine followed by the rate-limiting conversion of tyrosine by tyrosinase to form brown-black eumelanin and in the presence of cysteine or glutathione, the yellow-red pheomelanin, in the unique organelle of melanocytes - the melanosome (14, 15). In brief, melanogenesis involves a series of signals beginning with the rate-limiting oxidation of L-tyrosine to L-dihydroxyphenylalanine (L-DOPA), catalysed by tyrosinase (19, 20). Besides their positive regulatory role of melanogenesis, L-tyrosine and L-DOPA can act also as hormone-like regulators of other cellular functions. Upon stimulation by alpha-melanocyte-stimulating hormone (α-MSH) or adrenocorticotropic hormone (ACTH), melanocortin-1 receptor (MC1R) signaling can increase intracellular cyclic adenosine monophosphate (cAMP). The cAMP activates the response element-binding protein (CREB)-signaling pathway in melanocytes, to transcriptionally activate a variety of downstream targets, including microphthalmia-associated transcription factor (MITF). This crucial transcription factor activates essential melanin-forming genes such as the tyrosinase enzyme family and the melanosome matrix protein PMEL (14). Once melanogenesis is completed, eu-melanosomes assume an ellipsoidal form, while pheo-melanosomes retain their spherical shape with much lower structural organization (21). Mature melanosomes are transported (via multiple different modes) along dendrites and filopodia on their way to neighboring keratinocytes (22–25). This process represents an extremely rare example of organelle ‘donation’ from the cell type that makes it (i.e., melanocyte) to a wholly different histologic cell type (keratinocyte) that receives it. In the epidermis, melanin granules tend to aggregate as supranuclear caps within (often) proliferative keratinocytes, where it can protect their nuclear DNA from UVR-induced damage (26).

Melanocytes and keratinocytes exist in the human epidermis within a truly extraordinary and remarkably-stable unit of one post-mitotic melanocyte to approximately 36 viable keratinocytes - the so-called ‘epidermal melanin unit’ or ‘EMU,’ as first proposed by Breathnach and Fitzpatrick in 1963 (27). This unit is crucial for maintaining the integrity of the human epidermis and especially for its protective skin pigmentation (see Figure 2 ). Given that all human skin phototypes, from black to white, have the same melanocyte number per unit area of epidermis, much of the qualitative variation in human skin color is due to differences in a) the number of melanin granules per melanocyte; b) melanin type (i.e., eu-/pheo-melanin), and c) pattern of melanin granule distribution throughout various strata of the human epidermis (27). Within the EMU, the keratinocyte partner tightly controls several aspects of melanocyte behavior, including via a regulated balance of paracrine growth factors and cell-cell adhesion molecules. Distinct subpopulations of melanocytes seem to be differentially positioned within the basal layer of the human epidermis, generating at least two EMUs (28, 29) (see below).

Beyond the epidermis, the skin’s main appendage, the human hair follicle, contains a remarkable diversity of melanocyte subpopulations, the most apparent of which form a melanogenicially-active “follicular melanin unit” or “M-FMU” (30). These melanocytes, which engage principally with keratinocytes of the hair follicle pre-cortex, are located in the non-immunocompetent proximal anagen (growing phase) hair bulb and attach their cell bodies to the basal lamina separating the hair bulb epithelium from the mesenchymal (hair growth-inductive) follicular papilla (see Figure 3 ). Pre-cortical keratinocytes receive melanin from these M-FMU melanocytes to pigment the hair shaft (21, 30, 31). There also exist several other FMUs in anagen terminal (coarse) hair follicles, like those of the human scalp, operationally categorized on the basis of their melanogenic activity (see below). These include a) melanogenically-active melanocytes in the hair follicle infundibulum (In-FMU) and Sebaceous gland (Sb-FMU), and b) melanogenicially-inactive melanocytes located in the Stem cell niche of the hair bulge (Stem-FMU), Outer root sheath (O-FMU), and most Peripheral-Proximal hair bulb (PP-FMU) (See Figure 3 ).

Despite their shared neural crest origin and transient co-location in developing human epidermis [prior to the commencement of hair follicle morphogenesis (32)], clinical observations provide ample evidence for the relative independence of the EMUs and FMUs, especially in terms of how their respective melanogenesis is regulated. One can readily appreciate this in the striking snow-white hair of aged Black-Africans or the jet-black hair of alabaster, white-skinned youths of northwest European ancestry. Furthermore, there is selective and/or preferential EMU targeting in most cases of vitiligo, while M-FMU melanocytes alone are damaged by an immune-mediated pathology in acute alopecia areata (33). While some of these differences likely reflect distinct relationships with UVR (i.e., by virtue of their different locations in the skin), others are likely to involve differences in their respective immunological contexts and antigenic profiles in situ. The latter reflects, at least in part, the fact that FMUs in the anagen outer root sheath and hair bulb reside in a so-called ‘immune-privileged’ (MHC-I negative) part of the hair follicle epithelium (34, 35), contrasting with the immunocompetent MHC-I positive EMU. Interestingly, several of these melanocyte subpopulation differences can be retained in long-term culture (36, 37)

The mouse has been instrumental in providing significant insights into the regulation of mammalian pigmentation and, by extension, melanoma. However, readers should interpret these data with some caution, at least in the context of understanding how melanocyte-keratinocyte interactions in the epidermis are organized. This is so because mouse melanocytes are not present at the dermal-epidermal junction throughout most of their skin (except for ear, tail, nose, and food-pad), and mice do not develop melanoma without genetic manipulation. Moreover, induced melanomas in mice originate in the dermis, and while it is assumed that these tumors are derived from the proximal hair follicle, dermal melanocytes may also be a source of cutaneous melanoma in mice (38). Regardless, the very different melanocyte microenvironments in these two mammalian species will account for some of the many differences in melanoma development and progression between them. Furthermore, the heterogeneity of melanoma in outbred humans cannot be recapitulated in inbred laboratory mice, nor can the drivers for spontaneous melanoma development in humans. Indeed, the focus of mouse cutaneous pigmentation studies has been limited, in the main, to their hair follicle melanin unit (38).

Oncogenic transformation of a single melanocyte is the consequence of a multistage process of successive acquisition of (epi)genetic alterations affecting cell proliferation and survival (52). Like other solid malignant tumors, melanoma is characterized by uncontrolled proliferation, disruption of cellular and morphological differentiation, invasion, and metastatic spread to distinct organs. While different stages of melanoma progression are not easily distinguishable, pathological characteristics can be partially attributed to changes in intercellular communication between the transformed cell, its neighboring non-transformed cells, and their immediate microenvironment. One of the important early steps in melanoma development includes disruption of E-cadherin-mediated adhesive interaction between melanocytes and keratinocytes, accompanied by increased expression of N-cadherin, which facilitates proliferation and invasion of melanoma cells (53). Although changing interactions of melanoma cells with the tumor microenvironment are clearly very important, discussion of these phenotypic changes is beyond the scope of this review.

During oncogenic transformation, melanocytes in human skin begin to proliferate along the dermo-epidermal junction, progressively gaining independence from exogenous growth factors secreted by their keratinocyte partner cells in the EMU. Disruption in melanocyte-keratinocyte communication can trigger many phenotypic changes in both cell types, which may also include nesting into nevi and other benign lesions (54). With further damaging (genomic) alterations, single melanocytes or nevus nests of melanocytes may form dysplastic nevi or melanoma in situ subtypes ( Figure 4 ).

Melanoma cells usually adopt an epithelioid shape and lack many features of their normal antecedents e.g., the melanocyte’s dendricity, lone distribution among 5-8 normal basal keratinocytes, and pendulous distribution into the upper papillary dermis whilst remaining firmly anchored to the basement membrane zone that separates the stratum basale of the epidermis and dermis (30). Evidence of EMU disruption can be observed in distinct histological subtypes of melanoma. SSM and LMM exhibit differences in their growth location and in the morphology of their melanosomal components. In melanomas with a superficial spreading component, a pattern of intra-epidermal growth occurs, often with a marked pagetoid (i.e., upward) spread of enlarged malignant cells (55). These can cluster at both the dermo-epidermal junction as well as more superficial epidermal areas of a now disorganized epidermis, characterized by the loss of the stable EMU cell ratio with keratinocytes ( Figure 4 ).

On the other hand, the epidermis remains relatively thin in LMM, with proliferating malignant melanocytes restricted to the epidermal basal layer giving rise to a horizontal spreading phase referred to as Lentigo Maligna. There is typically other histologic evidence of severe solar damage (54) but only a modest increase in their EMU ratio with keratinocytes ( Figure 4 ). By contrast, NM is characterized by a marked vertical growth pattern, whereby malignant cells exhibit dermal invasion accompanied by a higher mitotic rate compared to thin SSM (56). NM cells are often enlarged spindle or epithelioid shapes and can be organized into aggregates of so-called cerebroid nests (57, 58). As such, NM typifies a marked dysregulation of cell communication of the EMU, as well as with cells of the dermis.

Although the main function of melanin is to protect and buffer against UVR and redox imbalance, this pigment can also affect melanoma cell behavior. There is evidence of dysregulated melanin synthesis in the early stages of melanomagenesis (59), with melanin synthesis and subtype (eu/pheomelanin) correlated with melanosome morphology and integrity. Clinico-pathological analysis has shown that dysregulated melanogenesis can shorten the overall survival time (OST) and disease-free survival time (DFS) of patients with advanced metastatic melanomas (60). It has been shown that induction of melanogenesis in metastatic melanoma cells is closely linked to an increase of the hypoxia-inducible factor (HIF)-1α expression, with concomitant upregulation of HIF-dependent pathways involved in the regulation of glucose metabolism, angiogenesis and stress responses (61). In addition to altered melanization, dysregulation of the EMU has been associated with changes in melanosome matrix components (62). For instance, SSM melanosomes typically display a round-oval architecture and high pheomelanin content, while in LMM, melanosomes retain their usual ellipsoidal shape, typical of non-malignant melanocytes and are highly melanized (eumelanin) (60). NM, by contrast, are often hypopigmented or amelanotic tumors (63, 64). Indeed, there is growing interest in the potential role of pheo-melanogenesis in melanoma, both as a risk factor that reduces UVR-protection in the skin but also potentially due to some intrinsic (i.e., UVR-independent) pro-carcinogenic characteristic of the pheomelanin polymer itself (65). Earlier studies reported that dysplastic nevi synthesize more pheomelanin than eumelanin (66) and that this may be linked to dysregulated and chronic oxidative stress (67).

The involvement of keratinocytes in the crosstalk with malignant melanocytes has been reported. For example, aberrant suprabasal expression of keratin 14 in the epidermis surrounding NM has been detected i.e., a marker of associated keratinocyte hyperplasia (68). This study highlights the importance of the keratinocyte microenvironment in melanoma biology, particularly the necessity of a balanced communication within the cell subpopulations of the EMU. Further, rare examples of recurrent LMM co-mingling with nests of basal cell carcinoma have also been reported (69).

Phenotypic heterogeneity is a defining characteristic of melanoma, in part because melanoma cells can dynamically and reversibly switch between differentiated and undifferentiated states, thereby exhibiting distinct proliferative, invasive, and tumor-initiating characteristics (70, 71). Without a precise understanding of the true status and nature of the melanoma ‘originator’ cell(s) in human skin, it remains difficult to delineate how a defined population of normal melanocytes can initiate a transformation process that ultimately gives rise to such a heterogeneous tumor. It may be reasonable to assume that single cells within the tumor occupy distinct cell states (e.g., of differentiation) or different positions on a dynamic landscape (i.e., continuum). The proportion of cells in different states and positions may influence the outcome and prognosis of this neoplasm.

The literature contains a plethora of data supporting the origin of cutaneous melanoma from either melanocyte stem cells (McSCs) or from mature differentiated melanocytes. Confounding this issue is that most data has emerged from model organisms, particularly melanocytes of the mouse dermis, rather than from research focused on the pigmentation of human epidermis, nor even murine pigmented epidermis (ear/tail). This furred and nocturnal species lacks melanogenically-active melanocytes in their adult truncal epidermis and exhibits several other key differences, including the enzymatic regulation of melanogenesis (28).

During embryogenesis, committed melanocyte precursors or melanoblasts, originating from a multipotent neural crest (71, 72), (some perhaps also from Schwann cell precursors), migrate through the dermis from which they arrive to ‘invade’ the developing epidermis. Subsequently, a subpopulation of these melanocytes change from E-cadherin to P-cadherin-expressing cells, thereby enabling them to enter the developing P-cadherin-rich hair follicles as the latter start to develop (30). Although the melanin units of the epidermis and hair follicle are thereafter distinct, they do remain open, especially when the skin tissue is stimulated.

Still, it is important to emphasize that a considerable fraction of cutaneous melanocytes reside outside of the epidermis in adult human skin; mostly located in multiple distinct compartments of the hair follicle, its associated sebaceous gland, and even in sweat glands [see Figure 3 ; (73)]. Thus, the question of the precise location of the melanoma-targeted pigment cell must be very relevant in any discussion of human melanomagenesis. For example, upon stimulation (e.g., UVR or wound healing), hair follicle McSCs can indeed migrate to the interfollicular epidermis and differentiate into pigment-producing melanocytes (73–75).

But the hair follicle and sebaceous gland are not the only sources of melanocytes in the interfollicular epidermis post-stimulation. For example, studies from mouse tail skin lacking appendages show that this epidermis contains a low frequency of immature amelanotic melanocytes, which may function as interfollicular McSCs (76–78). These cells are reminiscent of the rare immature melanocytes that survive in the leucodermic epidermis of long-duration human vitiligo (79) or of immature melanocytes in hair-less glabrous skin - the suggested source for repigmenting of the lip in vitiligo.

However, available data suggest that melanoma may not arise from immature melanocytes, melanoblasts, etc. In fact, recent lineage-tracing studies using mouse tail epidermis (containing melanocytes) reported that melanoma arises instead from mature melanogenically-active melanocytes of the inter-follicular epidermis (80). These fully-differentiated cells appear to be reprogrammed and thereafter de-differentiate into cancer-initiating cells in vivo. Thus, differentiated pigment cells may acquire multipotency like their embryonic ancestors, allowing them to form heterogeneous tumors.

Other data supporting a melanocyte dedifferentiation mode of melanomagenesis is the coincident loss of melanocyte differentiation markers during melanoma progression, as well as an upregulation of genes associated with earlier stages of their development (81). The well-accepted concept in cancer biology that tumorigenesis recapitulates embryonic development has also been validated in a zebrafish melanoma model, in which loss of melanocyte signatures and emergence of neural crest signatures precede the expansion of melanoma (82). The relevance of this in public sun care policy may be important given the common ‘mole-watch’ focus of many melanoma prevention strategies, while at the same time, most melanomas (>70%) arise not from such lesions but rather arise de novo. This latter finding was given a powerful explanation from a human study where melanocytes in healthy skin were found to commonly contain pathogenic mutations (albeit weakly oncogenic ones) (83). This probably explains why they do not give rise to discernible melanocytic lesions. It also highlights that elucidation of the genomic ‘landscapes’ of individual melanocytes can provide insights into the cause and origin of melanoma. Increasing awareness of the clinicopathological and body-site differences between melanomas arising de novo or in association with a pre-existing pigmentary lesion supports the divergent pathway model of melanomagenesis (84).

While much of the above data is derived from modified mice models and requires urgent confirmation in human melanoma, these data suggest overall that melanoma can arise either from immature or mature pigment cells. This may explain the variability in cutaneous melanoma presentation and their outcomes. Therefore, it may be reasonable to suggest that the very significant heterogeneity observed in melanoma results from the involvement of several distinct subpopulations of melanocytes in different stages of their life histories and within various distinct cutaneous melanin units.

A hallmark of the comparative biology of epidermal and hair follicle melanocytes is the observation that melanogenesis in the latter is stringently coupled to the hair growth cycle (anagen-specific melanogenesis), while melanogenesis in the former is continuous, albeit often augmentable (e.g., after exposure to UVR). Up to 90% of melanomas are said to be ‘caused’ by UVR from sunlight (85), and while much incident UVB and UVA readily reaches melanocytes in the superficial epidermis, hair follicle melanocytes are located more deeply in the skin, most located well below the level of UVB penetration (18, 29). That said, more superficially distributed melanocytes, e.g., in the upper infundibulum (In-FMU) and potentially also the bulge stem niche Stem-FMU), may receive carcinogenic doses of UVB. This may be particularly so in small hair follicles producing fine and vellus hairs (86).

However, it is remarkable that the skin’s main appendage, the hair follicle, appears to be largely resistant to melanomagenesis. This is despite the fact that keratinocyte neoplasms can occur in this skin appendage, although much less commonly than in the epidermis. If hair follicle tumors do occur, they are most often benign (e.g., pilomatrixoma); although some can be malignant (e.g., pilomatrix carcinoma) (86). This relative resistance of hair follicles to transform may be due to their inherent capacity to engage in life-long cycling (i.e., hair follicle cells naturally regenerate themselves). This may protect them against tumorigenesis and tumor growth, even in the presence of the well-known cancer-causing mutations (87). Melanomas do not typically arise from melanocyte/melanoblast subpopulations in the hair follicle, despite their life-long proliferative potential. Very rare exceptions, so-called follicular malignant melanoma, have been reported, including some arising from the sun-damaged skin of elderly individuals (88). However, even here, there is a strong possibility that this melanoma may not be primary in origin but rather folliculotropic i.e., migrate toward the hair follicle. Greater attention to these tumors may help us understand the origin of the melanoma cell, given that cycling hair follicles exhibit both the highest rate of epithelial cell proliferation (after the gut), as well as periodic bursts of melanoblast/melanocyte proliferation (28).

A contrary view has recently been proposed by Sun et al. These researchers genetically engineered a c-KIT promoter-driven mouse model so that these mice expressed oncogenic mutations (a combination of oncogenic Braf ^V600E induction and Pten loss) in their hair follicle McSCs (89). They were able to track their potential involvement in subsequent melanomagenesis, and found that these mutated melanoblasts, under the influence of Edn and Wnt signaling, migrated out of the hair follicle bulge into the overlying (melanocyte-free) epidermis and became invasive melanomas. Interestingly, these melanomas exhibited similar genetic and molecular characteristics to human melanomas. In another mouse melanoma model study, UVB-stimulated McSCs were translocated from the quiescent hair follicle niche into the epidermis in an inflammation-dependent manner (90). However, these studies are based on genetically-manipulated mouse models. It will need to be confirmed whether these are true models of human melanomagenesis.

Although the involvement of the upper hair follicle in spreading malignant melanoma has been reported, human melanoma characteristically progresses within the epidermis i.e., along the dermal-epidermal junction. This is in marked contrast to melanomagenesis in mice, which emerge from the dermis. In a study of 100 cases of primary human cutaneous melanomas, which also examined growth association with nearby hair follicles, most cases had melanoma tumor cells within at least one hair follicle. Of these, the vast majority of cases showed melanoma cells limited to the infundibulum . Less than a third of cases with hair follicle association showed melanoma cells extending a little deeper to the isthmus (the upper hair follicle between infundibulum and insertion site of arrector pili muscle). Remarkably, only in one exceptional case did the researchers detect melanoma cells below the level of the hair follicle bulge, located in the upper third of the hair follicle (91). These authors postulated that some ‘physiologic barrier’ might restrict the intra-epithelial spread of melanoma tumor cells at or beyond the level of the stem cell niche in the hair follicle bulge.

To our knowledge, clear evidence of melanoma originating solely within the M-FMU - the home of melanogenically-active and post-mitotic follicular melanocytes - is lacking (34). Together, these data suggest that the hair follicle contains a remarkably effective system of checks and balances, which may prevent melanocytes from going ‘feral’ (i.e., as in melanoma) or even from allowing migration of transformed pigment cells deep into the proliferative part of the growing hair follicle. The nature of such a physiological barrier remains unknown but may reflect the existence of an inhibitory ‘chalone,’ reminiscent of some such chalone that confers a ‘refractory’ state to some resting (telogen) hair follicles that fail to progress to a ‘competent’ growing stage during the hair growth cycle (92).