Developing agents for the therapeutic prevention of melanoma: can the assessment of cutaneous precursor lesions help?

Melanoma was the fifth most frequently diagnosed cancer in USA in 2019, excluding nonmelanoma skin cancers, and its incidence is increasing more rapidly than any other of the 12 most common cancers [1]. While primary prevention efforts including regular use of sunscreens and other sun-protective behaviors have traditionally been recommended to reduce the risk of melanoma, the continued rising incidence suggests that these have so far been insufficient to address the problem. Furthermore, although there has been immense and ongoing progress in therapy for advanced melanoma, new targeted antitumor therapies and checkpoint blockade immunotherapies have nontrivial toxicities and high costs. Finding additional ways to prevent melanoma that are effective and economical could help to reduce the burden of this disease on both patients and the healthcare system.

The equivalent terms ‘chemoprevention’ and ‘therapeutic prevention’ refer the use of natural or synthetic agents to prevent the initiation of carcinogenesis, or to delay or reverse the progression of premalignant lesions into invasive cancer [2]. A recent review by Jeter et al. [3] provides a useful summary of many of the agents currently being studied for therapeutic prevention of melanoma. In short, while a number of candidate agents have shown promising results in preclinical or population studies, including NSAIDs, nicotinamide, statins, vitamins and various phytochemicals; data from high-quality human trials are either not available or are incomplete, contradictory or negative.

Progress to definitive Phase III trials of agents for cancer prevention has been challenging for a number of reasons, ranging from the inadequate understanding of melanomagenesis to difficulties moving compounds forwards in the current regulatory/funding environment. Many of these issues are common to cancer prevention in general and are described in recent reviews by Meyskens et al. [4,5]. One particularly important factor in melanoma prevention is the difficulty of using cancer incidence as an endpoint: trials in a representative US population will likely require thousands of patients and many years of follow-up. It has been argued that conducting these trials in particularly high-risk patients may substantially lower the number of patients required [6]. However, the highest-risk patients have relatively uncommon risk factors such as multiple primary melanomas, confirmed CDKN2A or CDK4 germline mutations, or familial atypical nevus syndromes [7]; and these subpopulations may still be difficult to accrue in sufficient numbers. One potential solution is to identify surrogate endpoints that allow promising candidate agents to be evaluated in a meaningful way in earlier stage trials. However, there is no clear consensus about what these endpoints might be.

We and others have suggested that patients with multiple atypical/dysplastic nevi (A/DN) and a prior history of melanoma might be an appropriate high-risk population to enroll in melanoma prevention trials. These patients have an approximately ninefold increased risk of developing a second primary melanoma [8] and may be relatively easy to recruit for trials since they are already generally followed in dermatology and/or medical oncology clinics. The A/DN themselves provide plausible intermediate endpoints since they share a number of features with fully evolved melanoma, such as morphological features of larger size, border irregularity and color variegation. These A/DN lesions are thought to be biologically intermediate between simple banal nevi and fully evolved invasive melanoma [9]. An example of this biological intermediacy can be seen in the STAT3 pathway, which was shown to be crucial to the effect of neoadjuvant IFN-α-2b in metastatic melanoma [10]. Constitutive activation of STAT3 has also been demonstrated in atypical nevi and it correlates with the degree of pathologic atypia observed [11]. A/DN are also considered to represent nonobligate precursor lesions, which means that while any particular A/DN lesion itself is unlikely to transform into melanoma, at least 20–30% of melanomas are thought to arise from A/DN [9]. Finally, A/DN occur much more frequently than melanoma, which may enable augmented accrual and allow meaningful results from much smaller sample sizes than would be needed for a trial studying melanoma incidence.

The use of precursor lesions as a surrogate endpoint in cancer prevention has well-established precedents in other tumor types. For example, adenomatous polyps are precursor lesions for colorectal cancer, and the presence of adenomas conveys increased risk of developing colorectal cancer [12]. In a population of patients at moderately high risk for spontaneous adenomas, the combination of sulindac and eflornithine produced remarkable decreases in colon adenomas [13]; this motivated the large ongoing Phase III PACES trial being run by the Southwest Oncology Group (S0820) that includes the incidence of colorectal cancer in its primary endpoint. In nonmelanoma skin cancer, nicotinamide was first shown to decrease the numbers of actinic keratoses in high-risk patients in two separate Phase II trials [14], and was subsequently shown to substantially and significantly lower the incidence of nonmelanoma skin cancer in a Phase III trial [15]. In both of these examples, initial effects observed in precursor lesions provided a plausible and accelerated readout that motivated the design of larger studies with cancer incidence as a primary endpoint.

Despite the foregoing theoretical rationale, there are challenges to using atypical nevi as surrogate endpoints for melanoma prevention. There have been longstanding disagreements about the best histologic criteria to use to define dysplastic nevi [9]. Their natural history is variable: while the large majority of atypical nevi show very little overt change over time, some disappear spontaneously and a small number actually progress to become melanoma [16]. Furthermore, there are different general biologic subtypes of cutaneous melanoma that differ in their epidemiology, location on the body surface, suspected precursor lesions and mechanisms of carcinogenesis [17]. Thus, it is possible that modulating features of atypical nevi may not be equally relevant for the prevention of all types of melanoma. Finally, there is very little literature available to suggest systemic agents that can alter the features of atypical nevi and cause them to regress and/or develop less frequently into melanoma. This may be either because these treatments don't exist; or because identifying them has simply not been a priority in prevention research, given that the nevi themselves do not cause morbidity. These potential limitations were illustrated in two recent Phase II studies of sulindac and lovastatin for cancer prevention in patients with atypical nevi. Both failed to show clear effects upon the morphology and/or histology of atypical nevi [18,19]. Although these results are discouraging, it is not clear whether the general approach was misguided, or whether the agents studied were the right drugs to pursue.

To further assess whether atypical nevi may be relevant to melanoma prevention, it will be helpful to learn more about their behavior over time, which has recently become more feasible. Traditionally, it has been difficult to analyze gene expression changes in nevi over time, given that the associated traumatic changes associated with repeated biopsy of the same lesion might confound interpretation of subsequent biopsies. There is now a validated two-gene assay, based on noninvasive tape stripping of the stratum corneum, that has high sensitivity and specificity to distinguish atypical nevi from melanomas [20]. This assay is now in clinical use, and the technology could theoretically be applied to serially study the expression of a wider range of relevant genes in the same nevus over time. At the simplest level of morphological assessment of skin lesions, software has been developed and commercialized that uses image registration and artificial intelligence algorithms to identify the appearance of new pigmented lesions as well as to track changes in existing pigmented lesions over time [21]. Technology like this might be used both prospectively to answer specific research questions and also retrospectively to detect previously unrecognized changes in nevi present in total body digital images that are obtained routinely in some multidisciplinary melanoma centers. Finally, the analysis of skin lesions at specific snapshots in time has also improved significantly; for example, the rapidly evolving technology of multiparameter immunohistochemistry may help provide simultaneous information about both the melanocytes and the associated immune cells in a precursor lesion, as has been done in metastatic melanoma [22].

In conclusion, the development of agents for the therapeutic prevention of melanoma is a worthwhile challenge that will be facilitated by the identification of appropriate patient populations and valid surrogate endpoints. We have proposed A/DN as both risk markers for patient selection and as a possible source of surrogate endpoints for future trials. More research is needed to identify the most appropriate endpoints and biomarkers, and to more fully validate this approach.