The biology and treatment of Merkel cell carcinoma: current understanding and research priorities

Merkel cell carcinoma (MCC) is a rare and aggressive skin cancer associated with advanced age and immunosuppression. Over the past decade, an association has been discovered between MCC and either integration of the Merkel cell polyomavirus, which likely drives tumorigenesis, or somatic mutations owing to ultraviolet-induced DNA damage. Both virus-positive and virus-negative MCCs are immunogenic, and inhibition of the programmed cell death protein 1 (PD-1)–programmed cell death 1 ligand 1 (PD-L1) immune checkpoint has proved to be highly effective in treating patients with metastatic MCC; however, not all patients have a durable response to immunotherapy. Despite these rapid advances in the understanding and management of patients with MCC, many basic, translational and clinical research questions remain unanswered. In March 2018, an International Workshop on Merkel Cell Carcinoma Research was held at the US National Cancer Institute, at which academic, government and industry experts met to identify the highest-priority research questions. Here, we review the biology and treatment of MCC and report the consensus-based recommendations agreed upon during the workshop.

The specific host cell type for MCPyV infection has thus far remained elusive. Benign Merkel cells are not suffi ciently numerous to account for the MCPyV burden typically detected in skin 27 . Peripheral blood monocytes have been proposed to act as a reservoir of infected cells 24 . MCPyV reporter pseudovirus can enter many cell types, including keratinocytes 28,29 , although dermal fibroblasts and HEK 293 cells are the only cells in which productive viral infection has been demonstrated in vitro 28,30 .
The viral life cycle of MCPyV is similar to that of other polyomaviruses. The episomal viral genome possesses an early region (ER) and late region (LR), which contain genes encoding proteins that coordinate viral replication and viral capsid proteins, respectively 31 ( fig. 2). Genes in the ER of MCPyV encode the large T antigen (LT), small T antigen (ST) and 57 kT antigen transcripts ( fig. 2b). The middle T-like overprinting gene large T open reading frame (ALTO) is also proposed to be located in the ER [32][33][34] . The MCPyV LR encodes the viral protein 1 (VP1) and VP2 capsid proteins 31,35 and might enable expression of the microRNA MCV-miR-M1-5p (which is located on the LR strand despite mapping among genes in the ER) 33,34,36 . Productive viral infection is associated with host cell death rather than oncogenic transformation 30 . MCPyV replication is regulated by E3 ligase targeting of phosphorylated LT, as well as feedback inhibition by LT on its own promoter 30 and inhibition of LT production by viral microRNA 36 . These features act together to inhibit viral replication after entry into a host cell; the virus typically then defaults to a latent, nonreplicative state after infection 30,34 .
Oncogenic transformation by MCPyV is hypothesized to require two events: integration of the viral genome into the host genome and truncation of LT to render the viral genome incapable of replication 4,37 ( fig. 3). Viral integration into the host genome might occur by accidental genome fragmentation during MCPyV replication; the precise location of the integration site seems to be random, without consistent involvement of specific cellular tumour suppressor genes or oncogenes 4,[38][39][40][41][42] . In virus-positive (VP)-MCC, mutations in MCPyV gp3 encoding LT disrupt the DNA binding domain and the helicase domain distal to the retinoblastoma-associated protein (RB) binding motif ( fig. 2c). The resulting truncated LT retains its ability to bind to RB and promote cell cycle progression 37 but cannot mediate viral replication 43,44 . The integrated and mutated virome no longer produces MCPyV virions. The very low probability of this required combination of events might explain why MCC is rare despite the apparent ubiquity of MCPyV infection.
LT and ST have diverse activities that might contribute to oncogenesis. Evidence suggests that cultured VP-MCC cells are dependent upon the presence of both viral T antigens for proliferation and survival [45][46][47] . The common amino-terminal region of viral T antigens associates with heat shock cognate 71 kDa protein, with roles in the regulation of viral replication and possibly tumour cell proliferation 43,48 (fig. 2c,d). Similar to the SV40 polyomavirus large T antigen, the MCPyV LT directly binds with and inactivates the RB ( fig. 2c). This interaction has been shown to enhance expression of the oncoprotein survivin, which might be an important therapeutic target in patients with VP-MCC tumours 49 . Unlike the SV40 large T antigen, MCPyV LT has not been shown to directly bind with p53 (ref. 50 ). MCPyV LT is also able to associate with the vacuolar sorting protein Vam6/Vps39-like protein, which has potential importance for viral replication 51 . Additional functions of LT, which are likely to be lost after truncation, include DNA binding, bromodomain protein 4 (BRD4) binding, helicase activity and inhibition of cell growth 31,52 . The tumour-associated mutations that truncate LT in MCPyV are also likely to disrupt ALTO 32 .
Multiple lines of evidence suggest that ST has a critical role in VP-MCC oncogenesis. ST expression is sufficient to transform rat-1 fibroblasts in culture 53 . Data from transgenic mouse models indicate that ST expression is transformative in various organ systems, including in the epidermis [54][55][56][57] . In mouse models, MCPyV T antigens alone are generally not sufficient to induce the formation of neuroendocrine tumours 55 ; C o n S e n S u S S tat e m e n t b a , presenting as a rapidly growing nodule on an extremity ; such lesions can also be commonly observed on a patient's head or neck. b | Light micrograph (×400 magnification) of a sample stained with haematoxylin and eosin depicting the histological appearance of MCC, demonstrating the presence of round cells with scant cytoplasm, neuroendocrine chromatin and numerous mitotic figures (arrows). Trabecular patterning might also be prominent.
however, co-expression of ST with the transcription factor ATOH1 generates intraepidermal MCC-like proliferations 57 and, with the loss of TP53, undifferentiated anaplastic tumours 54 .
ST has a diverse range of cellular activities ( fig. 2d). Similar to other small T antigens, the MCPyV ST has a PP2A region that binds with and inhibits various components of the serine/threonine phosphatase complexes PP2A and PP4 (ref. 58 ). Interactions between ST and PP4 might also mediate inhibition of the nuclear factor-κB (NF-κB) pathway, as well as promote cell motility via effects on the actin and microtubular cytoskeleton 59  includes the retinoblastoma-associated protein (RB) binding motif responsible for direct inhibition of RB, thus enabling cell cycle progression to S phase. The MUR mediates binding to the vacuolar sorting protein VPS39. The carboxyl-terminal helicase domain is a critical mediator of viral replication, with contributions from the adjacent zinc-finger (ZN), leucine zipper (LZ) and origin binding domain (OBD). The integrated MCPyV genome present in MCPyV-positive MCCs harbours truncating mutations in LT that disrupt the helicase domain, resulting in a replication-incompetent mutant form of LT that nonetheless retains the ability to promote cell cycle progression. d | Cellular functions of MCPyV ST. ST recruits L-MYC to the EP400 chromatin remodelling complex in order to mediate changes in gene expression. The LT-stabilizing domain (LSD) is proposed to inhibit E3 ubiquitin ligase activity via interactions with F-box/WD repeat-containing protein 7 (FBXW7) and cell division cycle protein 20 homologue (CDC20), resulting in increased oncoprotein stability and cap-dependent mRNA translation, respectively. ST also interacts with the protein phosphatase complex PP4 to inhibit nuclear factor-κB (NF-κB) signalling. 4E-BP1, eukaryotic translation initiation factor 4E-binding protein 1; CR1, conserved region 1; NEMO, NF-κB essential modulator; NLS, nuclear localization signal.
C o n S e n S u S S tat e m e n t oncogenic proteins, including MYC, might be stabilized by the ST via a similar mechanism 63 . Interaction of the LSD domain with cell division cycle protein 20 homologue (CDC20), an activator of the anaphase-promoting complex (also known as the cyclosome) ubiquitin ligase, has been reported to trigger a series of signalling events culminating in the inhibitory phosphorylation of the eukaryotic translation initiation factor 4E-binding protein 1 (4E-BP1) 64 , leading to derepression of capdependent mRNA translation with potential pro-survival effects 65 . The LSD domain is required for oncogenic activity by ST in vitro and in vivo, thus supporting a critical role of this domain in the oncogenesis of MCPyV-induced MCC 55,57,63 .
ST has been shown to bind with L-MYC to regulate the EP400 histone acetyltransferase and chromatin remodelling complex 66 (fig. 2d). Additional activities involving ST include increased aerobic glycolysis, possibly via the MYC and/or NF-κΒ signalling pathways 67 .

Detection of tumour-associated MCPyV.
Methods used for the experimental detection of MCPyV in tumours include immunohistochemistry, PCR, RNA or DNA in situ hybridization and next-generation sequencing (NGS) 39,[68][69][70][71][72] . These assays vary substantially in terms of sensitivity and specificity for the detection of tumour-associated MCPyV. Measuring the expression of T antigen proteins determined using immunohistochemistry is a common approach for the detection of MCPyV. The most broadly used antibody for MCPyV LT detection, CM2B4, is commercially available and has an approximately 88% sensitivity and 94% specificity (compared with that of multimodal approaches combining PCR with immunohistochemistry) 73 . Nonspecific staining with CM2B4 can be observed in both tonsillar and lymphoid tissues 73,74 . Other antibodies recognizing the LT, ST or the common T antigen region have been reported and might also be useful in detecting viral proteins 73 .
PCR is another common method of MCPyV detection. The 5ʹ end of the second exon of LT is frequently targeted using this approach 70 . Compared with multimodal MCPyV detection, PCR-based amplification within the second exon of LT had 83% sensitivity and 81% specificity 73 . The sensitivity of PCR-based MCPyV detection can be improved by additional PCR assays targeting other amplicons in the ER, including MCPyV gp4 encoding ST 70 . Quantitative PCR (qPCR) enables estimation of the number of integrated MCPyV copies per host cell genome. Copy numbers can be quantified by comparisons to the reference MCC cell line MKL-2, which has the lowest relative MCPyV copy number among established VP-MCC cell lines and is thus estimated to have a single viral copy per cell 70 . Similar results are obtained by comparing MCPyV qPCR values to those of selected reference genes from the human genome 71,75,76 . MCPyV copy number estimates can range from extremely low (<1 copy per 100 cells) to thousands of copies per cell 70 . The reasons for extremely low MCPyV copy numbers are incompletely understood but might be caused by technical factors, including inefficient PCR amplification owing to mutations in the integrated MCPyV genome, low purity of tumour samples or the detection of infectious wild-type MCPyV in the adjacent nonmalignant skin 77 . In low-purity tumour samples, copy number estimates of tumour-associated MCPyV might be confounded by background signals from skin and/or non-MCC tumours 69,77 . qPCR does not allow for visual confirmation that positive results are associated with tumour cells; therefore, background MCPyV infection cannot be excluded in MCC tumours with low signal 69 . A multimodal approach incorporating , the cell of origin undergoes ultraviolet-mediated DNA damage, resulting in a high tumour mutational burden and inactivation of tumour suppressor genes, including RB1 and TP53. The high mutational burden might result in the expression of tumour neoantigens that represent potential targets for antitumour immunity. b | In virus-positive MCC (VP-MCC), the cell of origin is infected by wild-type Merkel cell polyomavirus (MCPyV), which undergoes episomal replication. Rarely , MCPyV can become integrated into the host cell genome and further acquires a truncating mutation of the large T antigen (LT), resulting in deficient viral replication with continued production of viral oncoproteins. The resulting tumour has a low burden of cellular genomic mutations. Hence, for patients with VP-MCCs, T antigen proteins (rather than neoantigens) might be better targets for treatments designed to promote antitumour immunity.
both PCR and immunohistochemistry is likely to be a more sensitive and specific method for confirming MCPyV status among commonly used assays 73 .
Other newer approaches for MCPyV detection have been less extensively investigated but might have advantages over immunohistochemistry and/or PCR. RNA in situ hybridization might provide similar levels of sensitivity to that of PCR while also enabling visual correlation with tissue morphology and the exclusion of background infection 69 . NGS can be effective in detecting MCPyV sequences, including tumour-specific truncating mutations. NGS using a hybrid capture approach can further demonstrate the presence of viral integration sites 39 and therefore provides the greatest level of specificity for confirming the presence of tumour-specific MCPyV. However, the time, expense and expertise required for hybrid capture NGS currently make this approach impractical for many diagnostic and research laboratories.

Non-viral changes in MCC.
Similar to other malignancies, MCCs typically harbour genomic aberrations including chromosomal copy number variations (CNVs) and other mutations. However, significant differences in the patterns of these various changes exist between VP-MCCs and virus-negative (VN)-MCCs. A standard terminology for designating the viral status of MCCs has not been established. We propose the terms VP-MCC and VN-MCC to describe virus-positive and virus-negative MCCs, respectively, as useful standard abbreviations for this purpose.
Most MCCs harbour chromosomal CNVs, regardless of viral status 78  VP-MCCs and VN-MCCs have distinct patterns of somatic mutation. VN-MCCs have a high mutational burden, an ultraviolet mutational signature and highly recurrent inactivation of tumour suppressor genes, including TP53, RB1 and genes encoding members of the Notch family of signalling proteins 5-7,82 ( fig. 3). By contrast, VP-MCCs tend to have a low mutational burden, no definitive mutational signature and an absence of TP53 and RB1 mutations [5][6][7] . Mutations in VP-MCCs are predominantly subclonal, suggesting that the majority of these mutations occur late in tumour evolution rather than as early or founder events 79 , possibly arising owing to transcription-coupled damage or in the setting of double-stranded DNA breaks 38 . By contrast, mutations in TP53 and RB1 in VN-MCCs are typically clonal and shared by primary tumours and their matched metastases 79 . Hotspot activating mutations of several different oncogenes are observed in both VP-MCCs and VN-MCCs, including HRAS, KRAS and PIK3CA 5-7,83-85 . Other hotspot activating mutations have been described in VN-MCCs, but not in VP-MCCs, including mutations in KNSTRN, RAC1, AKT1, CTNNB1 and EZH2 (refs 5,7,83,86 ). Most VN-MCCs have detectable activation of at least one known proto-oncogene.
The genomic changes observed in MCC have implications for the relative contributions of the major intracellular signalling cascades to tumorigenesis. A minority of MCCs harbour mutations or chromosomal copy number alterations predicted to result in activation of the AKT-PI3K signalling pathway, affecting genes including PIK3CA, PIK3CG, AKT, PTEN, PREX2 and TSC1 (refs 5-7,83,85,87 ). The MAPK signalling pathway is not constitutively activated in MCCs 88 , and reports have been mixed regarding whether MAPK activation promotes proliferation or apoptosis of cultured MCC cells in vitro 89,90 . HRAS hotspot mutations in MCC have been associated with increased ERK phosphorylation but not with sensitivity to MEK inhibitors in an analysis involving VN-MCC cell lines 7 ; the possibility that RAS might also activate other pathways, including the PI3K pathway 91 , has not been investigated in this context. Coding mutations in receptor tyrosine kinases and RAF family members can also occur in MCC, although classic activating mutations or fusions affecting the genes encoding these proteins have yet to be identified [5][6][7]86,88,[92][93][94][95] . Evidence suggests that activation of Wnt signalling is not a common event in MCC tumours [96][97][98] , although activating β-catenin (CTNNB1) mutations 86 or nuclear β-catenin accumulation 96,97 can both be detected in small subsets of tumours. Molecular studies have failed to detect activation of Hedgehog signalling in most MCCs 99 , despite histological detection of Hedgehog pathway proteins 100,101 . Functional studies of these and other oncogenic pathways in VN-MCCs have been confounded by a lack of transgenic mouse models 102 and the atypical biology of several MCPyV-negative cell lines 103,104 .
Epigenetic deregulation, including modifications of both DNA (especially promoter silencing by CpG island hypermethylation) and histones, can also contribute to tumour aggressiveness. Promoter hypermethylation in MCC can affect several genes, including DUSP2, CDKN2A and members of the RASSF family [105][106][107][108] . The polycomb group complex, which includes the histone-lysine N-methyltransferase EZH2, mediates gene silencing via histone H3 lysine 27 trimethylation and can also be deregulated in MCC 83,109,110 . Notably, epigenetic silencing of HLA genes by histone deacetylases might contribute to immune evasion in the development of MCC [111][112][113] .

MCC cell of origin.
Identifying the cell type from which a neoplasm arises has implications for ascertaining the mechanisms of tumour initiation, experimental modelling and possibly uncovering therapeutic susceptibilities. MCC is a poorly differentiated neuroendocrine carcinoma that lacks a recognized benign or dysplastic precursor. Furthermore, MCCs are most frequently found in the dermis but can arise from any layer of the skin (from intraepidermal to subcutaneous) 3 . Thus, fundamental evidence on the MCC cell of origin is currently lacking. MCC cells have immunophenotypic C o n S e n S u S S tat e m e n t and ultrastructural similarities to benign Merkel cells; however, Merkel cells are postmitotic in vivo, and the anatomical regions with the highest Merkel cell density are not the most frequent sites of MCC detection 3,31 . On the basis of these observations, most investigators do not consider mature Merkel cells to be a likely candidate for the MCC cell of origin 9 . Instead, MCCs are more likely to be derived from a population that recapitulates the Merkel cell differentiation pathway either before or during neoplastic transformation. In addition, given that VN-MCCs have ultraviolet mutational signatures whereas VP-MCCs do not, these tumour types might arise from distinct cells of origin with different levels of baseline DNA photodamage 102 .
Proposed candidates for the MCC cell of origin include pro-B cells or pre-B cells, fibroblasts, dermal mesenchymal stem cells and epidermal progenitor cell populations, among others 28,55,57,[114][115][116] . Currently, no consensus has been reached regarding the most likely candidate for the MCC cell of origin. A comprehensive discussion of the evidence related to each putative cell of origin is beyond the scope of this article.
Other virus-associated malignancies might be informative when considering candidates for the MCC cell of origin. In cervical squamous cell carcinomas, the activity of viral oncogenes in intraepithelial precursor lesions that progress to epithelial tumours supports human papilloma virus (HPV)-infected epithelial stem cells as the cells of origin 117 . For patients with Kaposi sarcoma herpesvirus (KSHV)-associated primary effusion lymphoma, recent observations have raised the possibility that mesothelial cells might be the cell of origin via mesenchymal-to-lymphoid transformation 118 ; therefore, tumour phenotype might not always be a reliable predictor of the cell of origin.
Clinical features of MCC Epidemiology. As of 2013, the annual incidence of MCC in the USA was 0.7 cases per 100,000 people 119 . The incidence of MCC in the USA almost doubled between 2000 and 2013 and is expected to exceed 3,000 cases per year by 2025 (ref. 119 ), with similar increases in Australia and many but not all European countries 120 . The basis for this increasing incidence is unclear but might be related to an ageing population and improvements in diagnostic recognition 119 . The frequency of MCC is higher closer to the equator and much lower among those of non-white ethnicities 120,121 , suggesting an association with sensitivity to ultraviolet radiation. Australia currently has the highest reported incidence of MCC (up to 1.6 cases per 100,000 person-years) 121 , notably with a higher percentage of VN-MCCs, probably reflecting a higher risk of environmental ultraviolet exposure [122][123][124] . MCC also has a higher incidence among immuno suppressed populations 125 . Apart from immuno suppressed individuals, MCC arises almost exclusively in those of an advanced age. Incidence estimates fall below zero in patients under age <40 years of age 119 , and 90% of patients are >50 years of age 126 . Unlike many other cancers, in which disease incidence peaks and then declines with increasing age, the incidence of MCC continues to increase even as patients reach 80 or even 90 years of age, possibly owing to immune senescence 119 . Thus, MCC affects populations that frequently have substantial comorbidities that could complicate patient management.
Presentation and diagnosis. MCC classically presents as a rapidly growing red or violaceous nodule on the sun-exposed skin of an elderly, fair-skinned individual. However, up to 15% of patients with MCC will present with a tumour-positive lymph node without an identifiable cutaneous tumour, presumably reflecting metastatic disease with regression of the primary skin tumour 3,9,127 .
Confirming a diagnosis of MCC requires evaluation of histopathological and immunohistochemical findings. The histopathological differential diagnosis might include lymphoma, small-cell melanoma, meta static small-cell lung cancer (SCLC) to the skin and other small round-cell neoplasms involving the skin 3 . MCC can also be misdiagnosed as basal-cell carcinoma 128 . Thus, immunohistochemical investigations are necessary for diagnostic confirmation. MCCs typically express neuroendocrine markers, including chromo granin A and/or synaptophysin, although these are not specific markers. Cytokeratin 20 (CK20) is expressed focally or diffusely in most MCCs, typically in a paranuclear dot-like pattern, a cytoplasmic and/or membranous pattern, or both. Staining for neurofilament, another intermediate filament, can also have a paranuclear dot-like pattern. CK20 expression and the para nuclear dot-like pattern of intermediate filament staining are highly suggestive of MCC 3 . Synoptic reporting of newly diagnosed lesions facilitates treatment decisions and prognostic studies and as a minimum should include gross tumour size (in centimetres), peripheral and deep margin status and the extent of both lymphovascular invasion and extracutaneous extension (bone, muscle, fascia and/or cartilage involvement) 129 .
MCC is morphologically indistinguishable from metastatic SCLC. Immunohistochemical markers that enable distinction between these two entities include CK20, TTF1, MCPyV LT and neurofilament proteins. The MCC immunophenotype is CK20 + , LT +/− , neurofilament + and TTF1 − . Metastatic SCLC to the skin is CK20 − , LT − , neurofilament − and TTF1 + (ref. 3 ). However, expression patterns vary, and no single marker used in isolation is sufficiently sensitive or specific for the robust distinction of MCC from metastatic SCLC. Thus, a panel of markers is necessary, particularly for the diagnosis of challenging cases such as CK20-negative MCC. In addition, certain non-lung small-cell carcinomas (SmCCs), such as parotid and uterine cervical primary tumours, also frequently express CK20 and might, therefore, be more challenging to distinguish from MCC, especially in the setting of metastatic MCC from an unknown primary site 3 . Distinguishing between primary parotid SmCC and metastatic MCC of unknown primary is especially problematic, as the parotid is frequently a site of regional MCC metastasis, and primary parotid SmCC is rare 130 . As NGS-based analyses enter more widespread use, MCPyV detection combined with analysis of mutational signatures might be useful for distinguishing MCC (MCPyV-positive or with ultraviolet signature mutations) from metastatic SCLC (MCPyV-negative with smoking signature mutations) 83 . For example, ultraviolet signature mutations detected in a parotid gland tumour confirmed a diagnosis of metastatic MCC of unknown primary origin rather than a primary parotid neuroendocrine carcinoma 5 .
Patients with MCC have the potential to develop distant cutaneous metastases. In patients who present with a rare second cutaneous MCC that is spatially and temporally separated from the initial primary MCC, such that the tumour is clinically designated a second primary, investigations of clonality based on copy number alterations, mutations and/or MCPyV sequencing might be useful in distinguishing metastatic disease from a second primary MCC 79,131 .
Prognostic findings. The most clinically relevant prognostic parameters for MCC include tumour size (defined as maximum tumour diameter) and the presence of locoregional or distant metastases. These factors form the basis of the American Joint Committee on Cancer staging system for MCC 132,133 . Larger primary tumour size correlates with an increased risk of metastatic disease, although MCCs of any size confer a substantial risk of occult metastasis, thus supporting the use of sentinel lymph node biopsy sampling for all patients with clinically node-negative disease 134 . In a large-cohort study with results published in 2015, Breslow depth (the distance in millimetres between the top of the granular layer of the epidermis and the deepest point of tumour invasion) was found to be an independent prognostic risk factor for overall survival and sentinel lymph node positivity and was moderately correlated with tumour diameter (r = 0.53, P < 0.0001) 135 . Additional features of the primary tumour, such as lymphovascular invasion and tumour growth pattern, have also been proposed to have prognostic significance but have not been evaluated on a large scale [136][137][138] . The presence of clinically detectable nodal disease is associated with worse outcomes than the presence of microscopic metastases 132 . Other findings associated with a worse prognosis include a sheet-like pattern of involvement in lymph node metastases and an increasing number of lymph nodes with metastatic involvement 133,139 .
Owing to the difficulties in predicting aggressive disease on the basis of clinical and morphological findings alone, prognostic biomarkers for MCC have been an area of intense investigation. Studies of the prognostic role of MCPyV status have, thus far, had mixed results, predominantly finding either a worsened prognosis in patients with VN-MCC tumours or no difference relative to VP-MCC tumours 140 . The largest study to date to incorporate both immunohistochemical and PCR-based evaluations of MCPyV status found a better prognosis in patients with VP-MCCs 73 . Nevertheless, both VP-MCCs and VN-MCCs can have clinically aggressive and fatal courses.
For patients with VP-MCC, MCPyV serology can be informative for prognosis and can, therefore, guide disease management. Higher anti-VP1 antibody titres and the presence of anti-ST antibodies in the serum at diagnosis have been associated with more favourable outcomes 141,142 . A finding of persistent or re-emergent anti-ST antibodies in sera has been associated with a poor prognosis and a high risk of disease recurrence 141 .
Interestingly, patients presenting with nodal or presumed metastatic MCC with no identifiable skin lesions can have better outcomes than patients with disease of the same stage with a known primary tumour 119,127,132 . An antitumour immune response has been proposed to underlie both the primary tumour regression and improved patient outcomes in such patients. Analogously, data from multiple studies have confirmed the importance of immune competence as a determinant of prognosis in patients with MCC; immunosuppressed patients (including those with chronic lymphocytic leukaemia) have not only a higher incidence of MCC but also markedly worse survival outcomes 125,143 .
Biomarkers associated with immune competence have been extensively investigated in patients with MCC. Patterns of immune-cell infiltration might reflect both MCPyV status and disease outcomes. Relative to VN-MCCs, VP-MCCs are more likely to be associated with evidence of a brisk inflammatory response and an increased number of CD8 + T cells [144][145][146] . Immunological findings that have been correlated with an improved prognosis among patients with MCC include increased CD8 + T cells (either tumour-infiltrating T cells or those located at the tumour periphery), tumour-infiltrating MCPyV-specific T cells and tumour programmed cell death 1 ligand 1 (PD-L1) expression 111,[145][146][147][148][149] .
Additional markers proposed to have prognostic significance in MCC include tumour protein 63 (p63), EZH2, survivin (nuclear pattern), CD34 (vascular density), VEGF, vascular E-selectin, sonic hedgehog protein, phosphorylated signal transducer and activator of transcription 5B, cell adhesion molecule 1 and myelin and lymphocyte protein, among others 3 . TP53 mutations have also been shown to have prognostic implications 6 , although whether this effect is independent of MCPyV status remains unknown.

Management of MCC.
Surgical excision with 1-2 cm margins, typically followed by radiotherapy, are the mainstays of the management of primary MCCs. Adjuvant radiotherapy to the primary tumour site is often recommended; however, the morbidity risks associated with radiotherapy can be avoided with a low local recurrence rate in a subset of patients (such as those with tumours <2 cm in size without other adverse prognostic features) 150 .
Owing to the risk of occult nodal disease, sentinel lymph node biopsy sampling is recommended by the National Comprehensive Cancer Network (NCCN) for patients without clinically detectable metastatic disease 8 . Any size of metastatic deposit is currently considered positive with regard to nodal staging; therefore, immunohistochemistry for broad-spectrum cytokeratins and/or CK20 is routinely used to improve the detection of micro metastases in sentinel lymph nodes 3,151 . The NCCN recommends the management of clinically detectable or occult nodal disease using imaging investigations for distant metastases, followed by lymph node dissection and/or radiotherapy to the nodal basin 129 .
C o n S e n S u S S tat e m e n t Systemic therapies for patients with MCC have traditionally included chemotherapy, such as platinum-based drugs, taxanes, anthracyclines and etoposide. Owing to a lack of durable responses and no established effects on survival, chemotherapy is currently considered to have a palliative role 152 .
In the past 2 years, the anti-PD-L1 antibody avelumab has been approved as a therapy for patients with metastatic MCC by the FDA, the European Medicines Agency (EMA), Swissmedic and the Japanese Ministry of Health, Labor and Welfare 153 . Other immunotherapies for MCC have also demonstrated efficacy in clinical trials 154 . The success of immune checkpoint inhibition (ICI) is a milestone in the management of advanced-stage MCC. However, not all patients have durable responses to ICI. Furthermore, patients requiring immunosuppression as recipients of solid organ transplants or those with autoimmune disease might not be optimal candidates for ICI. Therefore, predicting and improving responsiveness to immunotherapy and identifying alternative therapies for patients in whom ICI is contraindicated and/or ineffective are both current research priorities in patients with MCC.

Investigational therapies in MCC
Immunotherapies. Several observations suggested that ICI would be effective in patients with MCC. Spontaneous regression of MCC is an uncommon but well-documented event, presumably owing to immune activation. Patients with T cell-infiltrated MCCs have better outcomes than those with tumours lacking infiltrating lymphocytes. The abscopal effect of irradiation (in which localized radiotherapy is associated with regression of additional tumour foci located outside of the treatment field) has also been observed in patients with MCC, suggesting a modulated immune response 155 . MCCs have a higher incidence and worse prognosis in the setting of immune compromise. Finally, MCCs express either MCPyV antigens or ultraviolet-mutation-associated neoantigens, providing putative targets for antitumour immunity 9 ( fig. 3).
Both the innate and the adaptive arms of the immune system can have antitumour effects. Adaptive antitumour immunity is mediated by the activation of effector T cells by antigen-presenting cells and is modulated by proteins expressed on the surface of both immune cells and tumour cells (fig. 4). Activation of the immune system can be stimulated by multiple signalling pathways, including the OX40-OX40L and CD137-4-1BBL axes. Other signalling pathways (including the cytotoxic T lymphocyte antigen 4 (CTLA-4) and programmed cell death protein 1 (PD-1)-PD-L1 immune checkpoints) suppress immune responses and can be exploited by tumour cells as a mechanism of immune evasion 156 . Thus, immunotherapies promote antitumour immunity either by activating stimulatory pathways or by suppressing inhibitory pathways 156 . Of these, inhibition of the PD-1-PD-L1 immune checkpoint has been intensely investigated in MCC. Several open-label phase II clinical trials involving anti-PD-1 or anti-PD-L1 antibodies have demonstrated durable responses to these agents in a subset of patients with MCC. In 88 patients with metastatic MCC who had previously received chemotherapy, treatment with the anti-PD-L1 antibody avelumab resulted in a 32.8% objective response rate (ORR), including complete -and durable -responses in 9.1% of patients 157 and an associated improvement in quality of life 158 . In a study involving 26 patients, treatment with the anti-PD-1 antibody pembrolizumab resulted in a 56% ORR, including complete responses in 15.4% of patients, as first-line systemic therapy 154 . The anti-PD-1 antibody nivolumab has also shown promise 9,159,160 . In all of these studies, patients responded regardless of the viral status of their tumours, although the cohort sizes were too small to enable rigorous comparisons between VP-MCCs and VN-MCCs. Avelumab is currently the only FDA-approved treatment of metastatic MCC 153  In virus-positive-MCC tumour cells, antigen presentation by MHC complexes is commonly suppressed through epigenetic silencing. Multiple signalling pathways have the potential to either stimulate (such as OX40-OX40L) or suppress antitumour immunity (such as programmed cell death protein 1 (PD-1)-programmed cell death protein 1 ligand 1 (PD-L1)). These various immune signalling pathways should be considered potential therapeutic targets. CTL A-4, cytotoxic T antigen 4; GITR, glucocorticoid-induced TNFR-related protein; SIRP α, signal-regulatory protein α; TCR , T cell receptor. nivolumab as adjuvant therapies for patients without detectable tumour material after surgery are currently ongoing (NCT03271372 and NCT02196961).
Several immunotherapies that act through mechanisms other than inhibition of PD-1 or PD-L1 are currently under investigation for MCC (Supplementary  Table 1). Therapeutic combinations including anti-CTLA-4 antibodies are an area of active investigation 8 . In a phase I trial (NCT01307267), utomilumab, which might stimulate antitumour immunity by acting as an agonist of CD137 (4-1BB), has been shown to be well tolerated, with preliminary evidence of antitumour activity 161 . Proof of principle that adoptive cell transfer using in vitro expanded autologous anti tumour T cells in combination with IFNβ and radiotherapy (NCT01758458) was demonstrated by a finding of tumour regression in one patient with metastatic MCC 162 . A phase I/II trial combining adoptive cell transfer with IL-2 and avelumab after conditioning the tumour with radiotherapy or IFNβ is currently underway (NCT02584829) 8,163 . DNA vaccines are effective in mice implanted with B16 murine melanoma cells expressing LT or ST 164,165 . The cutaneous location of primary and satellite and/or in transit MCC renders these tumours amenable to injections of local therapies. The efficacy of intratumoural administration of IL-12 cDNA delivered via electroporation was investigated in a phase II trial involving 12 patients with MCC, and disease regression was observed in 12 of 30 treated lesions 166 . Finally, several phase II trials (NCT02819843 and NCT02978625) are currently recruiting patients to determine the safety and efficacy of intratumoural injections of the modified oncolytic herpesvirus talimogene laherparepvec 163 .
Combinations of immunotherapies with traditional treatments might also be effective in patients with MCC. Downregulation of MHC class I (MHC I) expression in MCC can be reversed by radiotherapy, IFNβ and chemotherapies, potentially increasing the sensitivity of tumours to subsequent immunotherapy 112,162 . The histone deacetylase inhibitor vorinostat in combination with the chemotherapeutic agent mithramycin A can also upregulate MHC I expression in MCC cell lines and xenografts 113 . Radiotherapy also has the potential to induce the release of tumour-associated antigens and increase inflammation, thus further synergizing with immunotherapy 8 .
Targeted therapies. Alternatives to ICI are needed for patients with advanced-stage MCC who are immunosuppressed (and thus ineligible) or who do not respond to ICI. Several types of targeted therapies have been investigated in MCC cell lines and xenograft models, and some are undergoing further testing in early phase clinical trials.
YM155, a small-molecule inhibitor of the antiapoptotic protein survivin, is a potent inducer of cell death in VP-MCC cell lines and xenografts 49,167 . Similarly, a small-molecule inhibitor of the apoptosis regulator BCL-2 family proteins, ABT-263, has been shown to induce apoptosis in the majority of MCC cell lines tested, regardless of MCPyV status 168 . Moreover, BCL-2 antisense oligonucleotides were effective in xenograft models 169 but failed to show benefit in a phase II trial involving 12 patients with MCC 170 .
MCCs express somatostatin receptors; therefore, somatostatin analogues are being investigated for both molecular imaging 171,172 and treatment of patients with MCC (NCT01204476, NCT01652547, NCT02351128, NCT02936323 and NCT03167164) 8 . Hedgehog signalling has a role in the development of nonmalignant Merkel cells [173][174][175] , although MCC cell lines are not sensitive to Hedgehog inhibitors 99 .
In line with the observation of recurrent activation of the PI3K pathway in MCC tumours, preclinical and limited clinical data support the efficacy of PI3K-mTOR pathway inhibition in MCC: MCC cell lines are sensitive to inhibition of PI3K and mTOR 84,85,176,177 , and a complete response to the PI3Kδ inhibitor idelalisib has been reported in one patient, although the response might have been modified by concurrent use of radiotherapy 178 . Several clinical trials are investigating the safety and efficacy of mTOR inhibition in patients with MCC (NCT01155258, NCT02514824, NCT00655655 and NCT01204476) 163 (Supplementary Table 1). In VP-MCCs with wild-type TP53, investigations of tumour biology suggest that MDM2 (HDM2) inhibitors can be effective 179 .
In general, activating tyrosine kinase mutations are not detectable in MCCs and, thus far, there has been little evidence that tyrosine kinase inhibition is an effective treatment approach for patients with MCC. Clinical remission following treatment with imatinib has been reported in one patient 180 , although a phase II trial failed to show any consistent benefit with this approach 181 . Pazopanib inhibits several tyrosine kinases, including VEGFR, PDGFR, FGFR and KIT; this agent has been associated with clinical benefit in case reports 182 and was investigated in a trial in the UK 8 . A clinical trial involving the multikinase inhibitor cabozantinib in patients with advanced-stage, platinum-refractory MCC was terminated prematurely owing to poor tolerability and a lack of responses in the first eight patients treated 183 . Of note, evidence for ERK activation in MCCs is lacking 88 , and HRAS mutations do not render MCC cell lines responsive to MEK inhibition 7 . Fundamental questions remain regarding the biology of MCPyV in human infection and tumorigenesis. The cellular reservoirs and mechanisms of transmission of infectious, wild-type MCPyV remain unknown. T antigens and other viral gene products have been shown to have diverse functions involving numerous cellular partners. In particular, the numerous activities and effects of ST are incompletely elucidated. The identity of the specific viral protein or proteins that are critical for viral latency, replication, transformation and maintaining an oncogenic phenotype is not well understood. Identification of the essential tumour-promoting pathways has implications for effective therapeutic targeting of VP-MCCs and might provide an explanation for the apparently unique status of MCPyV as the only polyomavirus with established oncogenic effects in humans.

Consensus recommendations
Despite molecular differences, VP-MCCs and VN-MCCs are almost indistinguishable, both clinically and in their response to ICI. Common putative driver mutations (in RB1 and TP53) are present in VN-MCCs. Studies comparing the biology of VN-MCCs to VP-MCCs might ultimately determine a minimum set of signalling pathways that must be disruptedeither by mutation or by viral infection -to initiate MCC tumorigenesis. Parallels can be drawn with head and neck squamous cell carcinoma, which also has a mixed viral and non-viral aetiology. Studies involving mouse models of MCC might help to address several key questions including the cell of origin, which viral T antigens (and T antigen functional domains) are sufficient to generate VP-MCC and which genomic changes are sufficient to generate VN-MCC.

Translational science
Great progress has been made with ICI such that approximately half of all patients with advanced-stage MCC benefit from anti-PD-1 or anti-PD-L1 antibodies. However, many patients have either primary or acquired resistance to these agents. Combination immunotherapy approaches must be explored, although patients with MCC might also benefit from targeted approaches that synergize with immunotherapies or could be used in patients who do not respond to ICI. Unbiased small-molecule screens for agents that selectively affect VP-MCC or VN-MCC cell lines are currently ongoing. The wild-type status of TP53 in most VP-MCC tumours suggests that small molecules that result in the activation of TP53, such as MDM2 antagonism with a nutlin-like agent 184 , can be effective and could also synergize with other therapies. Therapies targeting survivin or BCL-2 might also be effective, owing to the biology of most MCCs. Studies of these agents and others could be facilitated by the use of patient-derived xenograft models. Patient-derived xenografts are generally more easily established from MCC tumour specimens than are new cell lines. Many patients with VP-MCCs do not respond to ICI and, therefore, the probable defects in the number, diversity and avidity of MCPyV-specific T cells merit further investigation. Future research should consider therapeutic vaccines as neoadjuvant therapies for patients with MCC, such as using vaccinia viruses to deliver inactivated viral oncoproteins, as has been demonstrated in mouse models of HPV-driven pre-malignant disease 185 . Despite the striking molecular differences between VP-MCC and VN-MCC, a role of MCPyV status in guiding the clinical management of MCC has not yet been established. Translational investigations would benefit from routinely determining and reporting the MCPyV status of all tumours.

Clinical science
The consensus on the clinical knowledge gaps and research priorities in MCC clusters around three major themes: therapeutics, biomarkers and infrastructure. The current armamentarium for the management of MCC includes surgery, radiation therapy, cytotoxic chemotherapy, ICI with anti-PD-1 or anti-PD-L1 antibodies and experimental therapies administered in clinical trials. The advent of anti-PD-1 and anti-PD-L1 antibodies has provided great benefit for some patients with MCC; however, it remains too early to determine the long-term outcomes of these patients. Moreover, subsets of patients are either refractory to anti-PD-1 or anti-PD-L1 antibodies or develop acquired resistance over time. Immunosuppression, previous solid organ and/or haematological stem cell transplantation and haematological malignancies are known risk factors for MCC. Patients with these risk factors have been routinely excluded from clinical trials and thus the optimal therapeutic strategy for these individuals is unclear. Finally, conventional cytotoxic chemotherapy and radiation therapy are active therapies that predate the introduction of anti-PD-1 and anti-PD-L1 anti bodies, and the best use of these agents in conjunction with immunotherapy remains to be determined.
The fact that MCC is either polyomavirus-positive (VP-MCC) or negative (VN-MCC) presents both unique opportunities and a therapeutic conundrum. At this time, MCC viral status does not help stratify patients into those who are more or less likely to respond to any specific therapy, given the present state of our knowledge and the available therapeutic options. We anticipate that targetable oncogenic drivers, at least in those with VP-MCC, are within reach. VN-MCC tumours generally have a molecular signature similar to that of the prototypical pattern seen in melanoma, characterized by ultraviolet-induced DNA damage and the presence of tumour-associated neoantigens. Thus, clinical developments might likewise follow the path taken by other immunotherapy-responsive tumours, including melanoma.
The clinical management of MCC would be improved by the creation of a collaborative infrastructure that enables formal information sharing and the rapid design and implementation of clinical trials. This aspect is particularly critical in a disease with such a low prevalence. Current understanding of the natural history of MCC and outcomes related to therapeutic interventions is scant, fractionated among individual investigators, and thus a common standard is lacking. To further our knowledge of the natural history of MCC, the implications of the various prognostic factors, predictors of drug response, implications of MCPyV status in therapeutic response and rapid clinical trial design and implementation, a collaborative multi-institutional effort is needed.

Conclusions
Substantial advances have been made in our understanding of MCC biology, diagnosis and therapy; however, major research challenges remain. A striking molecular dichotomy exists between VP-MCC and VN-MCC, although the biological and clinical consequences of this difference -including critical pathways for tumorigenesis -are incompletely understood. Furthermore, the MCC cell of origin remains unknown. Finally, continued progress in improving patient outcomes will require the validation of additional therapeutic options in order to augment the gains provided by surgery, radiotherapy and immunotherapy.