Predominance of Central Memory T Cells with High T-Cell Receptor Repertoire Diversity is Associated with Response to PD-1/PD-L1 Inhibition in Merkel Cell Carcinoma

Merkel cell carcinoma (MCC) is a highly aggressive neuroendocrine skin cancer, which occurs predominantly in the elderly, fair-skinned population. The mortality rate after primary diagnosis is reported to range from 33% to 46% (1). Known risk factors for MCC include chronic UV exposure and any type of immunosuppression (2). Moreover, an association with the Merkel cell polyomavirus (MCPyV) has been established predominantly for cases occurring in the northern hemisphere (3). Key feature of MCC is its high immunogenicity based on either its virus- or UV-associated carcinogenesis, causing either presentation of MCPyV-derived epitopes (4) or neo-epitopes by UV-associated mutations (5). Indeed, a high therapeutic activity of immune checkpoint inhibitors (CPIs) resulting in durable objective responses (ORs) in about 50% of patients has been observed (6, 7). Predictive biomarkers of therapy outcome to differentiate responders from nonresponders at treatment baseline have not been established. On the basis of the experience in melanoma, clinical features like overall performance status, metastatic stage, and previous therapies, as well as blood-derived parameters, for example, neutrophil-to-lymphocyte ratio, elevated lactate dehydrogenase (LDH) and C-reactive protein, are currently discussed as possible predictive markers (8, 9). However, their predictive power could so far not been indisputably confirmed (10). Similarly, potential predictive biomarkers determined on tumor tissue such as PD-L1 expression or MCPyV status could not be confirmed either (6, 7).

This study provides an extensive workup of both, clinical as well as immunologic and molecular features of 41 patients with MCC treated with CPIs for advanced disease in the real-world setting. The latter were based on T-cell receptor (TCR) repertoire usage, as well as gene and protein expression determined in a subgroup of patients on formalin-fixed, paraffin-embedded (FFPE) tumor tissue obtained at baseline before start of anti–PD-1/PD-L1 therapy. These unbiased multidimensional analyses were performed to improve our understanding of MCC immunology and to identify new candidates for predictive biomarkers. To further extend this aim, some of the patients' sequential samples obtained before and under therapy were analyzed.

Our results demonstrate that among previously discussed clinical and molecular biomarkers, only an unpaired performance status and the absence of immune suppression were strongly associated with a favorable clinical response to immunotherapy. More important, however, not the mere density of the immune infiltrate, but rather its functional properties correlated with response to CPI treatment. Specifically, tumor-infiltrating lymphocytes (TILs) with a predominance of central memory T cells and a diverse TCR repertoire were associated with a favorable treatment outcome.

Patients treated at the Departments of Dermatology, University Hospitals of Essen, Heidelberg and Lübeck, were retrospectively identified for this biomarker study according to the following selection criteria: diagnosis of MCC confirmed by histopathology, metastatic disease not amendable to surgery, and systemic therapy with anti–PD-1/PD-L1 antibodies. Treatment response to anti–PD-1/PD-L1 CPI was categorized as best overall response according to RECIST v1.1 (11). The study was approved by the ethics committee of the University Duisburg-Essen (11-4715; 17-7538-BO) and was conducted in accordance with the Declaration of Helsinki—Ethical Principles for Medical Research Involving Human Subjects. Informed written consent was obtained from each subject.

Detection of MCPyV DNA was performed as described previously (12). Briefly, DNA from FFPE tissue samples was isolated by AllPrep DNA/RNA FFPE Kit (Qiagen) according to the manufacturer's instructions. Presence of MCPyV DNA was determined by TaqMan Real-Time qPCR using large T-antigen–specific primers and TaqMan probe: forward primer: CCA AAC CAA AGA ATA AAG CAC TGA; reverse primer: TCG CCA GCA TTG TAG TCT AAA AAC; and probe: FAM-AGC AAA AAC ACT CTC CCC ACG TCA GAC AG-BHQ1. The PCR reaction had a final volume of 10 μL, consisting of DNA (10 ng), primers and probe (5 μmol/L each), and reaction buffer (LuminoCt ReadyMix, Sigma-Aldrich). Annealing was performed at 60°C for 15 seconds. CFX Manager (Bio-Rad) was used for data analysis.

PD-L1 expression was assessed in FFPE tumor tissue sections with the use of a rabbit monoclonal anti-human PD-L1 antibody (clone 28-8) and an analytically validated automated IHC assay (PD-L1 IHC 28-8 pharmDx for Autostainer Link 48; Dako), as described previously (13). PD-L1 positivity was defined as at least 1% of living tumor cells showing specific cell surface staining of any intensity in a section containing at least 100 evaluable tumor cells. Positive staining of tumor-infiltrating inflammatory cells or other cells of the tumor stroma was excluded from evaluation.

The TCR repertoire usage of TILs was determined by amplifying the highly variable CDR3 of the different TCR beta-chain (TCRB) families by a multiplexed PCR and subsequent high-throughput sequencing using the immunoSEQ hsTCRB Kit (Adaptive Biotechnologies, catalog No. ISK10101). After DNA was isolated from FFPE tumor tissue in a first PCR reaction, all recombined TCRB CDR3 sequences were amplified by a mix of V- and J-gene primers. After labeling the obtained amplicons in a second PCR amplification, the resulting library was sequenced on an Illumina MiSeq using the MiSeq Reagent Kit v3 (Illumina, catalog No. MS-102-3001).

Multiplex immunofluorescence staining of FFPE tumor tissue was performed using the OpalTM Chemistry (PerkinElmer, catalog No. OP7TL4001KT) with two panels of antibodies, that is, against CD4, CD8, CD45RA, CD45RO, and CK20 (panel 1), or CD27, CD45RA, CD45RO, and Synaptophysin (panel 2). Briefly, after deparaffinization and fixation, 3-μm tumor sections were processed with retrieval buffers for 15 minutes in an inverter microwave oven. Thereafter, sections were incubated with the antibody diluent for 10 minutes at room temperature, followed by incubation with the primary antibody for 30 minutes. After applying Opal Polymer HRP secondary antibody solution for 10 minutes, antibodies were removed by microwave treatment before another round of staining was performed. The antibodies and retrieval buffers used are described in detail in Supplementary Table S1. Visualization of the different fluorophores was achieved on the Mantra Quantitative Pathology Imaging System (PerkinElmer).

mRNA from FFPE tissue samples was isolated using the AllPrep DNA/RNA FFPE Kit (Qiagen, catalog No. 80234) according to the manufacturer's instructions. Gene expression was quantified using the HuV1 Cancer Immune Panel (NanoString Technologies, catalog No. XT-CSO-HIP1-12). After 100 ng of mRNA was used for the hybridization reaction at 65° C for 24 hours, the complex was further processed in the nCounter Prepstation for immobilization to the cartridge, which was processed in the nCounter Digital Analyzer.

A Bayesian logistic regression model was applied for predicting PD-1/PD-L1 blockade treatment response. It was built on a dataset consisting of 41 patients and 15 clinical parameters (Supplementary Table S2). The observed data were described in a probabilistic manner, also known as likelihood function. The probability that a given sample belongs to one of the two possible outcomes, responding to CPI therapy or not, was computed via the following formula: |${\theta _j}\ = \ logistic({\beta _0} + {\beta _{parameterX}}\middot \ {x_{j,parameterX\ }} \ldots)$|⁠; in which θ_j is the probability of responding to the treatment for patient j, β₀ is the base frequency of responders in the dataset, β_parameterX is the effect of parameter X on the treatment response, and x_j,prameterX is 1 if for patient j parameter X is applicable and 0 otherwise (14). In addition, HRs for each clinical parameter were calculated as mean of the credibility (posterior) interval. Fitting the model to the dataset was done with the R software package “brms,” which utilizes “Stan” in the background. Missing values were approached with multiple imputation with the R package “mice” (15–17). For additional details, please see Supplementary Materials and Methods.

Normalization of the rank-abundance-distribution (RAD) of T-cell clonality was done by MaxRank normalization as implemented in R package RADanalysis (18). Here, the MaxRank is the minimum dimension of rank abundance vectors for all tested samples (18). The plotted normalized RADs are the results of 50-fold averaging. The distribution of T-cell numbers over T-cell clones was quantified by Pielou's Evenness Index J = H/H_max, with Shannon entropy H, and theoretically possible maximum Shannon entropy H_max. J ranges between 0 and 1, where 1 represents completely even distribution of T cells over clones (19). Observed richness indicates the number of unique T-cell clones, and the Chao index iChao1 is used as an estimator of TCR clone richness for rare clones (20). Simpson Diversity Index (also known as, Simpson D) represents the probability that two T cells taken at random from a specimen represent the same clone (19).

We implemented the Grouping of Lymphocyte Interactions by Paratope Hotspots (GLIPH) analysis method (21) to uncover TCR antigen specificities shared between clones and patients: CDR3 sequences of TCR clonotypes were clustered according to their local and global similarities; global similarity was assumed if only one amino acid was exchanged; and local similarity was assumed if specific motifs of three amino acids in the CDR3 region were more frequently present than in the reference database. The software R v3.4.1 and nSolver3 (NanoString Technologies) were used for gene expression data analysis. For P value adjustment the Benjamini–Hochberg algorithm was used. The respective correlation coefficients were calculated by the method of Pearson. Gene ontology (GO) analysis was performed using the online platform Metascape (http://metascape.org; ref. 22), where differentially expressed genes are assigned to a set of predefined terms (Kyoto Encyclopedia of Genes and Genomes Pathway, GO Biological Processes, Reactome Gene Sets, Canonical Pathways, and CORUM). For estimation of term similarity, the agreement calculator Cohen kappa coefficient was deployed; a kappa value > 0.3 was set as a threshold for selecting the terms for clustering.

For statistical testing of the TCR repertoire characteristics, gene expression analysis, and central memory T (T_CM) cells abundance in MCC tumor tissue, P values were determined using the unpaired two-tailed Student t test, calculated in GraphPad Prism 5.

Forty-one patients treated with PD-1/PD-L1–blocking antibodies for advanced MCC were identified in three clinical centers. To determine the impact of the clinical and standard immunologic parameters on therapy response, we performed chart review. The extracted parameters were correlated by Bayesian inference with clinical response to therapy. Patients showing an OR to anti–PD-1/PD-L1 therapy, that is, complete (CR) or partial response (PR), were classified as responders, and patients with a stable (SD) or progressive disease (PD) as nonresponders. At database closure (November 2018), the median follow-up time after onset of checkpoint inhibition was 13.5 months. Detailed patient characteristics are provided in Supplementary Table S2.

Bayesian inference demonstrated that only an unimpaired overall performance status [Eastern Cooperative Oncology Group (ECOG) = 0] is associated with a positive response to CPI treatment, whereas any form of immunosuppression was identified as a negative predictor (Fig. 1). However, even for the unimpaired performance status, the 95% credibility interval overlaps with the region of zero effect. For all other currently discussed predictive biomarkers for response to CPI such as metastatic stage, neutrophil-to-lymphocyte ratio, serum LDH and C-reactive protein, as well as PD-L1 expression and MCPyV status, we could not identify a relevant association to therapy response. This observation is in-line with recently published study on 37 patients with MCC treated with CPI. Their multivariate analysis of clinical parameters including molecular subtype, age, prior radiotherapy, and PD-L1 expression did not show any predictive function (23).

Given the limited value of clinical, blood-based, and standard tissue-based biomarkers, we next performed a comprehensive, unbiased analysis to elucidate the impact of immunologic processes and immune cells characteristics on response to CPI. Given the need for sufficient amounts and quality of pretherapeutic tissue samples, these analyses were restricted to a subgroup of patients. The flow of the patients is detailed in Supplementary Fig. S1. To this end, 18 of the above-described patients with MCC were identified to meet these selection criteria to allow an in-depth immunologic and molecular work-up; this group consisted of seven responders and 11 nonresponders. Three of the 18 patients with MCC were stage III, not amenable to surgery or radiation and the other 15 were stage IV. Patients' characteristics are given in Table 1.

On the basis of the T-cell function and its attainment, the TCR repertoire usage is a mirror of immune response and differences in its usage have been suggested as a possible biomarker for the efficacy of immunomodulating therapies. Thus, we established the TCRB repertoire of TILs from MCC tumor samples obtained before and under anti–PD-1/PD-L1 therapy by high-throughput sequencing of the preamplified highly variable CDR3 sequences of the different TCR beta families. With their diverse repertoire of TCRs, T cells can be regarded as a generalized community, in which diversity reflects both richness and evenness. Using the Normalized Rank Abundance Distribution (NRAD; ref. 18), a descriptor for quantitative comparison of generalized communities, we compared the TCR repertoire usage with respect to response to CPI, revealing that TILs of responders had a more even clone size distribution, whereas in nonresponders the T-cell communities were dominated by a limited number of strongly expanded clones, corresponding to a high clonality (Fig. 2A). This observation is consistent with a higher richness for both strongly as well as weakly expanded T-cell clonotypes, the latter measured by the Chao index (Fig. 2B and C, respectively) and a higher evenness (Fig. 2D). Consequently, using the Simpson diversity reciprocal index to account for the clonal dominance hierarchy within each patient, TILs of responders is characterized by a higher diversity than those of nonresponders (Fig. 2E). Even though differences between both groups were not reckoned as significant by frequentist statistics, applying Bayes inference model supported the predictive value of the TCR repertoire richness and diversity for a favorable response to CPI treatment (Supplementary Fig. S2). The observed TCR repertoire richness showed the highest probability to be associated with therapy response.

Differences in TCR repertoire usage are likely to reflect functional differences of T cells. Analogous to other tissue types, T cells live through various stages of their differentiation, which are associated with variation of their function and proliferative capacity. Notably, these stages are discernable by gene expression. Thus, we performed NanoString-based gene expression analysis on 770 genes, characteristic for different immune cell types and their differential activation. A major advantage of these techniques, it is providing robust and reproducible results from FFPE tissues; however, sufficient integrity of RNA is still required to generate valid results. Thus, those pretherapeutic FFPE tissue samples that failed quality controls were excluded from analysis (Supplementary Fig. S1). In the remaining samples from n = 6 patients, the expression of immune-related genes clearly separated responders (n = 3) from nonresponders (n = 3; Fig. 3A; Supplementary Table S3). This discrimination was largely driven by approximately 100 genes involved in adaptive immunity, lymphocyte activation, leukocyte migration, and cytokine signaling pathways as computed by GO analysis (Fig. 3B). In detail, genes related to T-cell attraction (e.g., CCL5, CXCL9, IL16, CXCL11, CCL3, CXCL10, CCL21, and CCL4; Fig. 3C) and T-cell activation (such as IL2RB, IL2RG, IL15RA, LCK, CD97, JAK3, and NFATC2; Fig. 3D) were highly expressed in tumor tissue from responders. Whereas the cell-cycle–related CDK1 (#103) and the apoptosis regulation gene BCL2 (#106) were strongly expressed in tumors of nonresponders (Fig. 3A). Notably, both genes have been linked to T-cell differentiation (24).

The TCR repertoire usage and gene expression pattern in responders and nonresponders suggest disparate differentiation states of the respective predominant T-cell infiltrates. While memory T cells are characterized by their ability to exert a fast and sustained proliferative response to stimulation, exhausted and terminally differentiated T cells show an impaired proliferative capacity (25). Notably, for patients with melanoma it has been reported that predominance of memory T cells among TILs is associated with a favorable outcome of CPI treatment (26–28). To test this hypothesis in our MCC patient cohort, we performed multiplexed immunofluorescence staining for memory (CD45RO⁺) and effector (CD45RA⁺) CD4⁺ and CD8⁺ T cells, revealing that in tumors of responding patients memory T cells had higher abundance and with respect to their spatial distribution they were located closer to the tumor cells, whereas in nonresponders such cells were rare and mostly present in the stromal compartments (Fig. 4A and B). A similar pattern was seen for effector T cells (CD45RA⁺; Fig. 4A and B). To scrutinize the memory T cells in more detail, we included also the central memory T-cell marker CD27 in the staining panel (n = 8). The colocalization of CD27 with CD45RO demonstrated that T_CM cells are more frequent in responders (Fig. 4C–E; Supplementary Fig. S3; ref. 29). This notion was backed by gene expression signatures demonstrating higher abundance of genes such as TCF-7, CCR7, CCL21, CD62L, and IL7R in responders (Fig. 4F; refs. 29, 30).

Clustering of genes due to their functional relevance demonstrated that in responders, genes important for lymphocyte activation, differentiation, and migration, as well as cytokine-mediated signaling of T cells were not only more abundant at baseline, but also during therapy, indicating that this gene expression pattern was maintained or even boosted by CPI treatment (Fig. 5A). Furthermore, reflected by increased gene expression of CCR7, CD62L, TCF-7, CD27, IL-7R, and IL-2RA a higher presence of T_CM cells was detected in tumor tissues, which also persisted upon therapy (Fig. 5B). In contrast, nonresponders showed no expression dynamics for the above-described genes upon treatment, that is, persistently low gene expression.

Next, we quantified the dynamics of TCR repertoire usage upon therapy (Fig. 6A–F). In accordance with the above-described results, the TCRB repertoire usage of TILs of a responder were distributed over a larger number of smaller clones than in a nonresponder. Notably, in the nonresponder the expansion of T-cell clones were up to one order of magnitude larger, indicating a substantial previous clonal expansion (Fig. 6A and D). On the other hand, upon PD-1/PD-L1 blockade, induced clonal T-cell expansions were more prominent among TILs of the responder (Fig. 6A, B, D, and E), suggesting a more pronounced proliferative capacity of T cells. Extending the comparative analyses of the TCRB repertoire from the dynamics of identical T-cell clones to newly emerging clones under therapy further confirmed that the number of newly emerging clones was higher among TILs of the responding patient (Fig. 6C and F).

An alternative way to construe the TCR repertoire usage is to cluster TCRs into convergence groups based on the CDR3 amino acid sequences that are predicted to bind the same or a similar MHC-restricted epitope using the recently published GLIPH algorithm (21). GLIPH predicts convergence groups calculating the probability that a cluster of similar TCRs has appeared by selection of a collectively recognized epitope/MHC complex. Representative examples for convergence group clustering by GLIPH for a responder and a nonresponder are depicted in Fig. 6G and H; for consistency, we chose the same two patients for which the multiplexed immunofluorescence staining results are presented. In the responder, three major TCRB clusters consisting of a multitude of different TCRs were present; this observation suggests that a large proportion of TILs characterized by a diverse TCR repertoire in responders are still recognizing a defined set of antigens. Notably, when we subjoined established MCPyV epitope-reactive TCR sequences, these were joined in two of these larger clusters (4). Of note, in both patients' tumors MCPyV DNA was detected (Supplementary Fig. S1). Upon treatment, there were no major changes in TCRB diversity and only some minor shifts in cluster formation (Fig. 6G). For the nonresponder, the limited TCR repertoire was characterized by almost complete absence of such convergence groups of TCR clonotypes. It should be noted that only TCRs within convergence groups, regardless of the individual expansion of a given TCR clonotype are depicted. After CPI treatment, the most obvious change was a further expansion of the two largest T-cell clones, which is represented by the increased size of the blue circles (Fig. 6H). Thus, in the patient with MCC with a favorable response to PD-1/PD-L1 blockade, we detected a substantial number of different T-cell clonotypes, which are recognizing a limited set of antigens. Results from the GLIPH analysis of further patients with MCC are depicted in Supplementary Fig. S4.

The introduction of immunotherapy with CPI dramatically improved the hitherto poor prognosis of patients with advanced MCC. For either PD-1- or PD-L1–blocking antibodies, OR rates between 30% and 60% have been reported. Conversely, this means that almost half of the patients do not benefit from CPI and predictive biomarkers of therapy response are still lacking (6, 7, 31). In this study, we compiled clinical characteristics of 41 patients with MCC treated with CPI and tested currently presumed biomarkers, such as performance status, immunosuppression, previous therapies, serum LDH, neutrophil/lymphocyte ratio, PD-L1 expression, and MCPyV status. We chose to apply a Bayesian logistic regression model instead of classical least squared regression due to the given study character. Specifically, the study comprises a limited sample size and a large number of variables to be analyzed for potential effects on treatment response. In classical statistics these features lead to overfitting (14). Among the analyzed parameters, only performance status and immune suppression correlated with response to therapy. Thus, we next performed a comprehensive immunologic work-up of MCC tumor tissues taken at therapy baseline and additionally in a subgroup of patients after initiation of therapy. This work-up comprised high-throughput sequencing-based TCR clonotype mapping, multiplexed immunofluorescence staining, and immune gene mRNA expression. Thereby, we established that a prevalence of T_CM cells expressing a highly diverse TCRB repertoire among TILs is associated with a favorable therapy response. Notably, even though these parameters were established from only a subgroup, Bayes inference revealed that their predictive value is comparable with good performance status and absence of immune suppression of the complete group (Supplementary Fig. S2).

Because clonal expansion is one fundamental event in effector T-cell development, the TCR repertoire reflects both the previous history and future prospects of adaptive cellular immune responses (32). The TCRB repertoire of responders was characterized by higher richness, an indicator accounting not only for the number of individual TCRBs, but also for their heterogeneity. Greater richness was associated with a higher evenness of TCRB clonotype, thus resulting in a high T-cell clonotype diversity. The nature of this T-cell diversity was further analyzed by clustering the respective TCRB CDR3 regions by similarity. GLIPH analysis revealed that in responders the highly diverse TCR repertoire was readily grouped in a limited number of convergence groups. Notably, when artificially joining in established TCR CDR3 sequences of T cells specifically reacting with MCPyV epitopes (4), these were also clustered within these convergence groups. In TILs of nonresponding patients, despite larger expansion of individual clones, such convergence groups were virtually absent. The positive predictive value of a diverse TCR repertoire for response to CPI therapy was recently also observed in melanoma (33). Consistent with the predictive value of a high TCR-repertoire evenness, our NRAD analysis demonstrated that TILs of nonresponders were dominated by a few, largely expanded T-cell clones. Thus, our findings indicate that in responding patients, immune-inhibiting conditions prevented clonal expansion of T cells, which could be abrogated by anti–PD-1/PD-L1 therapy, whereas in nonresponders' reactive T cells were previously expanded reaching their proliferative capacity, thus a state not amendable by immune checkpoint blockade. To test this, we additionally addressed dynamic changes of the TCR repertoire usage upon CPI therapy revealing that both size and number of T-cell clones increased in responders, whereas in nonresponders the already large T-cell clones at baseline did not undergo the same relative expansion and the number of newly emerging clones was substantially smaller. Moreover, the TCR-repertoire diversity in nonresponders remained low, suggesting an irreversible T-cell dysfunction, for example, by terminal differentiation. To this end, the age-related involution of the thymus is associated with a shift of the T-cell pool from naïve to effector memory T cells (34, 35). As a result, the naïve T-cell repertoire is increasingly curtailed, which is evidenced by a loss of TCR-repertoire diversity (36). The notion appears to be particularly important for MCC, characterized by an elderly patient population (1). These observations stress the importance of the TCR-repertoire diversity as a pivotal feature of a functional immune system. In-line with this concept, functional characterization of the immune microenvironment in MCC before initiation of CPI treatment by gene expression analyses demonstrated a very low or no expression of genes related to T-cell attraction or activation in nonresponders; on the other hand, these genes were highly expressed in tumors of responders. Similarly, recent studies of TILs in melanoma demonstrated a correlation of TCR clonality and the fraction of dysfunctional T cells (37).

Deconvolution of the molecular immune phenotype derived from gene expression data revealed higher numbers of T_CM cells in responding patients. Immunofluorescence detection of high numbers of CD27⁺ CD45RO⁺ T cells among TILs of responders confirmed this notion. The amount of T_CM cells remained stable upon CPI treatment, which was also observed in a preclinical therapy model of colon cancer (38). This maintenance of T_CM cells may be explained by the strong expression of the transcription factor TCF7, which is crucial for differentiation, self-renewal, and persistence of memory CD8⁺ T cells. Indeed, TCF7 expression remained high during treatment. Similarly, studies in melanoma demonstrated that TCF7 is linked to an effective CD8⁺ T-cell response upon immunotherapy and elevated frequencies of TCF7⁺CD8⁺ T cell are predictive for therapy response (38, 39). In contrast, TILs of nonresponders were characterized by a high prevalence of terminally differentiated, nonfunctional T cells, characterized by expression of BCL-2. Terminally differentiated T cells have to be distinguished from exhausted T cells. T-cell exhaustion is a consequence of continuous stimulation causing a gradual loss of effector capabilities and expression of inhibitory receptors. Exhausted T cells, however, can be readily reactivated by therapeutic interventions such as inhibition of PD-1/PD-L1 pathway (40). Terminally differentiated T cells, in contrast, are induced by overstimulation causing excessive proliferation resulting in critically shortened telomeres (41–43). Thus, terminally differentiated T cells have reached the final phase of their activation cycle, and can no longer be reactivated even by immune checkpoint blockade or epigenetic modifiers (29).

In conclusion, we identified immunologic and molecular characteristics measurable in tumor tissue at treatment baseline associated with clinical response to anti–PD-1/PD-L1 therapy of advanced MCC. In patients with a favorable response, TILs had a rich and diverse TCR repertoire as well as a phenotype of T_CM cells, whereas a predominance of largely expanded, terminally differentiated T cells was associated with an impaired response.

S. Ugurel is an employee/paid consultant for Bristol-Myers Squibb, Merck, and Merck Sharp and Dohme. P. Terheyden reports receiving other remuneration from Sanofi, Pfizer, Roche, Bristol-Myers Squibb, Novartis, Roche, and Biofrontera. J.C. Hassel is an employee/paid consultant for Merck Sharp and Dohme, Pierre Fabre, Sanofi, and Sun Pharma, and is an advisory board member/unpaid consultant for Bristol-Myers Squibb and Novartis. J.C. Becker is an employee/paid consultant for Sanofi, Merck Serono, ReProTher, 4SC, Amgen, eTheRNA, and Pfizer, reports receiving commercial research grants from Alcedis, IQVIA, Merck Serono, and Amgen, speakers bureau honoraria from Amgen, Merck Serono, Sanofi, Pfizer, and Recordati, and other remuneration from Incyte. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Spassova, S. Ugurel, J.C. Becker

Development of methodology: I. Spassova, L. Kubat, L. Peiffer, J.C. Becker

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Ugurel, P. Terheyden, A. Sucker, J.C. Hassel, C. Ritter, R. Kumar, D. Schadendorf, J.C. Becker

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Spassova, S. Ugurel, C. Ritter, D. Habermann, F. Farahpour, M. Saeedghalati, L. Peiffer, R. Kumar, D. Schrama, D. Hoffmann, J.C. Becker

Writing, review, and/or revision of the manuscript: I. Spassova, S. Ugurel, P. Terheyden, J.C. Hassel, C. Ritter, L. Peiffer, R. Kumar, D. Schrama, D. Hoffmann, D. Schadendorf, J.C. Becker

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Spassova, S. Ugurel, A. Sucker, J.C. Becker

Study supervision: R. Kumar, J.C. Becker

This work was funded by the DKTK site budget OE 0460 EDO3. We thank the patients who participated in this study, their families, and the staff members at the clinical centers who cared for them.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.