A total of 90 formalin-fixed paraffin embedded (FFPE) samples from 71 Japanese patients with histologically diagnosed MCC on the basis of biopsy or surgical resection samples obtained at nine facilities were collected as previously reported.3 The cohort is summarized in online supplemental table 1. Immunohistochemical analyzes were performed on these samples. Of these 90 FFPE samples, 44 samples were randomly selected for RNA sequencing using a next-generation sequencer (NGS). Three samples were dropped from the study because of low gene expression and 41 samples were used for further analyzes (summarized in table 1). Blood serum samples were collected from another cohort of patients diagnosed with MCC at three facilities, and are summarized in table 2.

Tumor tissue was carefully dissected from 3 to 5 undyed FFPE tissue sections (4 µm thickness) using a scalpel blade and deparaffinized in 640 µL deparaffinization solution (Qiagen, Hilden, Germany). Total RNA was refined using an AllPrep DNA/RNA FFPE Kit (Qiagen) according to the supplier’s instructions. The RNA integrity number and DV₂₀₀ values were measured using a Bioanalyzer (Agilent Technologies, Santa Clara, California, USA) to evaluate the quality of the extracted RNA. RNA samples confirmed to be of sufficient quality were reverse-transcribed to cDNA using a SuperScript VILO cDNA Synthesis Kit (Thermo Fisher Scientific, Waltham, Massachusetts, USA) after assessing the density using a Qubit 4 Fluorometer (Thermo Fisher Scientific). cDNA samples were amplified and applied to the NGS using a PTC-100 thermal cycler (MJ Research, Watertown, Massachusetts, USA) and Ampliseq for the Illumina Immune Response Panel (Illumina, San Diego, California, USA). After quantification of the library using a Bioanalyzer, NGS analysis was performed using the MiniSeq System (Illumina). Data were uploaded and analyzed on the cloud-based software application BaseSpace Sequence Hub (Illumina). All data were uploaded to the national center for biotechnology information gene expression omnibus database (GSE154938).

Gene set enrichment analysis (GSEA) was performed using the c5 Gene Ontology gene set collections as supplied by the Molecular Signatures Database10 and GSEA software (https://www.gsea-msigdb.org/gsea/).11

Undyed FFPE tissue slides (Nunc, Roskilde, Denmark) were processed for indirect immunofluorescence to detect the expression of signal transduction proteins using primary antibodies to the anti-PD-L1 antibody (1:100, ab205921, Abcam, Cambridge, UK) as previously described.3 Bound antibodies were visualized with the appropriate secondary antibodies (Alexa Fluor 594 goat anti-rabbit IgG, A11005; Invitrogen, Waltham, Massachusetts, USA) at 37°C for 30 min at 1:100 dilution with 5% goat serum. 4’,6-diamidino-2-phenylindol (Vector Laboratories, Burlingame, California, USA) was used as a counterstain. The red fluorescence produced by Alexa 594 and blue fluorescence produced by 4’,6-diamidino-2-phenylindol were observed and captured using a fluorescence microscope BZ-X800 (Keyence, Osaka, Japan). The fluorescence intensities of PD-L1 were calculated using ImageJ Software (NIH, Bethesda, Maryland, USA) from 10 randomly selected fields as previously described.3 5 The pixel value of more than 45 was defined as ‘PD-L1 high’ and less than 35 was defined as ‘PD-L1 low’. Immunostaining for Merkel cell polyoma virus (MCPyV) was performed with large T-antigen (CM2B4 antibody; Santa Cruz Biotechnology, Santa Cruz, California, USA). Our cases exhibited a 69% positivity rate. The prognostic values of PD-L1 expression and MCPyV infection were analyzed in our previous work using the same cohort.3 Immunohistochemical staining of glucose-6-phosphate dehydrogenase (G6PD) was performed using anti-G6PD Rabbit IgG (1:25, HPA000247, MilliporeSigma, St. Louis, Missouri, USA) and a DAB substrate kit (SK-4100, Vector). The positive cells were counted using BZ-X800 from 10 randomly selected fields. A mean positive ratio of 50% or more was defined as high, and a mean positive ratio of less than 50% was defined as low.

G6PD activity was measured from blood serum using a G6PD activity assay kit (ab176722, Abcam) according to the supplier’s instructions. Fluorescence was monitored on a fluorescence microplate reader (Spectra Max Gemini EM, Molecular Devices, San Jose, California, USA) in kinetic mode for 40 min.

NGS data were analyzed on the cloud-based software BaseSpace Sequence Hub (Illumina) using the RNA Amplicon application. A clustered heatmap of all samples was generated using the online tool iDEP.91 (http://bioinfomatics.sdstate.edu/idep/). Disease-specific survival was analyzed using the Kaplan-Meier method and log-rank test. The Kruskal-Wallis test was used to compare stages and a paired-t test was used to compare pretreatment and post-treatment in blood serum tests.

Total RNA was extracted from 44 samples and analyzed with NGS-targeted expression of 395 cancer biomarkers involved in tumor-immune system interactions and indicative of an immunotherapy response. Three samples were excluded due to low expression of all genes evaluated. Data from 41 samples were further analyzed. The characteristics of the 41 samples are summarized in table 1. The cohort included 13 men and 24 women with a median age of 76.57 (range 40–98) years. The most commonly affected site was the head and neck (25 samples, 67.6%), followed by the limbs (11 samples, 29.7%) and trunk (1 sample, 2.8%). Four samples (9.8%) showed spontaneous regression after biopsy. The mean RNA integrity number was 2.08 (1.0–2.6) and the mean DV₂₀₀ value was 53.0 (28–88). Hierarchical cluster analysis was conducted with the 41 samples and the samples were divided into two groups according to the expression pattern of all 395 genes evaluated (figure 1A). Group A comprised 23 samples and group B comprised 18 samples. Kaplan-Meier survival analysis revealed a poorer prognosis for patients in group A compared with patients in group B (p=0.035, log-rank test, figure 1B). Gene expression levels in each group are presented as volcano plots in figure 1C. Vertical and horizontal broken lines represent threshold of log2 fold change (−0.5 and +0.5) and p value (1.0×10^–5).

To interpret the classes of genes that were upregulated or downregulated in each group, GSEA was performed using the Gene Ontology resource including 10, 192 gene sets. The highest ranked and only gene set enriched in group A with a p<0.05 was the ‘cell division’ set. The normalized enrichment score was 1.63, and the p value was 0.047. In group B, we detected 49 gene sets with a p<0.05, and 7 gene sets with a p<0.01 (see online supplemental table 2). No gene sets with a false discovery rate (FDR) less than 0.25 were detected. Enrichment plots of representative gene sets in each group are shown in figure 2A. Graphs display the enrichment score (y axis) versus the gene rank in an ordered dataset (x axis); genes with high relative expression in group A were given a low rank order value (leftmost tail), and genes with high relative expression in group B were given a high rank order value (rightmost tail) of the gene rank representation. The rank of each gene in the respective gene set is indicated by the horizontal line below the enrichment plot. Normalized enrichment scores, p values and FDR q values for each analysis are shown. Gene expression heatmaps of the highest ranked gene sets for each group are presented in figure 2B. On the basis of these results, the two groups could be annotated as the ‘cell division type (group A)’ and the ‘immune active type (group B)’.

Significant differences in the gene expression between MCPyV infection positive (n=24) vs negative (n=10) (figure 3A), spontaneous regression after biopsy positive (n=4) negative (n=29) (figure 3B), cases have lymph node or distant metastasis during the follow-up positive (n=17) negative (n=17) (figure 3C), and PD-L1 expression in tumor cells high (n=10) vs low (n=18) (figure 3D) were analyzed. Red dots indicate significantly upregulated genes with p value <1.0×10⁻⁵ in group A ‘cell division type’ and blue dots indicate significantly upregulated genes with p value <1.0×10⁻⁵ in group B ‘immune active type’. Fisher’s exact tests revealed a significant association between groups A and B with regard to PD-L1 expression in MCC cells (p=0.032, OR=7.0, 95% CI 1.18 to 41.36, (see online supplemental table 3). High expression of PD-L1 is one factor of the upregulated immune activity observed in group B. High PD-L1 expression in MCC is associated with better clinical outcomes.4 PD-L1 expression is heterogeneous, however, even in the same case.5 6 This heterogeneity complicates the usage of this immune factor as a prognostic indicator. Actually, a previous immunohistochemical analysis using the same 90 samples revealed no significant correlation between PD-L1 expression in primary MCC lesions and clinical outcomes. PD-L1 expression correlates with patient prognosis only in skin metastatic lesions.3 We focused on G6PD as a substitute prognostic factor for PD-L1. G6PD was one of the highly and significantly upregulated genes in group A ‘cell division type’ (log2FC=−1.98, p=8.65×10⁻⁶, FDR=3.25×10⁻⁵, figure 1C). G6PD expression positively correlated with lymph node or distant metastasis during follow-up (log2FC=1.55, -value=1.22×10⁻⁴, FDR=0.020, figure 3C) and negatively correlated with PD-L1 expression (log2FC=−2.10, p value=4.84×10⁻⁵, FDR=0.017, figure 3D) and had the smallest p value among the 395 genes in both analyzes. G6PD can be used as an indicator for classifying 2 types of MCC and negatively correlates with PD-L1 expression and prognosis.

G6PD can be used as a promising prognostic marker. Kaplan-Meier survival curves comparing MCC with high and low expression of G6PD classified based on RNA expression (counts per million; CPM) calculated by NGS showed significant differences in disease-specific survival (n=40, p=0.027, log-rank test, figure 4A). The cut-off value (1072 CPM) was calculated by the receiver operating characteristics curve. Immunohistochemical staining of G6PD using all FFPE samples, including primary and skin metastatic lesions, also correlated with patient outcome (n=79, p=0.034, log-rank test, figure 4B). Staining of more than 50% of tumor cells was defined as ‘G6PD high’. Representative samples of G6PD-high and G6PD low are shown in figure 4C,D. Both cases underwent spontaneous regression after biopsy. The case shown in figure 4C, however, experienced distant recurrence 10 months later and died after a few days, as we previously reported.12 The case shown in figure 4D had a good prognosis with no recurrence. These clinical outcomes might be predicted by the high expression of G6PD in the primary lesion. Immunohistochemical expression of G6PD has less heterogeneity than that of PD-L1. Even in the same case, expression of PD-L1 is heterogeneous, as we previously reported.5 6 In that case, PD-L1 expression was low in a primary lesion (figure 4F, left). Although multiple skin and lymph node metastases occurred, PD-L1 expression dramatically increased in a skin metastatic lesion (figure 4F, left) and eventually the patient had a good prognosis. Contrary to the changing PD-L1 expression, G6PD expression was consistently low in the primary and metastatic lesions (figure 4E and F, right). G6PD expression is a more useful predictor than PD-L1 expression and reflects the potential of using measures of immune activity as a prognostic factor.

G6PD activity is measurable by a blood serum test. We collected 53 samples from 21 patients with MCC in various stages (summarized in table 2). In contrast to the immunohistochemical expression of G6PD, G6PD activity in the serum is constantly changing. Serum G6PD activity significantly increased as the tumor stage progressed in samples obtained from pretreated patients (n=19, p=0.0064, Kruskal-Wallis test, figure 5A) and decreased after treatment including surgery, radiation or both (n=8 paired, p=0.030, paired-t test, figure 5B). The change in G6PD activity in the two representative cases treated with the immune checkpoint inhibitor (ICI) avelumab is shown in figure 5C,D. The case shown in figure 5C responded to avelumab and the serum G6PD activity decreased during treatment. In contrast, the case shown in figure 5D did not respond to avelumab and the serum G6PD activity increased during administration of the ICI; the patient died a short while later. Serum G6PD activity may, therefore, be a useful tumor marker reflecting tumor progression and evaluating the response to treatment.