Characterization of immunosuppressive myeloid cells in Merkel cell carcinoma: correlation with resistance to PD-1 pathway blockade

Abstract

Purpose

Merkel cell carcinoma (MCC) is a highly immunogenic skin cancer. Although essentially all MCCs are antigenic through viral antigens or high tumor mutation burden, MCC has a response rate of only ~50% to PD-(L)1 blockade suggesting barriers to T cell responses. Prior studies of MCC immunobiology have focused on CD8 T-cell infiltration and their exhaustion status, while the role of innate immunity, particularly myeloid cells, in MCC remains underexplored.

Experimental design

We utilized single cell transcriptomics from 9 MCC patients and multiplex-immunohistochemistry staining of 54 patients’ pre-immunotherapy tumors, to identify myeloid cells and evaluate association with immunotherapy response.

Results

Single cell transcriptomics identified tumor-associated macrophages (TAMs) as the dominant myeloid component within MCC tumors. These TAMs express an immunosuppressive gene signature characteristic of monocytic myeloid derived suppressor cells and importantly express several targetable immune checkpoint molecules, including PD-L1 and LILRB receptors, that are not present on tumor cells. Analysis of 54 pre-immunotherapy tumor samples showed that a subset of TAMs (CD163+, CD14+, S100A8+) selectively infiltrated tumors that had significant CD8 T-cells. Indeed, higher TAM prevalence was associated with resistance to PD-1 blockade. While spatial interactions between TAMs and CD8 T cells were not associated with response, myeloid transcriptomic data showed evidence for cytokine signaling and expression of LILRB receptors, suggesting potential immunosuppressive mechanisms.

Conclusions

This study further characterizes TAMs in MCC tumors and provides insights into their possible immunosuppressive mechanism. TAMs may reduce the likelihood of treatment response in MCC by counteracting the benefit of CD8 T-cell infiltration.

Introduction

Merkel Cell Carcinoma (MCC) is a skin cancer that recurs in 40% of patients (1) with about new 3,000 new cases per year (2). MCC is driven by the Merkel cell polyomavirus (MCPyV) in 80% of cases, leading to numerous non-self viral antigens. The remaining 20% of MCC cases are caused by UV mutations that produce neoantigens with a high tumor mutation burden (3). Both etiologies for MCC thus lead to a highly immunogenic cancer, making MCC a model for studying immune oncology. Clinical trials in MCC of PD-(L)1 blockade have resulted in response rates of over 50% (4,5). However, many patients with antigenic MCC tumors do not respond and factors that are strongly predictive of immunotherapy response have not been identified. A key role for the immune system in controlling MCC is supported by the fact that intratumoral cytotoxic T cells strongly predict improved patient outcomes including survival (6,7). In addition, a greater diversity of T cell receptors that are specific for MCPyV within patient tumors also significantly predicts survival (8). Studies of MCC-infiltrating T cells have shown that these cells express multiple inhibitory receptors, with co-expression of PD-1 and TIM-3 being elevated in these lymphocytes (9). However, CD8 T cell infiltration and diversity has not been predictive of immunotherapy response (10), with some patients with highly inflamed tumors not responding. This suggests that other cells within the tumor microenvironment (TME) may play a role in PD-(L)1 resistance.

The innate immune system is essential in the TME as it can both enhance and suppress adaptive immune function. Myeloid cells that include dendritic cells (DCs), monocytes, macrophages, and neutrophils are among the innate immune cells that contribute to the TME. Monocytes and macrophages, in particular, can polarize into diverse functional phenotypes, that can either promote or inhibit tumor progression. In the TME, monocytes/macrophages often differentiate into tumor associated macrophages (TAMs), that can promote tumor progression by secreting mediators that remodel the TME in a manner that supports tumor growth. The presence of TAMs in the TME is often correlated with poorer prognosis and resistance to therapy, including immunotherapies (11). Recent studies have shown that TAMs are often derived from circulating monocytes that infiltrate tumors (12,13). In certain pathological conditions like cancer, monocytes can transform into immunosuppressive monocytic-myeloid derived suppressor cells (M-MDSC) (14). This transformation is a change in activation state rather than an irreversible distinct myeloid subset (15). When in the M-MDSC state, monocytes differentiate into suppressive TAMs (16,17). Key genes such as S100A8 and S100A9 are upregulated in M-MDSC and are associated with inflammation and cancer (18,19). Their high expression is also linked to the development of immunosuppressive TAMs. Moreover, S100A9-expressing macrophages have been associated with tumor progression after PD-1 blockade treatment in metastatic melanoma (18), and S100A8 was shown to induce PD-L1 expression in monocytes and macrophages, promoting immune evasion (19). Numerous therapeutic strategies targeting these immunosuppressive cells have shown promising outcomes in pre-clinical and clinical trials (16), thus providing a promising clinical approach to enhance immunotherapy responses and counteract immune evasion mechanisms.

The objective of our study was to characterize infiltrating myeloid cells and investigate their potential role in response to PD-(L)1 blockade in MCC. We hypothesized that immunosuppressive myeloid cells are involved in a tumor immune evasion mechanisms relating to PD-(L)1 blockade.

We utilized single cell technology to analyze specific myeloid signatures in MCC. Subsequently, we validated selected markers on tumor samples collected before PD-(L)1 blockade treatment analyzing myeloid spatial biology. This approach allowed us to identify and characterize the dominant myeloid cells in MCC: TAMs that express hallmarks of M-MDSCs such as upregulated S100A8 and S100A9 genes, as well as immunosuppressive cytokines like IL-10 and VEGF. Furthermore, our analysis demonstrated that higher levels of TAMs relative to CD8 T-cell infiltration in tumors prior to PD-(L)1 blockade treatment were associated with primary resistance to immunotherapy and with decreased progression-free survival. These findings suggest that the presence of immunosuppressive TAMs may play a significant role in limiting the effectiveness of immunotherapy by impairing the function of CD8 T cells present within MCC tumors.

Materials and Methods

Patient samples

Patient materials and clinical information were obtained from the MCC specimen and data repository, approved by Fred Hutchinson Cancer Center Institutional Review Board (IRB #6585), and followed the Declaration of Helsinki principles. All patients provided written informed consent.

Single cell RNA sequencing experiments:

MCC tumor biopsies were dissociated into single cells using Miltenyi Biotech Tumor Dissociation Kit, human (130–095–929) and gentleMACS dissociator, and cryopreserved for later use with 50% human serum, 40% RPMI, and 10% DMSO. Single cell RNA sequencing experiments (10x Genomics) were conducted on nine patient tumor samples.

Multiplex-immunohistochemistry experiments:

Tumor samples from 78 MCC patients were included in tumor microarrays, of which 56 were collected before first line immunotherapy. The remaining 22 patient tumors were obtained after initiation of immunotherapy. Among the 56 pre-immunotherapy tumors, two were excluded due to missing data. The study was conducted on the remaining 54 patients.

Response to Immunotherapy and Long-term Outcomes

Initial response to immunotherapy was determined based on response assessments within 200 days from the immunotherapy start date. The two hundred day window was selected to represent approximately 6 months of follow-up, ensuring patients would have had at least one follow-up evaluation during this period. Response was defined by the treating clinician with RECIST 1.1 criteria as the guideline. Initial response groups were defined by the latest response assessment within 200 days. If the last response assessment demonstrated progressive disease (PD), then the patient was classified as having PD. The same criteria were applied for complete response (CR), partial response (PR) and stable disease (SD) assessments. Overall survival (OS) was defined as the time from immunotherapy start until death from any cause. Progression-free survival (PFS) was defined as the time from immunotherapy start until first clinical detection of disease progression or death from any cause. OS and PFS were censored at the last contact date if no qualifying event was identified.

Single Cell RNA sequencing sample preparation

Frozen, dissociated MCC tumors were used for single cell RNA sequencing (scRNAseq) and cellular indexing of transcripts and epitopes by sequencing (CITEseq) experiments. Cells were labeled with hashtag antibodies to identify sample origin in subsequent pooling steps, fluorophore labeled antibodies, and DNA barcode CITEseq antibodies (see supplementary table 1). The cells were then sorted on an Aria II Cell sorter (BD Biosciences). Dead cells and debris were excluded, while cell types of particular interest were actively selected for subsequent scRNAseq analysis. Specifically, populations of interest were selected as follows: T cells (CD3⁺), B cells (CD19⁺, in some cases enriched for viral antigen-specificity), myeloid cells (CD3⁻CD19⁻CD56⁻ cells) and tumor and NK cells (both CD56⁺). The single cell suspensions were concentrated to 700–1,200 viable cells/µL, loaded onto chip G, and run through a Chromium controller to obtain Gel Beads-in-Emulsion (10x Genomics). A library preparation process was performed for scRNAseq using the 5′ transcriptome kit with feature barcoding (V1.1; 10x Genomics) per manufacturer guidelines. The DNA library was enriched for 1056 human immunology panel genes (Human Immunology Panel PN 1000259, Target Hybridization Kit PN 1000248, Library Amplification Kit PN 1000249) and sequenced using a NovaSeq instrument (Illumina) with 2 × 92 bp paired-end reads. The sequencing aimed to obtain an average of 20,000 reads/cell, following the 10x Genomics guidelines.

Single Cell RNA sequencing data analysis

Raw sequencing reads were aligned to the hg38 genome using Cell Ranger v.3.1. Filtered counts matrices of transcripts and feature barcoding counts were loaded into a SingleCellExperiment object for further analyses in R (v.4.1.2). Sample hash deconvolution was performed using DropletUtils (v.1.14.2). Doublets were detected using scds (v.1.10.0) and hash deconvolution, and subsequently removed.

Cells from various runs were combined into a single dataset using totalVI (20). This integration was conducted using the enriched 1056 immunity genes and 12 CITE-Seq proteins that were shared between all runs. The low-dimensional latent representation inferred by totalVI was then used to generate a UMAP projection for visualization, and clustered using Leiden (21). The myeloid cell clusters were identified using common markers, and then re-clustered separately from the rest of the data using Leiden. Upregulated genes in each TAM cluster were identified using totalVI, with a target false discovery rate of 0.05 and positive posterior mean fold change. Overrepresentation of upregulated genes (KEGG database, RRID:SCR_001120) pathways were tested using the GSEApy (https://pypi.org/project/gseapy/) interface to Enrichr (https://maayanlab.cloud/Enrichr/).

Tissue microarray (TMA) construction

TMA blocks were constructed on the fully automated TMA Grand Master (3DHistech) instrument. Staff pathologists reviewed scanned Hematoxylin & Eosin (H&E)-stained sections and identified areas with the highest tumor density, from which two 1.0 mm diameter tumor cores and two tumor margin cores were sampled per case. The Grand Master TMA control software overlapped the annotated digital slide with the image of the corresponding donor block. According to the digital annotations, cores were removed from the donor blocks using the TMA instrument and then relocated to a recipient block in a precise alignment. Replicate tumor and margin cores were placed into separate TMA blocks. TMA data were automatically archived and stored in a detailed grid map file. In total, 10 TMAs were constructed comprised of 100 MCC cases from 115 blocks and representing 78 patients. Subsequently, 4 μm sections were cut from the constructed TMA blocks, H&E-stained, and viewed for quality and cellularity.

Multiplex immunohistochemistry staining

Four-micron sections of the TMA blocks were stained on a Leica BOND Rx autostainer using the Akoya Opal Multiplex IHC assay (Akoya Biosciences, Menlo Park, CA) with the following changes: Additional high stringency washes were performed after the secondary antibody and Opal fluor applications using high-salt TBST (0.05M Tris, 0.3M NaCl, and 0.1% Tween-20, pH 7.2–7.6). TCT was used as the blocking buffer (0.05M Tris, 0.15M NaCl, 0.25% Casein, 0.1% Tween 20, pH 7.6 +/− 0.1). All primary antibodies were incubated for 1 hour at room temperature (RT). Slides were mounted with ProLong Gold and cured for 24 hours at RT in the dark before image acquisition at 20x magnification on the Akoya PhenoImager HT Automated Imaging System using the MOTiF filter set. Images were spectrally unmixed using Akoya Phenoptics inForm software.

Multiplex immunohistochemistry image analysis

HALO software (Indica Labs, Corrales, NM) High-Plex FL module was used to analyze each TMA slide image separately. Cells were identified based on the DAPI nuclear stain, and mean pixel fluorescence intensity was measured in applicable compartments for each cell. Cell settings were defined using real-time tuning and were verified on at least 6 representative tumor cores per TMA. The final algorithm parameters were applied to all tumor cores in the slide image. Two quality control layers were performed: 1) compare positive cell signals to positive object data results on at least three different cores, and 2) compare cell counts to images by a pathologist. Summary data was exported to CSV files, and cell counts were merged with corresponding patient metadata using advanced Excel functions. GraphPad PRISM 9 and R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) were used to generate plots and compare patient treatment groups.

Statistical analysis

Cell type percentages and their ratios to CD8 T-cells were log-transformed before statistical comparisons to reduce right-skewness. To avoid taking the log of zero, zero cell counts were replaced with 0.5 before calculating percentages and log-transformations (22). The cell percentages and ratios were compared between the PD and CR initial response groups using Student’s t-test, receiver operating characteristic (ROC) curve analysis, and the area under the ROC curve (AUC). Associations of cell percentages between different cell types were assessed using Spearman’s rank correlation coefficient.

Univariable and multivariable Cox regression models were used to evaluate associations of cell percentages and ratios with OS and PFS outcomes among all patients, regardless of their initial response category. The multivariable models included sex, immunosuppression status and age as covariates. Cell markers (log-transformed) were included in models as continuous variables. Univariable models were also fit using only tumors with a CD8 percent higher than the median (high CD8 tumors). Multivariable adjustments were not performed for the high CD8 tumor models due to the smaller number of events. Associations were summarized by hazard ratios (HRs) expressed as the HR per 2-fold increase in the cell marker. The Wald test of the HR was considered the primary test of association because it avoids needing to dichotomize the markers and is expected to maximize statistical power (23). To produce an approximate visualization of the continuous associations indicated by the Cox models, OS and PFS curves were estimated using the Kaplan-Meier estimator after stratifying patients by whether the cell marker was above or below the sample median. All statistical calculations were performed using R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) and GraphPad Prism version 9. All p-values are two-sided and statistical significance was defined as p < 0.05. To avoid reducing power in this initial study, P-values were not adjusted for multiple comparisons.

Data availability

The data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE227708 and in NCBI database at PRJNA1019891.

Results

TAMs are the predominant myeloid subset in MCC and exhibit an immunosuppressive gene signature.

Single-cell CITE-seq analysis of 9 virus-positive MCC tumor samples (see Table 1 for clinical characteristics) was performed to characterize the types of myeloid cells present in MCC. MCC tumor cells and immune cells were enriched by flow sorting prior to single cell analysis. Unbiased clustering identified tumor cells, B cells, CD4 and CD8 T cells, NK cells, myeloid cells, pDCs and fibroblasts (Figure 1A and supplementary table 2). Sub-clustering of myeloid cells demonstrated a small cluster of DCs, and 3 clusters of TAMs as the most prevalent myeloid cell subtypes (see differentially expressed genes in supplementary table 3). To further characterize TAMs, we used the recent consensus/universal gene signature system of major mononuclear phagocyte (MNP) subsets (12). Assessment of the six MNP universal signatures (migratory DCs, DC2/3, DC1, CD16⁺ monocytes, classical monocytes, and macrophages; see supplementary table 4 (12)) demonstrated that all three TAM clusters in our analysis expressed high levels of four universal genes associated with classical monocytes (S100A8, S100A9, S100A12 & CSF3R), and did not show expression of macrophage genes (Figure 1B and Figure S1). The fact that TAMs present in MCC have multiple monocyte patterns, suggest they may have originated from M-MDSC. To test this, we examined M-MDSC-related immunosuppressive genes (supplementary table 4 (14)) and found that all three TAM sub-clusters expressed M-MDSC-associated genes (Figure 1C and Figure S2A) and the canonical CD11b protein (Figure S2C). DC genes were also evaluated (Figure S2B), and their canonical protein CD1c was expressed (Figure S2C). Patient samples within each cluster are shown in Figure S2D. To identify candidate targets for multiplex-IHC, we ranked the highest expressed genes within each subcluster. Indeed, S100A8 and S100A9 were selectively upregulated in all three TAM subclusters (Figure S2E) compared to DCs, suggesting they can be used to identify TAMs in MCC. Furthermore, we investigated the expression of immune checkpoint molecules on myeloid cells in MCC. Consistent with previous reports, we observed variable expression of PDL1 and PDL2 on DCs and TAMs, but not on MCC tumor cells. Notably, we found high expression of inhibitory leukocyte immunoglobulin-like receptors (LILRBs) 1, 2, and 4 on all TAM and DC clusters (Figure S3). We further conducted pathway analysis (Figure 1D) to explore the potential functional role of MCC TAMs. This analysis found that the two most prominent TAM transcriptional pathways were ‘cytokine-cytokine receptor interaction pathway’ and ‘viral protein interaction with cytokine and cytokine receptor pathway’.

Table 1.

Patient demographics table for patients included in scRNAseq studies

	Overall (N=9)
Age
Mean (SD)	69.6 (11.4)
Range	56.0 – 93.0
Sex
Female	2 (22.2%)
Male	7 (77.8%)
Immunosuppression
N	8 (88.9%)
Y	1 (11.1%)
Sample location
Axillary lymph nodes	2 (22.2%)
Extremity	4 (44.4%)
Head/neck	2 (22.2%)
Trunk	1 (11.1%)
Recurrence/Primary
Primary	2 (22.2%)
Recurrence	7 (77.8%)
Viral status
Positive	9 (100.0%)

scRNAseq cohort

Figure 1: Transcriptomic analysis of myeloid cells in MCC reveals abundant TAMs demonstrating a strong M-MDSC signature.

Single cells from MCC tumors were processed for transcriptomics as described in methods. A) Unbiased clustering of cells of interest from 9 MCC tumors across 1056 innate and adaptive immune genes. B) Violin plots of relative gene expression of 4 genes (names under plots) that were selected as representative of the mononuclear phagocytes’ universal gene signature. C) M-MDSC gene expression is depicted within all 3 TAM clusters. D) KEGG_2021_Human pathway analysis of TAM clusters were used to identify the 45 most significant pathways.

The relative proportion of TAMs to CD8 as potential predictor of immunotherapy response

To assess the correlation between TAMs in MCC tumors and patients’ response to immunotherapy, we performed multiplex-IHC staining for 3 TAM markers identified transcriptomically (S100A8, CD14, CD163), and a cytotoxic T cell marker (CD8), on the immunotherapy TMA.

Patients (see Table 2) were stratified into 4 response groups, based on their primary response: 18 CR, 19 PR, 2 SD, and 15 PD. The tumor images in Figure 2A show three enlarged tumor cores that exhibited characteristic patterns. Cold tumors, characterized by low or no CD8 counts, were predominantly found in the PD group, whereas tumors with high CD8 infiltration were present in both the PD and CR groups.

Table 2.

Patient data in the Immunotherapy TMA group

	Overall (N=54)
Age
Mean (SD)	70.9 (8.5)
Range	50.0 – 89.0
Sex
Female	15 (27.8%)
Male	39 (72.2%)
Immunosuppression
N	47 (87.0%)
Y	7 (13.0%)
IMTX agent
Anti-PD-1 monotherapy	37 (68.5%)
Anti-PD-1/Anti-CTLA-4 therapy	2 (3.7%)
Anti-PD-L1 therapy	15 (27.8%)
Viral status
Positive	28 (51.9%)
Negative	22 (40.7%)
No data	4 (7.4%)

TMA cohort

Figure 2: Multiplex immunohistochemistry images of MCC tumor microarrays indicate significant myeloid cell infiltration.

A) Representative tumor-core images from the MCC immunotherapy tumor microarray. The tumor microarray slides were stained with 4 immune markers: CD8 (purple) indicates cytotoxic T cells, S100A8 (green) enriched in M-MDSC, CD14 (yellow) indicates monocyte lineage and CD163 (red) indicate macrophages lineage. CD8 T cell counts were divided by the median: tumors with CD8 < median are defined low, and tumors with CD8 > median are defined high. Responses defined as complete responders (CR) and progressive disease (PD). B) Magnified image of the Z1371 patient’s tumor exclusively displaying myeloid markers. White arrows indicate MDSC-MAC: CD14+CD163+S100A8+, light blue arrows indicate MDSC: CD14+CD163−S100A8+, and purple arrows indicate Mono-Mac: CD14+CD163+S100A8−.

An important difference was that, relative to the CR group, the PD group had markedly higher myeloid cell markers, particularly S100A8. To further investigate the underlying biology, we carried out a detailed analysis of the CR and PD groups. We did not include the intermediate groups because tumors isolated from patients with SD and PR have (by definition) variable responsiveness and would thus introduce noise into the biological analyses. Data representing the entire cohort is shown in Figure S4.

We compared the expression of selected markers using three different co-expression patterns representing different TAM phenotypes (Figure 2B): monocytic macrophages (Mono-Mac; CD14⁺CD163⁺S100A8⁻), MDSC macrophages (MDSC-Mac; CD14⁺CD163⁺S100A8⁺), and MDSCs (CD14⁺CD163⁻S100A8⁺). To evaluate the frequency of these markers among responders and non-responders, we compared the percentage of positive cells for each single marker and cell phenotype between the CR and PD groups (Figure 3A). Although prior studies have seen a non-significant trend between the presence of intratumoral CD8 T cells and response to immunotherapy (24,25), in our cohort, intratumoral CD8 T cells significantly correlated with clinical response to PD-1 blockade (Figure 3A). No significant difference was observed between the response groups regarding CD14, CD163, Mono-Mac, MDSC-Mac and MDSC (Figure 3A). Conversely, higher expression of S100A8 was detected in CR tumors, though it did not reach statistical significance (p = 0.056). This observation is surprising, as S100A8 has been previously associated with immune suppression. To investigate a possible correlation between CD8 T cells and S100A8-expressing cells, we performed a Spearman correlation test (Figure 3B). A positive correlation between S100A8 and CD8 (R = 0.62) was observed, as well as a stronger link between CD14 and CD8 (R = 0.7; left panel of Figure 3B). A correlation between all cell phenotypes and CD8 T cells was also observed (right panel of Figure 3B). Of note, the strongest positive correlation (R = 0.76) was observed between the MDSC-Mac phenotype and CD8 T-cell infiltration. Interestingly, while CD14 and myeloid cell phenotypes also showed correlations with CD8 T cells, S100A8 had the largest apparent difference between CR and PD tumors, similar to the pattern observed for CD8 T cells. To further explore the relationship between CD8 and S100A8, we analyzed the distribution of their counts in each tumor (Figure 3C). The distribution plot revealed that S100A8 tends to be present more in tumors that have some level of CD8 T cell infiltration. Furthermore, in comparison to PD tumors, CR tumors exhibited a higher presence of CD8 T cells relative to S100A8. We further stratified the tumors based on their CD8 levels. The high CD8 group was defined as having a CD8 T cell percentage greater than the median (CR=12, PD=4). We then measured the ratio of S100A8 to CD8, as well as the ratio of the other individual markers and cell phenotypes to CD8, in the high CD8 group (Figure 3D). Our findings showed that the S100A8 to CD8 ratio tended to be higher in PD patients (p = 0.054). Furthermore, we observed a significant elevation in the ratio of MDSC-Mac to CD8 in the PD group (p = 0.007), suggesting a potential direct immunosuppressive role of MDSC-Mac on CD8 T cells. We next explored whether the ratio of MDSC-Mac to CD8 could better predict response than CD8 T-cell infiltration levels alone. A receiver operating characteristic (ROC) analysis showed among all tumors, CD8 had an AUC = 0.71 and MDSC-Mac to CD8 ratio had AUC = 0.74 (Figure S5). However, among the high CD8 tumors, MSDC-Mac to CD8 ratio had an AUC = 0.96 (p = 0.008) while CD8 had an AUC = 0.60 (Figure 3E). These results suggest that the ratio of myeloid cells to CD8 T cells might be a better predicator of response.

Figure 3: The ratio of MDSC-Mac to CD8 within MCC tumors with higher CD8 infiltration is associated with resistance to treatment.

Analysis of cell counts obtained from multiplex-IHC images. A) Comparison of positive cell percentages for the indicated single marker or cell type phenotype within tumors from patients with complete response (CR, blue) or progressive disease (PD, orange). CD8: T cell marker, S100A8: MDSC marker, CD14: monocyte lineage marker, CD163: macrophages lineage marker, Mono-Mac: CD14+CD163+S100A8−, MDSC-MAC: CD14+CD163+S100A8+ and MDSC: CD14+CD163−S100A8+. B) Results of spearmen test correlation plots between intratumoral percentages of positive single markers (left matrix) and co-expressing cell type phenotypes (right matrix). C) CD8− and S100A8-positive cell counts listed by patient case code and response (PD vs CR). D) Comparison of ratios between infiltrated individual markers and cell type phenotypes to CD8 T cells in tumors with CD8 percentage higher than the median. E) Receiver operating characteristic analysis comparing between CD8 percentage (green line) and MDSC-Mac/CD8 ratio (blue line) as a classifier of response to treatment within tumors with CD8 percentage higher than the median.

Given the findings relating to TAMs in terms of initial response, we wanted to determine the correlation between TAMs and long-term prognosis in the entire cohort, including the middle group (PR and SD). We observed that the extent of CD8 T cell infiltration was significantly associated with PFS among all tumors (HR: 0.85 per 2-fold increase in CD8%, p = 0.0023), indicating a reduced risk for patients with higher CD8 infiltration (Figure 4A). However, when we focused on tumors with high CD8 only, the extent of infiltration no longer had a significant association with PFS (HR: 0.93 per 2-fold increase in CD8%, p = 0.694). When we tested the ratios of TAMs to CD8 we observed a significant positive association with each of the three TAM subtypes and shorter PFS among all patients (MonoMac/CD8: HR: 1.18 per 2-fold increase, p = 0.0175, MDSCMac/CD8: HR: 1.24 per 2-fold increase, p = 0.0074 and MDSC/CD8: HR: 1.14 per 2-fold increase, p = 0.0495). Ratios of individual TAM markers to CD8 demonstrated similar associations. Results were similar after multivariable adjustments (Figure S6). Furthermore, when evaluated in tumors with high CD8 only, the numerical value of the HR associated with both MDSC-Mac/CD8, and S100A8/CD8 further increased from 1.24 to 1.39, and from 1.20 to 1.38 respectively. For both ratios, the corresponding HRs became marginally significant (p=0.0742 and p=0.0538) as the confidence intervals (CIs) became wider due to a smaller number of events. These findings support the idea that these ratios have prognostic value, particularly in patients with CD8-infiltrated tumors. The Kaplan-Meier estimates of PFS are presented in Figure 4B-C. There were similar associations with OS (Figure S7). To test whether the ratios have an effect on the duration of response, we performed an exploratory sub-group analysis of responders only (CR and PR, n=37). A higher ratio of MDSC-Mac/CD8 showed a negative trend with response durability (HR: 1.20 per 2-fold increase, p = 0.092), while a higher ratio of S100A8/CD8 significantly correlated with a less durable response (HR: 1.28 per 2-fold increase, p = 0.006, Figure S8). These findings further support these ratios as prognostic biomarkers, specifically the S100A8/CD8 ratio.

Figure 4: Higher levels of MSDC-Mac cells relative to CD8 T cells in MCC tumors, and especially in tumors with high CD8 T cells infiltration, are associated with worse prognosis.

A) The hazard ratio (HR) for progression-free survival (PFS) was estimated for each marker, cell phenotype and ratio using a univariable Cox model based on all tumors (Black square, n=54) and based on tumors with high CD8 only (Red square, n=27). CD8% (p = 0.002) and MDSCMac/CD8 ratio (p = 0.007) were significantly associated with PFS among all tumors. After restricting to tumors with high CD8, the HR for MDSCMac/CD8 numerically increased (1.24 to 1.39) and was marginally significant (p = 0.074). B) Representative kaplan-Meier estimates of PFS in all tumors, and C) in high CD8 tumors only. * The median values in B and C are based on all 54 tumors, except for the CD8% graph in column C, for which the median of the 27 tumors with high CD8% was used.

Potential TAM-associated immunosuppressive mechanisms

To explore possible immunosuppressive direct interactions between TAMs and CD8 T cells, we performed spatial analysis on the tumors within the high CD8 group (n=16). We examined how frequently TAMs were the nearest neighbor cell to a CD8 T cell and also the number of TAMs within 40μM of CD8 T cells (Figure S9). Surprisingly, no significant differences were found between the CR and PD groups, meaning no preferred spatial interaction between these cell types was observed in either of the response groups. Given the absence of clear spatial interactions, we assessed whether gene expression pathways may help explain how there would be a functional relationship between cells that are not preferentially adjacent. Indeed, the most enriched pathway within all 3 TAM clusters was ‘cytokine-cytokine receptor interaction pathway’ (Figure 1D). The lack of proximal interaction between these cell types, coupled with transcriptional evidence of active cytokine signaling in the identified TAMs suggests that immunoregulatory cytokines (IL10, IL1β, IL6 were upregulated in TAMs) may affect CD8 T cells within the TME.

Discussion

The role of myeloid cells in cancer has been broadly studied and new therapeutics are being developed to target diverse myeloid subsets. Nevertheless, the complex biology, high plasticity, and diversity of myeloid cells between tumor types and within tumor niches limits our understanding of how these cells impact responses.

Previous studies in MCC have suggested potential immunosuppressive roles of myeloid cells in MCC tumors. Indeed, as PD-L1 and PD-L2 expression are largely restricted to myeloid cells within MCC tumors, it is these cells that are the likely targets of both FDA-approved agents for this cancer (26,27). Furthermore, a more comprehensive characterization of myeloid cells in MCC and their correlation with patient outcomes (28) showed that the infiltration of myeloid cells (macrophages: CD68+ CD163+ and MDSC: CD33+) is strongly associated with CD8 infiltration, and a high infiltration of both populations is linked to better outcomes.

In our study, we used single-cell transcriptomics to characterize the types of myeloid cells present in MCC tumors. This modern platform provides a more in-depth characterization than prior immunohistochemistry approaches and shows that most myeloid cells in MCC tumors are TAMs that display M-MDSC characteristics. The M-MDSC can express different markers at various times and locations within the TME, ranging from monocytes to M-MDSC to TAMs depending on their current pathological phase or differentiation stage (29). Therefore, although there were minor variations in the gene signatures of the TAMs, each of the three subtypes identified in our study shared a similar signature characteristic of immunosuppressive M-MDSC. Based on the transcriptomics data, genes that were most closely associated with TAMs were identified and their corresponding antibodies were used in multiplex-IHC to quantitate TAMs within tumor specimens. We initially showed a correlation between infiltration of CD8 T cells and the myeloid markers, particularly the S100A8 and CD14 markers, which is in line with a prior study of myeloid cells in MCC (28). This correlation may lead to the misinterpretation that TAMs may have an anti-tumor role analogous to CD8 T cells in their association with better outcomes. However, our transcriptomic analysis provides clear evidence that these TAMs express an immunosuppressive gene signature and are thus very unlikely to promote anti-tumor immunity. Indeed, we observed that although there is an association with CD8 infiltration, tumors that did not respond to PD-1 blockade despite high CD8 levels also had high levels of myeloid markers, particularly S100A8. To further characterize these relationships, we analyzed ratios of myeloid cells to CD8 T cells. Indeed, our findings indicate that these ratios provide greater insights than CD8 frequency alone. Specifically, in tumors with high CD8 infiltration, CD8 frequency was no longer strongly associated with PFS or OS, while the ratio of TAMs-to-CD8 continued to correlate with outcome. Notably, the ratios of MDSC-Mac/CD8, as well as S100A8/CD8, showed the greatest association with primary resistance and PFS.

The proinflammatory cytokine S100A8, which is usually found in a heterodimer with S100A9, is a calcium-binding protein that is highly expressed in classical monocytes and in MDSC. Our study found that both S100A8 and S100A9 were highly expressed in MCC TAMs. These proteins have been proposed as biomarkers in several types of tumors due to their association with patient outcomes.

Studies have shown that elevated S100A9 stromal expression in early-stage oral cancer is associated with higher risk of recurrence compared to lower S100A9 expression (30). In melanoma, high expression levels of S100A8/A9 were found to be associated with malignant tumors and metastasis (31,32). Furthermore, upregulated serum levels of S100A8/A9 were found to be linked to impaired survival in patients treated with PD-1 blockade compared to patients with lower serum levels (32). The suppressive mechanism of S100A8 is not fully understood. It has been shown that overexpression of S100A8/A9 in monocyte-derived macrophages increases reactive oxygen species (ROS) production by three-fold (33). Varying levels of ROS have regulatory effects on T cell activation (34), with high ROS levels in the TME reducing antitumor activity, proliferation, and promoting T cell apoptosis (34,35). Consequently, the regulation of ROS levels by S100A8/A9 in TAMs may contribute to their immunosuppressive function, warranting further investigation for potential therapeutic targets.

Spatial interaction analysis of TAMs and CD8 T cells within CR and PD tumors showed no significant differences, thus we could not link cell-cell interactions with a possible immunosuppressive mechanism.

Nevertheless, the most upregulated transcriptional pathway identified in TAMs was cytokine-cytokine receptor interactions, suggesting that TAMs may generate a cytokine network that contributes to an exhausted immune status within the tumor. The relative high number of TAMs compared to CD8 T-cells may lead to high levels of inhibitory cytokines within the TME that nullify the benefit of CD8 infiltration, as assessed by response to PD-(L)1 blockade. Previous studies have described the role of major cytokine release and crosstalk in the TME, generating an exhausted environment in esophageal cancer (36) and lung squamous cell carcinoma (37).

In recent years, LILRB1–5 receptors have gained increasing attention due to their prominent role in regulating the immunosuppressive function of MDSCs and TAMs (38). The function of each receptor varies depending on the specific ligands it binds and the immunoreceptor tyrosine-based inhibition motif that mediates its inhibitory effects (39,40). However, the underlying mechanisms of these pathways are still not fully understood. The high expression of LILRB receptors across myeloid cells in MCC (both TAMs and DCs), may contribute to the production of suppressive cytokines and other immune evasion mechanisms. The ligands for LILRB receptors can be expressed by different cells in the TME, and soluble ligands can also be produced by the same myeloid cells expressing LILRB receptors and can then initiate autocrine signaling that maintains immunosuppressive activity of these cells (39).

LILRBs 1, 2 and 4 are the most studied receptors to date, and they are being investigated as therapeutic targets, with a recent phase I trial reporting an impressive CR from treatment with anti-LILRB-2 in a patient with PD-1 refractory MCC (41). The high expression of these receptors on TAMs and DCs in MCC suggests LILRB receptors as appealing targets in this cancer.

Many strategies are being examined to target immunosuppressive TAM subsets. One successful approach to control the expansion of MDSC and reduce their frequency by modulating retinoic acid signaling via all-trans retinoic acid (ATRA). This treatment shifts MDSC towards dendritic cells and anti-tumor macrophages, supporting antigen-specific T cell responses (42). Moreover, it has been shown that ATRA reduces S100A8 expression in tumors (43). A clinical trial testing a combination of anti-PD-1 treatment with ATRA for metastatic melanoma has shown a high response rate (71%, with 50% of patients having a complete response) and reduction of the frequency of circulating MDSC (44). As mentioned above, targeting LILRB2 and LILRB4, has emerged as promising therapeutics for blocking the immune suppressive activity of TAMs. Clinical trials testing the efficacy of these antibodies, including NCT05309187 and NCT03564691, are currently ongoing. In a recent first-in-human trial targeting LILRB2, combination treatment with anti-LILRB2 and PD-(L)1 blockade resulted in a partial response in 45% of patients with solid tumors who had previously experienced disease progression on PD-(L)1 blockade alone (45).

There are several relevant limitations of this study. Although our sample size was relatively large for MCC (samples from 54 patients prior to 1st line immunotherapy) there is clearly a need to validate these findings in an independent cohort. Moreover, to specifically investigate tumors with moderate to high CD8 infiltration, we categorized patient samples based on their CD8 levels. This stratification resulted in a relatively small subset of CD8-high patients who did not exhibit a response to treatment. Although the effect of TAMs in this smaller sub-group did not achieve statistical significance, there was a readily apparent trend that the presence of TAMs in CD8-infiltrated tumors was associated with a higher risk of progression. An additional limitation is that only 31% of patients were non-responders (PD/SD, n=17), and this did not provide as much statistical power as a 50/50 distribution would have.

In summary, MCC tumors are frequently infiltrated by TAMs expressing S100A8. Their infiltration correlates with non-responding high-CD8 patients and suggests a potential mechanism that may nullify CD8-infiltration benefit in these patients. These findings, combined with prior studies, support the rationale of inhibiting immunosuppressive macrophages to enhance anti-tumor responses. The emerging and available approaches that target TAMs represent a promising avenue to overcome myeloid-mediated resistance mechanisms to PD-(L)1 blockade treatment. We suggest that treatments targeting myeloid checkpoints, with or without T cell inhibitory checkpoint blockade, should be prioritized for clinical trials of patients with refractory MCC. Furthermore, such approaches may show benefit in other cancers in which TAMs play a significant role in evasion and resistance to treatment.

Statement of translational relevance.

PD-(L)1 blockade has shown remarkable success in treating advanced Merkel cell carcinoma (MCC), but less than half of patients experience durable benefit. Identifying the components of the tumor microenvironment (TME) that determine response to immunotherapy is crucial for developing future therapies. Intratumoral CD8 T cells are only partially predictive of outcome, suggesting there are important additional factors within the TME that affect response to immunotherapy. Using single cell analytics, we identified tumor associated macrophages (TAMs) with an immunosuppressive phenotype. Infiltration of TAMs was much more prevalent in tumors with abundant CD8 T cells. Indeed, upon linking to clinical response data, an elevated presence of TAMs was associated with a lower likelihood of response even among CD8-infiltrated tumors. These findings suggest that TAMs may actively suppress T cell function in MCC and that modulating their activity could improve responses to immunotherapy.