Spatial biology reveals altered macrophage states in immunosuppressed non-melanoma skin cancer

Summary

Immunosuppressed patients with nonmelanoma skin cancer experience worse clinical outcomes, yet the tumor immune microenvironment associated with systemic immunosuppression remains incompletely defined. Using integrated single-cell, spatial transcriptomic, multiplex immunofluorescence, and spatial epigenomic profiling across immunocompetent and immunosuppressed tumors, we found that overall immune-cell composition was largely preserved despite differences in immune-cell distribution, spatial organization, and T cell clonality. Immunosuppressed tumors demonstrated reduced intratumoral macrophage densities, decreased T cell clonal diversity, altered antigen-presenting cell and T cell spatial interactions, and distinct fibroblast- and macrophage-associated spatial niches. Multi-cohort validation across complementary spatial and single-cell platforms identified consistent alterations in innate-adaptive immune organization in immunosuppressed tumors. Together, these findings define spatial and functional remodeling of the tumor immune microenvironment under systemic immunosuppression and provide a framework for future therapeutic investigation in high-risk patients.

In brief:

Spatial multi-omics reveals macrophage states that constrain adaptive immunity in immunosuppressed non-melanoma skin cancer.

Graphical Abstract

Introduction

Non-melanoma skin cancer (NMSC), primarily comprising basal cell carcinoma (BCC) and cutaneous squamous cell carcinoma (cSCC), is the most common malignancy in the United States. In particular, the incidence of cSCC, the second most frequent subtype and more aggressive form of NMSC, is rising.¹ The pathogenesis of NMSC is multifactorial, with chronic ultraviolet radiation exposure, advanced age, decreased skin pigmentation, and, critically, immunosuppression being key risk factors.^2,3 Although surgical excision is often curative, it can have notable medical costs and aesthetic and functional sequelae.

Immunosuppressed (IS) individuals are particularly at risk for NMSC. Notably, recipients of solid organ transplant are 65–250 times more likely than immunocompetent individuals to develop cSCC.^4,5 In addition to organ transplants, immunosuppression-inducing conditions such as hematological malignancies and chronic autoimmune diseases are associated with increased incidence of cSCC and poor disease outcomes. Immunosuppression not only predisposes individuals to NMSC but is also associated with worse clinical prognoses irrespective of the similarities in treatment modalities between Immunosuppressed (IS) and immunocompetent (IC) patients, suggesting that immunosuppression itself is an independent factor for unfavorable NMSC outcomes.^6,7

Locally advanced cases of NMSC are being increasingly managed with immunotherapeutic strategies. In fact, recent clinical trials of immune checkpoint inhibitors (ICIs) for advanced cSCC have demonstrated reduced surgical morbidity and durable efficacy.^1,8–10 However, systemic delivery of ICIs has increased risk of serious adverse effects for patients who are chronically immunosuppressed.^1,6,11 Owing to a lack of safety and efficacy data for ICIs in this patient population, they do not have access to immunotherapy despite being at higher risk for NMSC and often presenting with more advanced disease when compared with IC individuals. IS patients have often been excluded from immunotherapy trials due to concerns about reduced efficacy, as well as the risk of transplant rejection or worsening autoimmune disease. A deeper understanding of how their underlying immune deficiencies uniquely alter the tumor microenvironment (TME) may help overcome these challenges and lead to the development of safer and more effective therapies for this population. Chronic immunosuppression impairs both innate and adaptive immune functions, weakening the immunosurveillance system, which is instrumental in the early recognition and destruction of cancer cells.¹² The efficacy of ICIs is intrinsically linked with their ability to unleash a potent antitumor immune response, one that is largely dependent on the functional status of T cells and a supportive cast of antitumor immune cells. Whether IS patients exhibit a comparably deficient immune milieu within their TMEs remains an open question. Given the growing traction of local immunostimulatory treatments for cancer, which have been illustrated by ongoing trials and data on intralesional therapy, such modalities may be viable, low-toxicity treatment options for IS patients. Delving into the complexities of the TME of NMSC among IS patients is imperative for navigating this new frontier of localized immunotherapy delivery.^13–15

Overall, this study provides an integrative single cell, spatial, transcriptional, and clonotypic resource describing the TME of IS patients with NMSC. We sought to define the specific alterations in the spatial organization and transcriptional programs of immune cells within the IS TME and to determine whether these changes extend beyond differences in overall immune-cell abundance. Our results demonstrate reproducible alteration in the distribution of antigen presenting cells in the TME of NMSC is associated with changes in the T-cell receptor (TCR) repertoire, suggesting altered immune coordination rather than absolute depletion as a distinguishing feature of the IS state. These data provide insight into the antitumor immune responses for IS patients, which challenges the prevailing notion that immunosuppression is solely driven by reduced immune cell abundance or lymphocytic dysfunction; and provide new prognostic markers and clinically safer and actionable immune targets for this patient population. Our insights into the TME of IS patients may serve as a resource of fully annotated multiomics dataset, for other investigators aiming to perform mechanistic studies or validate translational findings to support strategic approaches to harness the power of immunotherapy for NMSC while curtailing the severe adverse effects of treatment in this sensitive patient population.

Results

Patient demographics and clinical characteristics

In this cohort study, we analyzed 150 specimens from 138 patients with BCC or cSCC in both head and neck and non–head and neck sites, comparing demographic and clinical characteristics between IC and IS individuals, and performing multi-omic profiling across all cancer specimens. (Figure 1A-B, Table S1). Overall, we included 55 IC patients (39.8%) and 83 IS patients (60.1%); 28 patients had BCC (20%), and 109 patients had cSCC (78.9%). One patient had combined features of BCC and cSCC. The patients’ ages ranged from 42 to 92 years-old, with a higher prevalence of male patients in both groups (89.1%) The mean age of the IC patients was 63 years-old (range, 43–92 years-old), and the mean age of the IS patients was 68.5 years-old (range, 42–90 years-old).

Figure 1. Immune landscape of NMSC in immunocompetent and immunosuppressed patients.

(A) Heat map of patient demographics. IS, immunosuppression; HM, hematological malignancy; OTR, organ transplant recipient; HN, head and neck; cSCC, cutaneous squamous cell carcinoma; BCC, basal cell carcinoma. (B) Schematic of the experimental approach and analytic platforms used. ScRNA/TCR sequencing was performed using CD45⁺-enriched cells. Number of total samples and number of samples from immunocompetent (IC) and immunosuppressed (IS) patients used for each methodology are listed. (C) Kaplan-Meier survival curves of 5-year overall survival and disease free survival comparing IC (blue) vs IS (red) patients (D) UMAP visualization of all immune cells analyzed by scRNA-seq, annotated by major immune cell type. (E) Bar graph comparing the proportions of immune cell types between IC and IS samples analyzed by scRNA-seq. (F) Dot plot showing the difference in immune cell proportions between IS and IC with corresponding FDR values.

Among IS patients, hematologic malignancies accounted for the majority of cases (n = 60, 43.4%), including chronic lymphocytic leukemia (n = 20), follicular lymphoma (n = 7), acute myeloid leukemia (n = 3), diffuse large B-cell lymphoma (n = 3), and multiple myeloma (n = 2). Organ transplant recipients comprised 14 patients (10.1%), most commonly kidney (n = 4) and liver (n = 2) transplants, while an additional 9 patients (6.5%) had other etiologies such as chronic corticosteroid use, autoimmune diseases (e.g., rheumatoid arthritis, SLE), or HIV.

We first sought to evaluate the impact of immunosuppression on survival outcomes in our cohort. Overall survival (OS) analysis revealed that IS patients had higher mortality rate with 5-years OS of 33.3% compared with 68.8% in the IC cohort, p<0.001 (Figure 1C). Disease free survival (DFS) was impacted by the immune state as well, with 5 years DFS of 55.2% vs 24.2% in IC patients compared to IS, respectively (Figure 1C). These findings demonstrate that systemic immunosuppression is associated with significantly poorer oncologic outcomes, highlighting the need to better understand the underlying immune mechanisms driving disease progression in this population.

Multimodal profiling reveals similar immune landscapes in NMSC patients regardless of immune status

Given a paucity of data regarding the use of immune checkpoint inhibitors in the IS patient population, we first sought to identify differences in immune cell populations between IS and IC patients with NMSC. Toward this, we unbiasedly defined the immune-cell composition of the TME of patients with NMSC through single-cell RNA sequencing (scRNA-seq) of isolated CD45⁺ cells. Specifically, we used a cohort of 36 samples obtained from 30 patients (18 IS and 12IC) who underwent either Mohs microscopic surgical or biopsy procedures (Table S1). Overall, we sequenced 226,170 single cells, which yielded 25 transcriptional clusters that were systematically consolidated into 10 major immune cell types (Figure 1D). We annotated the cell types within the clusters using CellTypist.⁶¹ We detected the entire range of immune cell populations in tumors from IC and IS patients (Figure 1D, Figure S1A). However, we observed no statistically significant differences in the relative proportions of the overall immune cell subtypes between the two groups (Figure 1E-F). Our scRNA-seq findings indicated overall similar composition and relative proportion of immune cells into the TME.

Differential macrophage subsets and spatial distribution in immunosuppressed patients with NMSC

Given that scRNA-seq of isolated CD45+ cells provides information on relative immune-cell composition but does not capture spatial compartmentalization or cell density per tissue area, we next used multiplex immunofluorescence (mIF) to quantify immune-cell densities within the tumor and stromal compartments. We analyzed samples obtained from 88 patients with cSCC (54 IS patients and 34 age- and stage-adjusted IC patients) using multiplex immunofluorescence (Figure 2A). We compared the distributions of CD4⁺ T cells, CD8⁺ T cells, B cells, and CD68⁺ macrophages in the TME within the intratumoral and stromal areas (Figure 2A, B). Whereas both compartments are important for an efficient antitumor immune response, an effective distribution between the tumor and stromal compartments is important.¹⁶ We did not observe a significant difference in the abundance of CD4⁺ T cells, CD8⁺ T cells, or B cells in the tumor or stromal compartment between IC and IS patient samples. Importantly, the IC cohort had significantly higher densities of intratumoral CD68⁺ macrophages than the IS cohort (p=0.022) (Figure 2B, Table S3). In both IC and IS cohorts, the CD68⁺ macrophage density in cytokeratin-negative stroma was significantly higher than that in cytokeratin-positive tumors (both p<0.0001, Figure S2A, Table S3). In patients with solid tumors, such as breast or lung cancer, elevated programmed death-ligand 1 (PD-L1) expression in macrophages has been associated with increased overall survival.^10,17–20 Given the significant differences in CD68⁺ macrophage distribution within the tumor compartments, we investigated the differences in PD-L1 expression on CD68⁺ macrophages between IS and IC patients. In the IS cohort, the tumoral CD68⁺PD-L1⁺ cell density was significantly lower than the stromal CD68⁺PD-L1⁺ cell density (p<0.0001) (Figure S2A, Table S3). Interestingly, the global relative proportions of CD68⁺ cells and CD68⁺CD274⁺ subsets assessed by scRNA-sequencing (scRNA-seq) revealed similar relative proportions of macrophage populations among captured CD45+ cells between the IC and IS samples (Figure S2B-C). Overall, scRNA-seq of isolated CD45+ cells was used to assess immune-cell composition based on relative proportions, whereas mIF enabled quantification of immune-cell density per area within tumor and stromal compartments. In this context, scRNA-seq revealed similar relative proportions of macrophage populations between IC and IS samples, while mIF identified differences in CD68+ macrophage density and compartmental distribution.

Figure 2. Integrated spatial and transcriptional profiling reveals altered myeloid and macrophage landscapes in immunosuppressed NMSC.

(A) Multiplex immunofluorescent analysis of cSCC tissue stained with antibodies against cytokeratin, CD4, CD8, CD68, FOXP3,LAG3 and PD-L1 and counterstained with DAPI. CK, cytokeratin. (B) Violin plots demonstrating the density distributions of immune cell phenotypes (cells/mm², log₂) in tumor and stromal regions of cSCC samples from immunosuppressed and immunocompetent patients. (C) Heat map clinical and demographics annotation plot of patients in the external validation cohort from Cleveland Clinic. IS, immunosuppression; HM, hematological malignancy; OTR, organ transplant recipient; HN, head and neck; cSCC, cutaneous squamous cell carcinoma; (D) Multiplex immunofluorescent analysis of cSCC tissue from IC and IS patients in the Cleveland Clinic cohort stained with antibodies against CD11b, CD11c, CD40, CD68, and HLA-DR, together with cytokeratin (Pan-CK) and DAPI counterstain. (E) Violin plots showing the density distributions (cells/mm², log₂) of total myeloid cells (CD11b⁺CD11c⁺CD68⁺), macrophages (CD68⁺), dendritic cells (DCs; CD11c⁺CD68⁻), and antigen-presenting DCs (CD11c⁺CD68⁻HLA-DR⁺) in IC and IS tumors as quantified by multiplex immunofluorescence. Mean ± SEM is indicated; p values were determined by two-tailed Student’s t-test. (F) UMAP visualization of macrophage populations from all IC & IS NMSC using scRNA-seq, annotated by transcriptionally defined TAM subtypes: resident-tissue macrophages (RTMs), proliferative TAMs (Prof-TAMs), IFN-responsive TAMs (IFN-TAMs), lipid-associated TAMs (LA-TAMs), ECM-degradation TAMs (ECM-Deg), and inflammatory cytokine-enriched TAMs (Inflam-TAMs).(G) Dot plot of the canonical gene markers used to identify the macrophages cell types (H) Bar graph comparing the proportions of TAM populations from IC & IS NMSCs (I) Dot plot showing the difference in the macrophage subtypes between IS and IC with corresponding FDR values. (J) Number of differentially expressed genes (DEGs) between IS & IC tumors across the macrophages subclusters. (K) Pathway enrichment analysis of DEGs between IS and IC TAM populations showing key immunoregulatory and inflammatory signaling pathways; circle size represents significance (p value) and color indicates z-score of pathway activation. Pos, positive; reg, regulation; Ag, antigen.

To evaluate the association between immune-cell densities and clinical outcomes, we performed multivariable Cox regression and Kaplan–Meier analyses in IC and IS patients (Figure S2D-E). In IC patients, higher intratumoral and stromal CD8⁺LAG3⁺ T-cell densities were significantly associated with improved 5-year overall survival (p < 0.001 and p = 0.031, respectively). Stromal CD68⁺ macrophage density was also associated with overall survival in IC patients (p = 0.040). In IS patients, increased stromal CD68⁺PD-L1⁺ macrophage density was associated with reduced survival (p < 0.001). Kaplan–Meier analyses corresponding to these findings are provided in Figure S2E.

Given our findings indicating altered spatial organization of CD68⁺ cells, we sought to better define the intratumoral myeloid cell architecture. An external cohort from Cleveland Clinic (n=54 cSCC specimens; 29 IS, 25 IC) was profiled by multiplex immunofluorescence to assess intratumoral myeloid composition (Figure 2C-E). Across markers of the myeloid lineage (CD11b/CD11c/CD68), IS tumors harbored decreased intratumoral myeloid cell densities compared with IC tumors, with a notable reduction in CD68⁺ macrophages (p=0.029). Together, these data support that immunosuppression is associated with a decrease of intratumoral myeloid populations, particularly macrophages, and reduced intratumoral myeloid cell densities relative to IC tumors (Figure 2D-E; Table S3).

Given the differences in CD68⁺ macrophage trafficking between the tumor compartments in NMSC samples of IS patients, we sought to leverage our cohort for a more robust functional analysis. We therefore performed detailed macrophage subclustering based on our scRNA-seq data. Macrophages, which are known for their phagocytic role and activation into diverse phenotypes in response to environmental cues,²¹ exhibit heterogeneity that contradicts the traditional M1/M2 dichotomy of tumor-associated macrophages (TAMs). Subclustering of macrophages in the TME of tumor samples derived from IC and IS patients with NMSC identified 6 distinct clusters of macrophages based on their functional gene expression (Figure 2F-G, Table S2)^22,27,28: inflammatory cytokine-enriched TAMs (Inflam-TAMs; IL1β⁺IL1RN⁺CXCL8/3/1/2⁺), extracellular matrix (ECM)-degradation TAMs (ECM-Deg TAMs; MMP9⁺SPARC⁺), lipid-associated TAMs (LA-TAMs; APOC1⁺APOE⁺LPL⁺),proliferative TAMs (Prolif-TAMs; MKI67⁺STMN ⁺TYMS⁺), interferon (IFN)-responsive TAMs (IFN-TAMs; CXCL9/10 +IFITM1/3⁺), and resident-tissue macrophages (RTMs;CD163⁺MS4A7 ⁺MAF⁺MRC1⁺). The composition of macrophage subtypes differed between IC and IS tumors (Figure 2H-I). Resident-tissue macrophages were relatively more represented in IS samples, whereas proliferative TAMs were relatively more represented in IC samples; however, these differences were not statistically significant (Figure 2H-I). Ingenuity pathway analysis of differentially expressed genes within each macrophage subset identified pathway-level differences between IS and IC tumors that varied across TAM populations (Figure 2J-K, Table S4). Across multiple subsets, several innate immune and cytokine-associated pathways showed negative z-scores in IS relative to IC, including FCGR-dependent phagocytosis and IL-6, IL-17, and IL-33 signaling. In contrast, macrophage activation programs were not uniformly reduced, with classical activation showing positive z-scores in select TAM subsets. The magnitude and direction of pathway enrichment differed across RTMs, Inflam-TAMs, and ECM-Deg/IFN-responsive populations, highlighting the heterogeneity of pathway activity across macrophage subsets in IS tumors.

T-cell clonal diversity is diminished in immunosuppressed TMEs

To assess whether the altered cellular distribution of CD68⁺ cells in the TME of NMSC in IS patients results in a decrease in T-cell activation, we analyzed the clonal diversity of T cells using single-cell T-cell receptor (TCR) sequencing. First, we performed detailed T-cell subclustering, identifying multiple functionally distinct subsets (Figure 3). We extracted T and natural killer (NK) cells from the CD45⁺ population isolated from tumors of IC and IS patients with NMSC and then further subclustered distinct T-cell populations.

Figure 3. T-cell subset analysis identifies distinct populations of CD8⁺ and CD4⁺ T cells in both immunocompetent and immunosuppressed NMSC patient.

(A) UMAP visualization of CD8⁺ T cells cells from immunocompetent (IC) and immunosuppressed (IS) patient tumors with T-cell subset annotations. (B) Dot plot of the canonical gene markers used to identify CD8⁺ T-cell types. (C) Bar graph comparing the relative proportions of CD8⁺ T-cell and CD4⁺ T-cell subsets between IC and IS patient tumors. (D) Dot plot showing the differences in CD8⁺ T-cell subsets between IS and IC tumors, with circle size indicating FDR significance. (E) Stacked histogram showing the number of differentially expressed genes (DEGs) between IS and IC CD8⁺ T cells showing the proportion of genes upregulated, downregulated, or unchanged in IS tumors. (F) Pathway enrichment analysis of DEGs between IS and IC CD8⁺ T cells highlighting key immune activation and signaling pathways; bar color represents predicted activation state (z-score) in IS samples. (G) UMAP visualization of CD4⁺ T cells from IC and IS patient tumors with T-cell subset annotations. (H) Dot plot of the canonical gene markers used to identify CD4⁺ T-cell types. (I) Bar graph comparing the relative proportions of CD4⁺ T-cell subsets between IC and IS patient tumors. (J) Dot plot showing the differences in CD4⁺ T-cell subsets between IS and IC tumors, with circle size indicating FDR significance. (K) Stacked histogram showing the number of DEGs between IS and IC CD4⁺ T cells showing the proportion of genes upregulated, downregulated, or unchanged in IS tumors. (L) Pathway enrichment analysis of DEGs between IS and IC CD4⁺ T cells highlighting key immune activation and signaling pathways; bar color represents predicted activation state (z-score) in IS samples.

Among CD8⁺ T cells, we identified 11 clusters spanning the full spectrum of differentiation states: naïve cells, memory populations distinguished by migratory versus metabolic profiles, effector memory cells marked by granzyme K (GZMK⁺), chemokine XCL1⁺, and cytotoxic (CTL) signatures, an interferon-stimulated gene-high (ISG-high) population, stress-responsive cells expressing heat shock protein HSPA6⁺ or nuclear receptor NR4A1⁺, NK-like cells, and terminally exhausted (Tex) cells (Figure 3A-B, Table S2). Similarly, CD4+ T-cell analysis identified 14 distinct clusters including naïve cells, memory subsets with specialized CREM⁺ TGFB1⁺ signatures, regulatory T cells differentiated by FOXP3⁺ TNFRSF18⁺, IKZF2⁺, and FOXP3⁺ GNLY⁺ expression patterns, proliferating cells, antigen-presenting-like (APC-like) cells, follicular helper populations including a specialized G0S2⁺ subset, Th17 cells, stress-responsive HSPA6⁺ and NR4A1⁺ populations, and exhausted cells, collectively representing the diverse functional landscape of tumor-infiltrating T lymphocytes (Figure 3G-H, Table S2). Notably, the abundance of T-cell subsets was not different between the IC and IS cohorts for both CD8⁺ and CD4⁺ T cells (Figure 3C-D, I-J).

To functionally evaluate potential T cells perturbation associated with immunosuppression, we ran a comprehensive differential gene expression (DEG) and pathway analyses comparing IS versus IC populations (Figure 3E-F, K-L). This analysis revealed 2,947 differentially expressed genes in CD8⁺ T cells (FDR<0.05,Table S5), with IS samples exhibiting downregulation of cytotoxic effector pathways (GZMB, PRF1, IFNG) and upregulation of exhaustion markers (PDCD1, CTLA4, TIGIT). Similarly, CD4⁺ T cells demonstrate 2,181 DEGs with enrichment of regulatory and immunosuppressive programs in IS samples (Table S6). Ingenuity pathway analysis (IPA) pathway analysis confirmed suppression of anti-tumor immunity pathways in IS samples across multiple cell types (CD8: p<0.001 for “Cytotoxic T Cell-mediated Apoptosis”; CD4: p<0.001 for “Th1 Pathway”; Myeloid: p<0.001 for “Dendritic Cell Maturation”).

Next, we evaluated the clonotype diversity for both CD8⁺ and CD4⁺ T-cell subsets (Figure 4A-D). Antigen-presenting cells, including macrophages and cDC2 dendritic cells, present antigens to CD4⁺ and CD8⁺ T cells, facilitating their clonal expansion and diversification.^23–25 Comprehensive evaluation of T-cell clonal expansion in IC and IS patients with NMSC revealed a significantly lower number of hyperexpanded CD8⁺ T-cell clonotypes in IS tumors than in IC tumors (Fisher exact test, p=0.0005) (Figure 4A-D). TCR diversity was comprehensively evaluated across distinct T-cell populations using multiple diversity indices (Figure 4E-F). IS patients demonstrated significantly diminished TCR diversity relative to IC patients across both major T-cell subsets (CD8⁺: Shannon p = 0.014, CD4⁺: Shannon p < 0.001).

Figure 4. Single-cell TCR sequencing reveals decreased T-cell clonal expansion and clonotype diversity in immunosuppressed NMSC patient tumors.

(A-D) UMAP visualization of CD8⁺ (A) and CD4⁺ (C) T-cell clusters and comparison of the proportion of CD8⁺ (B) and CD4⁺ (D)T cell with expanded TCR clonotypes from tumor tissue derived from immunocompetent (IC) or immunosuppressed (IS) patients. The magnitude of the clonal expansion is overlaid across the different T-cell populations in the UMAP visualization. The p-value (Fisher exact test) is displayed above the proportion plot. (E,F) Comparison of TCR clonal diversity index scores for IS and IC tumors for CD8⁺ and CD4⁺ T-cell clusters, respectively. Shannon and Inverse Simpson indices were used to estimate baseline diversity, and the Chao index was used to estimate baseline richness. P-values (Bonferroni-corrected Wilcoxon signed rank test) are displayed. (G,H) Clonal overlap of cell clusters for each sample group using the Morisita index for CD8⁺ and CD4⁺ T cells, respectively.

Using these data, we next sought to identify antigen sharing among T-cell populations (Figure 4G). Clonal overlap analysis using the Morisita index revealed differences between IC and IS tumor environments. IC tumors demonstrated moderate TCR sharing among CD8⁺ T-cell subsets, with notable clonal overlap between EM_GZMK and Stress_NR4A1 clusters, and moderate sharing among Stress_HSPA6, NK-like, Naive, and Mem_migratory populations. IS tumors maintained the EM_GZMK-Stress_NR4A1 correlation but exhibited weaker inter-cluster TCR similarity overall, with reduced sharing patterns particularly evident among the NK-like clusters compared to IC samples. The generally weaker clonal sharing across most CD8⁺ T-cell subsets in IS tumors, while preserving some core stress-effector connections, demonstrates a more constrained TCR repertoire consistent with restricted clonal diversification in response to chronic immunosuppression in the NMSC tumor microenvironment.

Clonal overlap analysis of CD4⁺ T cells revealed a contrasting pattern to CD8⁺ populations (Figure 4H). IC tumors displayed minimal inter-cluster similarity, with weak correlations consistent with compartmentalized CD4+ differentiation and maintenance of discrete functional identities. In contrast, IS tumors exhibited increased inter-cluster TCR similarity, with enhanced sharing particularly evident among Stress_NR4A1, Stress_HSPA6, Prolif, and Naive clusters, and stronger correlations within the Treg populations compared to IC samples. This increased clonal overlap in IS tumors among naive, proliferative, and stress-associated subsets suggests limited clonal diversity and expansion of shared clones across multiple activation states, indicating compromised CD4+ T-cell compartmentalization in the immunosuppressive tumor microenvironment.

Immunosuppression is associated with disrupted immune cell distribution and interactions in the TME

Our results demonstrated that certain populations of innate immune cells in the TME of NMSC samples from IS patients are distributed differently from those in samples from IC patients, which is associated with a skewed TCR repertoire with decreased T-cell diversity and alterations in TCR clonal expansion. Spatial proximity of cells within the TME for processes such as antigen presentation or recognition of co-stimulatory molecules is necessary for an effective immune response.²⁶ To assess whether alterations in the spatial distribution of molecularly and functionally distinct populations of immune cells within the NMSC TME leads to a poor antitumor immune response, we performed spatial transcriptomic profiling at single-cell resolution using the Xenium platform with 81 samples obtained from 45 patients with cSCC (23 IC and 58 IS, Table S7). We designed a customized targeted panel of 350 genes for multiple cell types that includes tumor cells, fibroblasts, endothelial cells, and immune cells, including T, B, NK, dendritic, and myeloid cells (Figure 5A, Table S8). Afterward, hematoxylin and eosin (H&E) staining of the samples confirmed the spatial maps with pathologist annotations (Figure 5B, Figure S3A, Table S7).

Figure 5. Xenium-based spatial mapping of immunosurveillance in NMSC tumors at single-cell resolution delineates distinct niches and CCIs in immunosuppressed and immunocompetent patients.

(A) UMAP embedding of unsupervised clustering analysis combining targeted expression of 350 genes for 81 sample cores in a tissue microarray. Nineteen distinct clusters are overlaid with cell type annotations. (B) H&E stain of tissue microarray cores performed after the Xenium workflow and spatial maps of tissue microarray cores generated using Xenium Explorer v2 with cell group–based affiliations highlighting the concordance between histological and molecular features in representative samples of invasive SCC from immunocompetent (IC, sample 11I) and immunosuppressed (IS, sample 15C) patients. (C) Integration of scRNA-seq immune cell type signatures with Xenium-derived clusters. UMAP visualization and Xenium spatial maps of representative samples of invasive SCC from IC (sample 11I) and IS (sample 15C) patients with cell type annotations derived from the integration of scRNA-seq and Xenium data. (D) Intratumoral niches in representative samples of invasive SCC from IC (sample 11I) and IS (sample 15C) patients listed in decreasing order of cell-type frequency. (E) Box-whisker plot of the proportions of macrophage subtypes within tumor-rich niches between IC (n=20) and IS samples (n=46). Samples with sparse tumor regions were excluded. Tumor-exclusive niches were excluded from the analysis. (F) Point distribution of transcripts specific to tumor, fibroblast, and macrophage subtypes in spatial maps of representative samples of invasive SCC from IC (sample 11I) and IS (sample 15C) patients (see panel D). Transcripts for the following markers are visualized in distinct colors yellow: CD44 (tumor), COL6A1 (stoma), CD68 (TAMs), CTSB (ECM-Deg TAMs), CXCL10 (IFN-TAMs), APOE (LA-TAMs), CXCL1 (Inflam-TAMs), MKI67 (Prolif-TAMs/tumor), and CD163 (RTMs). IC, immunocompetent; IS, immunosuppressed. (G) Combined bar plot and line graph of the cell-type frequencies and numbers of CCI–ligand receptor (LR) interactions and (H) chord plots representing predicted CCIs by stLearn in representative IC (sample 11I) and IS (sample 15C) samples of invasive SCC from IS and IC patients. (I) Cohort (n=81) analysis of the number of CCIs among immune cells, stromal and tumor cells between IS and IC patient samples. (J) Box-whisker plot of proportions of antigen presenting cells (APC) subsets in CD8⁺ T cell–enriched niches (>20% CD8⁺ T cells) across IC (n = 11) and IS (n = 23) samples.

Unsupervised clustering of cells from all 81 cSCC sample cores identified 19 UMAP clusters, which were annotated based on the most differentially expressed gene markers and represented the different components of the TME (Figure 5A). The 19 UMAP clusters encompassed 8 tumor clusters with keratinocytic, cancer stem cell, and neural crest–like features. Each core predominantly harbored tumor clusters (e.g., proliferating keratinocytic-NEFL^high and cancer stem cell-PROM1^high tumor clusters), with specific clusters varying among cores (Figure 5B). To assess the validity of the data across samples, we first confirmed that each cluster was represented by both IC and IS patient samples (Figure S3B). An analysis of high-resolution spatial maps demonstrated that transcriptome-based spatially annotated tumor areas accurately corresponded to histologically defined tumor areas on cSCC samples (Figure 5B, Figure S3A). We identified two macrophage clusters containing CD163⁺CD68⁺CXCL9^high and CD163⁺CXCL12^high cells and one T-cell cluster (Figure 5A, B). We further subtyped the immune cells including macrophages and T cells in these clusters by interpolating the markers used for scRNA-seq annotation to generate UMAP clusters based on cell type (Figure 5C, Figure S4A). We applied the same macrophage subclustering approach that we used in our scRNA-seq data analysis and were able to identify all 6 macrophage subsets (Figure S4B,C) where ECM-deg TAMs were the predominant population (Figure S4D).^27,28 These analyses further underscore the importance of multi-omics approach to assess and cross-validate the nuances of cell-cell interactions within the TME.

To determine how immune cells are spatially confined in specific neighborhoods inside the tumor, we defined spatial niches in the NMSC TME using positional information for each cell (Figure 5D). Using niche analysis, we quantified the proportions of non-tumoral cells within tumor-rich niches and compared distributions between IC and IS samples. As expected, given their higher abundance when compared with other macrophage subsets (Figure S4D), the intratumoral niches of both IC and IS patients predominantly harbored ECM deg-TAMs (Figure 5E). Within intratumoral niches, lipid-associated TAMs were enriched in IC patients (p=0.048), consistent with their known roles in tissue remodeling, lipid metabolism, and local immunosuppression. RTM macrophage abundance was numerically higher in the IS population; however, this difference did not reach statistical significance (p=0.13). These findings highlight distinct macrophage-dominated niches depending on immune status, with RTMs potentially supporting an immunosuppressive environment in IS, and lipid-TAMs reflecting metabolic adaptation and activation in IC tumors. These niche patterns may contribute to, or reflect, altered innate–adaptive cross-talk in this patient population. Analysis of CD8⁺ T-cell markers in cSCC samples demonstrated that exhausted (LAG3⁺) and cytotoxic (GZMA⁺) T cells were more likely to be located within the tumor in IC samples and along the tumor periphery in IS samples (Figure S5).

To directly assess the cell-cell interactions (CCIs) between immune, stromal, and tumor cells, we used the stLearn approach, which relies on co-expression of genes within a spatial context (Figure 5F-J).²⁹ This analysis revealed significant interactions between ligand-receptor (LR) pairs and neighborhoods harboring these cell types which were calculated using permutation tests of spatially co-expressed ligand–receptor pairs, assessing whether the observed co-expression in neighborhoods is higher than expected by chance (Figure 5G-H). Patient-level quantification of CCI counts across immune and stromal populations in the entire cohort (n=81) revealed that B cells, Lipid associated TAMs, Prolif TAMs and monocytes showed significantly more cell-to-cell interaction nodes in IC samples than in IS samples (p<0.05, Figure 5I). Though not significant with current sample size, CD4+ T cells, CD8+ T cells and IFN TAMs showed reduced CCI counts in IS samples compared to IC samples (Figure 5I). Direct cell–cell interaction analysis revealed that immune populations, including ECM-degrading TAMs, IFN TAMs, proliferative TAMs, and CD8⁺ T cells, engaged in diverse heterotypic interactions with other immune cells, endothelial cells, fibroblasts, and tumor cells. In contrast, tumor cells predominantly exhibited homotypic interactions rather than cancer-immune cell interactions (Figure 5H, Figure S6). Among the significant ligand–receptor pairs identified (Table S9), interactions such as LGASL3–LAG3, LTB–CD40, and CXCL12–CD4 between APCs and exhausted T cells showed an increasing trend in IS compared to IC samples. The lack of statistical significance may be attributable to the markedly lower abundance of immune cells in the spatial dataset relative to tumor and fibroblast populations. Limited immune-cell representation within the spatial dataset relative to tumor and fibroblast populations may reduce statistical power for detecting interaction differences. To explore interactions between APCs and CD8⁺ T cells, we focused on niches with CD8⁺ T cell abundance greater than 20%, encompassing 7 IC and 20 IS samples. Within these niches, IS samples showed trends of reduced proportions of ECM-degrading TAMs (p = 0.06), IFN TAMs (p = 0.116), monocytes (p = 0.05), and reduced RTMs (p = 0.018), compared to IC samples (Figure 5J). These data indicate differences in the composition of macrophage populations interacting with CD8⁺ T cells in CD8⁺ T cell–rich niches between IC and IS tumors. Overall, immunosuppression in NMSC is characterized by a reduced number of interactions of macrophages and T cells with the tumor cells in spatial niches in the TME, likely contributing to immune dysfunction.

Spatial ATAC sequencing reveals epigenetic regulation and differential chromatin accessibility across immunocompetent and immunosuppressed TMEs

Epigenetic dysregulation of the antigen presentation machinery is a key mechanism by which cancer cells evade immune surveillance. To identify the epigenetic regulatory mechanisms that may contribute to tumor innate-adaptive cross-talk, we performed spatial assay for transposase-accessible chromatin sequencing (spatial-ATAC-seq) using the deterministic barcoding in tissue sequencing (DBiT-seq) approach. Spatial-ATAC-seq profiles chromatin accessibility with embedded spatial information, providing insight into the regions of the genome that are actively involved in gene regulation in their native spatial contexts. We performed spatial-ATAC-seq with 14 samples obtained from NMSC patients with cSCC and BCC (7 IS and 7 IC). Our quality metrics indicated successful sequencing, including a median of 100,000 peaks per sample, 70,000 average fragments on tissue, transcription start site score of 3.5, and 10% of high-quality fragments overlapping peaks (Figure S7A).

Unsupervised clustering identified 12 spatial epigenetic (spEP) clusters (Figure 6A). We annotated open chromatin peaks identified in each cluster based on genome-wide localization and found the peaks to be distributed across promoters, exonic and intronic regions of genes, and distal to the transcription end sites (Figure S7B). Normalized proportions of the clusters were distributed across most of the samples included in the study (Figure S7C). In samples that undergo spatial epigenetic analysis on a microfluidic chip, the analysis generates an in situ x-y coordinate system from the intersection of the microfluidic row and column. This yields a square-shaped element of tissue called a tixel. Mapping the spEP clusters onto the tixel coordinates of NMSC tumor samples revealed distinct spatial patterns in tumor-rich and peritumoral regions (Figure 6B, Figure S7D). Normalized proportions of the tumor and TME spEP clusters did not show significant differences between IS and IC groups except for fibroblast 2/CD8⁺ cytotoxic T cell cluster (C5) (Figure 6C). Importantly, we did not observe significant differences in the proportions of clusters among different regions of the samples (Figure S7E). To validate these findings, we integrated the spatial-ATAC-seq data with our spatial transcriptomic data. This analysis demonstrated significant concordance of gene activity and gene expression with the most differential cell type–specific gene signatures exhibiting consistent epigenetic and transcriptional activation (Figure 6D,E, Figure S8).³⁰

Figure 6. Spatial epigenomic analysis identifies distinct immune interactions within tumor and stromal regions of NMSC.

(A) UMAP embedding of unsupervised clustering analysis for chromatin accessibility overlaid with cluster identities. spEP, spatial epigenomic; KC, keratinocytic. (B) The spatial distribution of spatial epigenomic clusters in representative BCC and SCC sections from immunocompetent (IC) and immunosuppressed (IS) patients (n= 14) as represented by tixels. Each tixel is a square-shaped element of tissue generated by the intersection of a microfluidic row and column. The color of each tixel corresponds to the clusters in Figure 6A, demonstrating the spatial proximity of the clusters in tumor-rich and peritumoral regions. The sizes of the tixels were scaled for better visualization. Bright-field images and H&E-stained tumor tissue demonstrate cellular localization. (C) Normalized cell proportions in intratumoral and peritumoral clusters in IS and IC tumor samples. (D) Integration of Xenium cell-type signatures with UMAP clusters derived from spatial-ATAC-seq data. (E) Heat map of peak-to-gene Z scores used to integrate the ATAC and RNA features of cell-type signatures in spatial data. (F) Spatial mapping of the GASs for KRT5 (tumor marker), PDGFRA (stromal marker), CD68 (TAM marker), and CD163 (RTM marker) in representative IC and IS samples of BCC.

We further analyzed the gene accessibility scores (GASs) of the unsupervised spEP identified 12 clusters for known and differential cell type–specific markers (false discovery rate < 0.05, log2-transformed fold change ≥ 0.5) (Figure S9A,B) based on their spatial patterns and identified 5 distinct tumor clusters and 7 distinct TME clusters (Figure 6A). Of those, we identified two stromal macrophage types with macrophage 1 (CXCL12^high) enriched in C6 and macrophage 2 (CXCL10^high) enriched in clusters 7 and 9. We noted two different fibroblast clusters with fibroblast 1 (fibroblast rich) in proximity to CD8⁺ cytotoxic T cells (C5 and C8) and fibroblast 2 (fibroblast-macrophage rich) in proximity to macrophage 2 (C9). KRT14, a marker of keratinocytic tumor cells, had a high GAS in clusters 2 and 12, whereas the SOX15 GAS was increased in cluster 11. Cluster 11 primarily contains tumor cells and SOX15 has potential roles as a mediator of pluripotency and stemness and as a tumor suppressor.³¹ Cluster 10 had a high GAS for differentiating basal cell markers such as HOXC10, and cluster 3 had a high GAS for epithelial-to-mesenchymal transition markers such as SNAI2 (Figure S9C,E).^32,33 Also, the GASs for markers for stromal and immune components such as fibroblasts (CDH2), T cells (LAG3), and CD163 (M2 stromal macrophages) were associated with clusters 1, 4, 5, 6, 7, 8, and 9 in the peritumoral regions of the tissues analyzed (Figure S9D,F). Spatial patterns of gene accessibility for selective markers of tumor cells (KRT5), stromal fibroblasts (PDGFRA), TAMs (CD68), and RTMs (CD163) confirmed spatial-ATAC-seq was able to localize macrophages to tumor and stromal regions; and their chromatin accessibility and potential regulatory changes, in responses to their interactions with other cells within the TME (Figures 6F, S9G).

To systematically analyze these interactions and how they might affect transcription in the different niches within the NMSC TME, we carried out a neighborhood analysis by computing the propensity of cluster-specific spatial co-localization.^{34 42} We observed that a cancer-initiating cell–like tumor cluster (C10) had the highest homotypic cluster scores, suggesting minimum interaction with other clusters and the formation of homotypic niches (Figure S10A). Pairwise heterotypic neighborhood analysis using samples from both IC and IS patients determined the spatial relationships between clusters with different epigenetic features (Figure S10B) and suggested that tumor pairs such as C11-C12 and C2-C3 had the highest heterotypic cluster associations (Figure S10C). Among the TME clusters, the endothelial/immune cluster (C1) and the fibroblasts 2/macrophages 2 cluster (C9) had the greatest propensity for forming spatial niches by positioning close to each other, suggesting that endothelial immune niches are present among fibroblasts. We also identified that the TME fibroblast 1 (C4) and fibroblast 1/CD8⁺ cytotoxic T-cell (C5) clusters frequently were positioned with another similar cluster (C8), indicating that CD8⁺ T cells are mainly associated with the fibroblast 1 cell type (Figure S10C). Differential neighborhood analysis identified differences in heterotypic clusters between IS and IC samples (Figure 7A). Epigenetically defined macrophage 2 cells (C9) were likely to be closer to proliferating KC tumor 1 cells (C12) in IC patients than in IS patients (Figure 7A,B). Similarly, macrophage 2 cells (C7) adjacent to tumor clusters (C10/C11) were higher in number in IC patients than in IS patients (Figure 7A). The presence of macrophages with epigenetic features consistent with inflammatory, proliferative, and interferon-responsive TAM states, including enrichment near complement (C1QA/C1QC) and antigen-processing (CTSS) loci,, as well as increased gene activity of MHC class II loci (HLA-DQA1, HLA-DQB1, HLA-DRB1), in proximity to tumor cells was more frequently observed in IC tumors (Figure S10D).

Figure 7. Differential neighborhood relationships and TF motif enrichment are observed in immunocompetent and immunosuppressed TMEs.

(A) Heat map of the pairwise neighborhood relationships between spatial clusters in immunocompetent (IC) and immunosuppressed (IS) NMSC patients as quantified using the heterotypic score. (B) Neighborhood relationships are based on spatial data from IC and IS patients. Significant cluster interactions that differed in the IC and IS groups are highlighted by a red asterisk. (C) Volcano plot of differentially enriched transcription factor (TF) motifs in IS and IC samples. FDR, false discovery rate; FC, fold change. (D) Box plot of CD68 gene activity scores across clusters. A high CD68 gene activity score is shown in tumor cluster 12. GAS, gene activity score. (E) Volcano plot of differentially enriched TF motifs in IS and IC NMSC samples for cluster 12. (F) Spatial mapping of ATF3 TF motif deviation scores on representative IC and IS NMSC samples with tumor regions have been confirmed by high gene activity scores for KRT14. (G) Violin plot of the gene activity scores for IC and IS NMSC samples Spatial mapping of the gene activity scores for ATF3 on representative IC and IS samples. NS, not-significant.

To gain deeper insight into the key regulatory factors that determine the epigenetic state of immune cells and how they differ based on the immune status of NMSC patients, we examined enrichment in transcription factor–binding motifs at open chromatin regions IC and IS samples and identified distinct regulatory programs. Cluster 12, which harbors CD68⁺ TAMs along with tumor cells, had enriched motifs of transcription factors (TFs) including MAF, a key driver of immunosuppressive functions of TAMs (Figure S11A).^35,36 In IC samples, epigenetically-defined CD68⁺ TAMs (C12) were in close proximity to fibroblasts and other macrophages (C9) (Figure 7A,B). Contrary to this finding, IS samples had a higher abundance of clusters with epigenetically defined fibroblast 1 (C4 and C8) in close proximity to CD8⁺ cytotoxic T cell clusters (C5), CD8⁺ exhausted T cell clusters (C2) or cancer-initiating cell clusters (C10) (Figure 7A,B). Cluster 8 had enrichment of TCF7, which is a transcription factor regulating T cell development and differentiation, whereas C4 and C5 had enrichment of NANOG and E2F (Figure S11B). Cluster 1 had enrichment of CUX1, which mediates the inflammatory response in stromal fibroblasts; cluster 7 had enrichment of RUNX1 motifs, which regulates cancer-associated fibroblast functions; and cluster 9 had enrichment of SOX2, which promotes cancer-associated fibroblast–mediated tumor growth (Figure S11B).^37–40 These data suggested that macrophages with an active MAF regulatory program may underlie CD68⁺ TAMs’ immunomodulatory function in IC patients, whereas an epigenetically defined inflammatory fibroblast population with high CUX1 activity may contribute to poor immune function in IS patients by spatially regulating TCF7-expressing CD8⁺ T cells.

Comparing the TF motifs enriched in pseudobulk data for IS and IC NMSC patients demonstrated differential enrichment of motifs for ETS, HNF4, SPI1, and ETV4 in IS patients and for GRHL1, TP53, TP63, POU2F2, and FOXF2 in IC patients (Figure 7C). The spatial distribution of these top motifs identified that GRHL1-enriched motifs were associated with tumor clusters and that HNF4-enriched motifs were associated with the TME (Figure S12A-D). Interestingly, C12 tumor cluster had the highest abundance of CD68^high macrophages (Figure 7D). In this cluster, we found that IC, but not IS patients had ATF3 motifs that were highly enriched at the open chromatin regions (Figure 7E). ATF3 forms a hub of cellular adaptive-response networks and regulates the expression of proinflammatory genes.⁴¹ Spatial mapping confirmed high ATF3 motif enrichment in areas associated with tumor-rich regions (Figure 7F). Of note, the global GAS for ATF3 was higher in IS patients than in IC patients (Figure 7G), suggesting that the CD68⁺ TAMs in IS patients have increased ATF3 signaling; this might shed light on the mechanistic anti-inflammatory phenotype of CD68⁺ TAMs in IS patients.

Overall, the spatial-ATAC-seq analysis identified regulatory programs that underlie specific spatial niches in the TME. These data corroborate our multiplex immunofluorescence and transcriptomic (spatial and single cell) results demonstrating spatial alterations of CD68⁺ TAMs in the TME of IC and IS NMSC patients.

Multi-Cohort and multi-modal spatial and single-cell analyses reveal subtle remodeling of the tumor immune microenvironment under systemic immunosuppression

To further examine the conclusions of spatial analysis derived from different techniques, we used a second cohort (cohort 2, Table S1) of 12 samples (IS=9, IC= 3; BCC = 4, cSCC = 8) and performed histopathological, Xenium-based spatial transcriptomic and spatial ATAC analysis in serial consecutive sections. We first assessed the technical concordance between spatial ATAC fragment counts; this analysis showed strong concordance between IC and IS samples (R² > 0.9; Figure S13A). Unsupervised clustering identified 24 spatial epigenetic (spEP) clusters that are consistent with Xenium derived cell labels and histopathology (Figure S13B-G).

Within this cohort, Xenium data showed lower representation of intratumoral antigen-presenting cell subsets (ECM-deg TAMs, IFN-TAMs) and exhausted CD8⁺ T cells in IS tumors (Figure S14A). Spatial multiomic analysis also identified differences in APC–T-cell interaction patterns (LGALS3–LAG3, LGALS1–CD69) and ligand–receptor connectivity (Figure S14B-C, S15). Statistical comparisons for cohort 2 are provided in Supplementary Table S9.

Spatial ATAC analysis of cohort 1 samples identified an epigenetically defined inflammatory fibroblast population in IS patients that associated with CD8⁺ T cells. Extending this analysis to cohort 2, which contained six fibroblast-rich clusters (C4, C10, C12, C13, C14, C16), we identified an inflammatory fibroblast cluster (C14) enriched for NFKB1 transcription factor motifs (Figure S16A-B). Analysis of the corresponding serial section used for Xenium spatial transcriptomics confirmed the presence of inflammatory fibroblast markers (VIM, PDGFRA, THY1, CXCL12, CCL19, IL1B) in cells from the matching tissue region (Figure S16C-E). We also identified niches with dendritic cells (DC) abundance greater than 30%, which included 7 IC and 20 IS samples. Analysis of cell type composition within these niches revealed a lower percentage of CD8⁺ T cells in IS compared to IC samples (p = 0.125) (Figure S17A-B). This reduction was accompanied by fewer DC–CD8⁺ T cell interactions in IS patients, a trend that could have functional implications for antitumor immunity. To further assess DC–T cell interactions, we examined spatial ATAC data from cohort 2. Classical markers of cDC1 and cDC2 were enriched in mixed immune clusters C3 and C9, and we identified the samples enriched with these clusters (Figure S17B). Spatial mapping of DC and T cell markers nonetheless demonstrated their close proximity in representative samples, supporting their potential interactions within the tumor microenvironment (Figure S17C and S17D).

To further validate our spatial ATAC findings, we performed scMultiome (scATAC + scRNA) profiling on eight samples, including four overlapping with spatial ATAC. Integrated analysis identified 21 cell type–specific clusters, with macrophage (C20), keratinocytic (C21), and epithelial (C19) populations consistently shared across single-cell and spatial datasets (Figure S18A-D). The strong concordance between single-cell and spatial modalities confirmed the reproducibility and validity of the spatial ATAC annotations, demonstrating that key cellular identities and relationships were preserved across platforms (Figure S18E).

The Effect of Clinicopathologic Variables on the Tumor Immune Microenvironment in Immunosuppressed Patients

We next aimed to assess whether systemic immunosuppression imposes a convergent suppressive phenotype across histologies or preserves intrinsic immune distinctions. Towards this aim, we performed a stratified analysis of our scRNA-Seq data. This analysis demonstrated that immune-cell composition and clustering patterns were consistent between cSCC and BCC in IC and IS tumors (Figure S19). Myeloid subclustering revealed similar proportions of RTMs, LA-TAMs, IFN-TAMs, Inflam-TAMs, ECM-Deg TAMs, and Prolif-TAMs across both histologies (Figure S20). Likewise, T/NK-cell subclustering showed conserved distributions of naïve, memory, effector, and regulatory subsets, indicating that tumor histology did not independently alter immune architecture under immunosuppressive conditions (Figure S21). Interestingly, TCR diversity analyses revealed significantly reduced clonal diversity in IS cSCC samples, across both CD8⁺ and CD4⁺ subsets, but not in BCC (Figure S22).

We also sought to assess whether the etiology of immunosuppression drive differences in the immune microenvironment. We compared tumors from organ transplant recipients (OTRs) and patients with hematologic malignancies (HM). In the mIF cohort, stratified analysis revealed that CD4⁺ T-cell infiltration, both intratumoral and stromal, was significantly lower in HM patients compared with OTRs (p = 0.006 and p = 0.034, respectively; Figure S23). Other immune subsets, including CD8⁺ T cells, B cells, and CD68⁺ PD-L1⁺ macrophages, remained comparably altered across both groups. Consistent with these results, scRNA-seq analysis demonstrated reduced overall CD4⁺ T-cell abundance in HM tumors relative to OTRs, whereas macrophage and dendritic cell proportions were similar across IS etiologies (Figure S24, Table S10).These findings are consistent with the systemic lymphodepletion observed in hematologic-malignancy, whereas myeloid compartment are more agnostic to the etiology of IS with preservation of the T cell compartment in transplant-related immunosuppression.^42,43

Discussion

Our comprehensive multi-omic analysis of the TME in NMSC patients provides needed insight into the immune landscape of IC and IS patients. Our large cohort of over 200 NMSC samples, including a population of IS individuals from The University of Texas MD Anderson Cancer Center and Cleveland Clinic, offers a robust foundation for understanding the complex interplay between tumor cells and the immune system in NMSC patients. Our findings reveal that the primary difference between IC and IS NMSC patients lies in the innate immune compartment. Differences in intratumoral CD68+ macrophage density within both the tumoral and stromal compartments of IS patients, coupled with the downregulation of key genes involved in antigen presentation pathways, indicate a disrupted innate immune activation program that may compromise coordination with adaptive responses. Rather than a uniform depletion of immune cells, patient-level analyses reveal that immunosuppression in NMSC manifests primarily through altered spatial organization and functional reprogramming rather than global immune cell loss.

These macrophage alterations likely influence downstream adaptive immunity, as altered antigen presenting cell-T cell crosstalk can limit the breadth and quality of tumor-specific immune responses. Consistent with recent works showing that TAMs can induce CD8⁺ T-cell dysfunction and demonstrating that classically activated macrophages are required for optimal immunotherapy efficacy,^44,45 our findings emphasize the multifaceted roles of macrophages in regulating antitumor immunity. Together with altered cDC2 function, these results suggest that impaired coordination between macrophages and dendritic cells may underlie the disrupted antigen presentation network observed in IS NMSC tumors. In line with that, our analysis revealed reduced TCR diversity in IS patients and increased clonal sharing within effector-memory populations. However, the functional significance of these repertoire constraints remains unclear: both IC and IS patients developed tumors despite these differences, indicating that altered TCR diversity alone does not prevent tumor formation. Whether these constraints contribute to worse clinical outcomes in IS patients or represent a consequence of chronic immunosuppression will require functional validation studies examining T-cell cytotoxicity, proliferative capacity, and response to immunotherapy. While both IC and IS patients developed clinically evident tumors, indicating that neither mounted a fully preventive antitumor response, IS patients experienced significantly worse outcomes, including higher recurrence and reduced survival. These findings suggest that the functional programs rather than merely a depletion of tumor antigens-specific T-cell responses.

These functional alterations may arise from several non–mutually exclusive mechanisms: chronic antigen stimulation driving oligoclonal expansion, impaired T-cell priming due to innate immune dysfunction, or homeostatic proliferation compensating for reduced thymic output in transplant recipients. Together with the transcriptional evidence of cytotoxic dysfunction and regulatory skewing in IS tumors, these repertoire patterns suggest a coordinated shift toward immunosuppressive programming rather than a loss of immune competence. Determining whether such repertoire constraints actively contribute to failed tumor immune surveillance or instead reflect secondary consequences of broader immune dysregulation will require functional validation. Future studies assessing T-cell responses to tumor-associated antigens, proliferative capacity upon stimulation, and response to immune checkpoint blockade across distinct repertoire patterns will clarify the biological significance of these clonal differences.

Neighborhood analysis demonstrated that macrophages and T cells had fewer interactions with tumor cells in IS patients, which may be driven by epigenetic regulation of transcription factors that regulate inflammation and innate immune response. These findings are supported by a recent study that applied spatial mapping to define macrophage populations in breast and colon cancer. The authors demonstrated that spatial localization of TAMs within discrete multicellular niches was associated with distinct clinical outcomes.⁴⁶ Overall, our results provide a mechanistic explanation for the poor clinical outcomes often observed in IS patients with NMSC.

Despite major advances in immunotherapy for cutaneous malignancies, the immune mechanisms underlying poor outcomes in IS patients with NMSC remain poorly defined. Our data suggest that these patients would benefit from alternative therapeutic approaches that specifically target and enhance innate immunity. For instance, strategies aimed at activating the innate immune system locally within the tumor microenvironment could potentially overcome systemic immunosuppression while minimizing toxicity. Several promising avenues for local immune activation warrant further investigation based on our findings. For example, Toll-like receptor agonists can be used to stimulate antigen-presenting cells and promote a proinflammatory TME.⁴⁷ Furthermore, oncolytic viruses such as talimogene laherparepvec (T-VEC) not only directly lyse tumor cells but also induce local immune responses that can trigger systemic antitumor immunity.^48,49 Additionally, cancer vaccines designed to enhance the presentation of tumor-specific antigens may compensate for the deficient antigen presentation machinery in immunosuppressed NMSC patients.^50–52 By focusing on local activation of innate immunity, the approaches described above may induce robust antitumor responses while potentially avoiding the systemic toxic effects associated with immune checkpoint inhibitors. This is particularly crucial for organ transplant recipients and other IS populations for whom systemic immune activation could lead to severe adverse events or graft rejection.

Our study provides the critical biological premise that has been missing to justify clinical investigation of the localized immunotherapeutic strategies described herein in IS NMSC patients. The detailed characterization of the immune deficits in these patients that we report offers a rational basis for designing targeted clinical trials. Such trials could explore the efficacy and safety of innate immune activators, potentially opening new therapeutic avenues for a patient population with limited treatment options. Moreover, the spatial and epigenetic analyses presented herein offer insight into the regulatory mechanisms governing immune cell behavior within the TME. These findings could guide the development of more precise and effective immunotherapeutic strategies tailored to the specific immune landscapes of individual patients.

Strengths and limitations

Clinical outcome analyses revealed that IS patients experienced worse outcomes associated with spatial distribution of CD68⁺PDL1⁺ cells compared with IC patients, prompting us to consider how immune architecture and repertoire features may influence disease progression. The TCR repertoire differences we observe, i.e. reduced diversity, increased clonal sharing, and altered clonotype distribution across T-cell subsets, represent descriptive features of the adaptive immune landscape in IS patients. However, the significantly worse clinical outcomes in IS patients suggest that the quality of the immune response, reflected in TCR diversity constraints and spatial alterations in the myeloid cells, may influence tumor behavior and treatment responsiveness rather than initial tumor development and requires additional functional assessment.

Our study has several important strengths. The use of multi-omic approaches, including scRNA-seq, TCR sequencing, spatial transcriptomics, multiplex immunofluorescence, and spatial epigenomics, in a large cohort enriched for patients with a secondary malignancy or significant comorbidity (e.g., organ transplant recipients) provides a comprehensive view of the TME in this unique population. This enabled us to identify significant differences between IC and IS patients that may not have been detectable in smaller or less deeply profiled studies. The integration of single-cell and spatial data types yields a more nuanced understanding of the immune landscape in NMSC, while the parallel inclusion of IC and IS patients provides comparative insights that are crucial for developing targeted therapies. The identification of specific regulatory mechanisms and candidate therapeutic targets offers a promising direction for future hypothesis-driven research and clinical translation. Furthermore, the harmonized analytic pipeline, publicly shared datasets, and clearly defined quantitative metrics create a platform that can be applied and refined in multi-center, prospective studies.

These strengths notwithstanding, several limitations should be considered when interpreting the findings, each of which also highlights opportunities for future work. First, although our multimodal profiling provides complementary insight into immune composition and spatial organization, these platforms were not designed for direct quantitative benchmarking. Specifically, scRNA-seq of isolated CD45+ cells provides relative immune-cell composition among captured cells but does not quantify absolute immune-cell density or spatial compartmentalization. In contrast, multiplex immunofluorescence enables compartment-specific quantification of immune-cell densities within intratumoral and stromal regions. Accordingly, while scRNA-seq revealed similar relative proportions of macrophage populations between IS and IC tumors, mIF identified differences in compartmental macrophage density. These findings highlight the importance of interpreting each platform within its technical scope, and future studies specifically designed for cross-platform benchmarking will be valuable. Second, immunosuppressed patients constitute a biologically heterogeneous group with diverse etiologies, treatments, and disease histories, which can blur signals driven by any single cause of immunosuppression; however, this heterogeneity also allowed us to identify macrophage and T-cell programs that are recurrent across multiple forms of immunosuppression, suggesting common pathways that can now be dissected in more narrowly defined clinical subsets. Third, analyses rely on tumor specimens obtained at single timepoints, limiting direct assessment of temporal evolution in spatial organization and clonal dynamics; nonetheless, static snapshots across many patients at different clinical stages generate testable hypotheses about trajectories of macrophage–T-cell interactions that can be addressed in future longitudinal and interventional studies. Fourth, although links between macrophage programs, spatial organization, and T-cell clonal diversity are inferred from integrative computational analyses and ligand–receptor modeling rather than direct perturbation, this multi-layered framework prioritizes specific cell states and interactions for mechanistic validation. In addition, the classification of TAM subsets is based on transcriptional signatures derived from single-cell RNA-seq data, and no direct functional assays were performed to validate antigen presentation capacity or effector function in these subsets. Thus, functional properties are inferred from molecular signatures rather than experimentally confirmed, providing a roadmap for future mechanistic and translational studies. Finally, as a two-center study with partially retrospective components and validation in a single external cohort, the generalizability of findings to broader NMSC populations remains to be established; prospective, multi-center replication using the shared analytic infrastructure described here will be an important next step.

In conclusion, this study not only advances our understanding of the immune dynamics in NMSC patients but also challenges the current paradigm of excluding IS patients from immunotherapy trials. By highlighting the central role of innate immunity and antigen presentation in shaping antitumor responses, our work lays the foundation for a new era of personalized immunotherapy for NMSC, potentially benefiting a broader spectrum of patients, including those previously considered poor candidates for such treatments.

RESOURCE AVAILABILITY

Lead contact:

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Moran Amit MD PhD.

Materials availability:

This study did not generate new unique reagents.

Data and code availability:

All data generated in this study will be made publicly available prior to publication in accordance with Cell Resource policies:

• Sequencing data: scRNA-seq, scTCR-seq,scMultiome, spatial transcriptomics (Xenium), and spatial ATAC-seq datasets have been deposited in the Gene Expression Omnibus (GEO) under accession numbers GSE316922 (scRNA-seq), GSE318638 (scTCR-seq), GSE306129 (spatial ATAC), GSE306130 (Xenium), and GSE308361 (scMultiome).

• Protein-based assays: Raw and processed multiplex immunofluorescence (mIF) datasets are available through https://zenodo.org/records/20561071?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjJiZjdkMjBmLTE4MGUtNDMwZS05MTgxLWQ4OTBmMWE4YTJmMCIsImRhdGEiOnt9LCJyYW5kb20iOiI3MTVjNDIwZWE5OTdmMmExYzE3MTA5NTgzZGIyZTAyNCJ9.zoRh0-Qc6WVfgBF7bWclgFjP2yu6D1exs4NM8zuIdS7WfaYalwiwHl0Suy7cHz5nmV4DSZFHBM9SquNLVBU5Vw

• Code availability: All custom analysis code used for spatial and cross-platform integration is available on GitHub (https://gitlab.com/railab/spatial-biology-reveals-macrophage-dysfunction-in-immunosuppressed-non-melanoma-skin-cancer) and archived on Zenodo. The spatial ATAC-seq analysis pipeline is accessible at: https://github.com/atlasxomics/ATX_epigenomics/tree/218fdc0a961bbad738d7b92cf8f1e71832735c89.

• Xenium maps and H&E images are available via figshare.com at: https://figshare.com/account/articles/30296713

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Human participants and clinical cohorts:

This study was approved by the institutional review boards at MD Anderson and the Cleveland Clinic. the University of Pennsylvania, and the Moffitt Cancer Center. A waiver of informed consent was granted owing to its retrospective nature.

Using the REDCap-based cSCC registry in the department of Head and Neck Surgery at MD Anderson, all patients 18 years old or older with a histopathological diagnosis of cSCC and BCC who had undergone primary surgery at MD Anderson from January 1, 1998 to August 2025 were identified. We included all systemic treatment-naïve patients who had available data, a follow-up duration of at least 6 months, and available tissue samples for analysis. We excluded patients who received any treatment other than biopsy or local excision before beginning their care at MD Anderson, patients with a follow-up duration shorter than 1 month, and patients for whom tissue samples were not available. scRNA-seq analyses, we included consecutive patients who underwent treatment of NMSC at the MD Anderson Mohs clinic for non-melanoma skin cancer, who agreed to participate, and who provided informed consent (MD Anderson Institutional Review Board, Lab08–0848).

We retrieved demographic, clinicopathological, and treatment data from the electronic medical record. We categorized the patients into 2 cohorts according to their immunological status: IS or IC. The IS cohort included patients who underwent solid organ or stem cell transplantation, patients with active hematopoietic malignant disease or other myeloproliferative disorders, patients with autoimmune disease requiring immunosuppressive therapy or chronic corticosteroid use, patients living with HIV/AIDS, and patients with insulin-dependent diabetes mellitus.⁶ Insulin-dependent diabetes mellitus represents a myriad of well-documented immune deficiencies, such as impairment of cytokine production, leukocyte recruitment inhibition, and macrophage dysfunction.^53–57 For the samples used in the multiplex immunofluorescence and spatial transcriptomic analyses, the IC cohort comprised patients with NMSC whose age and T-stage distributions matched those of the immunosuppressed cohort. Staging was performed according to the guidelines of the American Joint Committee on Cancer (8th edition).⁵⁸ Similarly, we identified additional patients who underwent solid organ transplants at the University of Pennsylvania from January 1, 2015 to December 31, 2020 using the inclusion and exclusion criteria described above.

Cleveland Clinic validation cohort

All studies were approved by the institutional review board at Cleveland Clinic. 54 cSCC tissues were obtained from IC and IS patients and reviewed by a trained pathologist. Patients demographics and immunosuppression status was stratified similar to the main cohort as described above.

METHOD DETAILS

Pathological analyses

H&E-stained tissue sections were reviewed by two dermatopathologists (PN and KT). For each tissue sample, two representative sections that included both epithelial and stromal components were selected for the analysis by the dermatopathologist. Selected samples were then analyzed via immunofluorescent staining, Xenium HD, and spatial-ATAC-seq to profile cutaneous lesions and immune infiltration in the tumor and stroma (spatial analysis).

Immunofluorescence staining

Immunofluorescence staining of samples obtained from IC and IS patients with cSCC to identify immune cell populations in the TME was performed using the Opal 7-color multiplex staining kit (PerkinElmer, Waltham, MA). All samples were stained together. Sections of tissue were deparaffinized in xylene and hydrated in ethanol prior to antigen retrieval with EDTA buffer at 98°C for 20 min. Each sample was incubated with 2 panels of primary antibodies (Bethyl Laboratories, Montgomery, TX). Panel 1 included primary antibodies against cytokeratin (A500–019A) 1:250, CD68 (A500– 018A) 1:200, CD8 (A700–044) 1:300, CD4 (A700–015) 1:400, CD20 (A500–017A) 1:500, and PD-L1 (A700–020) 1:250. Panel 2 included primary antibodies against cytokeratin (A500–019A) 1:250, CD8 (A700–044) 1:300, LAG3 (A700–027) 1:200, CD4 (A700–015) 1:400, and FOXP3 (A700–034) 1:250 .After each cycle of antigen retrieval, the slides were incubated with peroxidase and protein blocking solutions (Dako, Carpinteria, CA). A secondary antibody conjugated with horseradish peroxidase and the EnVision+ System (Dako) were used for signal detection and amplification. The secondary antibody was associated with the tyramide-conjugated fluorescent dyes Opal 480, Opal 520, Opal 570, Opal 620, Opal 690, and Opal 780. After each staining cycle, the sections were heat-treated to disengage the antigen-antibody complex while the covalently bound, fluorescence-labeled tyramide remained on the tissue. Counterstaining of the cell nuclei was performed using DAPI (PerkinElmer). Coverslips were placed onto each stained slide using fluorescence mounting medium (Dako).

Multiplex immunofluorescence imaging and analysis

The slides were prepared as described above were scanned using the Vectra Polaris multispectral scanning system (PerkinElmer) at 4x to 40x magnification (Figure 2A). The scanned images were annotated using Phenochart software (PerkinElmer, Hopkinton, MA) and analyzed using inForm version 2.4.8 software (Akoya Biosciences, Marlborough, MA). After excluding necrotic areas and tissue artifacts (e.g., tissue tears or folds, immunofluorescence artifacts), tissue segmentation was performed, followed by cell segmentation and phenotyping. For tissue segmentation, tumor (cytokeratin-positive) and stromal (cytokeratin-negative) areas were annotated using the inForm software. Cell segmentation was determined based on the presence of the nucleus (DAPI⁺) using cytoplasmic and membranous markers. Lastly, threshold-based phenotyping was performed by automatically classifying cell phenotypes using machine-learning algorithms with the inForm software. Phenotyping was completed after training without a priori determination of the thresholds. Cartographic coordinates of each cell were obtained to determine cells’ location (i.e., within the tumor or stroma).

Immunofluorescence Staining (Cleveland Clinic Cohort)

Multiplexed immunofluorescence staining to identify immune cell populations in the TME was performed using tyramide signal amplification. All samples were stained together. Sections of tissue were deparaffinized in xylene and hydrated in ethanol prior to antigen retrieval with Citrate buffer (BioSB, BSB-0021) at 115°C for 15 minutes. Each sample was incubated with 2 panels of primary antibodies (Bethyl Laboratories, Montgomery, TX | Abcam, Cambridge, UK | Cell Signaling Technology, Danvers, MA). The myeloid panel included primary antibodies against CD40 (Bethyl, A700–056), CD11c (Abcam, ab52632), HLA-DR (Bethyl, A500–022A), CD68 (Abcam, ab213363), CD11b (Abcam, ab52478), and pan-cytokeratin (Cell Signaling Technology, 67306S). After each cycle of antigen retrieval, the slides were incubated with 5% BSA for protein blocking. Secondary antibodies conjugated with poly-horseradish peroxidase targeting rabbit (ThermoFisher, 32260) and mouse (ThermoFisher PI32230) were used for signal detection and amplification, along with tyramide-conjugated fluorescent dyes (Biotium, Fremont, CA) CF430 (96053), CF488A (92171), CF550R (96077), CF583R (96085), CF680R (92196), and CF754 (96090) in the presence of manual amplification buffer (Akoya Biosciences, Marlborough, MA). After each staining cycle, the sections were heat-treated in antigen retrieval buffer, either pH 6.0 Citrate or pH 9.0 EDTA (BioSB, BSB-0031), to disengage the antigen-antibody complex while the covalently bound, fluorescence-labeled tyramide remained on the tissue. After all rounds of staining were completed, tissues were treated with TrueBlack lipofuscin autofluorescence quencher (Cell Signaling Technology). Counterstaining of the cell nuclei and coverslipping was performed using VectaShield Antifade Mounting Media with DAPI (Vector Laboratories, Newark, CA). The slides were scanned using the PhenoImager Fusion (Akoya Biosciences) at 20x magnification. Single-stained slides were used to generate a spectral library using inForm 3.1 (Akoya Biosciences) for spectral unmixing. Exported images were analyzed using MATLAB R2022b (Mathworks). Bulk tissue was detected based on a maximum intensity projection across all channels. Nuclei were segmented using a bandpass filtered followed by difference of gaussians, then segmented by Otsu’s method. Local maxima indicative of nuclei centers were identified and used for a seeded watershed filter to separate overlapping nuclei. Nuclei were further used for a seeded watershed filter to segment individual cells, with a maximal dilation restricted to the nuclear diameter. Mean fluorescence intensity of individual cells was used to determine marker positivity.

scRNA-seq preprocessing and analysis

After samples from patients with NMSC were obtained from biopsy or surgical resection, specimens were immediately dissociated by using Human whole skin dissociation kit and Red Blood Cell Lysis Solution (10x) (Miltenyi Biotec, Germany). Dissociated cells were stained with APC conjugated CD45 (H130) mouse mAb (Cell Signaling Technology, Danvers, MA). Then CD45⁺ cells were sorted using CytoFLEX SRT Cell Sorter (Beckman Coulter, Indianapolis, IN) to obtain a CD45⁺ enriched population for single-cell sequencing. Prior to microfluidic chip loading, CD45⁺ enriched single-cell suspensions were centrifuged at 500 rpm for 5 min and washed twice with phosphate-buffered saline (21–040-CM; Corning, NY) and 0.04% bovine serum albumin (AM2616; Thermo Fisher Scientific, Waltham, MA). Cell and viability counts for each sample were taken using a Countess 3 FL Automated Cell Counter (AMQAF2000; Thermo Fisher Scientific). Single-cell emulsions and cDNA were generated using the Chromium Next GEM Single Cell 5' HT v2 reagent kit (user guide CG000423 Rev C; 10x Genomics, Pleasanton, CA). Gene expression and V(D)J library generation were automated using the Chromium Connect user guide (CG000180 Rev D; 10x Genomics) starting from cDNA. Single-cell libraries were pooled and sequenced at 150PE using a NovaSeq X Plus (Illumina, San Diego, CA). Cell Ranger software (v7.1.0; 10x Genomics) was used through 10x Genomics cloud services to process sequencing data and generate gene expression count matrices, TCR contigs, and V(D)J annotations.

Raw data output from Cell Ranger for 36 samples of tumor tissue obtained from immunocompetent and immunosuppressed patients with NMSC was imported into R and analyzed using Seurat software (version 5.0.1).⁵⁹ Genes that were detected in fewer than 3 cells and cells having up to 500 features and at least 10% mitochondrial reads were excluded from the downstream analysis. This filtration was aimed at mitigating the inclusion of dying cells or those with poor-quality RNA. Furthermore, potential doublets were identified and removed using the DoubletFinder package (version 2.0.3).⁶⁰ DoubletFinder first simulates artificial doublets and mixes them with existing scRNA-seq data. Next, it distinguishes real doublets from singlets by identifying cells with high percentages of artificial neighbors in gene expression spaces.⁶⁰ The Seurat function SCTransform was used to normalize the data and identify the top 3,000 most highly variable features. Data integration of samples was performed using the Harmony to correct for batch effects and align cell populations across data sets.

For analysis of the integrated data comprising all 36 samples described above, we applied dimensionality reduction using principal component analysis to identify the most significant principal components followed by searching for the nearest neighbors of data points and clustering of cells with different resolution parameters to discern distinct cell populations. Further dimensional reduction was performed using UMAP. The clusters of cells were initially annotated using CellTypist, a machine learning tool for rapid and precise cell type annotation.⁶¹ For each of the main clusters identified, such as T/NK cells, Macrophages/DCs, and B/plasma cells, we subset the cells out and re-clustered them for detailed cell type annotation using cluster-specific differentially expressed genes (DEGs) and known marker genes. Specifically, the Seurat FindAllMarkers function was used to identify DEGs for each cell cluster. Cells were clustered into 25 transcriptional groups and consolidated into 10 major immune cell types through a single standardized pipeline. In addition, a curated set of canonical T-cell markers was used to determine different T-cell types.^62,63

For macrophage subclustering, cells clusters from tumor samples that were annotated as macrophages were extracted from the integrated data set of all 36 samples for re-clustering. Cell type annotation was performed using curated marker genes and previously established differentially expressed genes.^27,28 Differential gene expression and Ingenuity Pathway Analysis (IPA) was performed for all six macrophage subtypes (RTM, Inflam-TAM, IFN-TAM, LA-TAM, ECM-Deg TAM, and Prolif-TAM) comparing IS vs IC groups.

T-cell clustering and TCR analysis

T-cell clustering and TCR repertoire analysis were conducted as previously outlined.⁶² Clusters of T and NK cells were identified based on the expression of established marker genes, including CD3D, CD3E, CD3G, NCAM1, CD16, and KIR. The TCR gene expression as well as the TCR data were used as references for T-cell identification. Next, T and NK cells were selected and further subclustered to better reflect different cell populations. We repeated the Seurat integration steps as described above. To prevent bias of TCR gene expression, TCR genes were excluded from the highly variable genes. The subsequent dimension reduction and clustering steps were similar to those used for the scRNA-seq analysis. CD4⁺ and CD8⁺ T cells were further distinguished by the expression of the main markers CD4, CD8A, and CD8B. Again, different cell types for each of the CD4⁺ and CD8⁺ T cells were identified using differentially expressed genes as well as canonical marker genes from previous studies.^62–64

Assembly of TCR V(D)J sequences and the identification of paired clonotypes were carried out using Cell Ranger (v7.1.0). The scRepertoire R package (version 1.12.0) was employed to create images depicting basic T-cell repertoire data and to integrate single-cell TCR and scRNA-seq data into a single Seurat object. The getCirclize function was used to extract unique and shared clonotype information from this object, which were visualized using Circos/Chord plots. Diversity in the TCR repertoire was analyzed using the clonalDiversity function in the scRepertoire package to compute Shannon, Inverse Simpson, Chao, and ACE indices. Clonotype sharing between cell clusters were evaluated using the Morisita Index of Dispersion, which quantifies data set similarity with values ranging from 0 (no overlap) to 1 (complete similarity of clonotypes).

Xenium panel design

A custom Xenium human skin probe panel targeting 350 genes, including those related to skin tissue, epithelial, melanocyte, immune, fibroblast, neuronal, neural crest, mesenchymal, and stem cell phenotypes and negative controls for cell type identification, was designed (Table S6).

Xenium sample preparation

5-μm formalin-fixed, paraffin-embedded normal and tumor tissue from patients with cSCC or premalignant lesions as part of a tissue microarray were sectioned onto a Xenium slide. After deparaffinization and permeabilization, a custom probe panel and control probes were added to the slides at a probe concentration of 10 nM, and overnight hybridization was done at 50°C. After stringent posthybridization washes to remove unhybridized probes, probes were ligated at 37°C for 2 h along with annealing of a rolling circle amplification primer. Amplification of circularized probes was carried out enzymatically for 2 h at 37°C, and copies of the gene-specific barcode for each RNA-binding event were generated. The samples were then washed and chemically quenched of background fluorescence. Next, nuclei were stained with DAPI, and sections were loaded onto a Xenium Analyzer.

Xenium Analyzer

Tissue microarray sections were incubated with reagents and fluorescent-labeled probes for detecting RNA that was automatically cycled in, imaged, and removed by the Xenium Analyzer for 15 rounds. Z-stack images spanning a 0.75-μm step size were acquired across the entire tissue thickness.

Image processing and analysis

Z-stack images taken by the Xenium Analyzer in every cycle and channel were processed and stitched on a DAPI image to create a single image representative of 1 region of interest (ROI). Using this image, we built a spatial map of the transcripts across the tissue section. Every punctum, which is measured in units of observed photons, in every cycle, image, and color to observe all potential mRNA was detected by the Xenium Onboard Analysis pipeline. The observed emitted light from localized puncta and those from different cycles from the same spatial locations were registered to one another and fitted into a Gaussian distribution. An optical signature unique for each gene was then generated after 15 Xenium cycles using fluorescent intensity in each of the 4 Xenium color channels measured in each cycle. A Phred-scaled quality value called the Q score was assigned for each decoded transcript, and transcripts with Q ≥ 20 were used in the Xenium spatial gene plots. The Xenium cell segmentation algorithm used DAPI images that were acquired and calculated in a 15-μm radius from the nucleus outward or until another cell boundary was reached. The on-instrument pipeline output included files containing the feature-cell matrix, the transcripts, and a CSV file of the cell boundaries (a differentially enriched gene list for each cluster). Further downstream analysis and data integration for all samples were performed off-instrument and visualized using Xenium Explorer.

Post-Xenium histology

Following acquisition of Xenium images, H&E staining was performed on the same tissue microarray sections that underwent Xenium analysis. A Keyence (Itasca, IL) microscope was used to obtain images of the tissue microarray sections before and after H&E staining at 20x magnification. The images were then stitched together to generate a composite bright-field image at high resolution. Xenium Explorer was used to align the post-Xenium images with the Xenium images.

Xenium spatial transcriptomic data analysis

All files were retrieved from the Xenium analyzer to generate the gene-by-cell matrix. After QC filtering, 81 cSCC samples remained, totaling 353,153 cells, which were used for the downstream analysis. The raw matrix was normalized using Seurat.⁶⁵ Principal component analysis (PCA) was conducted on all genes. An elbow plot was created using Seurat's ElbowPlot function, which helped determine the number of significant principal components (PCs). The FindNeighbors function was used to construct the shared nearest neighbor (SNN) graph based on unsupervised clustering performed with the FindClusters function. Dimensionality reduction and two-dimensional visualization of cell clusters were performed using UMAP.^66,67 Custom cell type annotation was carried out using marker genes for major cell types. Multiple rounds of clustering and sub-clustering analyses were conducted to identify major cell types. Niche analysis was performed using Seurat v5.⁶⁸ The optimal number of niches for each TMA sample was identified using silhouette scores and the 'local neighborhood' for each cell, consisting of its k.neighbors. Known ligand-receptor gene pairs were loaded from the stlearn Python module.²⁹ Spots with significant interactions between these pairs were identified. For each ligand-receptor pair and each cell type-cell type combination, instances where neighbors of a significant spot for that ligand-receptor pair linked two given cell types were counted. Significant interactions with p < 0.05 were identified from cell type information permutation. Xenium codes are available at: https://gitlab.com/railab/spatial-biology-reveals-macrophage-dysfunction-in-immunosuppressed-non-melanoma-skin-cancer

Spatial-ATAC-seq

Tissue preparation:

For spatial-ATAC-seq, fresh frozen tumor tissue from 14 immunocompetent and immunosuppressed patients with NMSC was obtained from the Institutional Tissue Bank at MD Anderson with approval from the MD Anderson Institutional Review Board and with written informed consent from patients. An experienced pathologist and primary clinical investigator rereviewed the pathology of all samples used in this study to confirm cSCC and BCC diagnosis and identify the tumoral and peritumoral areas of interest. Fresh frozen tissues were then embedded in optimal cutting temperature compound and cryosectioned into 10-μm-thick slices, affixed to Superfrost Plus slides (12–550-15; Thermo Fisher Scientific), and stored at −80°C. Samples were then shipped to AtlasXomics (New Haven, CT) in dry ice for spatial-ATAC-seq. The frozen slides were warmed at room temperature for 10 min. The tissue was then fixed with formaldehyde (0.2% for 5 min) and quenched with glycine (1.25 M for 5 min) at room temperature. After fixation, the tissue was washed twice with 1 ml of 1x Dulbecco's phosphate-buffered saline, cleaned in deionized water, and dried. The samples were then used for spatial-ATAC-seq after optimization of assay conditions and flow testing. Deterministic Barcoding in Tissue sequencing (DBiT-seq), a commercial, first-generation, 50-channel platform (Portal50, AtlasXomics), was used for spatial-ATAC-seq.

Bulk ATAC for optimization of parameters:

Ahead of spatial-ATAC-seq, optimal permeabilization and tagmentation parameters were determined by performing bulk ATAC using an AtlasXpress bulk insert and bulk chips on additional slides for each cancer type by varying the NP-40 concentrations and timing for tagmentation on fixed tissues. Bulk Barcodes A and B were used for this optimization. Tissue lysis and libraries from genomic DNA were prepared as described below in the main experiment protocol, and library traces were run using TapeStation software (Agilent, Santa Clara, CA) to view the nucleosome-free and mononucleosomal, dinucleosomal, and trinucleosomal peaks.

Flow test:

To confirm the microfluidic compatibility, we conducted a flow test using AtlasXpress bulk insert and bulk chips on fixed tissue sections. Fluorescent-tagged scrambled barcodes were passed on to the sections using a flow chip containing 50 inlets, which was placed on the section with the flow barcoding region covering the ROI. A vacuum hood was set over the outlet wells of the chip with the gauge set to 50 inches of water for 6 min, allowing the barcodes to traverse the tissue. Fluorescent images were acquired to ensure that leaking or merging of barcodes in adjacent lanes did not occur and that problematic debris did not block the microfluidic flow.

Permeabilization and tagmentation:

Each fixed tissue section was mounted on an AtlasXpress with bulk insert, and a bulk chip was placed on the slide so that the tissue ROI was centered. Next, 500 μl of permeabilization buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 0.01% Tween-20, 0.01% NP-40, 0.001% digitonin, 1% bovine serum albumin) was added to the slide, which was then incubated for 15 min at 37°C. Slides were washed with 500 μl of wash buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 1% bovine serum albumin, 0.1% Tween-20) for 5 min. Next, 100 μl of transposition mix (50 μl of 2x tagmentation buffer, 33 μl of 1x Dulbecco’s phosphate-buffered saline, 1 μl of 10% Tween-20, 1 μl of 1% digitonin, 5 μl of transposome, and 10 μl of nuclease-free water) was added, followed by incubation at 37°C for 30 min after covering the bulk insert opening. After removing the transposition mix, 500 μl of 40-mM EDTA was added for incubation at room temperature for 5 min to stop transposition. Finally, EDTA was removed and the tissue section was washed with 500 μl of 1x NEBuffer 3.1 for 5 min. The appliance was unclamped, and the slide was removed, cleaned with nuclease-free water, and dried.

Spatial barcoding:

Premade barcode A and B plates were thawed to room temperature. Ligation master mix was prepared in a reagent reservoir on ice and added to barcode A and B plates. For barcode A in situ ligation, chip A was applied on tissue slides in a clean polymerase chain reaction (PCR) hood so that the center of the flow barcoding region covered the ROI. We mounted the slide onto the AtlasXpress and tightly clamped the assembly. Using a multichannel pipette, 6 μl of Ligation master mix with barcode A was loaded to each of the 50 inlets of chip A, and the whole assembly was mounted onto a Keyence BZ-X800 microscope. A vacuum hood was set over the outlet wells of the chip with the gauge set to 50 inches of water for 6 min, allowing the barcodes to traverse the tissue. After 6 min of flow, the vacuum was turned off, and the slides were imaged under the GFP and RFP channels of a fluorescent microscope. The appliance-containing slide and chip were incubated at 37°C for 30 min. The appliance was then unscrewed, the slide/chip was quickly submerged in 1x phosphate-buffered saline, and chip A was removed. The tissue was washed with NEBuffer 3.1 for 5 min, cleaned with nuclease-free water, and dried. For barcode B in situ ligation, chip B was applied on tissue slides as described above, and 6 μl of Ligation master mix with barcode B was loaded to each of the 50 inlets of chip B. The protocol for ligation described above was used. After ligation of barcodes A and B, a 10x image of the ROI in bright-field and fluorescence was captured for analysis.

Lysis and genomic DNA purification:

A Portal lysis chip was placed on a slide so that the reservoir opened directly over the barcoded tissue ROI. Next, the assembly was clamped on the AtlasXpress. Next, 100 ml of lysis solution (50 mM Tris-HCl, pH 8.0, 1 mM EDTA, 1% SDS, 200 mM NaCl, 0.4 mg/ml proteinase K) was added to the reservoir over the ROI. The slide was taped and incubated at 40° C overnight. The lysate from the lysis reservoir was transferred into a 1.5-ml microcentrifuge tube, and DNA from the lysate was eluted in 21 μl of buffer using a Zymo DNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA).

Next-generation sequencing library preparation and sequencing:

DNA from barcoded, lysed tissue was mixed with PCR reagents (2 μl of 10-μM i5 PCR primer, 2 μl of 10-μM i7 PCR primer, and 25 μl of 2x NEBNext Master Mix), and PCR was performed using the following program: 72°C for 5 min, 98°C for 30 s, and cycling 5 times at 98°C for 10 s, 63°C for 10 s, and 72°C for 1 min. To determine additional cycles of quantitative PCR needed to reach one third of the saturated signal, 5 μl of a preamplified mixture of DNA was mixed with quantitative PCR mix (5 μl of NEBNext Master Mix, 3.25 μl of nuclease-free water, 0.75 μl of 20x EvaGreen Dye, 0.5 μl of 10-μM Universal i5 primer, and 0.5 μl of 10-μM Universal or barcoded i7 primer), and the quantitative PCR reaction was performed under the following conditions: 98°C for 30 s and then 20 cycles at 98°C for 10 s, 63°C for 10 s, and 72°C for 1 min. Finally, the remaining 45 μl of the preamplified DNA was amplified by running the required number of additional cycles of PCR (i.e., the number of cycles needed to reach one third of the saturated signal in quantitative PCR). To remove PCR primer residues, the final PCR product was purified using 1x SPRI beads (45 μl) according to the standard protocol and eluted into 20 μl of nuclease-free water. Before sequencing, a Qubit dsDNA HS Assay (Thermo Fisher Scientific) and TapeStation (Agilent) were used to quantify the concentration and size distribution of the library. Next-generation sequencing was performed using a NovaSeq 6000 sequencer (Illumina, San Diego, CA). We used the recommended sequencing parameters, which were single-index (8 bp), paired end (150 bp), and 15% PhiX, with a target number of reads per library of 200–300M.

Spatial ATAC data preprocessing:

Filtered sequences (read filtering and barcode filtering details) were transformed to the Cell Ranger ATAC format (10x Genomics), which had genome sequences in new read 1, and barcodes A and B in the new read 2. The resulting FASTQ files were aligned to the human reference genome (hg38). We then filtered the reads to remove duplicate reads and counted reads using Chromap38/Cell Ranger ATAC v.1.2. BED-like fragment files were generated for downstream analysis. The fragment files contained coordinates of sequenced DNA molecules mapped to tixel barcodes.

Data visualization:

In summary, tixels on tissue samples were identified using microscopy images with the AtlasXomics software program AtlasXBrowser to create metadata files that were compatible with the Seurat workflow for spatial data sets. After reading the fragment file into ArchR, nontissue tixels were removed based on the generated metadata. Data normalization and dimensionality reduction were performed using ArchR’s implementation of iterative latent semantic indexing (iterations = 2, resolution = 0.2, varFeatures = 25,000, dimsToUse = 1:30, and n.start = 10) followed by graph clustering and UMAP embedding (nNeighbors = 30, metric = cosine, and minDist = 0.5). Gene accessibility scores and matrices were generated with ArchR's Gene Score mode. Marker regions and genes used to define cell clusters were identified with the getMarkerFeatures and getMarkers functions in ArchR (testMethod = “wilcoxon”, cutoff = “FDR≤0.05”). To assess enrichment of binding sites for specific transcription factors at marker peaks or differential peaks, the addMotifAnnotations() and peakAnnoEnrichment() functions of ArchR were used. Gene-score imputation was implemented with addImputeWeights for data visualization. ArchR results were loaded into Seurat v.3.2.3 to map the data back to the tissue section, and tixel size was scaled using the pt.size.factor parameter in the Seurat package for enhanced clarity in data visualization. Genome browser tracks were generated in ArchR using the plotBrowserTrack function and normalized to the recommended default value (normMethod = “ReadsInTSS”), which simultaneously normalizes tracks based on sequencing depth and sample data quality. Minus strand genes were represented in blue, and plus strand genes were represented in red. The peak to gene interactions were shown as loops along with the Pearson correlation between peak accessibility and gene expression. ATAC peaks were called with macs2 using the addReproduciblePeakSet function in ArchR. To project bulk ATAC-seq data, the reads in 5-kb-tiled genomes were counted using the getCounts function in chromVAR59, subsamples in pseudosingle cells (n = 250), and projected onto spatial-ATAC UMAPs using the projectBulkATAC function in ArchR. Finally, processed data in the arrow file format used by ArchR were loaded into a custom web interface (RShiny) for easy visualization.

Integration of Xenium and spatial-ATAC-seq

Cell-type identification and pseudo-scRNA-seq profiles from Xenium were added to the spatial-ATAC-seq data through integration with Xenium reference data. The FindTransferAnchors function (Seurat v.3.2 package) was used to align tixels from spatial-ATAC-seq with cells from Xenium by comparing the spatial-ATAC-seq gene score matrix with the scRNA-seq gene expression matrix from Xenium. The GeneIntegrationMatrix function in ArchR was used to add cell identities and pseudo-scRNA-seq profiles to the spatial-ATAC-seq data. Pseudobulk group coverages based on cluster identities were generated using the addGroupCoverages function and peak calling was performed using the addReproduciblePeakSet function in ArchR. To compute per-cell motif activity, we ran chromVAR59 with addDeviationsMatrix using the cisbp motif set after a background peak set was generated using addBgdPeaks. Cell type–specific marker peaks were identified using the getMarkerFeatures (bias = c(“TSSEnrichment”, “log10(nFrags)”, and testMethod = “wilcoxon”) and getMarkers (cutOff = “FDR≤0.05 and Log2FC≥0.1”) functions. Pseudotemporal reconstruction was implemented via trajectory analysis using ArchR. We used the cell-type definitions from the label transfer from the Xenium pseudo-scRNA-seq profiles as described above. A trajectory backbone was first created in the form of an ordered vector of cell group labels. We then used the addTrajectory function to create a trajectory and added the pseudotime to Seurat spatial object to map the data back to the tissue section. Dynamics of the selected gene score along the pseudotime were plotted with the plotTrajectory function using the default values. A correlation analysis was conducted by calculating the Pearson correlation coefficient in R. P values were calculated using the cor.test function in R and were adjusted for multiple comparisons using the Benjamini-Hochberg method.

Single-cell multiome library preparation

From frozen tumor tissues, nuclei were isolated according to the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression protocol (10x Genomics). Single-cell multiome libraries were generated using the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression kit (10x Genomics) following the manufacturer’s instructions. Briefly, isolated nuclei were loaded onto Chromium microfluidic chips to generate Gel Bead-In Emulsions (GEMs), enabling simultaneous profiling of gene expression and chromatin accessibility from individual nuclei. Following GEM generation, barcoded cDNA and transposed chromatin fragments were amplified separately and used for construction of gene expression and ATAC libraries. Libraries were quantified using a TapeStation 4200 system (Agilent Technologies) and pooled for sequencing. Gene expression and ATAC libraries were sequenced on an Illumina NovaSeq 6000 platform using the manufacturer-recommended paired-end sequencing configuration. Raw base call files were demultiplexed and processed using Cell Ranger ARC (10x Genomics) with alignment to the GRCh38 human reference genome to generate gene expression count matrices and chromatin accessibility fragment files for downstream analyses.

Single-cell multiome integrated analysis

Single-cell multiome data analysis for all samples was done using the ArchR framework in R. Fragment files generated from the ATAC modality were converted into arrow files using the createArrowFiles() function, while corresponding gene expression matrices were imported from Cell Ranger using import10xFeatureMatrix(). An ArchR project was constructed from the arrow files, and only cells with matching barcodes between the ATAC and RNA datasets were retained to ensure paired multiome analysis using strict barcode matching. Quality control filtering was performed to retain high quality cells with transcription start site (TSS) enrichment scores greater than 5, more than 2500 unique ATAC fragments, and valid RNA UMI counts. Dimensionality reduction was carried out independently for chromatin accessibility and gene expression data using iterative latent semantic indexing (LSI). ATAC profiles were analyzed using the TileMatrix with fragment-depth normalization, whereas RNA profiles were analyzed using the GeneExpressionMatrix with the top 2,500 variable genes and RNA UMI-based normalization. The resulting ATAC and RNA latent spaces were integrated using addCombinedDims() to generate a multimodal low-dimensional representation. Uniform Manifold Approximation and Projection (UMAP) embeddings were generated for the ATAC, RNA, and integrated datasets, and graph-based clustering was performed on the combined latent space to identify cellular populations.

Cluster-specific marker genes were identified using the GeneScoreMatrix and the Wilcoxon rank-sum test while correcting for TSS enrichment and sequencing depth (log10(nFrags)). Significant markers were defined using a false discovery rate (FDR) ≤ 0.01 and log2 fold-change (Log2FC) ≥ 1.25 and were visualized using marker heatmaps. To characterize chromatin accessibility landscapes, pseudo-bulk coverage tracks were generated for each cluster and reproducible peaks were identified using MACS2 through ArchR’s addReproduciblePeakSet() function. A peak accessibility matrix was subsequently constructed, and differentially accessible peaks were identified using the Wilcoxon rank-sum test with the same bias correction strategy. Peaks exhibiting FDR ≤ 0.01 and Log2FC ≥ 1 were considered significant and were visualized through accessibility heatmaps. To investigate the regulatory mechanisms underlying cluster-specific chromatin states, transcription factor binding site enrichment analysis was performed using ENCODE TFBS annotations, and motif enrichment analysis was conducted using HOMER motif databases.

QUANTIFICATION AND STATISTICAL ANALYSIS

Experimental design considerations

This study was retrospective in nature and therefore no randomization was performed. Investigators were not blinded to cohort assignment during data collection and analysis. Sample size was determined by the availability of eligible patient samples meeting predefined inclusion and exclusion criteria, and no formal statistical sample size calculation was performed.

Participant demographics

Age, sex, and clinicopathological characteristics were collected from the electronic medical record and are summarized in the manuscript. Data regarding ancestry, race, ethnicity, gender identity, and socioeconomic status were not uniformly available for all patients included in this retrospective study and were therefore not analyzed. The absence of these variables may limit the generalizability of the findings across diverse patient populations.

Statistical analysis

For multiplex immunofluorescence, the cell counts obtained for each primary antibody were normalized by the analyzed tissue area and described as cell density per square millimeter. Cell density comparison between immunocompetent and immunosuppressed patient samples was adjusted for multiple testing across patient populations and tissue subcategories (i.e., tumor and stroma).Associations between categorical and continuous variables were assessed using the Kruskal-Wallis test. Relationships between continuous variables were assessed using the Pearson correlation coefficient. A p-value of less than 0.05 was considered significant. Survival was analyzed using the Kaplan–Meier method and compared using the log-rank test.

For each cell subtype, patient-level densities were compared between IC and IS groups using two-sided Wilcoxon rank-sum tests with FDR correction. Ligand–receptor (LR) co-expression was quantified across all 81 samples using the stLearn framework, with significance assessed by permutation testing of neighborhood-level co-expression (p < 0.05 considered significant).

Supplementary Material

Supplementary Tables:

Table S1. Summary of clinical samples sorted by methodology. Related to Figure 1.

Table S2. DEG for marker genes used to define the immune cell clusters in scRNA-seq. Related to Figure 2.

Table S3. Immune and tumor cell densities in immunocompetent and immunosuppressed patients. Related to Figure 2.

Table S4. Macrophage cells DEG and functional analysis in IS vs IC patients. Related to Figure 2.

Table S5. CD8⁺ T cells DEG and functional analysis in IS vs IC patients. Related to Figure 3.

Table S6. CD4⁺ T cells DEG and functional analysis in IS vs IC patients. Related to Figure 3.

Table S7.TMA cores used for Xenium analysis. Related to Figure 5.

Table S8. Xenium 350 gene panel. Related to Figure 5.

Table S9. Ligand-receptor interactions in immunocompetent and immunosuppressed patients. Related to Figure 5.

Table S10. DEG of IS patients HMvsOTR. Related to Figure 1.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-Cytokeratin Monoclonal Antibody Purified [AE1/AE3]	Bethyl Laboratories	A500-019A
Mouse anti-CD68 Recombinant Monoclonal Antibody Purified [KP-1]	Bethyl Laboratories	A500- 018A
Rabbit anti-CD8 alpha Recombinant Monoclonal Antibody [BLR044F]	Bethyl Laboratories	A700-044
Rabbit anti-CD4 Recombinant Monoclonal Antibody [BL-155-1C11]	Bethyl Laboratories	A700-015
Mouse anti-CD20 Monoclonal Antibody Purified [L26]	Bethyl Laboratories	A500-017A
Rabbit anti-PD-L1 Recombinant Monoclonal Antibody [BLR020E]	Bethyl Laboratories	A700-020
Rabbit anti-LAG3 Recombinant Monoclonal Antibody [BLR027F]	Bethyl Laboratories	A700-027
Rabbit anti-FOXP3 Recombinant Monoclonal Antibody [BLR034F]	Bethyl Laboratories	A700-034
Rabbit anti-CD40 Recombinant Monoclonal Antibody [BLR056F]	Bethyl Laboratories	A700-056
Mouse anti-HLA-DR Monoclonal Antibody Purified [LN3]	Bethyl Laboratories	A500-022A
Anti-CD11c antibody [EP1347Y] - C-terminal	Abcam	ab52632
Anti-CD68 antibody [EPR20545]	Abcam	ab213363
Anti-CD11b antibody [EP1345Y]	Abcam	ab52478
Pan-Keratin (AE1/AE3) Mouse Monoclonal Antibody	Cell Signaling Technology	67306S
Bacterial and virus strains
Biological samples
Chemicals, peptides, and recombinant proteins
Critical commercial assays
Opal 7-Color Multiplex Staining Kit	PerkinElmer/Akoya Biosciences	#NEL861001KT/#NC 1769715
Whole Skin Dissociation Kit, human	Miltenyi Biotech	130-101-540
Chromium Next GEM Single Cell 5' HT v2 Reagent Kit	10X Genomics	PN-1000356
Chromium Single Cell V(D)J Amplification Kits Human, TCR	10X Genomics	PN-1000252
Xenium Slides & Sample Prep Reagents kit	10X Genomics	PN-1000460; 1000644
Spatial ATAC	AtlasXomics	N/A
Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits	10X Genomics	PN: 1000283
Deposited data
scRNA-seq	This paper	GSE316922
scTCR-seq	This paper	GSE318638
spatial ATAC	This paper	GSE306129
Xenium	This paper	GSE306130
scMultiome	This paper	GSE308361
multiplex immunofluorescence (mIF) datasets	This paper	https://zenodo.org/records/20561071?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjJiZjdkMjBmLTE4MGUtNDMwZS05MTgxLWQ4OTBmMWE4YTJmMCIsImRhdGEiOnt9LCJyYW5kb20iOiI3MTVjNDIwZWE5OTdmMmExYzE3MTA5NTgzZGIyZTAyNCJ9.zoRh0-Qc6WVfgBF7bWclgFjP2yu6D1exs4NM8zuIdS7WfaYalwiwHl0Suy7cHz5nmV4DSZFHBM9SquNLVBU5Vw
Xenium maps and HE images	This paper	https://figshare.com/account/articles/30296713
Experimental models: Cell lines
Experimental models: Organisms/strains
Oligonucleotides
Recombinant DNA
Software and algorithms
MATLAB R2022b	Natick, Mathworks	https://www.mathworks.com.
Cell Ranger software (v7.1.0)	10X Genomics, Zheng GXY et al.	doi:10.1038/ncomms14049
Seurat (v5.0.1)	Hao, Y., et al. 65	DOI: 10.1038/s41587-023-01767-y
DoubletFinder (v2.0.3)	McGinnis et al. 60	DOI: 10.1016/j.cels.2019.03.003
Harmony	Korsunsky et al.	DOI: 10.1038/s41592-019-0619-0
CellTypist	Domínguez Conde et al. 61	DOI: 10.1126/science.abl5197
Ingenuity Pathway Analysis (IPA)	QIAGEN	RRID:SCR_008653
scRepertoire (v1.12.0)	Borcherding et al.	DOI: 10.12688/f1000research.22139.2
Xenium Onboard Analysis Pipeline	10x Genomics	https://www.10xgenomics.com/products/xenium
Xenium Explorer (v3)	10x Genomics	https://www.10xgenomics.com/products/xenium
UMAP	McInnes et al. 66; Becht, et al. 67	https://doi.org/10.1038/NBT.4314
stLearn	Pham et al. 29	https://doi.org/10.1101/2020.05.31.125658
Chromap	Zhang et al.	DOI: 10.1038/s41467-021-26865-w
Cell Ranger ATAC (v1.2)	10x Genomics	RRID:SCR_021160
AtlasXBrowser	AtlasXomics	https://www.atlasxomics.com
ArchR	Granja et al.	DOI: 10.1038/s41588-021-00790-6
MACS2	Zhang et al.	DOI: 10.1186/gb-2008-9-9-r137
chromVAR	Schep et al.	DOI: 10.1038/nmeth.4401
RShiny	Posit	RRID:SCR_001626
Other
Vectra Polaris multispectral scanning system	PerkinElmer	CAT NO #CLS143455
PhenoImager Fusion	Akoya Biosciences	CAT NO # 240204
Xenium Analyzer	10X Genomics	CAT NO# 1000481
ChromiumX	10X Genomics	CAT NO# 1000331
DBiT-seq platform for high-resolution spatial epigenomics	AtlasXomics	https://www.atlasxomics.com/

Highlights.

• Immune alterations in immunosuppressed patients are associated with poor survival.

• Immunosuppression is associated with restricted T-cell repertoire diversity.

• Differential macrophage subsets and spatial distribution highlight immune regulation.