Single-Cell Lymphocyte Heterogeneity in Advanced Cutaneous T-cell Lymphoma Skin Tumors

Cutaneous T-cell lymphomas (CTCL) are a heterogeneous group of malignancies characterized by chronic inflammation and accumulation of malignant T lymphocytes in the skin (1). CTCL encompasses diverse presentations including Sezary syndrome where patients present with erythroderma, lymphadenopathy, and circulating malignant T lymphocytes, as well as mycosis fungoides in which malignant cells reside primarily in the skin (2). Mycosis fungoides is the most common form of CTCL and typically runs an indolent course with an excellent 5-year survival rate in early stages, but significantly decreased survival in advanced disease (3). In the early stages, most T cells reside in the skin and only a few circulate in peripheral blood and lymph nodes. However, a small number of patients progress, and tumor cells may involve other sites of the body with a fatal outcome (4). About 20% of patients progress to advanced-stage mycosis fungoides (stages IIB to IV; ref. 5), and the prognosis for patients with widespread CTCL manifestation beyond the skin is poor with a 5-year survival rate of only 40% (6). Large cell transformation occurs in 56% to 67% of patients with advanced stage mycosis fungoides (6) and is accompanied by clinically aggressive disease and shortened survival. Diagnosis of mycosis fungoides is difficult, especially in the early stages, due to the absence of specific markers for malignant lymphocytes that distinguish them from nonmalignant tumor infiltrating T lymphocytes (TIL). Diagnosis is usually based on clinicopathologic correlation, and the average time-to-diagnosis is 7 years (7). Delays prevent timely treatment and result in poorer clinical outcomes, while the treatment options for patients with aggressive forms of mycosis fungoides are limited, reflecting our poor understanding of disease pathogenesis.

Lymphocyte proliferation in CTCL is largely restricted to the skin, implying that malignant cells are dependent on their specific cutaneous microenvironment. Cytokines and other immunomodulator factors produced by malignant lymphocytes and TILs (8, 9) as well as by other immune and stromal cells (10) affect cutaneous inflammation (1, 8) and are important constituents of tumor local microenvironments, fostering survival, proliferation, and suppression of tumor cell immunosurveillance (8, 11). In this context, reactive TILs are exposed to multiple immunosuppressive pressures, including negative regulatory pathways and upregulation of inhibitory receptors such as PD1, CTLA4, LAG3, TIM3, and TIGIT that render them dysfunctional (12–14) and unable to elaborate their full effector functions for ultimately killing tumor cells (12, 13). However, the heterogeneity and immunologic state of malignant and reactive lymphocytes within CTCL skin tumors remain incompletely characterized.

Recent advances in single cell transcriptome technology, including droplet-based single-cell RNA-sequencing (scRNA-seq; ref. 15), profile gene expression across thousands of individual cells from a large heterogeneous population (16, 17) such as a patient biopsy. This high-resolution analysis of cellular heterogeneity reveals individual cell functions in the context of their microenvironment and provides striking insights into the complex cellular composition of normal and diseased tissue. Here, we report scRNA-seq analysis of skin tumor cells from patients with advanced stage CTCL. This analysis provides an unprecedented view of lymphocyte heterogeneity within the skin-microenvironment of individual CTCL tumors by identifying molecular signatures that are unique for each tumor. We also established a common gene expression signature in highly proliferating lymphocytes as well as the immunological state of TILs within each tumor. Together, these data provide important implications for personalized disease management.

Skin samples were obtained at the Comprehensive Skin Cancer Center, Columbia University Medical Center, from 10 patients with confirmed diagnoses of advanced CTCL (stage IIB–IVA; described in Supplementary Methods; Supplementary Table S1) and staged according to the most recent consensus (4, 5). Five patient samples were used for scRNA-seq and all ten for IHC. Participants gave written informed consent. Human research protocols were approved by the Institutional Review Board, Columbia University. Controls included human normal skin (NS, n = 8; 4 each for scRNA-seq and IHC) and atopic dermatitis (AD, n = 4; all for IHC) obtained from The Health Sciences Tissue Bank, University of Pittsburgh (Pittsburgh, PA). This study was conducted in accordance with the Declaration of Helsinki. Experimental procedures followed established techniques using the Chromium Single Cell 3′ Library V2 Kit (10x Genomics; ref. 18). Briefly, cell suspensions from enzymatically digested skin biopsies were loaded into the Chromium instrument (10X Genomics), and the resulting barcoded cDNAs were used to construct libraries. RNA-seq was performed on each sample (approximately 200 million reads/sample). Cell-gene unique molecular identifier counting matrices generated were analyzed using Seurat (19) to identify distinct cell populations using Louvain clustering (15). See Supplementary Methods for details. All scRNA-seq data have been deposited in the GEO database (accession GSE128531).

Single and dual antibody staining using tyramide signal amplification (Thermo Fisher Scientific) were performed on formalin-fixed, paraffin-embedded skin samples as described previously (18). Antibodies were all purchased from Sigma. IHC images were obtained with an Evos FL Auto microscope (Life Technologies). Confocal images were captured on an Olympus Fluoview 1000 confocal microscope using an oil immersion 100× objective.

We used scRNA-seq to profile gene expression in cells obtained from the enzymatically digested skin of five advanced-stage CTCL skin tumors and four healthy control skin samples (Supplementary Table S1). Fig. 1A depicts histologic features of the tumors studied, as described in Supplementary Table S2. A 3-mm skin biopsy from each donor yielded 3,607 to 9,272 cells from skin tumor samples and 2,200 to 4,847 cells from healthy skin. After reverse transcription from each cell, we constructed cDNA libraries and performed massive parallel sequencing, obtaining an average of 47,894 mapped reads per cell and a median of 1,261 unique genes detected per cell, comparable with previous studies (18). Cells were grouped according to their expression profiles by principal components analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) dimensional reduction (20). Comparison of whole skin cell distribution from each tumor sample with the four control skin samples shows overlap in the transcriptional profiles of cells from healthy skin samples but generally no overlap between cells from the tumor samples and the healthy samples (Fig. 1B). Strikingly, the combination of five tumors and four controls show that the tumor sample profiles do not overlap with each other, thus exhibiting significant intertumor heterogeneity (Fig. 1C). Unsupervised graph-based Louvain clustering by Seurat (19) identified 26 clusters of cells (Fig. 1D) whose types were identified by the expression of cell-specific marker genes (18) (Fig. 1E). We found the greatest heterogeneity between tumors and controls, as well as across tumors, was at the level of lymphocytes, keratinocytes, fibroblasts, and macrophages. In addition to PCA, canonical correlation analysis showed comparable results (19). Thus, scRNA-seq analysis characterizes details of the large intertumor cell transcriptional heterogeneity in advanced CTCL skin samples.

T lymphocyte transcriptional profiles of the tumors overlapped minimally with profiles of lymphocytes purified from healthy skin (Fig. 2A). Strikingly, we also observed impressive inter-tumor heterogeneity by the marginal overlap between transcriptional profile of tumor-derived lymphocytes (Fig. 2B). Comparison of the transcriptomes of each lymphocyte subset from the tumors and control skin samples identified 11 clusters (Fig. 2C). Some lymphocyte clusters were unique to individual tumors, such as cluster 8 (CTCL-2), clusters 2 and 3 (CTCL-5), cluster 4 (CTCL-6), cluster 6 (CTCL-8), and cluster 5 (CTCL-12), while clusters 1 and 10 included lymphocytes derived from all tumor and healthy skin samples.

We determined the differential expression (DE) of genes in each of the unique clusters from each tumor by comparing gene expression from each cell in the cluster to that of all other cells in the dataset, using a cut-off of P < 0.05 and further requiring expression of the gene from >25% of cells in the cluster. Thus, DE-identified genes are expressed either uniquely or by a large proportion of cells within each cluster compared to all other clusters. Examples of the most highly significant DE genes for each cluster are highlighted by heatmap (Fig. 2D) and the proportion of cells and the scaled average expression of these genes by all tumors and controls show strong and specific expression within individual tumors (Fig. 2E). Uniquely expressed genes included RDH10, CXCL13, SCG2 (CTCL-2), FGR, IGFBP2/P6, NEFM (CTCL-5), ANO1, TNP1, CES4A, ZDHHC14 (CTCL-6), LGALS7, SERPINB3/B4, SPRR2A (CTCL-8), NTRK2, TMPRSS3 (CTCL-12). From these and other DE genes (Supplementary Table S3), we identified distinct gene expression signatures for each tumor-specific cluster, including expression of eukaryotic initiation factors (eIF) and oncogenes (CTCL-2); NK-cell receptor and signaling molecules (CTCL-5); genes associated with tumor cell survival, proliferation, and metastasis (CTCL-6); members of the serpin, S100, and galectin families (CTCL-8), and genes associated with increased cell motility and invasiveness (CTCL-12). Ingenuity Pathway Analysis (IPA) (21) identified activation of key molecular pathways in these patient-specific tumor clusters. Highly significant examples of distinct pathways activated in each tumor were unrelated between patient samples (Fig. 2F) but followed the gene expression signatures identified above. These included eIF2, eIF4, and mTOR signaling (CTCL-2); NK-cell signaling and virus entry via endocytic pathways (CTCL-5); tumorigenic pathways common to glioma and non-small cell lung cancer (CTCL-6); pathways related to skin inflammation and skin-barrier dysfunction (CTCL-8); and pathways associated with epithelial–mesenchymal transition (CTCL-12).

TOX (thymus high-mobility group box) is considered a marker of malignant lymphocytes in CTCL tumors (9, 22, 23). Strikingly, clustering of the TOX⁺ cells from each tumor sample (Supplementary Fig. S1) revealed a large overlap of gene expression with the corresponding tumor-specific clusters identified above (Fig. 2G). In addition, in these tumor-specific clusters, we found significant but heterogeneous overexpression of genes associated with tumorigenesis, tumor-cell proliferation, and resistance to apoptosis (Supplementary Fig. S2). Although the single-cell gene expression approach employed (3′ Library Kit) does not allow TCR repertoire profiling, and therefore the identification of malignant lymphocytes by alpha-betaTCR clonality (24), we still could detect strong expansions of TRBC1 or TRBC2 genes in these tumor-specific clusters (Supplementary Fig. S3), consistent with the occurrence of alpha-betaTCR clonality. Newly available scRNA-seq tools will allow characterization of this clonality in future work as was demonstrated in a recent study on CD4⁺ T cells from the peripheral blood of a Sezary patient (25). Nonetheless, we conclude from our lines of evidence that the heterogeneous but tumor-specific signatures confirmed in TOX⁺ cells represent patient-specific gene expression of malignant lymphocytes that may have implications for personalized therapy focusing on specific pathways.

Cell-cycle analysis identified actively proliferating lymphocytes by expression of G₂–M and S-phase genes (Fig. 3A and B). The fifty highest DE genes for each cluster of the five tumors analyzed (Fig. 3C) show specific clusters characterized by strong gene expression signatures, such as clusters 1 and 7 (CTCL-2), cluster 3 and 4 (CTCL-5), cluster 4 (CTCL-6), cluster 3 (CTCL-8), and cluster 7 (CTCL-12). Strikingly, these clusters corresponded to the highly proliferating lymphocytes identified in Fig. 3A and B and highly expressed genes involved in cell-cycle progression (e.g., PCNA, CDK6, CCND1, NUSAP1, CENPE, CCNA2, HMMR, CDCA8, CDK1, CENPM, CDC20, ATP5C1), proliferation (e.g., MIK67, KIAA0101, TOP2A, NPM1, IGF2, PLK1, MYC, FOS, NPM1, PRDX1, PIM2, RAN), and survival (e.g., BCL2, BIRC5, BIRC3, BCL2L12, MCTS1, TSC22; Supplementary Table S4), therefore likely representing highly proliferating malignant lymphocytes. Comparison of these clusters identified a 17-gene expression signature common to all five tumors tested (Fig. 3D). High expression by these genes was detected in all tumors while only few positive cells were found in the lymphocyte clusters of healthy controls for most genes identified (Supplementary Fig. S4). Strikingly, this 17-gene expression signature was also found in TOX⁺ cells from all patient samples but not from controls (Fig. 3E and F). We focused on three of these common genes, PCNA, ATP5C1, and NUSAP1, that presented low expression in normal lymphocytes. These were further investigated by IHC in the five tumors analyzed by scRNA-seq (Fig. 4A) as well as in additional samples from patients with advanced stage CTCL (Supplementary Fig. S5). Staining in tumor samples was compared with normal (NS) and atopic dermatitis (AD) skin. Results showed that apart from scant PCNA⁺ cells in the epidermis, NS and AD skin were negative for expression of these markers, while all tumor samples tested exhibited high numbers of positive cells both in the epidermis and in the dermis for all three markers tested. By multicolor immunofluorescence microscopy, we next demonstrated that these markers colocalized with TOX (Fig. 4B). Thus, we have identified a gene expression signature of highly proliferating malignant lymphocytes that is common to all tumors tested and could be developed as a marker for the diagnosis of CTCL.

TILs, particularly CD8⁺ T cells, are the major effector cell-type for fighting and killing cancer cells (12, 13). To define the molecular signature of CD8⁺ TILs in the tumor microenvironment of advanced stage CTCL skin tumors, we examined gene expression of effector molecules, checkpoint receptor inhibitors, and markers of T regulatory (Treg) cells in CD8⁺ T cells from the tumor and control skin samples. Transcriptome profiles were distinct and nonoverlapping for CTCL-5 and CTCL-6 (Fig. 5A), while most CD8⁺ T cells from tumors and controls appear to overlap in cluster 1 (Fig. 5B). The cell composition of each CD8⁺ cluster is reported in Supplementary Table S5. Thus, the cells in cluster 1 appeared to reflect reactive CD8⁺ T cells, that is, TILs, as they did not express tumor-associated genes (Fig. 5C, bottom, and D), while CD8⁺ T cells from the other clusters expressed genes associated with tumors likely indicating malignant lymphocytes but which are also associated with dysregulation. CD8⁺ TILs in cluster 1 expressed markers of skin-residency, such as CD69 and ITGAE, and of memory cells (CD27). Multiple coinhibitory receptors were expressed by cells in this cluster, although we found variability across tumors (Fig. 5C, top, and 5E), including expression of PD1, CTLA4, TIM3, LAG3, and TIGIT by most CD8⁺ TILs from CTCL-6; PD1 and LAG3 expression by several cells from CTCL-2, and expression by few cells from the other tumors. However, coinhibitory receptors were also expressed by CD8⁺ lymphocytes from tumor-specific clusters such as PD1 and TIM3 (clusters 0, 2, and 5: CTCL-5) as well as TIM3 and TIGIT (clusters 3 and 4: CTCL-6; Fig. 5C). Similarly, we observed variable expression of coinhibitory receptors by CD3⁺CD4⁺ T cells across all samples and within tumor-specific clusters (Supplementary Fig. S6). Multicolor immunofluorescence microscopy shows representative expression of coinhibitory receptors by CD8⁺ lymphocytes in advanced CTCL tumors (Fig. 5F). No cells in any CD8⁺ clusters expressed Treg markers such as FOXP3 and only cells in clusters 0, 2, and 3 expressed IL2RA. Conversely, we were able to identify specific CD3⁺CD4⁺ lymphocyte clusters that coexpressed FOXP3, IL2RA, PD1, CTLA4, LAG3, and TIGIT, which likely identified Treg TILs (Supplementary Fig. S6C).

Analysis of effector molecule expression indicated that several CD8⁺ TILs in cluster 1 expressed granzyme A (GZMA), while only few cells expressed granzyme B (GZMB) and perforin (PRF1). CTCL-6 CD8⁺ T cells from clusters 3 and 4 highly expressed GZMB and PRF1, or PRF1 only, respectively. None of the other CD8⁺ clusters contained significant numbers of cells expressing cytolytic molecules (Fig. 5C). However, we detected a variable and modest up-regulation of FASL in most clusters, potentially providing an alternate cytolytic mechanism. CD8⁺ lymphocytes from clusters 1 and 2 expressed IFNG while cells from most clusters expressed TNFA and IL1B. Interestingly, specific CD4⁺ clusters expressed IFNG and TNFA as well as immunosuppressive cytokines such as IL10, TGFB1, IL4 and IL13 (Supplementary Fig. S6C). Finally, we detected no IL2 production by any clusters of CD8⁺ or CD4⁺ T cells.

Together, these results reveal a complex landscape of CD8⁺, CD4⁺, and Treg TIL gene expression characterized by different levels of effector molecules and a variable combination of coinhibitory receptors likely impairing an effective antitumor response.

Tumor cellular heterogeneity poses challenges to cancer diagnosis and treatment. Advances in single-cell gene expression profiling of patient samples open new avenues for dissecting this heterogeneity, which is a central feature of precision medicine. We employed scRNA-seq technology to profile the transcriptomes of thousands of individual cells from advanced-stage CTCL skin tumors. Our analysis revealed a large inter- and intratumor gene expression heterogeneity, particularly in the T lymphocyte subset, as well as a common gene expression signature in highly proliferating lymphocytes that was validated in multiple advanced-stage skin tumors. In addition, we established the immunological state of TILs and found heterogeneity in effector and exhaustion programs across patient samples. Thus, single-cell analysis provides an unprecedented view of all major cellular components simultaneously and their individual gene expression states, with important implications for diagnosis and personalized disease treatment.

Large numbers of malignant and nonmalignant reactive lymphocytes are often found infiltrating CTCL skin tumors. However, specific markers have been lacking for identifying the malignant lymphocytes since the tumor cells cannot be reliably isolated from the lesional skin of patients. This limitation prevents characterizing the transcriptional profile and heterogeneity of malignant lymphocytes or distinguishing them from benign reactive lymphocytes that might block tumor growth. Furthermore, it delays the diagnosis of CTCL and complicates development of effective treatments. Characterizing single-cell transcriptomes overcomes these problems while providing an unbiased and comprehensive map of rare lymphocyte populations and cell states within each tumor sample. We found that the transcriptional profiles of lymphocytes isolated from healthy control skin were similar to each other, while those from tumor skin appeared mostly distinct from the healthy skin profiles and overlap only partially with each other. This variation reflects both the different subtypes of the samples studied as well as any tumor-specific expression unique to individual patients. Significantly, we could identify at least one unique lymphocyte cluster for each CTCL tumor sample analyzed, and their malignant phenotype was confirmed by the overlap in gene expression with cells from the same tumor expressing TOX, a marker previously shown to identify malignant lymphocytes in CTCL. Thus, we have demonstrated a novel basis for identifying tumor cell heterogeneity that may be developed for personalized therapies.

The unique transcriptional pattern of tumor-specific lymphocyte clusters observed for each tumor sample indicates activation of specific tumor-associated signaling pathways. Some expressed genes had not been previously associated with CTCL while the expression of others that had been linked to CTCL varied across patient samples from high level expression by many cells in some tumors to little or no expression in others. For example, the cluster unique for CTCL-2, a sample from a stage IVA Sezary syndrome patient, overexpressed chemokine genes such as CXCL13, CCR7 and CCR4 that confer enhanced migratory ability to memory Sezary cells (26). Furthermore, multiple eIF proteins were upregulated in this tumor sample, showing that the eIF2 and eIF4 signaling pathways are activated as well as the mTOR signaling cascade, a major regulator of eIF4 and ribosomal protein S6 kinase (27). Interestingly, deregulation or altered expression of eIFs leads to translational reprogramming and promotes several oncogenic processes, including tumor cell survival, proliferation, neovascularization, and metastasis (27). In contrast, lymphocytes of the two clusters unique to CTCL-5 (representing a CD8⁺ aggressive cytotoxic CTCL) expressed genes involved in NK cell-signaling as well as several NK-cell receptors, including killer-cell immunoglobulin-like receptors and CLEC12A that negatively regulate NK-mediated cytotoxicity against tumor cells (28, 29). We also detected upregulation of several genes involved with virus entry via endocytic pathways, which is intriguing in view of the potential role for persistent viral infections in the etiology of CTCL (30). A third pattern characteristic of cells from the CTCL-6–specific cluster was expression of genes involved with tumor cell survival, proliferation, and metastasis, some common to tumorigenic pathways specific to glioma and non-small cell lung cancer (e.g., CDK6, KRAS, PA2G4, PIK3R1, RB1, RRAS2, RXRA, TFDP1).

Parallel to upregulation of genes associated with carcinogenesis and metastasis such as TPT1 (31) and MALAT1 (32), cells from the CTCL-8–specific cluster expressed the cysteine-protease inhibitors SERPINB3 and SERPINB4, which are expressed by various tumors and involved in inactivating granzyme M, an enzyme that kills tumor cells (33). Moreover, SERPINB3/B4 promotes tumor-cell transformation, migration, and drug resistance (33) and contributes to inflammation and barrier dysfunction in inflammatory skin diseases (34). Indeed, we found that cells in this cluster upregulated expression of genes from the psoriasis-like pathway as well as those associated with skin-barrier dysfunction, which is a striking match to the histopathologic characterization of the CTCL-8 sample (Supplementary Table S3). Previous studies have shown that malignant T lymphocytes drive the morphological and histopathological changes observed in mycosis fungoides skin lesions, including keratinocyte hyperproliferation and compromised skin barrier function (35). These changes lead to downregulation of keratinocyte differentiation markers and increased intercellular distance, enhancing skin permeability and contributing to the increased susceptibility to skin infections in CTCL patients, particularly in advanced-stage disease (36). The cluster unique to CTCL-12, from a patient with mycosis fungoides and large-cell transformation, overexpressed genes associated with epithelial–mesenchymal transition (EMT). This process allows tumor cells to acquire migratory, invasive, and stem-like properties, promoting tumor infiltration and metastasis (37, 38). Although originally associated with epithelial-derived tumors, EMT is also implicated in hematologic cancers (39). During ETM, cancer cells change morphology by disruption of intercellular junctions, loss of cell polarity, reorganization of the cytoskeleton, and increased cell motility necessary for invasion (37, 38). In particular, we observed upregulation of genes associated with the remodeling of adherens junctions, changes in cell adhesion, activation of small GTPases of the Rho family such as RAC1, CDC42, and RHOA, and reorganization of the actin cytoskeleton (Supplementary Table S3; refs. 37, 38, 40). We also found upregulation of TMPRSS3, a type II transmembrane serine protease that contributes to EMT in other human cancers via activation of the ERK1/2 signaling pathway (41).

While characterizing this vast intertumor heterogeneity in gene expression that may be essential for tailoring personalized medicine, we also searched for gene expression common to all tumors that could be used to guide diagnosis, design new medications that treat all CTCL tumor subtypes, and monitor treatment efficacy. We found that highly proliferating T lymphocytes in each tumor expressed a very defined and strong gene expression signature involving cell-cycle progression, proliferation, resistance to apoptosis, and metabolic processes. Strikingly, we found that these signatures had 17 genes in common that we found to be also expressed by TOX⁺ cells in all tumors. We validated the protein coexpression of three of them (PCNA, ATP5C1, NUSPA1) with TOX in multiple patients with advanced-stage CTCL. Thus, these results strongly indicate that both the common and heterogeneous patterns of gene expression can be exploited for diagnosis and treatment of CTCL.

Expression of checkpoint inhibitory receptors renders CD8⁺ TILs incapable of mounting an efficient antitumor response, as manifested by impaired degranulation and reduced proinflammatory cytokine production (12, 13). Thus, recent cancer immunotherapies have focused on enhancing CD8⁺ T-cell antitumor responses by targeting the inhibitory receptors (12), yielding major clinical benefits (42). However, only a subset of patients exhibited clear long-term responses, while most patients with different types of tumors failed to respond. This failure likely results from the expression of multiple inhibitory receptors on TILs that may synergistically modulate antitumor responses by different pathways. Preclinical and clinical (43) studies have indicated that full antitumor immunity may require several inhibitory receptors to be blocked. Consistent with these studies, our analysis demonstrates that co-inhibitory receptors are simultaneously but heterogeneously expressed on both CD8⁺ and CD4⁺ T lymphocytes. A particularly striking example is the overexpression of TIGIT, LAG3, and TIM3 by CD8⁺ T cells in CTCL-6, and LAG3 in CTCL-2, indicating a strong patient-specific signature that may be exploited for individualized targeting. Although co-inhibitory receptors appeared to be expressed mostly by reactive TILs, we also observed their expression by T lymphocytes expressing tumor-associated genes. In some cases, these cells were clearly identifiable as malignant, while in others they may instead represent exhausted T lymphocytes acquiring a malignant phenotype or nonconventional Tregs such as CD4⁺ Tregs lacking FOXP3 and/or IL2RA expression.

TIGIT expression by CD4⁺ T cells from the peripheral blood of patients with advanced stage Sezary syndrome was recently associated with reduced IFNG and IL2 production (44). We found that TIGIT is also highly expressed by both CD8⁺ and CD4⁺ TILs from advanced stage CTCL skin tumors, and in parallel with expression of other coinhibitory receptors, most notably TIM3, PD1, LAG3, and to a lesser extent CTLA4. Because TIGIT can foster an immunosuppressive tumor microenvironment by promoting Treg function and maintenance as well as by inhibiting cytotoxic T-cell activity (45), we confirmed TIGIT expression by Tregs in the samples tested and found a correlation with lack of GZMB and perforin expression in cluster 1 CD8⁺ TILs that characterizes exhaustion. Interestingly, we found that CTCL-6–specific CD8⁺TIGIT⁺ lymphocytes from cluster 3 expressed GZMB and perforin as well as tumor-associated genes, likely representing cytotoxic malignant lymphocytes. CD8⁺ TILs showed no IL2 expression while most cells expressed TNFA and IFNG, which was associated with EOMES expression, consistent with an exhausted CD8⁺ T-cell phenotype (46). Such cells are defective in IFNG production and cytotoxicity but continue expressing IFNG and granzyme mRNAs (46), consistent with our findings. Likewise, we observed that CD4⁺TILs also expressed IFNG and TNFA as well as TGFB1, which is likely produced in combination with IL10 by FOXP3⁺ cytotoxic Tregs (47). Furthermore, variability observed in TGFB1, IL10, IL4, and IL13 cytokine production from CD4⁺ T cells of the CTCL-12–specific and CTCL-8–specific clusters likely reflects FOXP3⁻Tregs or malignant lymphocytes, or CD4⁺FOXP3⁺ malignant lymphocytes with suppressive activity (48), respectively. We conclude that multiple coinhibitor receptors are expressed by malignant and reactive lymphocytes in advanced CTCL skin tumors, conferring to the latter a dysfunctional phenotype. Understanding the heterogeneity in coinhibitory receptor expression may be essential for developing patient-specific therapy and for guiding checkpoint inhibitor blocking to permit effective killing of tumor cells by TILs.

In conclusion, single-cell transcriptome profiling provides novel insights into CTCL disease heterogeneity by revealing patient-specific landscapes of malignant and reactive lymphocytes. Recent reports applying single-cell RNA-seq to peripheral blood of Sezary patients (25, 49) also demonstrated heterogeneity in the malignant T-cell population and, in addition, genetic heterogeneity within the same patient over time has been observed (50). Although studies with small subsets of patients need to be expanded to confirm and extend general trends, we nonetheless demonstrate the ability to detect gene expression patterns among single skin tumor cells that can provide a framework for improving CTCL diagnosis and treatment, thus realizing the goal of precision medicine.

L.J. Geskin reports receiving speakers bureau honoraria from Helsinn, is a consultant/advisory board member for Helsinn and Therakos, and reports receiving commercial research grants from Mallinckrodt and Actelion. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L.J. Geskin, R. Lafyatis, P. Fuschiotti

Development of methodology: R. Lafyatis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.J. Geskin, C.-A. Bayan, P. Fuschiotti

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.M. Gaydosik, T. Tabib, L.J. Geskin, J.F. Conway, R. Lafyatis, P. Fuschiotti

Writing, review, and/or revision of the manuscript: A.M. Gaydosik, T. Tabib, L.J. Geskin, C.-A. Bayan, J.F. Conway, R. Lafyatis, P. Fuschiotti

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Gaydosik, T. Tabib, L.J. Geskin, C.-A. Bayan, J.F. Conway, R. Lafyatis, P. Fuschiotti

Study supervision: L.J. Geskin, P. Fuschiotti

We thank Christina Morse for technical support for immunofluorescence microscopy. This work was supported by NIH/NCI grant R21 CA209107-02 to P. Fuschiotti.

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