SLAMF7 Signaling Reprograms T Cells toward Exhaustion in the Tumor Microenvironment

The biological process of T cell exhaustion (or dysfunction) resulting from chronic T cell stimulation and activation has emerged as one of the most important topics in immunology and cancer biology (1–4). T cells undergoing persistent TCR stimulation, as occurs in the tumor microenvironment (TME), begin a process whereby they assume a cellular state less amenable to proliferation, cytokine production, and cytotoxicity (1, 3, 4). Conversion of infiltrating T cells to an exhaustion phenotype is a well-known hallmark tumors use to evade the host immune response and continue unchecked growth (5–7).

These exhausted or dysfunctional T cells are characterized at the cellular level by epigenetic remodeling (3, 8, 9) and expression of a number of inhibitory receptors, also known as checkpoint receptors, including PD-1, CTLA-4, LAG3, Tim3, TIGIT, VISTA, CD38, 2B4 (10), and others (1, 2). Through both known and unknown mechanisms, these inhibitory receptors directly inhibit regulatory and effector T cell functions and are responsible for the characteristic loss of effector function observed in exhausted T cells (3). Abs targeting these inhibitory receptors are showing great promise in generating responses in cancer patients. However, responses vary by tumor type; the majority of patients do not respond effectively to checkpoint blockade of PD-1, PD-L1, or CTLA-4 (11); and a combination of PD-1 and CTLA-4 blockade results in substantial risk of serious adverse events (12, 13). Consequently, there is a need to understand why some patients do not respond to checkpoint blockade, what additional signals besides repetitive TCR signaling induce inhibitory receptors on T cells in the TME, and if T cell exhaustion can be prevented from occurring in the first place (3).

Renal cell carcinoma (RCC) is the most common cancer of the genitourinary system, with over 400,000 new cases and over 175,000 deaths each year, and it continues to increase in prevalence (14). The clear cell RCC (ccRCC) subtype is the most common form of RCC, and although it is a heterogeneous disease, it has a unique TME, making it an attractive target for immunotherapy (14, 15). In contrast to most tumors in which the TME is dominated by M-2 macrophages, the ccRCC TME is composed primarily of T cells, with a large portion consisting of CD8⁺ T cells (15, 16). However, the infiltrate of CTLs does not portend a favorable prognosis because of extensive levels of T cell exhaustion (17–19). ccRCC is notable in that, of all the cancers in The Cancer Genome Atlas (TCGA), immune evasion occurs in ccRCC almost entirely via T cell exhaustion (19). This has paved the way for trials of various checkpoint inhibitors in ccRCC, with the most recent findings of the CheckMate 214 trial establishing combination ipilimumab (anti-CTLA-4) and nivolumab (anti-PD-1) as first-line treatment for intermediate or poor-risk ccRCC patients (14, 20). However, this regimen lacks effectiveness in all ccRCC patients (11) and can cause significant adverse effects (12, 13, 20), signifying the need to better understand T cell exhaustion mechanisms in ccRCC.

The SLAMF7 (CRACC, CD319, CS-1) receptor is a member of the signaling lymphocytic activation molecules (SLAM) family of receptors (21), with expression restricted to hematopoietic cells, and is present at differing frequencies on various types of immune cells (21–23). SLAMF7, and other SLAM family members (except 2B4), are unique in that they function as homotypic receptors, which once activated, recruit various SH2 domain-containing proteins to their cytoplasmic immunoreceptor tyrosine-based switch motifs (21, 24). In doing so, SLAMF7 is able to modulate a host of immune cell–specific functions across various immune cell types (21, 23–25). SLAM family receptors are increasingly being linked to T cell exhaustion, with expression of SLAMF6 recently discovered as a marker to identify progenitor exhausted CD8⁺ T cells (26) and both SLAMF6 and SLAMF4 (2B4) being shown to function as inhibitory receptors on CD8⁺ T cells (10, 27). Notably, SLAMF7 has also recently been linked to a CD8⁺ T cell subset enriched in melanoma patients who fail to respond to checkpoint blockade (28) and expressed on certain memory-precursor and effector CD8⁺ T cells, which respond indirectly to checkpoint blockade (29). In this study, we set out to define the role of SLAMF7 on T cell regulation in the context of T cell exhaustion. We uncovered a novel role for SLAMF7 in regulating the expression of various T cell inhibitory receptors, exhaustion-promoting transcription factors (TFs), and STAT1/3 phosphorylation in a TCR-independent manner. Also, we find that SLAMF7 self-ligation between tumor-associated macrophages (TAMs) and T cells is sufficient to induce the upregulation of these inhibitory receptors. Our data identify SLAMF7 as a novel regulator of T cell inhibitory programs, with potential clinical implications and therapeutic opportunities.

The following human Abs were used: CD3-FITC (HIT3a), CD4-V450 (RPA-T4), CD8-Alexa700 (HIT8a), SLAMF7-PerCp-eFluor710 (162), LAG-3-APC-eFluor780 (T47-530), TIM-3-PE-Cy7 (F38-2E2), EZH2-PE (11/EZH2), PD-1-Alexa488 (EH12.2H7), FOXP3-BV421 (206D), YY1-Alexa594 (H-10), Blimp-1–DyLight650 (3H2-E8), CTLA4-FITC (14D3), IL-2-APC (MQ1-17H12), TNF-⍺-Alexa488 (Mab11), IFN-γ-PE (4S.B3), STAT1 (pY701)-Alexa647, STAT2 (p690)-PE, STAT3 (p704)-Alexa647, STAT5 (p694)-Alexa647, STAT6 (p641)-Alexa647, P38 MAPK (pT180/pY182)-Alexa647, ERK1/2 (pT202/pY204)-Alexa647, ZAP70 (p319/p352)-PE, and SLAMF7 (162.1) (used for cross-linking [CL]). Mouse Abs used include the following: CD3-APC (145-2C11), CD38-BV510 (90/CD38), CD8a-Alexa700 (53-6.7), LAG3-PE (C9B7W), Tim3-APC-Fire750 (B8.2C12), PD-1-PerCp-eFluor710 (J43), 2B4-FITC (eBio244F4), SLAMF7-BV421 (4G2), CD11b-BV570 (M1/70), CD19-PerCp-Cy5.5 (1D3), Ly6G-BV711 (1A8), Ly6C-BV421 (HK1.4), MHC-II-BV785 (M5/114.15.2), CD206-Alexa647 (C06862), CD45-Alexa532 (30-F11), NK1.1-PE-Cy7 (PK136), SLAMF7-APC (4G2), CD11c-PE-CF594 (HL3), CCR2-BV750 (475301), LAG3-BV785 (C9B7W), CD3-BUV737 (17A2), TOX-eFluor660 (TXRX10), SLAMF6-BUV395 (13G3), CD69-PE-Cy7 (H1.2F3), and CD4-eFluor450 (RM4-5). All Abs were purchased from BioLegend, BD Biosciences, or Thermo Fisher Scientific. Recombinant human IL-2 (Peprotech) was used at 300 IU/ml. Recombinant murine IL-2 (R&D Systems) was used at 300 IU/ml. CD3/28 Dynabeads (Thermo Fisher Scientific) were added at a 1:1 cell/bead ratio.

Fresh PBMCs were obtained from buffy coats and processed as previously described (30). CD3⁺ T cells were isolated from PBMCs via negative selection using Dynabeads Untouched Human T Cells Kit (Thermo Fisher Scientific) per manufacturer’s instructions.

Isolated, fresh, primary human CD3⁺ T cells were plated at 2 × 10⁵ cells per well in 96-well plates. Cells were cultured in complete RPMI 1640 (RPMI 1640, 10% FBS, 1× penicillin, streptomycin, and fungizone). For CL experiments, 8 μg/ml anti-SLAMF7 (162.1) mAb was added to a sterile, high-binding 96-well cell culture plate overnight at 4°C. Wells were washed with PBS three times before addition of cells. Cells were cultured for indicated times at 37°C and 5% CO₂. For 6 d time points, cells were cultured in the presence of stimuli for 3 d before being moved into a fresh SLAMF7 mAb cross-linked plate with fresh media and rhIL-2. For restimulation experiment, cells were washed and all stimuli removed after 6 d of culture. Cells were replated in fresh complete media with 300 IU/ml IL-2 and CD3/28 Dynabeads at a 0.2:1 bead/cell ratio and cultured for an additional 3 d. GolgiPlug was added for the last 4 h of culture, and intracellular staining of cytokines was performed with BD CytoFix/Perm kit per manufacturer’s instructions.

Isolated, primary human CD3⁺ T cells were rested for 24 h in 40 IU/ml rhIL-2 in complete RPMI 1640. Cells (300,000 per well) were stimulated in 96-well plates as indicated for various time points at 37°C and 5% CO₂ before being fixed in BD CytoFix/Perm buffer at 37°C and 5% CO₂. Permeabilization was performed with BD Phosflow Perm Buffer IV (0.5×) per manufacturer’s instructions. Cells were stained with anti-CD4, anti-CD8, and an mAb against a single phosphorylated protein at the same time. Samples were run on a Cytek Aurora Spectral Cytometer.

All mice used in experiments were bred in house. SLAMF7^−/− genotype was verified by PCR. Eight- to twelve-week-old wild-type (WT) (C57BL/6J) or SLAMF7^−/− (C57BL/6J background) mice were used for all experiments. SLAMF7^−/− mice were generated at the Michigan State University (MSU) Transgenic and Genome Editing Facility using CRISPR-Cas9 with guide RNAs targeting introns between exon 2/3 and exon 5/6 of SLAMF7. Genomic deletion of this region of SLAMF7 completely abrogates SLAMF7 expression on all cells. A single founder mouse with homozygous deletions in both introns was bred to a WT mouse, and the F1 progeny were bred together to establish a stable line of SLAMF7^−/− mice.

Investigators were blinded to mouse genotypes when measuring tumors. Either 7 × 10⁵ B16 tumor cells (for experiments involving isolation of TAMs) or 4 × 10⁵ B16 tumor cells (for tumor growth experiments) were injected into the hind flank of WT and SLAMF7^−/− mice. For TAM isolation experiments, tumors were harvested 25 d later and manually dissociated into a single-cell suspension. For tumor growth and tumor-infiltrating leukocyte (TIL) phenotyping experiments, tumors were harvested at completion of tumor measurements. Cell suspension was pelleted and resuspended in RPMI 1640 with 0.5 mg/ml Collagenase IV (Millipore-Sigma) and 1000 IU/ml DNaseI (Millipore-Sigma) under constant, gentle agitation at 37°C for 1 h. Digestion was stopped with EDTA, and cell suspension was filtered through a 40-μm cell strainer. Cells were then subject to a Ficoll-Paque gradient centrifugation step to enrich for immune cells. Cells were washed three times with complete RPMI 1640 before immune cell isolation procedures. TAMs were isolated from dissociated B16 tumors using a mouse CD11b positive selection kit (StemCell) per manufacturer’s instructions. Murine immune cells were either stained directly for spectral cytometry analysis or plated for coculture experiments. For TAM-CD8 T cell coculture experiments, 5 × 10⁴ TAMs and 1.5 × 10⁵ CD8⁺ T cells (1:3 ratio) were plated in a 96-well cell culture plate in complete RPMI 1640 supplemented with an additional 1% PSF and 50 IU/ml rmIL-2. Cells were cultured for 6 d at 37°C and 5% CO₂ with a media change at 3 d. For murine in vitro CD3/28 stimulation experiment, mouse total splenocytes were cultured for 6 d in complete RPMI 1640 and 50 IU/ml rmIL-2 with or without murine CD3/28 Dynabeads at a 1:1 cell/bead ratio. Bone marrow–derived macrophages were generated as previously described (31).

Cells were prepared and stained as previously described (23). TF and cytokine staining were performed with the Transcription Factor Buffer Set (BD Biosciences) per manufacturer’s instructions. For IL-2 intracellular staining, GolgiPlug (BD Biosciences) was added for the final 4.5 h of cell culture. Viability staining was performed with either Zombie NIR Fixable Viability Kit (BioLegend) or LIVE/DEAD Aqua fixable viability dye (Thermo Fisher Scientific) and was included in all experiments. Fc receptors were blocked in all samples during staining with human Fc block (BD Biosciences) or mouse Fc block (BD Biosciences). Samples were acquired on a 5-laser Cytek Aurora Spectral Cytometer. Spectra were unmixed using SpectroFlo software, and data were analyzed using FlowJo version 10.6.1 (Tree Star). SPICE plots were generated using Pestle version 2.0 and SPICE version 6.0 (32). For Fig. 2E, t-distributed stochastic neighbor embedding (tSNE) dimensionality reduction was performed in FlowJo on a concatenated FCS file containing Mock and SLAMF7 CL samples, in duplicate, from two healthy donors with the following tSNE parameters: iterations = 7000 and perplexity = 70. FlowSOM clustering was performed in FlowJo with the number of meta-clusters set to 10. Clusters with very few cells and high similarity to existing clusters were manually merged with those clusters to obtain a final consensus of four CD8 T cell clusters and two CD4 T cell clusters. Merged FlowSOM clusters were color coded and overlaid onto the tSNE map of all T cells. The same approach was repeated for (Fig. 3E).

Primary human CD3⁺ T cells were plated at 2 × 10⁵ cells per well in a 96-well plate with indicated stimuli. Exogenous rhIL-2 was not added. Supernatant was collected at indicated time points, and IL-2 concentration was assessed by sandwich ELISA (BioLegend).

Fragments per kilobase of exon per million reads mapped RNA sequencing (RNA-seq) expression values for genes of interest and survival data were obtained from the TCGA-TARGET-GTex dataset using Xena (http://xena.ucsc.edu). Gene expression values were normalized to account for differences in total immune cells by dividing by PTPRC (CD45) and plotted using matplotlib in Python 3.7. Kaplan-Meier curves were generated using the lifelines package. Survival metrics used in each Kaplan-Meier plot were chosen based on previously published work (33). Tumor Immune Dysfunction and Exclusion (TIDE) analysis was performed on the TIDE webpage (http://tide.dfci.harvard.edu/query/) by querying the SLAMF7 gene. Figures were taken directly from the resulting TIDE analysis.

The entire ccRCC CyTOF dataset from Chevrier et al. (15) was downloaded from the Bodenmiller laboratory webpage (https://cytobank.org/bodenmillerlab/index.html). All immune cell subsets were identified based on predetermined Phenograph-assigned cluster identifiers present in the raw FCS files and were colored in figures in the same manner as in Chevrier et al. (15). SLAMF7 expression on various immune cell types was manually gated based on a priori knowledge of expected SLAMF7 expression patterns, and it was determined that a gate set at “1” was most appropriate and accurate for all immune cell subsets. This threshold was used consistently for SLAMF7 gating on all immune cell types. Although this threshold is lower than what some investigators use for CyTOF analysis, to our knowledge, SLAMF7 has never been assessed or manually gated on CyTOF. Furthermore, although SLAMF7 is expressed at a very high Ag density on certain immune cells (i.e., dendritic cells and plasma cells), it is also present on many other immune cell types, albeit at a lower Ag density. For this reason we used manual gating to allow us to accurately assess marker expression, whereas some other techniques can make appreciating cells with a low density of SLAMF7 receptor difficult. Samples from healthy controls and RCC no. 17 (too few cells) were excluded.

Violin plots were generated using the ggplot2 package in R. Heatmap correlation matrix was generated using the corrplot package in R. FlowJo v10.6.1 software was used for all spectral cytometry analysis and high-dimensional analyses. Pie charts were made in Prism v7 (GraphPad). All t tests were paired and performed in GraphPad Prism v7. Log-rank test was used to analyze all Kaplan-Meier plots. Two-way ANOVA was used to compare tumor growth curves. Details on statistical analyses used for TIDE analysis can be found in the original publication (19). SPICE plots were compared using the permutation test included in the software. Groups in Figs. 2A, 3B, and 3D were compared by fitting to a mixed model before performing multiple comparison test. This was accomplished in Prism v8 and performed because of the presence of missing values due to random chance. No mice or human samples were censored from any analysis for any reason (except the CyTOF samples listed above).

All animal procedures were approved by the MSU Institutional Animal Care and Use Committee.

To examine associations between SLAMF7 and various inhibitory receptors on T cells, we used a chronic, polyclonal, in vitro T cell activation assay. CD3⁺ human T cells were cultured with only rhIL-2 (Mock) or rhIL-2 + CD3/CD28 Dynabeads, and exhaustion markers were assessed over a period of 6 d using spectral cytometry (Supplemental Fig. 1A–C). We observed time-dependent increases in expression of LAG3, PD-1, and Tim3 on CD3/28-activated CD8⁺ and CD4⁺ T cells (Fig. 1A, 1B). SLAMF7 expression was minimal on both CD8⁺ (1.12%) and CD4⁺ (0%) T cells initially but increased over time with CD3/28 stimulation, more in CD8⁺ than CD4⁺ T cells, and interestingly, was robustly induced in a time-dependent manner in the absence of CD3/28 activation on CD8⁺ T cells (Fig. 1A, 1B). To assess coexpression patterns of SLAMF7 with LAG3, Tim3, PD-1, and CTLA-4, we used SPICE (34) and observed that SLAMF7 is highly coexpressed in cells expressing any combination of exhaustion markers and is consistently present on CD8⁺ T cells expressing multiple exhaustion markers (Fig. 1C). This effect was also observed on CD4⁺ T cells, but less pronounced, as these cells do not upregulate SLAMF7 expression to the levels seen on CD8⁺ T cells (Fig. 1D). Additionally, we repeated this assay with mouse splenocytes and observed the same pattern of coexpression (Supplemental Fig. 1D–H).

Because we identified that SLAMF7 is coexpressed with multiple exhaustion markers, we decided to examine if inducing SLAMF7 signaling in T cells impacts the expression level of T cell inhibitory receptors. We used a similar in vitro T cell stimulation assay as before, but instead activated the SLAMF7 receptor on T cells with an agonistic Ab via cross-linking, as previously described (23). We found that activating the SLAMF7 receptor on human T cells induced expression of PD-1, Tim3, and LAG3 in a time-dependent manner on CD8⁺ T cells (Fig. 2A, 2B) but to a lesser extent on CD4⁺ T cells (Fig. 2C, 2D). We also noted considerable numbers of T cells coexpressing multiple exhaustion markers by 3 and 6 d (Fig. 2B, 2D). This finding was not observed when cells were cultured in the absence of IL-2 (data not shown).

To gain a more in-depth understanding of the effects SLAMF7 activation had on T cell exhaustion, we employed unbiased, neural-network-based, computational clustering using FlowSOM (35) and dimensionality reduction using tSNE, which revealed six distinct T cell clusters in Mock and SLAMF7 activated samples following 6 d of stimulation (Fig. 2E). We identified two CD4⁺ T cell clusters (CD4 1–2) and four CD8⁺ T cell clusters (CD8 1–4) (Fig. 2E). Importantly, the CD4 2, CD8 2, and CD8 3 clusters were virtually absent in Mock stimulation and only appeared in SLAMF7 activated samples (Fig. 2F, 2G). Comparison of these clusters with clusters present during Mock stimulation revealed an inhibitory receptor and TF expression profile characteristic of exhausted/dysfunctional T cells (32, 36–38) (Fig. 2H, Supplemental Fig. 2A, 2D, 2E). Specifically the CD4 2 cluster had high expression of PD-1 and the epigenetic-reprogramming TFs YY1, Blimp1, and EZH2 (Fig. 2H). The two SLAMF7 CL-induced CD8⁺ T cell clusters shared an exhaustion phenotype but differed in expression of various markers, such as EZH2, Tim3, and YY1 (Fig. 2H). Together, this suggests that SLAMF7 activation on human T cells is able to induce expression of inhibitory receptors and TFs linked to control of T cell exhaustion, independent of TCR signaling.

Because of the fact that regulatory T cells (Tregs) are known to express high levels of certain exhaustion markers (39, 40), we wanted to rule out the possibility that SLAMF7 activation was merely inducing CD4⁺ and CD8⁺ Tregs. FOXP3 staining showed minor but consistent increases in FOXP3⁺ CD4⁺ T cells (Supplemental Fig. 2B) and CD8⁺ T cells (Supplemental Fig. 2C) following SLAMF7 activation, but these few cells did not account for the far greater percentage of cells expressing multiple exhaustion markers.

Although the T cells induced following SLAMF7 activation have the appearance of exhausted T cells, we wished to confirm this by performing a functional assay. Primary human T cells were stimulated as described above for 6 d and then restimulated with fresh CD3/28 Dynabeads for an additional 3 d. We then assessed the ability of these cells to produce proinflammatory cytokines following restimulation. We found that T cells first stimulated for 6 d with CD3/28 Dynabeads were unable to produce IL-2, TNF-⍺, and IFN-γ upon restimulation, consistent with their inhibitory marker profile, identifying them as bona fide exhausted T cells (Fig. 3A–D). In contrast to these cells, cells receiving Mock stimulation (only IL-2) first were able to produce significant quantities of all three cytokines (Fig. 3A–D). Interestingly, however, we found that cells first stimulated with SLAMF7 CL for 6 d were still able to produce cytokines equivalent to Mock samples (Fig. 3A–D). This indicates that although in vitro SLAMF7-activated T cells upregulate exhaustion markers, they still retain the ability to produce proinflammatory cytokines, suggesting they have not reached terminal exhaustion yet.

We next compared the inhibitory receptor and TF profile of SLAMF7-activated T cells to terminally exhausted T cells (CD3/28 stimulation) in an unbiased manner. FlowSOM clustering of Mock, SLAMF7 CL, and CD3/28 stimulated T cells (all stimulated for 6 d) revealed three CD4⁺ T cell clusters and five CD8⁺ T cell clusters (Fig. 3E). Each of the CD4 T cell clusters was strongly enriched in one of the stimulation conditions, with Mock being primarily composed of CD4 A, SLAMF7 CL composed mostly of CD4 B, and CD3/28 stimulation composed primarily of CD4 C (Fig. 3F, 3G). Similarly, Mock samples primarily contained the CD8 A cluster, whereas SLAMF7 CL samples contained mostly the CD8 B cluster, and CD3/28 stimulation contained mostly CD8 C and CD8 D clusters (Fig. 3F, 3G). Comparison of inhibitory receptors and TFs in these clusters revealed important distinctions between SLAMF7-activated T cell clusters and terminally exhausted T cell clusters. We found that exhausted T cell clusters (CD8 C, CD8 D, and CD4 C) have high expression of PD-1 and LAG-3, along with lower expression of the TFs YY1, EZH2, and Blimp1 (Fig. 3H). SLAMF7 CL samples still coexpressed multiple inhibitory receptors, but at a lower level than exhausted T cells, along with much stronger expression of all three TFs (Fig. 3H). Together, this reveals that although in vitro SLAMF7-activated T cells resemble exhausted T cells, they are not functionally (terminally) exhausted yet still have an increased susceptibility to multiple inhibitory signals because of their inhibitory receptor profile. As they show high expression of TFs involved in T cell exhaustion, in vitro SLAMF7-activated T cells may be progressing along the path to terminal exhaustion through a pathway independent from TCR activation.

To mechanistically understand how SLAMF7 signaling in CD8⁺ T cells modulates the expression levels of multiple inhibitory receptors and their associated TFs, we performed phosphoflow on human T cells using a panel containing common intermediary signaling proteins implicated in T cell responses (41). We found that SLAMF7 activation induced a progressive increase in STAT3 phosphorylation over time and a burst of STAT1 phosphorylation at 90 min postactivation, without affecting numerous other known T cell signaling intermediaries (Fig. 4A–H). Additionally, we observed that SLAMF7 activation induced CD69 expression and minimally altered CD44 expression on CD8⁺ T cells (Fig. 4I, 4J)—an important observation as CD69 has been previously shown to drive inhibitory receptor expression on T cells (42). SLAMF7 activation on CD8⁺ T cells also shifted the subset distribution away from naive and terminal differentiated effector memory re-expressing (CD45RA) cells toward effector memory cells (Supplemental Fig. 2F). Finally, as high levels of IL-2 can drive both T cell inhibitory receptor expression (43, 44) and Treg formation (43, 45), we assessed IL-2 production and found that SLAMF7 activation does not stimulate IL-2 production or secretion from T cells (Fig. 4K, 4L). Together, these findings tie STAT1 and STAT3 phosphorylation to SLAMF7 activation in CD8⁺ T cells, suggesting that these factors might be involved in SLAMF7-dependent T cell signaling.

To begin to identify if the SLAMF7 receptor functions to modulate T cell responses in cancer patients in vivo, we assessed coexpression of SLAMF7 mRNA with mRNA from various exhaustion markers in patient tumor samples. Using TCGA RNA-seq data, we first assessed all cancers to see which had the highest levels of CD8⁺ T cells and SLAMF7 expression (Supplemental Fig. 2G, 2H). We focused our further analyses on ccRCC because of its high levels of CD8⁺ T cells, large sample size, and the knowledge that immune dysfunction in ccRCC occurs primarily via T cell exhaustion (19). Correlation analyses between SLAMF7 with CTLA-4, LAG3, and PD-1 revealed very strong coexpression in ccRCC (Fig. 5A). Coexpression was weaker with Tim3 (Fig. 5A), an effect possibly due to the expression of Tim3 on various myeloid cell subsets (46). Importantly, stratifying ccRCC patients into the highest and lowest SLAMF7-expressing quartiles revealed reduced disease-specific survival in patients with high expression of SLAMF7 (Fig. 5B).

To further determine if SLAMF7 expression in ccRCC was linked to T cell exhaustion and dysfunction, we used the TIDE tool (19). TIDE analysis of a separate, smaller cohort of ccRCC patients confirmed impaired survival in patients with high SLAMF7 expression (Fig. 5C). It also showed that high levels of CTLs are moderately protective, except when SLAMF7 expression is high, in which case there is reduced survival (Fig. 5C). Analysis of other cancers in the TCGA revealed this phenomenon was not conserved (Supplemental Fig. 2I–L). In cancers in which immune dysfunction is not primarily occurring via T cell exhaustion, such as ovarian cancer (Fig. 5D–F) and squamous cell lung carcinoma (Supplemental Fig. 2K) (19), this trend disappears. Together, these results suggest that high SLAMF7 mRNA expression is linked to poor survival, only in cancers dominated by T cell exhaustion.

Although the correlations and effects on patient survival noted using TCGA data are informative, it does not address immune cell–specific expression patterns of SLAMF7 in the ccRCC TME. To address this, we reanalyzed a previously published CyTOF dataset of ccRCC patient tumors, which contained expression data for SLAMF7 (15). As predicted by CIBERSORT analysis of bulk RNA-seq data (Supplemental Fig. 2G), the ccRCC TME is composed primarily of T cells, followed by myeloid cells (Fig. 6A) (15). The SLAMF7⁺ compartment largely mimics that seen in the periphery, with major contributions from plasma cells and NK cells (Fig. 6A). Looking at SLAMF7 expression on major immune cell types, as well as various predetermined TAM subsets (15), we observed considerable heterogeneity in SLAMF7 expression on numerous immune cell types across patients (Fig. 6B, Supplemental Fig. 3A). In particular, NK cells, plasmacytoid dendritic cells, and various TAM subsets show the greatest heterogeneity (Fig. 6B).

To begin to identify which SLAMF7-expressing cell types in the TME might be responsible for activating SLAMF7 on T cells and inducing exhaustion, we performed a correlation analysis of various SLAMF7⁺ immune cells to all of the predetermined T cell subsets (15) (Fig. 6C). We found strikingly strong positive correlations between subsets of exhausted T cells (T-0, T-1, T-7, T-16, and T-18) (15) and SLAMF7⁺: total immune cells, total myeloid cells, M-0 TAMs, M-1 TAMs, M-2 TAMs, M-3 TAMs, and M-10 TAMs (Fig. 6C, Supplemental Table I). Importantly the “exhausted” T cell subsets were labeled as such based on high expression of inhibitory receptors and were never functionally characterized (15). We also noted relatively strong positive correlations of CD4⁺ Tregs (T-6) (15) to SLAMF7⁺ total immune cells (Spearman r = 0.3) and SLAMF7⁺ total myeloid cells (Spearman r = 0.38) (Fig. 6C). Furthermore, we observed strong negative correlations between the above-listed SLAMF7⁺ TAM subsets and subsets of effector memory T cells (T-3 and T-4) (15) (Fig. 6C, Supplemental Table I). It is recognized that TAMs are responsible for driving TILs toward exhaustion phenotypes (47). Our results in this study suggest that SLAMF7–SLAMF7 interactions between TAMs and TILs may be a novel mechanism through which tumors are able to create an inhospitable immune niche by the induction of T cell exhaustion, Treg formation, and shifts away from beneficial effector memory T cells (48).

Previous analysis of this dataset concluded that high CD38 expression marks pathogenic TAMs (15). We discovered a population of SLAMF7^highCD38^high TAMs present at varying levels between patients (Fig. 6D). Hypothesizing that this may be a pathogenic TAM subset linked to T cell exhaustion, we correlated the percentage of this cell population per total immune cells in the TME of each patient to T cell subsets in our correlation matrix. Levels of this unique TAM subset showed the strongest positive correlations with the above-listed exhausted T cell subsets and Tregs (Fig. 6C, Supplemental Table I). Further analysis of this TAM subset revealed it is composed primarily of M-5 and M-11 TAMS (Fig. 6D), both previously linked to T cell exhaustion and poor survival in this cohort (15). To identify which SLAMF7-expressing immune cell types were linked to poor survival, we performed a series of Kaplan-Meier analyses by splitting patients up based on high or low frequencies of various SLAMF7⁺ cell types. We found that high levels of SLAMF7⁺ total immune cells, CD8⁺ T cells, dendritic cells, and NK cells were not linked to poor survival (Fig. 6E). Patients with high levels of SLAMF7⁺ plasma cells (Fig. 6E) had poor survival; however, as nearly all plasma cells express SLAMF7, further analysis showed this effect was merely due to differences in numbers of plasma cells (Supplemental Fig. 3B). Unsurprisingly, we also noted that patients with high levels of Tregs (T-6) had markedly worse survival (Supplemental Fig. 3C). Most interestingly, we found that stratifying patients by high levels of the SLAMF7^highCD38^high TAM subset was the most effective parameter for identifying patients with poor survival (Fig. 6E). This effect was not just due to this population containing high levels of M-5 TAMs [previously linked to T cell exhaustion (15)], as patients with high levels of M-5 TAMs showed no decrease in survival (Supplemental Fig. 3D). Finally, patients in the high and low SLAMF7^highCD38^high TAM groups were otherwise balanced in available demographic and clinical parameters (Supplemental Fig. 3E). Together, these findings suggest that SLAMF7^highCD38^high TAMs may be driving T cell dysfunction and immune evasion in the ccRCC TME.

To validate our findings from human samples in vivo, we generated SLAMF7^−/− mice and assessed growth of B16-F10 tumors. We found tumor growth to be significantly attenuated in SLAMF7^−/− mice compared with WT mice (Fig. 7A). We also observed high coexpression of SLAMF7 with various inhibitory receptors on CD8⁺ TILs from WT mice (Fig. 7B, 7C). In comparing WT to SLAMF7^−/− TILs, we observed significantly fewer CD8⁺ T cells expressing multiple exhaustion markers (Fig. 7D) and individual markers (Fig. 7E) in SLAMF7^−/− mice. To validate if SLAMF7 was modulating levels of bona fide exhausted T cells, we compared expression of the exhaustion-defining TF, TOX (49–51). SLAMF7^−/− CD8⁺ intratumoral T cells expressed significantly fewer TOX⁺ cells (Fig. 7F) and PD-1⁺/TOX⁺ cells (Fig. 7G). Taking a more granular approach, we used the markers SLAMF6 (Ly108) and CD69 to track where CD8⁺ TILs from WT and SLAMF7^−/− mice were on the exhaustion developmental trajectory, as recently described (52). We found that CD8⁺ T cells from WT mice tended to be at either end of the exhaustion spectrum (Tex^prog1 and Tex^term) (Fig. 7H), whereas SLAMF7^−/− CD8⁺ T cells occupied the intermediate stages more often (Tex^prog2 and Tex^int) with significantly less terminally exhausted cells (Fig. 7H). We also observed SLAMF7 expression to be highest on the two progenitor stages (Tex^prog1 and Tex^prog2) (Fig. 7I), which together with the above findings, suggests that SLAMF7 signaling may help to propel CD8⁺ T cells through the exhaustion developmental pathway to terminal exhaustion.

Interestingly, we also noticed a significantly decreased CD8/CD4 T cell ratio in TILs from SLAMF7^−/− mice (Fig. 7J). Returning to our in vitro human data, we observed a corresponding reverse trend in the presence of SLAMF7 activation (Fig. 7K). Thus, although SLAMF7 appears to drive T cell exhaustion, it may also be playing a role in CD8⁺ T cell survival or ability to maintain tissue residency.

To determine if SLAMF7–SLAMF7 self-ligation between TAMs and tumor-infiltrating T cells induces T cell inhibitory receptor expression, we developed an ex vivo TAM–T cell coculture model (Fig. 8A). B16-F10 tumors were grown in WT and SLAMF7^−/− mice and TAMs were isolated from established tumors at day 25 and cocultured with CD8⁺ T cells from splenocytes of non–tumor-bearing mice (Fig. 8A). We observed that T cells cocultured with SLAMF7⁺ TAMs expressed multiple inhibitory receptors (Fig. 8B, 8C). Notably, T cells cultured with SLAMF7^−/− TAMs had significantly fewer cells positive for various exhaustion markers, with dramatic decreases in T cells coexpressing multiple markers appearing similar to CD8⁺ T cells cultured in the absence of TAMs (Fig. 8C). Importantly, this finding is specific to TAMs, as CD8⁺ T cells cocultured with bone marrow–derived macrophages from WT or SLAMF7^−/− mice do not show alterations in inhibitory receptor expression (Supplemental Fig. 3F, 3G). Additionally, comparison of TAMs from WT and SLAMF7^−/− mice revealed that they have similar expression of numerous markers, except for SLAMF7 (Fig. 8D). This finding suggests that direct SLAMF7–SLAMF7 interactions between TAMs and CD8⁺ T cells are sufficient to induce expression of multiple inhibitory receptors on T cells.

The impressive response rates and instances of complete remission following checkpoint inhibitor therapy have catapulted immunotherapy and T cell exhaustion to the forefront of cancer research. Although therapies targeting PD-1 (and PD-L1) and CTLA-4 have proven efficacious, there are still a great many patients who do not respond or develop resistance to checkpoint blockade, suggesting the need for a greater understanding of the biological processes regulating T cell dysfunction. To this end, a host of additional inhibitory checkpoint receptors has been identified and found to be expressed on exhausted T cells (2), with blocking Abs against these receptors now in phase I, II, and III clinical trials for various cancers (2). However, this approach of targeting additional T cell inhibitory receptors does not address the root problem, which is, what are the TCR-independent mechanisms responsible for inducing inhibitory receptor expression in the TME in the first place (3)? In this study, we have begun to answer this question in ccRCC and provide early evidence that SLAMF7–SLAMF7 self-ligation between CD8⁺ T cells and SLAMF7⁺ TAMs may be driving T cell inhibitory receptor expression and exhaustion.

Based on the temporal expression patterns we observe, it would seem that SLAMF7 is actually one of the first receptors expressed on activated CD8⁺ T cells following any sort of proinflammatory stimuli. This observation has been noted at the single-cell level (29) and lends credence to the idea that SLAMF7 may be one of the earliest countermeasures T cells use to rein in overactivation. It accomplishes this by providing an early TCR-independent signal to begin upregulating expression of multiple inhibitory receptors immediately upon initial activation or entrance into an inflammatory microenvironment. It also appears to reprogram these cells by inducing numerous TFs known to positively modulate T cell exhaustion programs. Notably, although SLAMF7 signaling on T cells does not impair their ability to produce cytokines, it does lend them highly susceptible to signaling from multiple inhibitory receptors, which would still result in the induction of nonfunctional T cells in the TME where these signals are abundant. This is an important subtlety to the biology of SLAMF7 signaling on T cells and suggests that interruption of this signaling could result in effective anti-tumor responses by T cells.

Already we have taken the first steps at understanding the underlying mechanism by finding that SLAMF7 activation on CD8⁺ T cells induces STAT1 and STAT3 phosphorylation. Phosphorylated STAT3 in intratumoral T cells has been shown to induce PD-1 expression (53), inhibit effector T cells (53), and promote expansion of Tregs (53), findings that we also observe following SLAMF7 activation on T cells. Furthermore, there is evidence that EZH2 is able to induce phosphorylation of STAT3 (54) and that CD38 is able to modulate EZH2 activity (55), suggesting SLAMF7 signaling in T cells may be inducing expression of inhibitory receptors via a CD38-STAT3-EZH2 axis.

Contrary to STAT3, the role of STAT1 in modulating CD8⁺ T cell signaling and exhaustion marker expression is less well understood. There is evidence showing the potential for pSTAT1 to induce expression of PD-1 in T cells and other immune cells (56). Further confirming that SLAMF7-mediated STAT1 phosphorylation is implicated in driving T cell exhaustion is the finding that Tex^term CD8⁺ T cells are enriched with open chromatin regions bound by STAT1 (52), and we find that in the absence of SLAMF7, there are fewer of these cells. Additionally, SLAMF7-mediated increases in the expression of the exhaustion-controlling TFs YY1 (38), Blimp1 (32, 36), and EZH2 (37, 38), are likely potential mechanisms regulating T cell exhaustion. The choice to study these TFs specifically was made not only because they are linked to T cell exhaustion but because both Blimp1 and YY1 have been found to control SLAMF7 expression (57, 58), which implies there may be reciprocal regulation involved. Interestingly, Tex^term CD8⁺ T cells are enriched in open chromatin regions bound by Blimp-1, we find that SLAMF7 activation induces Blimp-1, and in the absence of SLAMF7, there are fewer Tex^term cells, implicating another potential mechanism behind SLAMF7-specific effects on T cells.

The restricted B16-F10 tumor growth in SLAMF7^−/− mice can potentially be explained by the CD8⁺ T cell population shifts we see along the exhaustion development spectrum. Tex^int cells (which we find significantly more of in the tumors of SLAMF7^−/− mice) have been found to have high cytotoxic capacity and lower expression of inhibitory receptors compared with the other subsets recently described (52). Having more of these cells is likely to provide a beneficial anti-tumor response. How exactly SLAMF7 signaling drives cells away from Tex^int toward Tex^term and if SLAMF7 mediates changes in TOX expression remain to be discovered.

Our discovery that high expression of both CD38 and SLAMF7 on TAMs marks a unique subset strongly linked to T cell exhaustion and poor survival helps to mechanistically explain why a previous study found M-5 TAMs to be so strongly associated with exhausted CD8⁺ T cells (15). M-5 TAMs have the highest expression of SLAMF7 of any of the other TAM subsets that were identified, and the SLAMF7^highCD38^high TAM subset is composed of 70% M-5 TAMs (15). It will be important for future studies to determine if this unique TAM subset is present in the TME of other cancers, and if so, if it also induces T cell inhibitory receptor expression. Along the same lines, the results from our ex vivo TAM–T cell coculture model support this proposed mechanism but do not provide conclusive, in vivo proof.

An important concept to consider as future studies are conducted to both better understand the physiology of SLAMF7 signaling and attempt to therapeutically modulate it, is the sensitivity of SLAMF7 to different methods of activation or inhibition. By targeting SLAMF7 with a ligand attached to a solid surface (such as that of a cell culture plate or a cell membrane) as we have done in this study and previously (23), the receptor becomes activated, and downstream signaling is initiated. However, adding soluble SLAMF7 ligands to the system might result in blocking of the receptor and achieve effects directly opposite to that of receptor activation (23). This makes the study of SLAMF7 signaling very dependent on the experimental methods used and may explain why other studies (59) reached different conclusions from this study.

Our analysis of the roles played by SLAMF7 on both T cells and TAMs, incorporating a combination of multiparametric spectral cytometry and targeted analysis of existing datasets, provides early evidence implicating a novel immune evasion mechanism employed by tumors.

We thank the MSU Transgenic and Genome Editing Facility, Elena Demireva, and Huirong Xie for generating SLAMF7^−/− mice. We appreciate assistance performing spectral cytometry provided by Matthew Bernard and the MSU South Campus Flow Cytometry Core Facility. Thanks to Austen N. Grooms for assistance with IL-2 ELISA experiments.

The authors have no financial conflicts of interest.