CD8⁺PD-1^–ILT2⁺ T Cells Are an Intratumoral Cytotoxic Population Selectively Inhibited by the Immune-Checkpoint HLA-G

Mechanisms allowing for the evasion of cancer cells from immune surveillance can be targeted in order to restore the host's antitumor immune response. This has been universally accepted in light of the therapeutic efficacy of immune-checkpoint inhibitors. Those currently most advanced in their development target exhausted tumor-infiltrating lymphocytes (TIL), which are chronically stimulated T cells fallen into a state of anergy maintained by the engagement of inhibitory receptors, such as PD-1 or CTLA-4. PD-1 antibodies are effective in multiple cancer types, albeit only in a subset of patients, as exemplified by metastatic renal cell carcinoma (mRCC): the PD-1 antibody nivolumab (1) is effective in about one-fourth of patients with pretreated mRCC, and the combination of nivolumab with the CTLA-4 antibody ipilimumab in about half of the patients with treatment-naïve, intermediate- or high-risk mRCC (2). Although this illustrates the interest of targeting multiple checkpoints, not all patients benefit from immunotherapy. One explanation is that several checkpoints are active in a single tumor, as we showed in localized clear-cell renal cell carcinoma (ccRCC; ref. 3): in a series of 19 tumors, we observed that expression of PD-L1 was heterogeneous and could coexist with expression of the immune-checkpoint HLA-G.

HLA-G is a nonclassic MHC class I molecule primarily found at the maternal–fetal interface (4). It exerts direct inhibitory functions on B, T, and NK cells through its receptor ILT2 (immunoglobulin-like transcript 2, product of the gene LILRB1) and on monocytes/macrophages through receptors ILT2 and ILT4 (5), as well as indirect immunosuppressive effects through induction of regulatory T cells (6) or myeloid-derived suppressor cells (MDSC; ref. 7). HLA-G is frequently neoexpressed in immunotherapy-sensitive cancer types such as ccRCC (3, 8–10), melanoma (11–13), or non–small cell lung cancer (14, 15), especially in advanced-stage disease, and is associated with poor prognosis (13–15). These elements suggest a role in tumor escape from immune surveillance, which was confirmed by animal studies (7, 16). We previously showed that TILs expressing ILT2 were present in the tumor microenvironment (TME) of ccRCC (3).

ILT2 is an inhibitory receptor with high affinity for HLA-G and lower affinity for classic MHC class I ligands (17). In the peripheral blood, ILT2 is expressed by a subset of CD8⁺ T cells: typically about 20% in younger healthy subjects (18–21); however, the proportion of ILT2⁺ CD8⁺ T cells may rise to over 50% with age and chronic viral infections (22, 23). Previous reports have associated ILT2 expression by peripheral blood CD8⁺ T cells with a differentiated phenotype (CD28^–CD27^–CD57⁺, CCR7^–CD45RA⁺; refs. 21–25) and perforin expression (21, 25). However, thorough characterization of CD8⁺ILT2⁺ TILs is needed, and the impact of ILT2 on effector T-cell functions remains unclear (21, 23, 26).

Here, using transcriptomics and flow cytometry, we characterized peripheral blood and tumor-infiltrating ILT2⁺CD8⁺ T cells as a differentiated cytotoxic population distinct from PD-1 expressing, exhausted T cells. Using ex vivo assays, we demonstrated that their effector functions were directly inhibited by target-expressed HLA-G through ILT2. Finally, we discuss HLA-G tumor expression as a mechanism of resistance to current cancer immunotherapy.

T cells used for our experiments were isolated from cancer patients as well as control patients. Cancer patients were either patients who underwent transurethral resection for non–muscle-invasive bladder cancer (NMIBC patients) or patients who underwent nephrectomy for localized ccRCC (ccRCC patients) at our center (Urology Department, Saint-Louis Hospital, Paris, France); control patients were patients ages over 40 with no personal history of cancer, and admitted for planned, noncarcinologic, surgery. The study design was approved by the ethics committee of the St. Louis Hospital and was conducted according to the criteria set by the declaration of Helsinki. Each subject signed an informed consent before participating in the study.

Blood sampling was performed upon admission to the Urology Department prior to surgery. After sampling, peripheral blood mononuclear cells (PBMC) were isolated using Ficoll gradient (Ficoll-Paque, LifeSciences) as per the manufacturer's instructions and stored at −150°C.

For the sorting of CD8⁺ILT2^– and CD8⁺ILT2⁺ subpopulations, PBMCs were labeled with antibodies specific for CD3, CD4, CD8, CD19, and ILT2. The CD8⁺ subpopulations were then sorted according to ILT2 expression on a BD FacsAria II cell sorter, then immediately lysed in RNA WIZ reagent (Ambion), and total RNA was extracted using the RNeasy micro kit (Qiagen). Samples were amplified and labeled according to the manufacturer's protocol for hybridization to Affymetrix Human Gene 2.0 ST arrays. Sample preparation, hybridization, washing, staining, scanning, and quality control were performed by the Institut Curie Genomics core facility, Paris, France.

We extracted the expression matrix of the differentially expressed genes between ILT2⁺CD8⁺ and ILT2^–CD8⁺ T cells. We then imported the matrix on Cytoscape software version 3.5.1. Analysis was performed in parallel for the ILT2⁺ and the ILT2^– upregulated gene expression matrix. Network inference was performed using ARACNe application. After selecting genes of the output network from ARACNe, we utilized the ClueGO and CluePedia Applications to determine pathway enrichment. Public data sets of Gene Ontology (GO)—Biological process-GOA, ImmuneSystemProcess-GOA, Molecular Function-GOA—KEGG, Reactome, and WikiPathways were used. Only pathways with a “Bonferroni step down” corrected P value below 0.05 were kept.

TILs were extracted from ccRCC specimens. Fresh tumor samples were selected on nephrectomy specimens by a pathologist, rinsed with phosphate-buffered saline and placed in RPMI culture medium for 30 minutes. Each tumor sample was then subjected to mechanical disruption as previously described (3). Extemporaneous counting and phenotyping of TILs was performed using a MACSQuant 10 flow cytometer (Miltenyi Biotec), and cells were then stored at −150°C for further phenotyping and functional assays. To rule out contamination with peripheral blood cells, a simultaneous flow cytometry analysis of PBMCs was performed.

Formalin-fixed paraffin-embedded samples from the same ccRCC specimens were analyzed by IHC for tumor expression of HLA-G (clone 4H84) and PD-L1 (clone E1L3N); immune infiltration was detected by immunohistostaining of CD3 (polyclonal rabbit) and CD8 (clone C8/144B) as previously described (3). Percentages of PD-L1 positivity in tumor cells and tumor-infiltrating cells were reported, and HLA-G expression was estimated by the percentage of tumor cells positive for membrane staining.

Monoclonal antibodies (mAb) with the following specificities were used for cell-surface staining and analysis: from Miltenyi Biotec: CD3-PerCP, CD3-VioBlue, CD4-PE-Vio770, CD4-VioBright-FITC, CD4-APC-Vio770, CD45RA-VioGreen, CD8-APC-Vio770, CD8-PerCP-Vio770, CD8-VioGreen, CD11b-FITC, CD38-PE-Vio770, CD45RA-VioGreen, CD57-VioBlue, CD127-PE-Vio770, CCR7-APC, IFNγ-FITC, KLRG1-FITC, NKp80-APC, Perforin-VioBlue, Granzyme B-PE, PD-1 PE-Vio770; from Beckman Coulter: CD27-PE, CD62L-FITC, CD69-PE, CD127-PE; from BD Pharmingen: CD28-FITC, CD137-PE; from eBioscience: ILT2-PE or ILT2-APC (clone HP-F1), Tim-3-FITC; from BioLegend: PD-1-BV421.

Intracellular staining was performed using the Inside Stain kit (Miltenyi Biotec), according to the manufacturer's instructions. Acquisition was made on a MACSQuant 10 flow cytometer (Miltenyi Biotec); analysis was performed using the MACSQuantify software (Miltenyi Biotec) and FlowJo software using appropriate isotypic controls to account for nonspecific background staining of the cell subsets of interest.

For functional studies, the monocytic cell line THP-1 (ATCC) transduced or not to express membrane-bound HLA-G1 (THP1-HLA-G1) was used as target cells. Details on the lentiviral vector and transduction protocol used for transduction can be found in ref. 27. Cell-surface expression of HLA-G on THP-1-HLA-G1 cells was confirmed, using a PE-conjugated HLA-G mAb (clone MEM-G9, Exbio), and lack of ILT2 cell-surface expression was confirmed, using a PE-conjugated ILT2 mAb (clone HP-F1, eBioscience) by flow cytometry (Supplementary Fig. S1).

Cell lines used in these experiments were tested monthly for Mycoplasma contamination. Original THP-1 cells were obtained in 2005 and vials from working banks were not maintained in culture more than 2 months. No reauthentication of the cells was performed in the past year.

THP1 or THP1-HLA-G1 cells were placed in a 96-well culture plate in RPMI culture medium (Sigma) supplemented with 10% fetal calf serum (Sigma), l-glutamine (300 μg/mL), gentamicine (50 μg/mL), and amphothericin B (2.5 μg/mL, Gibco), and containing phorbol 12-myristate 13-acetate (PMA, 50 ng/mL, Sigma). Cells were cultured with PMA for 72 hours in order to obtain a confluent macrophage layer (mTHP1/mTHP1-HLA-G1). Then, cells were coated for 15 minutes with CD3 mAb (clone OKT3, Orthoclone). Coating concentrations of OKT3 for CD107a and interferon-gamma (IFNγ) secretion assays were 20 and 10 ng/mL, respectively. PBMCs from NMIBC patients were incubated for 20 minutes at 37°C with 20 μg/mL of a blocking ILT2 antibody (clone GHI/75, BioLegend) or a control antibody (28, 29). PBMCs were then added to the OKT3-coated mTHP1/mTHP1-HLA-G1 target cells in culture medium supplemented with monensin and brefeldin A (Protein Transport Inhibitor Cocktail, eBioscience) in the presence of PE-conjugated CD107a antibody (clone H4A3, BioLegend) or isotype control.

After 4 hours of coincubation at 37°C, cells were washed and stained for flow cytometry analysis. For IFNγ secretion assays, intracellular staining was then performed using the Inside Stain kit (Miltenyi Biotec) according to the manufacturer's instructions.

The cytolytic degranulation and IFNγ secretion of CD3⁺CD8⁺ T-cell subsets were evaluated using the percentage of CD107a^– and IFNγ⁺ cells respectively. CD8⁺ T-cell subsets were defined by the expression of ILT2 and CD57. Because of interference between the ILT2 mAb GHI/75 used for blocking and the ILT2 mAb HP-F1 used for staining, we used the CD57^high subset, as a surrogate population for ILT2⁺ T cells in blocking experiments. Indeed, as shown in this report, this subset consistently comprised more than 75% ILT2⁺ T cells in selected patients, compared with the CD57^– subset, which was more than 90% ILT2^– T cells.

THP1 or THP1-HLA-G1 cells were differentiated into macrophage target cells as described above. TILs were incubated with the ILT2 mAb or isotype control and added to the target cells in culture medium supplemented with monensin and brefeldin A (Protein Transport Inhibitor Cocktail, eBioscience) and PE-conjugated anti-CD107a or control isotype as described above. TILs stimulated by PMA and ionomycin (cell stimulation cocktail, eBioscience) served as positive controls, and TILs incubated with non-OKT3–coated target cells served as negative controls. After a 4-hour coincubation, cells were washed and stained for flow cytometry analysis as described above. Results from a preliminary phenotypical characterization of TILs served to select the best surrogate markers for the CD8⁺ILT2⁺ TIL population in each sample (either CD57^high or CD45RA⁺ TILs).

For transcriptomic experiments, gene-expression data were normalized using RMA algorithm on custom Brainarray CDF. We selected the 500 most variant genes by IQR (30) to perform unsupervised analyses. Differential subgroups were identified by hierarchical clustering using Pearson correlation metric and Ward distance. Differentially expressed genes between two groups were defined using limma R-package (P ≤ 0.05 adjusted with Benjamini and Hochberg and |log fold change|> 1.5). All these analyses were performed with R software environment.

For flow cytometry phenotyping, comparisons between T-cell subsets were made using t tests paired by sample.

For functional studies, differences in terms of cytolytic degranulation or IFNγ production between different T-cell subsets from the same PBMC or TIL samples were evaluated with t tests paired by sample. The impact of HLA-G expression by the target cells and of ILT2 blocking on a specific T-cell subset was evaluated using unpaired t tests.

In this study, two types of cancers, NMIBC and ccRCC, were investigated. First, for transcriptomic studies on PBMC, we used NMIBC, because the pathologic relevance of CD8⁺ILT2⁺ T cells had already been demonstrated for this context (3). Second, for functional analysis of intratumoral CD8⁺ILT2⁺ T cells, we used ccRCC tumors that are larger and provided the required cell numbers.

First, we performed transcriptomic analysis to identify key features of the ILT2⁺ subpopulation. Peripheral blood CD8⁺ T cells from 4 NMIBC patients were sorted with respect to ILT2 cell-surface expression and transcriptome analysis was performed on the CD8⁺ILT2⁺ versus CD8⁺ILT2^– T-cell subpopulations. Hierarchical clustering demonstrated that these two cell populations could easily be distinguished (Fig. 1A). There were 71 genes specifically upregulated in the CD8⁺ILT2⁺ T-cell population, whereas 113 upregulated genes characterized their CD8⁺ILT2^– counterparts (Fig. 1B). A list of these genes is provided in Supplementary Table S1. Functional network inference revealed that the genes upregulated in the CD8⁺ILT2⁺ population belonged to effector function– and effector function regulation–related categories (Fig. 1C), centered around NK cell–type categories (natural killer cell–mediated cytotoxicity/immunity) and immune regulation (immune response–inhibiting cell-surface receptor signaling pathways/regulation of αβ T-cell activation/immunoregulation interactions between a lymphoid and a nonlymphoid cell). Comparatively, the functional network inference of downregulated genes in ILT2⁺ versus ILT2^–CD8⁺ T-cell populations was less restricted, but nevertheless centered around function-related categories (e.g., Cytokine–cytokine receptor interaction/regulation of cell–cell adhesion, positive regulation of chemotaxis), and differentiation (T-cell differentiation/myeloid leukocyte differentiation, binding of TCF/LEF:CTNNB1 to target gene promoters).

The aim of this analysis was originally to characterize the CD8⁺ILT2⁺ T-cell subpopulation using phenotyping in the context of an antitumor function. Thus, we next focused on cell-surface–expressed markers and function- or differentiation-related genes. Out of the 184 genes found to be differently expressed between CD8⁺ILT2⁺ and CD8⁺ILT2^– subsets, 46 matched these criteria, including 19 upregulated and 27 downregulated in CD8⁺ILT2⁺ T cells (Fig. 1D). As suggested by the functional inference network data, upregulated genes in the CD8⁺ILT2⁺ population were mostly involved in NK and/or cytotoxic T-cell functions. Cytotoxic T/NK-related genes were TBX21 (T-bet), GNLY (granulysin), GZMB and GZMH (granzymes B and H), ITGAM (CD11b), and FCRL6. Genes usually associated with NK cells were NCAM1 (CD56), cytotoxicity-triggering receptors NCR1 (NKp46), KLRF1 (NKp80), and CD244, lectin-like receptors KLRK1, KLRC3, and KLRC4 (NKG2D, NKG2E, and NKG2F), and immunoglobulin-like receptors KIR2DL1, KIR2DL3, and KIR3DL2.

Downregulated genes in the CD8⁺ILT2⁺ population were more diverse, but two categories stood out: genes involved in T-cell stimulation/costimulation (CD28, CD28H (TMIGD2), CD27, TNFRSF8, CD40LG, TESPA1), and genes associated with T-cell differentiation (CD28, CD27, CCR7, CCR4, SELL, IL7R, TCF7).

These data indicated that ILT2⁺ CD8⁺ T cells were likely to be highly cytotoxic, antigen-experienced T cells expressing NK cell-surface molecules (activating and inhibitory).

Frozen PBMCs from 4 NMIBC patients, 4 ccRCC patients, and 2 control patients were used for phenotypical validation of transcriptomics findings using flow cytometry. No differences were observed across these various clinical settings regarding the phenotype of CD8⁺ILT2⁺ T cells.

In accordance with transcriptomics findings, flow cytometry showed significantly lower expression of surface markers CD28, CD27, and CD127 on ILT2⁺ CD8⁺ T cells (Fig. 2A and B). These cells also frequently expressed CD57, and virtually all ILT2⁺ cells expressed KLRG1, perforin, and granzyme B. Conversely, the proportion of ILT2⁺ cells was significantly higher in CD28^–, CD57⁺, or KLRG1⁺ subpopulations (Fig. 3A). Most NKp80⁺ and perforin⁺ CD8⁺ T cells also expressed ILT2. Progressive acquisition of ILT2 during CD8⁺ T-cell differentiation was apparent through the rising proportion of ILT2⁺ cells between the CD27^high, CD27^low, and CD27^– subpopulations, in that order. ILT2⁺ cells were consistently CCR7^– (Fig. 3B and C), consistent with antigen-experienced effector-memory CD8⁺ T cells (31); ILT2 expression was most frequent in the CCR7^–CD45RA⁺ subset of CD8⁺ T cells (TEMRA or effector T cells; Fig. 3C).

After having characterized their phenotype, we studied the cytotoxicity of CD8⁺ILT2⁺ T cells through the assessment of IFNγ secretion and cytolytic degranulation of peripheral blood T cells after ex vivo coincubation with anti-CD3–coated target cells.

As the above results show, ILT2⁺ CD8⁺ T cells consistently showed higher degranulation than their ILT2^– counterparts (Fig. 4A and B). Expression of HLA-G1 by the target cells significantly reduced degranulation of CD8⁺ILT2⁺ T cells by a mean 33% (range, 18%–47%), whereas CD8⁺ILT2^– cells were unaffected (Fig. 4B). In restoration experiments, when ILT2 staining was impossible because of previous ILT2 blocking, this could also be observed in surrogate T-cell subpopulations defined by CD57 expression: expression of HLA-G1 by the target cells reduced degranulation in the ILT2-enriched CD8⁺CD57^high subpopulation by a mean 34% (range, 21%–43%), whereas the ILT2^–CD8⁺CD57^– T-cell subpopulation was unaffected (Fig. 4C). ILT2 blocking significantly increased the degranulation of CD8⁺CD57^high T cells in the presence of HLA-G1, reversing HLA-G1–associated inhibition by a mean 86% (range, 63%–112%), whereas CD8⁺CD57^– T cells were unaffected.

Similarly, CD8⁺ILT2⁺ T cells showed higher IFNγ secretion than their ILT2^– counterparts (Fig. 4D and E), which was reduced by a mean 30% (range, 13%–56%) in the presence of HLA-G1, whereas CD8⁺ILT2^– cells were unaffected (Fig. 4E). Expression of HLA-G1 by the target cells reduced IFNγ secretion by CD8⁺CD57^high T cells by a mean 40% (range, 24%–56%), CD8⁺CD57^– T cells were typically unaffected (Fig. 4F). ILT2 blocking significantly restored IFNγ secretion by CD8⁺CD57^high T cells in the presence of HLA-G1 by a mean 113% (range, 57%–177%) whereas CD8⁺CD57^– T cells were unaffected (Fig. 4F).

In Fig. 5, expression of HLA-G by the tumor cells and CD8⁺ or CD3⁺ infiltrates is shown. As was published earlier (3), it can be seen that HLA-G and infiltration were heterogeneous. For instance in patient #24, HLA-G was highly expressed; CD3⁺ T-cell infiltrate was present, but not CD8⁺ T cells, suggesting a CD4⁺ T-cell infiltration. One area of tumor from patient #25 showed a diffuse HLA-G positivity and diffuse CD3⁺CD8⁺ infiltrate, whereas another tumor area showed CD3⁺CD8⁺ infiltration but no HLA-G positivity. Finally, in tumor from patient #27, focal HLA-G expression and focal CD3⁺CD8⁺ T-cell infiltrate next to each other can be observed. No correlation was determined between HLA-G expression and T-cell infiltration.

After examining the peripheral blood T cells of patients, tumor-infiltrating cells from 9 ccRCC specimens were investigated. Expression of immune-checkpoint ligands HLA-G and PD-L1 in these tumors as well as the distribution of PD-1⁺ and ILT2⁺ TILs are summarized in Table 1. Blood contamination was deemed minimal in all TIL samples, as shown by clear phenotypical discrepancies between PBMCs and TILs, such as the absence of specific subpopulations in the tumor, e.g., CD8⁺CCR7⁺ T cells (Supplementary Fig. S2).

As we previously reported (3), the proportions of CD8⁺PD-1⁺ and CD8⁺ILT2⁺ TILs varied among specimens. Most strikingly, we observed that PD-1 expression and ILT2 expression by tumor-infiltrating CD8⁺ T cells defined two subpopulations that were mutually exclusive. The individual results obtained for all 9 patients are shown in Fig. 6. As seen in all patients but patient #29, PD-1⁺ infiltrating CD8⁺ T cells did not express ILT2 (ILT2^–) and ILT2⁺ infiltrating CD8⁺ T cells did not express PD-1 (PD-1^–). As is also indicated in Table 1, PD-1⁺ILT2^– and PD-1^–ILT2⁺ CD8⁺ T cells were present in varying proportions: in patients #24 and #25, PD-1^–ILT2⁺ T cells were almost absent, whereas in patients #23, #26, #27, and #42, PD-1⁺ILT2^– and PD-1^– ILT2⁺ populations were present in comparable proportions. Overall, PD-1⁺ T cells were generally more numerous than ILT2⁺ T cells. PD-1⁺ILT2⁺ double-positive CD8⁺ T cells were also detected. One such rare situation is illustrated in Fig. 6, patient #29.

CD8⁺ILT2⁺ TILs harbored a phenotype similar to that of their peripheral blood counterparts, most of them displaying an effector phenotype (CCR7^–CD45RA⁺; Fig. 7A), highly expressing CD57, and being strictly KLRG1⁺CD28^–CD27^– (Fig. 7B). Perforin expression was again a specific feature of ILT2⁺ cells (Fig. 7B). Conversely, CD8⁺PD-1⁺ TILs displayed a less mature phenotype with negative-to-low expression of CD57, CD28, and CD27, and no expression of KLRG1 or perforin (Fig. 7B); they typically exhibited an effector-memory phenotype (CCR7^–CD45RA^–; Fig. 7A). In accordance with previous large-scale studies (32), expression of exhaustion-associated markers Tim-3, CD38, CD69, and 4-1BB/CD137 was seen only on PD-1⁺ cells and never on ILT2⁺ cells (Fig. 7B). Conversely, most cells expressing CD11b were ILT2⁺ (Fig. 7B). Flow cytometry figures for the most characteristic markers (CD57, KLRG1, and perforin) in PD-1⁺ and ILT2⁺ TILs from 6 patients are shown in Supplementary Fig. S3.

Finally, to compare with the results obtained on PBMCs, the effector functions of tumor-infiltrating CD8⁺ T cells were investigated. Again, cytolytic degranulation after ex vivo coincubation with anti-CD3–coated target cells was higher in CD8⁺ILT2⁺ TILs than in CD8⁺PD-1⁺ TILs (Fig. 8A). HLA-G1 expression by the target cells reduced the degranulation of CD8⁺ILT2⁺ TILs by a mean 32% (range, 17%–44%), whereas no such effect was observed with CD8⁺PD-1⁺ TILs. When considering ILT2-enriched surrogate TIL subsets selected for each individual sample (namely, CD8⁺PD-1^–CD45RA⁺ or CD8⁺PD-1^–CD57^high cells), expression of HLA-G1 by the target cells reduced degranulation by a mean 27% (range, 25%–28%). Although this was not apparent in our preliminary assays with PBMCs, ILT2 blocking in the absence of HLA-G1 resulted in an increase in the degranulation of these surrogate ILT2⁺ TILs, consistent with the removal of an HLA-G-independent lower-level inhibition due to the engagement of ILT2 by target-expressed classic MHC class I molecules. Regardless, ILT2 blocking fully counteracted HLA-G1–mediated inhibition, as shown by a mean 197% reversion (range, 127%–318%).

As with cytolytic degranulation, IFNγ secretion by CD8⁺ILT2⁺ TILs was higher than that of CD8⁺PD-1⁺ TILs (Fig. 8B). HLA-G1 expression by the target cells reduced the degranulation of CD8⁺ILT2⁺ TILs by a mean 34% (range, 28%–42%) and of surrogate ILT2⁺ populations by a mean 34% (range, 19%–46%). Again, ILT2 blocking resulted in an increase in IFNγ secretion by surrogate ILT2⁺ TIL subsets in the absence of HLA-G1 as well as in full reversion of HLA-G1–mediated inhibition (mean 243% reversion; range, 222%–256%).

Since we described the presence of CD8⁺ILT2⁺ TILs in the stroma of ccRCC (3), their functional nature was a crucial question. The phenotypical and functional studies described here definitely associate ILT2⁺ TILs and their peripheral blood counterparts with late differentiation and strong cytotoxic capacity. A most striking finding was the clear dichotomy between ILT2⁺ and PD-1⁺ TILs in the tumor microenvironment, which was especially relevant considering that the latter are the prime target of current cancer immunotherapy despite being less cytotoxic and sometimes less numerous than the former. Such separation between PD-1⁺ and ILT2⁺ T-cell subsets is likely to be tumor-specific because substantial overlap in expression of both checkpoint molecules has previously been found in virus-specific T cells (33). Although upon viral infection, continuous antigen stimulation may favor the dynamics of T-cell differentiation and especially the intermediate PD-1⁺ ILT2⁺ stage (34), intratumoral T cells evolve in an immune-inhibitory microenvironment lacking activating signaling that may maintain T cells in the PD1⁺ or ILT2⁺ stages.

Our transcriptomics and flow cytometry analyses provide a definitive characterization of both peripheral blood and tumor-infiltrating CD8⁺ILT2⁺ T cells as late-differentiated (CD28^–CD27^–CD57⁺) T cells in accordance with previous reports (19–22), ILT2 expression being a specific feature of effector-memory (CCR7^–) cells, even more prevalent in terminally differentiated (CCR7^–CD45RA⁺) TEMRA/effector T cells (35). The CD8⁺ILT2⁺ T-cell subpopulation overlaps with those defined by translineage cell-surface markers of cytotoxicity, KLRG1 (36) and NKp80 (37), and indeed these cells display high expression of perforin and granzyme B at the transcript and protein levels. Consistent with these cytotoxic and innate-like phenotypical traits, CD8⁺ILT2⁺ T cells exhibit the highest degranulation upon CD3 engagement, appearing to be “ready-to-kill” effector cells. The cytotoxic functions of CD8⁺ T cells expressing the costimulatory receptor NKp80 have also been shown to be triggered without engagement of the TCR/CD3 complex (38) through NK cell–like costimulation. Whether such T cell–borne antigen-independent cytotoxicity is a component of antitumor immunity, possibly kept in check through ILT2, remains to be determined.

It has long been known that the HLA-G/ILT2 could inhibit CD8⁺ T-cell functions (39). However, the historical demonstration that HLA-G could inhibit T-cell cytotoxicity was made using a cultured virus-specific cytotoxic T lymphoid (CTL) clone. Because the culture and cloning of T cells alters their properties and phenotype, including ILT2 expression (21), this demonstration left some uncertainties as to the inhibitory potency of the HLA-G/ILT2 interaction in CD8⁺ T cells in vivo.

Thus, we chose to perform CD3-mediated polyclonal activation of uncultured PBMCs or TILs ex vivo, which allowed us to observe the immediate effects of HLA-G on its native T-cell targets.

In contrast, many previous studies of the impact of ILT2 on T-cell functions did not test its engagement by HLA-G (21, 23, 26), and we believe that in such conditions the blocking of ILT2 could only lift the lower-level inhibition consecutive to its engagement by classic MHC class I molecules (17). We only inconsistently observed such an effect with uncultured PBMCs, as already reported (23), but we did observe it with TILs here. We hypothesize that TILs within the tumor microenvironment, although having been exposed to prolonged stimulation, had also been rendered more sensitive to classic MHC class I–mediated inhibition. This would reconcile our observations with the enhancement of in vitro cytotoxicity through ILT2 blocking, which was observed in functional assays that used T cells expanded through repeated in vitro stimulation (21, 25). In any case, our experiments clearly showed evidence of superior ILT2-mediated inhibition of T-cell cytotoxicity when HLA-G was expressed by the target cells.

Apart from a single observation of circulating melanoma-specific CD8⁺ILT2⁺ T cells (19), a possible role for CD8⁺ILT2⁺ T cells in antitumor response has been overlooked. We previously observed that CD8⁺ILT2⁺ T cells were abundant in the microenvironment of ccRCC (3), and we show here that they are effective cytotoxic effectors, functionally and phenotypically distinct from exhausted T cells, and readily inhibited by target-borne HLA-G.

Although HLA-G expression by normal adult tissue is extremely restricted, it is frequently neoexpressed in immunotherapy-sensitive cancer types (40), including ccRCC (3, 10). Our findings suggest that tumor-expressed HLA-G, through ILT2, could be a major inhibitory checkpoint for the effector functions of the naturally occurring CD8⁺ILT2⁺ cytotoxic T cells in the tumor microenvironment. Apart from our own observations, the relevance of this phenomenon may be inferred from a mass cytometry study of 77 ccRCC cases by Chevrier and colleagues, who described 22 TIL clusters as well as their frequent associations (32). Based on our flow cytometry data, we postulate that ILT2⁺ effector CD8⁺ TILs may be found in several of their PD-1^– clusters, namely, the T-11 (CD11b⁺), T-14 (CD45RA⁺), and T-4 (PD-1^– with no other positive discriminatory marker) clusters. These clusters segregate together and may be more represented than PD-1⁺ clusters in about one-third of ccRCC cases: it still remains to be clarified whether this constitutes an “effector-infiltrated” subset of ccRCCs in which targeting HLA-G may be of particular interest.

Therapeutic targeting of the PD-1/PD-L1 checkpoint is the backbone of modern cancer immunotherapy. In metastatic renal cell carcinoma, anti–PD-1 monotherapy yields clinical responses in about one-fourth of patients (1), despite frequent PD-1 expression on TILs. We show here that naturally occurring CD8⁺PD-1⁺ TILs display an incompletely differentiated phenotype, less mature than that of CD8⁺ILT2⁺ TILs, and a much lower expression of perforin. We postulate that efficient PD-1 blockade and release from exhaustion may allow CD8⁺PD-1⁺ TILs to undergo proliferation and full differentiation toward ILT2⁺ effectors, as suggested by Choueiri and colleagues (41), who performed transcriptomics analyses on ccRCC biopsy specimens from patients before and on treatment with nivolumab. Their data showed that anti–PD-1 treatment led to higher expression of CD3 and CD8 transcripts, suggestive of CD8⁺ T-cell proliferation, as well as higher expression of PRF1, GZMB, or IFNG, suggestive of full cytotoxic differentiation. Among other genes overexpressed under PD-1 blockade were KLRG1 and LILRB1, the ILT2 gene. Regarding KLRG1, Choueiri and colleagues suggested that PD-1 blockade may lead to NK cell proliferation because KLRG1 is primarily known to be expressed by NK cells; LILRB1, which they classify as a myeloid cell gene, is not discussed. When complemented by our present observations, these results suggest the expansion of differentiated cytotoxic CD8⁺KLRG1⁺ILT2⁺ TILs following PD-1 blockade. From this perspective, we hypothesize that HLA-G expression by cancer cells could be a mechanism of resistance to PD-1/PD-L1 blockade in ccRCC patients, because these newly generated effectors would eventually be inhibited by tumor-borne HLA-G. Studies are ongoing at our center to confirm this hypothesis and suggest that combined PD-1/PD-L1 and HLA-G/ILT2 blockade in selected patients may be beneficial.

In conclusion, ILT2 expression is a key feature of differentiated cytotoxic CD8⁺ T cells, rendering them susceptible to HLA-G–mediated inhibition. ILT2⁺, HLA-G–sensitive, effector T cells are present in the tumor microenvironment of ccRCC, a solid cancer type in which HLA-G is frequently neoexpressed. This suggests that direct effector inhibition through the HLA-G/ILT2 checkpoint could be a mechanism of tumor escape from immune surveillance as well as of resistance to current checkpoint blockade therapy. This strengthens the rationale for targeting HLA-G concomitantly with other immune checkpoints in selected cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: C. Dumont, F. Desgrandchamps, E.D. Carosella, N. Rouas-Freiss, J. LeMaoult

Development of methodology: C. Dumont, A. Jacquier, J. Verine, C.-L. Wu, T.-M. Hung, J. LeMaoult

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Dumont, A. Jacquier, J. Verine, A. Goujon, F. Desgrandchamps

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Dumont, A. Jacquier, J. Verine, F. Noel, S. Culine, N. Rouas-Freiss, J. LeMaoult

Writing, review, and/or revision of the manuscript: C. Dumont, A. Jacquier, J. Verine, S. Culine, N. Rouas-Freiss, J. LeMaoult

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Dumont, J. Verine, A. Goujon

Study supervision: C. Dumont, E.D. Carosella, N. Rouas-Freiss, J. LeMaoult

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.