Plasma CD27, a Surrogate of the Intratumoral CD27–CD70 Interaction, Correlates with Immunotherapy Resistance in Renal Cell Carcinoma

This article is featured in Highlights of This Issue, p. 4833

Renal cell carcinoma (RCC) is classified into different histologic groups, with clear-cell RCC (ccRCC) accounting for approximately 70% of cases. Immunotherapy based on an anti–programmed cell death protein 1 (PD-1) mAb has demonstrated clinical efficacy in metastatic RCC (mRCC) both as a first-line treatment in combination with antiangiogenic drugs or anti–CTLA-4 or as a second-line monotherapy after VEGF pathway-directed therapy failure (1). Biomarkers [programmed cell death ligand 1 (PD-L1) and the tumor mutational burden] that predict the response to immunotherapy in some other tumors have no impact in mRCC (2). Other candidate markers have been proposed by our group and others (coexpression of inhibitory receptors, high number of expanded T-cell receptor (TCR) clones pretreatment, B-cell abundance, resident memory T cells (T_RM), human endogenous retrovirus expression, PBRM1 mutations, and AXL expression), but to date, they have not been validated in a prospective clinical setting (2–8).

In the majority of cancers, intratumoral CD8⁺ T-cell infiltration is associated with a good prognosis (9), and the preexistence of these cells in tumors seems to be a prerequisite for the response to immunotherapy (10). Unlike in other tumors, in RCC, CD8⁺ T cells predict neither clinical outcome (11, 12) nor response to immunotherapy (13, 14). Coexpression of inhibitory receptors or exhaustion-related molecules (CXCL13) by CD8⁺ T cells has been hypothesized to explain the absence of a favorable prognostic role or even a negative effect mediated by these cells (4, 5, 15–17). M2 macrophages, regulatory T cells, and 9p21.3^DEL were shown to be enriched in CD8⁺ T-cell–infiltrated tumors and may influence the impact and activity of CD8⁺ T cells in mRCC (2, 13). Expression of the nonclassical MHC molecule HLA-G or HLA-E by tumor cells (18), as well as multiple metabolic impairments (defects in glucose uptake, glycolysis, and mitochondrial function; ref. 19), may also negatively interfere with T-cell function in RCC. These different results classify RCC as a particular tumor type based on the role of CD8⁺ T cell as an antitumor factor.

CD27 is a costimulatory molecule that binds to CD70 and is constitutively expressed on naïve and central memory T cells. The CD27–CD70 interaction triggers a series of additional costimulatory signals during the priming phase, leading to the expansion and differentiation of memory and effector T cells with prolonged lymphocyte survival (20). Following this interaction, soluble CD27 (sCD27) is derived from proteolytic cleavage of the transmembrane molecule on activated T cells (20, 21).

However, persistent delivery of costimulatory signals via CD27–CD70 interactions, as may occur during chronic active viral infections, can exhaust the T-cell pool, deplete the naïve compartment, and promote apoptosis (22, 23). In line with this observation, CD70 blockade during chronic lymphocytic choriomeningitis virus (LCMV) infection results in increased CD8⁺ T-cell responses (24).

Interestingly, CD70 is aberrantly overexpressed in hematologic malignancies and solid tumors, especially in ccRCC secondary to the presence of Von Hippel–Lindau (VHL) mutations directly involved in the regulation of this marker (25, 26).

To better understand the roles of CD27 and CD70 in renal cancer, we characterized the phenotypes and functions of intratumoral CD27⁺ T cells, hypothesizing that chronic interaction with CD70-expressing tumor cells induces dysfunction in CD27⁺ T cells and leads to the release of sCD27. In this situation, we further evaluated the predictive value of sCD27 in patients with mRCC treated with anti–PD-1.

Thirty-five patients diagnosed with mRCC (4 patients with papillary RCC and 31 patients with ccRCC) and treated with anti–PD-1 (nivolumab) were included in this study (see the flow chart in Supplementary Fig. S1). The clinical characteristics of this cohort (age, sex), which was named Colcheckpoint, are documented in Supplementary Table S1. This cohort was approved by the ethics committee [CPP Ouest I SI CNRIPH n°18.11.21.67518 (05.03.2019)]. The correlation between sCD27 and the clinical response to anti–PD-1 therapy in the patients with mRCC included in this training cohort was examined.

The "BIONIKK" cohort (27), used as a validation cohort, is composed of patients with mRCC treated with first-line anti–PD-1 (n = 46) or sunitinib/pazopanib (investigator choice; n = 36) according to molecular classification into four randomized groups and for whom a plasma sample collected before treatment was available. This protocol was previously approved by the Ile de France ethics committee 8 (ref. 16.10.69) and registered at ClinicalTrials.gov under the NCT number 02960906.

For in situ analysis of the tumor microenvironment (TME), formalin-fixed, paraffin-embedded tumor tissues were available for 25 patients with ccRCC derived from the Colcheckpoint mRCC cohort and the exhauCRF cohort composed of patients with primary nonmetastatic ccRCC (registered at ClinicalTrials.gov under # NCT03149523, authorized by French Health authorities (ANSM n°2016A01458–43, CPP Ile-de-France II n°2016–07–08, IRB 00001072).

Apart from these retrospective cohorts, an independent prospective cohort of 14 fresh tumors was collected on the day of surgery for flow cytometric analysis (n = 10 ccRCC) and single-cell analysis [n = 3 ccRCC and 1 papillary RCC (pRCC)]. The tumor tissues were verified by clinical pathologists.

All clinical investigations were conducted according to the principles of the Declaration of Helsinki. Informed written consent was collected from all patients involved in this study.

Fresh tumors were digested with 25 mL Hank's Balance Salt Solution (HBSS) with calcium and magnesium (Lonza), 1-mg/mL collagenase D (Roche), and 30-IU/mL DNase (Roche) at 37°C for 1 hour. The digested tumors were filtered and washed with HBSS containing EDTA (Sigma–Aldrich) at 1,000 rpm for 10 minutes to stop the enzymatic reaction. After removal of the supernatant, the residual cells were resuspended in HBSS. Cell numbers were counted with Trypan Blue. A total of 1 million cells per tube were used for staining. Cells were first stained with the viability dye Zombie NIR (BioLegend) for 30 minutes. After washing with cell staining buffer (BioLegend), the cells were stained with various monoclonal antibodies labeled with fluorophores (see the list of antibodies in Supplementary Table S3). Samples were acquired on a cytometer (Navios 10 colors, Beckman Coulter), and the data were analyzed with Kaluza software (version 1.2).

Peripheral blood mononuclear cells (PBMC) were collected from 5 healthy donors using Ficoll (PAN-Biotech) and frozen with CryoStor CS10 (STEMCELL Technologies). Frozen PBMCs were thawed and cultured with RPMI1640 GlutaMax medium (Thermo Fisher Scientific) supplemented with 10% heat-inactivated FBS (GE Healthcare), 1-mmol/L sodium pyruvate (Thermo Fisher Scientific), 1% nonessential amino acids (Thermo Fisher Scientific), and 1% penicillin–streptomycin (Thermo Fisher Scientific) in a 48-well plate at 0.5×10⁶/cells per well. PBMCs were stimulated with precoated 5-μg/mL anti-CD3 (OKT3; Miltenyi), 5-μg/mL soluble anti-CD28 (Miltenyi), 10-U/mL or 1,000-U/mL IL2 (PeproTech), and 10-μg/mL anti-CD70 blocking antibody (Abcam) for 4 days. On Day 4, the supernatant was collected for sCD27 measurement by ELISA, and the remaining cells were stained with antibodies for specific markers for phenotypic characterization by flow cytometry (LSRFortessa X-20, BD). Data were analyzed with FlowJo software (TreeStar Inc.; RRID:SCR_008520). The antibodies used for flow cytometry are listed in Supplementary Table S2.

The plasma sCD27 concentrations in patients with RCC and healthy donors were determined with the CD27 (soluble) Human Instant ELISA Kit (Thermo Fisher Scientific) according to the manufacturer's instructions. sCD27 levels were measured in duplicate. Data were acquired with an MRX Revelation Microplate Reader (DYNEX Technologies).

One-way ANOVA was performed with the UCSC Xena tool for the comparison of CD27 and CD70 gene expression among different types of solid tumors. Other statistical analyses were performed using GraphPad Prism software version 8.0 (RRID:SCR_002798). Two-tailed unpaired and paired t tests were used to compare quantitative data between two groups, as appropriate. For correlation analyses, Pearson's correlation coefficient (r) and the P value were calculated. For survival analyses, Kaplan–Meier plots were generated, and significant differences were identified using the log-rank Mantel–Cox test. The test applied to determine the significance of differences is specified in detail in the corresponding figure legend. A P value < 0.05 was considered statistically significant.

For other techniques and methods, see the Supplementary Methods.

Single-cell RNA sequencing (scRNA-seq) data sorted from RCC samples are available from the NCBI Gene Expression Omnibus (GEO) archive platform (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE160243.

First, we studied CD70 and CD27 expression in RNA-seq data for a series of 31 types of cancers from The Cancer Genome Atlas (TCGA) Pan-Can Atlas (n = 9,575; Fig. 1A; Supplementary Table S3). As expected, kidney renal clear-cell carcinoma (KIRC = ccRCC) was the tumor with the highest CD70 expression, followed by mesothelioma and papillary renal cancer (KIRP = pRCC). These results emphasize the particularly aberrant expression of CD70 in renal cancer, in line with the respective frequencies of CD70 expression in ccRCC (85%) and pRCC (35%) reported in the literature (25, 28). CD27, the ligand of CD70, was also detected at high prevalence in ccRCC and at a more moderate level in pRCC (Fig. 1B; Supplementary Table S3). Interestingly, a correlation between CD70 and CD27 expression was found in the TCGA database, but we did not confirm it at the protein level in our series of patients (Supplementary Fig. S2)

Given the frequent mRNA expression of CD70 and CD27 in RCC, we carried out an in-depth study of the interactions between CD70-expressing tumor cells defined by the PAX8 marker and CD27-expressing T cells at the protein level by multiplex immunofluorescence in situ analysis (Fig. 2A and B; Supplementary Fig. S3A and S3B). A clear interaction was observed between CD27 and CD70 in the TME (Fig. 2B; Supplementary Fig. S3B). To measure this interaction, cells were phenotyped and quantified using HALO software (Fig. 2C). Each cell was assigned coordinates. The distances between cells expressing CD27 and CD70 were calculated to determine the CD27–CD70 interactions (spatial analysis; Fig. 2D). We defined CD27⁺ cells within 30 μm of CD70⁺ cells as “interacted CD27⁺ cells” (29). A representative image reflecting significant CD27–CD70 interactions is shown in Fig. 2D. In our series of RCC patients, 68.2% of CD27-expressing cells interacted with CD70⁺ cells. CD27 was expressed by CD4⁺ T cells (55.4%), CD8⁺ T cells (27.1%), and some non–T cells (17.5%; Fig. 2E). To more precisely define the interacting cells, we labeled tumor cells with PAX8 and cytokeratin 11 (CK11), as we observed that separate use of these markers underestimates the number of tumor cells identified by morphology and size. With this approach, CD70 expression appeared to be present in the majority of tumor cells (78%; Supplementary Fig. S3C). Moreover, most CD27⁺CD4⁺ T cells and CD27⁺CD8⁺ T cells interacted with PAX8⁺ and/or CK11⁺ tumor cells (Supplementary Fig. S3D). However, interactions between CD27-expressing T cells and CD70-expressing nontumor cells were also observed (Supplementary Fig. S3D), which could be explained by the possible expression of CD70 by activated T and B lymphocytes, mature dendritic cells, and fibroblasts (20, 30). The respective roles of the various cell subsets in the interactions with CD27⁺ T cells need to be investigated in future work. In terms of the specificity of interaction, we found a greater interaction between CD70 and CD27⁺CD4⁺T cell, than with CD27⁻CD4⁺T cell, but no difference in the interaction of CD70 with the different populations of CD8⁺T cells (Supplementary Fig. S3E)

Following reports that a chronic CD27–CD70 interaction could lead to apoptosis in T cells in vitro (31), we compared the expression of cleaved caspase 3 (a classic marker of apoptosis) between fresh CD27⁺ and CD27⁻ tumor-infiltrating lymphocytes (TIL). We observed that CD27⁺CD4⁺ and CD8⁺ T cells exhibited higher expression of cleaved caspase 3 (shown as the MFI and percentage), than CD27⁻ T cells (Fig. 3A–C). We then conducted a more detailed phenotypic analysis of these CD27⁺ T lymphocytes and found that they expressed CD103, which characterizes T_RM, with 67.5% of CD4⁺CD103⁺ T_RM cells and 76.2% of CD8⁺CD103⁺ T_RM cells expressing CD27 (Fig. 3D). The PD-1 molecule, which is associated with exhaustion and/or activation, was also expressed at higher levels in CD27⁺ T cells, than in CD27⁻ T cells (Fig. 3E). The proliferation marker Ki67 was more frequently detected in CD27⁺CD8⁺ T cells (Fig. 3E). To confirm these data, we performed scRNA-seq analysis of CD8⁺ T cells sorted from four fresh renal tumor samples. We merged all data for each donor and performed unsupervised community detection to cluster the CD8⁺ subpopulations based on highly variable genes and projected cells in two dimensions using Uniform Manifold Approximation and Projection (UMAP; Fig. 4A). Differential gene expression results derived from the scRNA-seq data resolved the CD8⁺ T-cell subsets into 11 clusters (Fig. 4A) with differentially expressed genes (Supplementary Table S4). Interestingly, we found that Cluster 11, corresponding to effector memory CD8⁺ T cells, was enriched in the expression of CD27 (Fig. 4B), as well as T_RM and exhaustion gene signatures (Supplementary Fig. S4 and Supplementary Table S4). When we performed gene ontology (GO) analysis of the genes highly expressed in Cluster 11, we found 8 significant GO biological process terms related to apoptosis with a P value < 0.0001, confirming our previous results (Fig. 4C). We then generated a heatmap based on the differential expression of genes in the scRNA-seq data between the CD8⁺ T-cell populations with or without CD27 expression (Fig. 4D). CD8⁺CD27⁺ TILs had an expression profile of enriched genes associated with apoptosis based on the GO database composed of BAX, BAD, CASP2, CASP3, CASP7, TRAF5, CDKN2A, and PHLDA3. In addition, we confirmed the T_RM phenotype of CD27⁺CD8⁺ T cells, as they more frequently expressed transcription factors associated with the T_RM phenotype (BATF, RBPJ, and RGS1). Furthermore, scRNA-seq analysis also confirmed that CD27⁺CD8⁺ T cells expressed markers of exhaustion (PDCD1, HAVCR2, LAG3, TIGIT, SIRPG, and CD38; Fig. 4D). Increases in proliferation marker (TOP2A, CCNB1, and MYBL2) and cytotoxicity-associated transcripts (GZMA, TIA1, and RAB27A) were also observed in CD27⁺CD8⁺ T cells. Finally, we extended our analysis to the phenotype of CD27⁺CD8⁺ T cells using previously published datasets. Gene set analysis performed on the Braun and colleagues (ref. 11; left) and Krishna and colleagues datasets (ref. 7; right) using ssGSEA revealed that the CD8⁺ T-cell population enriched in CD27 was apoptotic and exhausted (Supplementary Fig. S4B).

Taken together, our results demonstrate that CD27⁺CD8⁺ T cells are more apoptotic and display a phenotype of dysfunction/exhaustion.

Previous studies have shown that the CD27–CD70 interaction leads to the release of sCD27 (32). We first confirmed these results by showing that activation of PBMCs via TCR stimulation, but not by IL2 treatment alone induced the release of sCD27 (Fig. 5A). Interestingly, blocking the CD27–CD70 interaction partially inhibited this secretion (Fig. 5A). Blocking the interaction also increased CD27 expression on CD8⁺ T cells, likely through decreased CD27 membrane cleavage (Fig. 5B and C). Thus, it appears that TCR signaling and the CD27–CD70 interaction are important for the release of sCD27.

We then investigated the correlation between the CD27–CD70 interaction in situ and circulating sCD27 in patients with RCC. We quantified the number of interacting CD27 molecules in situ and measured the plasma sCD27 concentration in the corresponding patients with RCC. We first considered that the CD27–CD70 interaction in situ may occur when CD27⁺ cells are located within 30 μm of CD70-expressing cells. As shown in Fig. 5D, interacting CD27⁺ cells in situ significantly correlated with sCD27 in the peripheral blood (P = 0.0017; Fig. 5D, left). These results were confirmed, when a distance of 15 μm was selected to evaluate the interaction, which reinforced the strength and robustness of this correlation (P = 0.0061; Fig. 5D, right). These results suggest that the elevated levels of sCD27 in the peripheral blood may predominantly be derived from CD27–CD70 interactions in the TME.

To evaluate the clinical relevance of sCD27, we compared the sCD27 levels in patients with primary nonmetastatic (n = 22) or metastatic (n = 62) kidney cancer (total n = 84) with those in healthy subjects (n = 15). We found that patients with RCC had significantly higher plasma sCD27 concentrations than healthy subjects (P < 0.001; Fig. 6A). This increased sCD27 level was especially marked in patients with mRCC (Fig 6A). Since sCD27 could be a surrogate marker of the CD27–CD70 interaction leading to T-cell apoptosis, we wondered whether sCD27 could predict the clinical response to anti–PD-1/PD-L1 immunotherapy. As shown in Fig. 6B, we found that in patients with mRCC treated with anti–PD-1 or anti–PD-L1 after the first-line setting (Colcheckpoint cohort), high levels of sCD27 (> 91.8 U/mL) predicted poor survival, as patients with a level of sCD27 > 91.8 U/mL had a median survival time of 20.6 months, while patients with a low sCD27 level (< 91.8 U/mL) did not reach a median survival time [P = 0.037; HR = 2.63; 95% confidence interval (CI), 1.05–6.54; Fig. 6B]. However, in this series of patients, the plasma sCD27 concentration did not correlate with progression-free survival or objective radiologic response according to the RECIST 11 criteria (data not shown). Although the limited number of patients made it difficult to perform multivariate analyses, we tried to analyze whether sCD27 correlates with other clinical or laboratory parameters. We could not find any correlation between sCD27 and inflammatory biomarkers (CRP and the NLR; Supplementary Fig. S5A and S5B) or clinical prognostic criteria [Eastern Cooperative Oncology Group performance status, Memorial Sloan Kettering Cancer Center (MSKCC), or Heng classification, number of metastatic sites, and number of lines of treatment; Supplementary Fig. S5C–S5F and data not shown].

To confirm these preliminary results, we selected patients treated in the first-line setting with anti–PD-1 or sunitinib/pazopanib [tyrosine kinase inhibitor (TKI)] in the BIONIKKi clinical protocol (27). In our first analysis, we found that the median sCD27 concentrations in the Colcheckpoint cohort (103.2 U/mL) and BIONIKK cohort (Arm anti–PD-1: 69.1 U/mL and Arm TKI: 76.07 U/mL) were different, most likely due to the different stages of disease in the 2 cohorts (first-line treatment for BIONIKK and ≥ second-line treatment for Colcheckpoint). By adjusting the optimal cutoff value in the BIONIKK cohort considering the different distribution of sCD27, we found that patients with a high sCD27 concentration (≥112.71 U/mL) had an improved overall survival time of 17.8 months (14.8–NE), while patients with a low sCD27 concentration (<112.7 U/mL) did not reach the median survival time (P = 0.0044; HR = 5.02 95% CI, 0.94–26.55; Fig. 6C). In contrast, the sCD27 levels in patients in the BIONIKK cohort treated with the first-line TKI with the same cut-off value (112.71 U/mL) did not correlate with overall survival (P = 0.35; HR = 0.28–15.8; Fig. 6D). Similarly to the previous cohort, we did not find any correlation between sCD27 and clinical prognostic criteria (MSKCC or Heng classification; Fig. S5G–S5I). In addition, high sCD27 has been reported for patients with non-RCC with kidney dysfunction (33). To eliminate a confounding factor in the interpretation of the results, we measured sCD27, creatinine, and creatinine clearance in this cohort. We found no significant correlation between sCD27 and creatinine clearance and creatinine concentrations (Supplementary Fig. S6). sCD27 levels in sera have been previously reported to be linked to autoimmune comorbidities. In our study, pretreatment sCD27 levels were not correlated with the occurrence of immune-related adverse events including autoimmune diseases (data not shown).

This study reports that CD27-expressing T cells mostly interact with CD70-expressing tumor cells in the RCC TME. We found that CD27-expressing T lymphocytes had an apoptotic signature, indicated by both cytometry and scRNA-seq, compared with CD27-negative T cells. This feature was also confirmed by reanalyzing a public dataset. This result may seem paradoxical, as the interaction of CD27 with its cognate receptor CD70 triggers NF-κB signaling in T cells, which results in the expansion and differentiation of memory and effector T cells (34). Nevertheless, it is necessary to distinguish the costimulatory role of this interaction between the acute and chronic phases of the immune response. Indeed, blockade of the CD70–CD27 interaction during the later, but not the early, phase of a T-cell response to influenza virus or LCMV prevents dysfunctional apoptosis in antigen-specific CD8 T cells (22, 35). In striking contrast to the observations for acute LCMV infection, CD27-deficient mice are protected against chronic LCMV infection (36). In vitro, CD70-positive glioma or renal cell tumor cell lines also induce PBMC apoptosis (31, 37). It has been reported that CD27-induced apoptosis is mediated by the proapoptotic protein Siva (23, 31). In addition to the acute or chronic nature of the interaction between CD27 and CD70, CD27 increases T-cell survival after stimulation with a low-affinity Ag or low doses of Ag, whereas CD27 induces T-cell apoptosis after stimulation with a high-affinity Ag or high doses of Ag, as observed in a tumoral context (38, 39).

Nevertheless, our study did not allow us to determine whether CD27 is directly involved in T-cell apoptosis or is simply a marker of T-cell dysfunction. We showed, in this study, that CD27-positive T lymphocytes in the TME had a phenotype of resident memory T lymphocytes and expressed higher levels of exhaustion markers (PD-1, Tim-3, Lag3, TIGIT, etc.), than CD27-negative T cells (Fig. 4D).

We have observed a greater interaction between CD70 and CD27⁺CD4⁺T cell, than with CD27⁻CD4⁺T cell, but no difference in the interaction of CD70 with the different populations of CD8⁺ T cells expressing or not CD27 (Supplementary Fig. S3E). This can be explained by the fact that CD8⁺ T cells not expressing CD27 express other molecules such as PD-1, albeit at lower levels, but which can interact with the tumor cell co-expressing CD70 and PD-L1 (Fig. 3E). It would have been interesting to evaluate the predictive therapeutic impact of the in situ interaction between CD27 and CD70 in comparison with sCD27.

These results may explain the poor prognosis associated with CD70 expression by tumor cells in mRCC, particularly in the presence of T-cell infiltration (25). Since the deleterious effects of persistent CD27–CD70 interactions on T cells require activation of TCR by the cognate antigen (22), it can be hypothesized that this interaction between CD27 and CD70 is an escape mechanism set up by cancer cells to protect themselves from attack by tumor-specific lymphocytes.

We showed that CD27 expression by CD8⁺ T cells was associated with an exhausted phenotype (Fig. 3D and Fig. 4D). These results are in line with previous works showing that continuous CD27–CD70 signaling as observed in chronic infection or cancer results in T-cell exhaustion (22, 40, 41). In different preclinical mouse models, the administration of an agonist anti-CD27 antibody, which stimulates the CD27 pathway, has been shown to inhibit tumor growth (42, 43). In most cases, the tumor models used, such as transplanted tumor cell lines, did not favor chronic interaction between the tumor and host cells. Not surprisingly, these models favored the costimulatory activity of the CD27–CD70 interaction, which promotes antitumor activity. In humans, the development of an anti-CD27 agonist antibody has, thus far, shown little efficacy (44). Our results suggest that in the context of chronic stimulation, such as in cancer, an antibody blocking the interaction between CD27–CD70 rather than an antibody promoting CD27 signaling in T cells could be interesting to evaluate to reverse the dysfunction of these cells. Moreover, as anti–HIF-2α inhibitors are indicated in kidney cancer with VHL disease, it would be interesting to evaluate their impact on CD70 expression (45).

Our study focused on ccRCC, but other types of tumors (mesothelioma, stomach adenocarcinoma, skin cutaneous melanoma, thyroid carcinoma, and testicular germ cell cancer) were found to overexpress CD70 and CD27 in the TCGA analysis (Fig. 1). Furthermore, Chen and Shalper's group showed that a population of apoptotic CD8⁺ T cells existed in the TME of patients with non–small cell lung cancer who strongly expressed CD27 and other markers of exhaustion and proliferation, thus resembling the population identified in our study. Expansion of this population correlated with resistance to immunotherapy (46).

In this work, we also showed that plasma sCD27 concentrations correlated with the level of the CD27–CD70 interaction in the TME. Previous studies have shown that the extracellular domain of CD27 is released after lymphocyte activation and engagement of CD70 (21). In vitro studies also reported that coculture of PBMCs with RCC cells resulted in a CD70-dependent increase in sCD27 in the culture supernatant (25). Our study supports that sCD27 may be a surrogate marker of the CD27–CD70 interaction in the TME.

In patients with metastatic renal cancer, sCD27 levels were increased and correlated with resistance to anti–PD-1 therapy. Although we showed that sCD27 did not predict the response to antiangiogenic therapy, our results are too preliminary to define whether sCD27 is a predictive marker for the response to checkpoint inhibitors or a prognostic marker. One study showed that sCD27 was not associated with disease stage or recurrence, but with survival in 182 patients with localized ccRCC treated with surgery or chemotherapy (47).

In terms of the value of sCD27 as a pharmacological surrogate to anti–PD-1 therapy, it has been shown that administration of anti–PD-1 antibodies increases the sCD27:sCD40 ratio suggesting that sCD27 may also reflect immune activation (48).

There are many limitations to the development of biomarkers based on tumor tissue, such as the difficulty of obtaining biopsies from certain locations, the semiquantitative nature of in situ assays that are often difficult to standardize, and the spatial heterogeneity within a single tumor or between primary and metastatic tumors, which raises concerns about the representativeness of any tissue analyzed. Soluble forms of checkpoint inhibitors, cytokines, costimulatory molecules, and their ligands have been detected in the blood of renal cancer patients. Some molecules (sTim-3, sLag-3, sPDL-1, sPD-L2, BTN3A, BTN2A1, and IL8) have been associated with a poor prognosis, and some of them have been shown to predict clinical response, although these associations have not been observed consistently (47, 49–53). An attractive hypothesis would be that these soluble molecules reflect the corresponding expression in the TME (54). sCD27 may represent a new class of biomarkers as a surrogate marker of the interaction between tumor cells and host cells.

In conclusion, we demonstrated that sCD27, a surrogate of T-cell dysfunction in tumors induced by the CD27–CD70 interaction, is a predictive biomarker of resistance to PD-1/PD-L1 blockade in mRCC. To our knowledge, this study is the first to report that a blood biomarker may reflect certain aspects of the tumor–host interaction in the TME. This CD27–CD70 interaction may constitute a new tumor escape mechanism, which could be a therapeutic target for blocking antibodies. Given the frequent expression of CD70 and CD27 in solid tumors, our findings may be applicable to other types of tumors.

N. Benhamouda reports a patent for WO2022023379A1, 2022-02-03 issued and a patent for US2022107323A1, 2022-04-07 issued. I. Sam reports a patent for WO2022023379A1, 2022-02-03 issued. N. Epaillard reports a patent for US2022107323A1, 2022-04-07 issued. N. Chaput reports personal fees from AstraZeneca; grants from Sanofi, Bristol Myers Squibb, and GlaxoSmithKline; and other support from Cytune outside the submitted work. L. Albiges reports other support from Bristol Myers Squibb, Novartis, Pfizer, Ipsen, Astellas, MSD, Merck & Co., AstraZeneca, and Eisai, as well as grants from Bristol Myers Squibb outside the submitted work. J. Adam reports personal fees from Amgen, MSD, and AstraZeneca, as well as grants from Bayer outside the submitted work. Y.A. Vano reports grants from Bristol Myers Squibb and ARTIC during the conduct of the study; Y.A. Vano also reports personal fees and other support from Bristol Myers Squibb, MSD, Ipsen, Pfizer, and Roche, as well as personal fees from Novartis and Merck outside the submitted work. E. Tartour reports a patent for US2022107323A1, 2022-04-07 issued and a patent for WO2022023379A1, 2022-02-03 issued. No disclosures were reported by the other authors.

N. Benhamouda: Conceptualization, formal analysis, validation, methodology. I. Sam: Conceptualization, software, formal analysis, validation, methodology. N. Epaillard: Conceptualization, formal analysis, validation, investigation. A. Gey: Software, formal analysis, validation, methodology. L. Phan: Data curation, formal analysis, methodology. H.P. Pham: Software, formal analysis, validation, methodology. N. Gruel: Software, formal analysis, validation, methodology. A. Saldmann: Formal analysis, validation, methodology. J. Pineau: Software, formal analysis, validation, methodology. M. Hasan: Software, formal analysis, validation. V. Quiniou: Software, formal analysis, validation, methodology. C. Nevoret: Formal analysis, validation, methodology. V. Verkarre: Formal analysis, validation. V. Libri: Software, formal analysis, validation, methodology. S. Mella: Data curation, software, formal analysis, validation, methodology. C. Granier: Formal analysis, validation. C. Broudin: Formal analysis, validation, visualization. P. Ravel: Formal analysis, validation, methodology. E. De Guillebon: Data curation, formal analysis, validation. L. Mauge: Resources, formal analysis, validation. D. Helley: Resources, formal analysis, validation. B. Jabla: Software, formal analysis, validation, methodology. N. Chaput: Resources, formal analysis, validation. L. Albiges: Resources, formal analysis, validation. S. Katsahian: Data curation, software, formal analysis, validation, methodology. J. Adam: Resources, formal analysis, validation. A. Mejean: Resources, formal analysis, validation. O. Adotevi: Resources, formal analysis, validation. Y.A. Vano: Resources, data curation, formal analysis, validation. S. Oudard: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, writing–review and editing. E. Tartour: Conceptualization, resources, data curation, supervision, funding acquisition, validation, investigation, writing–original draft.

This work was funded by the Fondation ARC pour la Recherche sur le Cancer (Grant number SIGN'IT20181007747 and PGA12019110000946_1581 to E. Tartour), the INCA (Institut National du Cancer) (Grant number 2016–1-PL BIO-05–1 to E. Tartour), the Agence Nationale de la Recherche (Labex Immuno-Oncology to E. Tartour), the Institut National du Cancer (Grant SIRIC CARPEM to E. Tartour, S. Oudard), FONCER (to E. Tartour and S. Oudard), IDEX Université de Paris (Grant number AMI-CD27CD70BLOC to E. Tartour), and Inserm Transfert (Grant number: MAT-PI-20278–1-02 to E. Tartour).

We thank the staff of the tumor banks of HEGP (B. Vedie and D. Geromin) for providing the sample materials and the Histology platform of PARCC (C. Lesaffre).

We thank Franck Pages and Florence Marliot for providing access to the Halo software platform.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.