PD-1 Expression on Intratumoral Regulatory T Cells is Associated with Lack of Benefit from Anti-PD-1 Therapy in Metastatic Clear Cell Renal Cell Carcinoma Patients

Abstract

Purpose:

PD-1 expression on CD8⁺TIM-3⁻LAG3⁻ tumor-infiltrating cells predicts positive response to PD-1 blockade in metastatic clear cell renal cell carcinoma (mccRCC). Since inhibition of PD-1 signaling in regulatory T cells (Tregs) augments their immunosuppressive function, we hypothesized that PD-1 expression on tumor-infiltrating Tregs would predict resistance to PD-1 inhibitors.

Experimental design:

PD-1⁺ Tregs were phenotyped using multiparametric immunofluorescence in ccRCC tissues from the CheckMate-025 trial (nivolumab: n=91, everolimus: n=90). Expression of CD8, PD-1, TIM-3, and LAG-3 was previously determined (Ficial et al, 2021). Clinical endpoints included progression-free survival (PFS), overall survival (OS), and objective response rate (ORR).

Results:

In the nivolumab (but not everolimus) arm, high percentage of PD-1⁺ Tregs was associated with shorter PFS (3.19 versus 5.78 months; p=0.021), shorter OS (18.1 versus 27.7 months, p=0.013) and marginally lower ORR (12.5% versus 31.3%, p=0.059). An integrated biomarker (PD-1 Treg/CD8 ratio) was developed by calculating the ratio between percentage of PD-1⁺Tregs (marker of resistance) and percentage of CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells (marker of response). In the nivolumab (but not everolimus) arm, patients with high-PD-1 ratio experienced shorter PFS (3.48 versus 9.23 months, p<0.001), shorter OS (18.14 versus 38.21 months, p<0.001) and lower ORR (15.69% versus 40.00%; p=0.009). Compared to the individual biomarkers, the PD-1 ratio showed improved ability to predict outcomes to nivolumab versus everolimus.

Conclusion:

PD-1 expression on Tregs is associated with resistance to PD-1 blockade in mccRCC, suggesting that targeting Tregs may synergize with PD-1 inhibition. A model that integrates PD-1 expression on Tregs and CD8⁺TIM-3⁻LAG3⁻ cells has higher predictive value.

Introduction

Anti-PD-(L)1-based therapies are standard regiments for patients with metastatic clear cell renal cell carcinoma (mccRCC) (1–3). However, predictors of response or resistance to these agents remain unclear as there are currently no FDA-approved biomarkers that can guide treatment selection for patients with mccRCC.

There is increasing evidence that classification of tumors based on their immune milieu can help explain patient response or resistance to immune checkpoint inhibitors (ICI). We recently showed that pre-treatment levels of antigen-experienced but not terminally exhausted CD8⁺ effector T cells (i.e. CD8⁺ cells expressing PD-1 but not TIM-3 and LAG-3) within the tumor are associated with high levels of T-cell activation and improved clinical outcome in patients with mccRCC treated with the anti-PD-1 antibody nivolumab in the context of several clinical trials (4–6). Our findings support the concept that patients with inflamed tumors containing antigen-experienced, mildly exhausted tumor-specific T cells that can be effectively re-activated by PD-1 blockade are most likely to respond to the treatment. Although these results are encouraging, it should be noted that a substantial proportion of patients with an active anti-tumor immune response (i.e. high levels of non-terminally exhausted CD8⁺ effector T cells) in their pre-treatment tumors do not respond to anti-PD-1 therapy, indicating that other tumor-intrinsic or -extrinsic factors constitute important determinants of outcome to ICI. Therefore, understanding mechanisms of resistance to PD-1 blockade represents a critical step to better define the patient population that benefits from anti-PD1 treatment and to develop more effective anti-PD-1-based combination therapies for mccRCC.

Regulatory T cells (Tregs) play a central role in autoimmunity, chronic infection, and transplant rejection (7–9). Tregs also suppress anti-tumor immune responses by reducing the activation and recruitment of both CD4⁺ T helper cells and CD8⁺ effector T cells (10,11), and by secreting various inhibitory cytokines (including IL-10, IL-35, and TGF-β) (12) that promote the exhaustion of CD8⁺ cells (13). It is well recognized that a subset of Tregs express PD-1, and work from our group and others recently demonstrated that PD-1 signaling in Tregs restrains their immunosuppressive function (14,15). Indeed, PD-1-deficient Tregs were shown to have an activated phenotype and increased immunosuppressive function both in vitro and in vivo. Specifically, selective deletion of PD-1 in Tregs led to less severe disease in a mouse model of autoimmune encephalomyelitis and conferred protection from diabetes in nonobese diabetic (NOD) mice (15). Importantly, it was also shown that Treg-specific PD-1 deficiency or blockade suppressed anti-tumor response by PD-1-deficient effector T cells and promoted tumor growth in mice (14). Taken together, these data suggest that in patients treated with anti-PD-1 therapy, inhibition of PD-1 signaling in Tregs enhances their immunosuppressive function and thus may limit the efficacy of the treatment. In support of this hypothesis, Kumagai and colleagues recently reported that PD-1 expression on intratumoral Tregs (assessed by flow cytometry) was associated with lack of response to PD-1/PD-L1 blockade in small uncontrolled cohorts of patients with gastric cancer, lung cancer and melanoma. Of note, this study also suggested that the balance between PD-1⁺ effector T cells and PD-1⁺ Tregs cells would best predict clinical outcomes (16).

Here, we investigated the role of PD-1 expression on tumor-infiltrating Tregs (alone or in combination with non-terminally exhausted CD8⁺ effector T cells) as a biomarker of resistance to anti-PD-1 therapy in a randomized trial of nivolumab versus everolimus in previously treated patients with mccRCC (17).

Materials and Methods

Study design and clinical endpoints

Tumors samples from patients included in the randomized phase 3 clinical trial CheckMate 025 (CM-025) were analyzed (17). This study compared nivolumab with everolimus in patients with advanced or metastatic ccRCC with measurable and progressive disease according to the Response Evaluation Criteria in Solid Tumors (RECIST version 1.1) after at least one anti-angiogenic therapy. At the time of enrollment, the study sponsor collected archival formalin-fixed paraffin-embedded (FFPE) tissue sections which were transferred to the investigators upon approval of a material transfer agreement. Institutional review board approval was obtained. Written informed consent to study participation, tissue collection, and tissue analyses was provided by all patients before enrollment, according to the principles of the Declaration of Helsinki.

Objective response rate (ORR) was assessed by the investigators and defined as the proportion of patients with a complete response (CR) or a partial response (PR) per RECIST 1.1. Progression-free survival (PFS) was defined as the time from randomization to the first documented tumor progression per RECIST 1.1 or death from any cause (patients were censored at the date of last disease assessment). Overall survival (OS) was defined as the time from randomization to the date of death (patients still alive were censored at the date of the last contact).

Multiplex immunofluorescence assays and image analysis

A 4-plex immunofluorescence (IF) assay including CD4, FOXP3, PD-1, and DAPI was optimized and performed on a Bond RX Autostainer (Leica Biosystems) using the Perkin Elmer Opal tyramide signal system. Performance of antibodies against CD4, FOXP3, PD-1 was validated on non-neoplastic tonsil tissue to obtain a staining pattern in accordance with available published data. The anti-CD4 antibody (1:50; 4B12; mouse monoclonal antibody; Agilent; Cat# M7310, RRID:AB_2728838) was detected using the Opal 520 fluorophore (1:50, FITC); The anti-FOXP3 antibody (1:250; D2W8E^™; rabbit monoclonal antibody; Cell Signaling Technology; Cat# 98377, RRID:AB_2747370) was detected using the Opal 570 fluorophore (1:100); the anti-PD-1 antibody (1:5,000 of 1.3 mg/mL, EH33 mouse monoclonal antibody, Dr. Freeman laboratory, Dana-Farber Cancer Institute, Boston, MA (18)) was detected using the Opal 690 fluorophore (1:50, Cy5). The slides were scanned using the Vectra 3 automated quantitative pathology imaging system (PerkinElmer), and whole slide multispectral images were acquired at 100x magnification. At least five tumor regions of 931×698 μm characterized by the highest Treg infiltration (hotspots) were selected for each slide by a pathologist (TD or SM) using the Perkin Elmer Phenochart v 1.0 software. Each region was then acquired at 200x magnification using the Vectra 3 instrument. The Inform 2.2 software was then used in order to deconvolute the multispectral images, as previously described (4,5). Deconvoluted images of hotspots in TIFF format were uploaded into Indica Lab HALO platform version 3.0. For each image, the tumor area was manually annotated by two pathologists (TD, SM). Empty spaces, necrosis, erythrocytes, and fibrous septa were excluded. CD4⁺ FOXP3⁺ cells and PD-1⁺ CD4⁺ FOXP3⁺ cells were phenotyped using the Indica Lab High-Plex FL v3.x module. A unique image analysis algorithm was created for each case by a pathologist, and the performance of the algorithm was then independently validated by a second pathologist. In case of disagreement, the two pathologists (TD, SM) reviewed the data together, and reached an agreement on the optimal algorithm to be used.

Data on tumor-infiltrating CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells was generated as part of a previous study (5). As described in the Results and in the Supplementary Materials and Methods sections, this data was utilized here to create an integrated biomarker (called PD-1 Treg/CD8 ratio) that was developed by calculating the ratio between the percentage of PD-1⁺ Tregs and the percentage of CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells.

For spatial analyses, tumor samples from ten patients enrolled in the CM-025 trial were stained with a 7-plex immunofluorescence assay including CD8, PD-1, TIM-3, LAG-3, CD4, FOXP3, and DAPI. The assay was similarly optimized and performed on a Bond RX Autostainer (Leica Biosystems) using the Akoya Biosciences Opal tyramide signal system. First, the anti-CD8 antibody (1:5000; clone C8/144B, mouse monoclonal antibody; Agilent; Cat#M7103 RRID:AB_2075537) was detected using opal Polaris 480 fluorophore (1:150), the anti-PD-1 antibody (1:5,000 of 1.3 mg/mL; EH33 mouse monoclonal antibody; Dr. Freeman laboratory, Dana-Farber Cancer Institute, Boston, MA) was detected using the Opal Polaris 690 fluorophore (1:50), anti-TIM-3 antibody (1:750, polyclonal goat antibody, R&D systems; cat# AF2365 RRID:AB_355235) was detected using Opal Polaris 520 fluorophore (1:100), the anti-LAG3 antibody (1:3,000 of 1mg/ml; 17B4, mouse monoclonal antibody, LifeSpan Biosciences, inc. Cat#LS-C18692, RRID:AB_1650048) was detected using opal Polaris 620 fluorophore (1:50), the anti-CD4 antibody (1:80; 4B12; mouse monoclonal antibody; Agilent; Cat# M7310 RRID:AB_2728838) was detected using Opal Polaris 570 fluorophore (1:100), and finally, the anti-FOXP3 antibody (1:50; D2W8E^™; rabbit monoclonal antibody; Cell Signaling Technology; Cat# 98377, RRID:AB_2747370) was detected using opal Polaris 780 (1:25)/TSA-DIG (1:100). Whole-slide multispectral images were acquired at 20x magnification using the PhenoImager HT automated quantitative pathology imaging system (Akoya Biosciences). Phenochart was used to divide the whole-slide image into hot spots (stamps). Inform 2.6 software was then used to deconvolute the multispectral images of hot spots. Deconvoluted images of hotspots in TIFF format were uploaded into Indica Lab HALO platform version 3.1.1076.423 and fused into a whole-slide image. For each whole-slide image, the tumor area was manually annotated by two pathologists (NSA, SM). Empty spaces, necrosis, erythrocytes, and fibrous septa were excluded. CD4⁺ cells (i.e. CD4⁺ FOXP3⁺, PD-1⁺ CD4⁺ FOXP3⁺, PD-1⁻ CD4⁺ FOXP3⁺ cells) and CD8⁺ cells (CD8⁺PD-1⁺TIM-3⁺LAG-3⁻, CD8⁺PD-1⁺TIM-3⁻LAG-3⁺, and CD8⁺PD-1⁺TIM-3⁺LAG-3⁺, CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells) were phenotyped using the Indica Lab High-Plex FL v3.x module. A unique algorithm was created for each case, and its accuracy was validated through visual inspection by two pathologists (NSA, SM). Nearest neighbor analysis was performed using the Indica Lab HALO spatial analysis module.

Statistical analysis

In this study, 1-sided p-values are reported unless otherwise stated and 1-sided p-values ≤0.05 that are consistent with pre-stated hypothesis are considered statistically significant. For interaction tests, 2-sided p-values are reported and values ≤0.10 (i.e., 10% significance level) are considered statistically significant. Detailed statistical methods are available as Supplementary Materials and Methods.

Data availability statement

Relevant data are available from the corresponding authors and/or are included in the manuscript.

Results

Patient characteristics

From October 2012 through March 2014, 803 patients with previously-treated mccRCC were enrolled, randomized, and treated in the CM-025 trial (nivolumab=406, everolimus=397). Multiplex-immunofluorescence (mIF) analysis for the evaluation of PD-1 expression on CD4⁺ FOXP3⁺ cells (i.e. Tregs) (Supplementary Figure 1) was performed on available tumor tissue samples from 214 patients and evaluable mIF data were obtained for 181 patients (nivolumab= 91, everolimus= 90) (CONSORT diagram, Supplementary Figure 2).

A comparison between patients with mIF data and those without showed no imbalance with regards to gender, age, treatment arm, Karnofsky Performance Status or clinical outcomes. There was a slight enrichment of patients with unfavorable risk profile (i.e. intermediate and poor risk per MSKCC criteria (used as study randomization stratification factor) in the group with mIF data (Supplementary Table 1).

Association between expression of PD-1 on tumor-infiltrating Tregs and clinical outcomes

As studies from our group and others recently suggested that inhibition of PD-1 signaling in Tregs can augment their immunosuppressive function and thus limit the efficacy of PD-1 inhibitors (15,16), we hypothesized that high levels of PD-1 expression on intra-tumoral Tregs would correlate with worse clinical outcomes in patients with metastatic ccRCC treated with the anti-PD-1 antibody nivolumab (but not in those treated with the mTOR inhibitor everolimus).

To test our hypothesis, we first examined the presence of a relationship between the percentage of tumor-infiltrating Tregs expressing PD-1 (i.e. CD4⁺FOXP3⁺PD-1⁺ cells/ total CD4⁺FOXP3⁺ cells), measured as a continuous variable, and clinical outcomes in each treatment arm. In nivolumab-treated patients, the percentage of PD-1⁺ Tregs was correlated with worse OS (HR 3.12; 90% CI 1.41–6.89; p-value= 0.009) and showed a trend in the correlation with worse PFS (HR 2.07; 90% CI 0.83–5.18; p-value = 0.096) and lower ORR (OR 0.39; 90% CI 0.06–2.57; p-value = 0.206). In everolimus-treated patients, no correlation was observed between percentage of PD-1⁺ Tregs and either PFS or OS. The correlation with ORR was not assessed due to the low rate of response observed in patients treated with everolimus (Table 1). Similar results were obtained when controlling for MSKCC groups (Supplementary Table 2). In addition, we also tested for the presence of a relationship between the total levels of tumor-infiltrating Tregs (i.e. density of CD4⁺FOXP3⁺ cells), measured as a continuous variable, and clinical outcomes. This analysis demonstrated no correlation between Treg density and PFS, OS, or ORR in either treatment arm (Table 1) and similar findings were observed when controlling for MSKCC groups (Supplementary Table 2), suggesting that total Tregs infiltration is not a predictor of outcome to anti-PD-1 therapy. The distribution of % of PD-1+ Tregs and density of Tregs in the patient cohort is shown in Supplementary Figure 3.

Table 1.

Association between percentage of PD-1⁺ Tregs or density of total Tregs (measured as continuous variables) and clinical outcomes.

Treatment Arm	Biomarker	PFS HR (90% CI; p-value)	OS HR (90% CI; p-value)	ORR OR (90% CI; p-value)
Everolimus	% of PD-1+ Tregs	0.92 (0.38–2.23; 0.565)	0.77 (0.32–1.83; 0.692)	NA
	Density of Tregs	1.15 (0.93–1.43; 0.13)	1.06 (0.86–1.31; 0.317)	NA
Nivolumab	% of PD-1+ Tregs	2.07 (0.83–5.18; 0.096)	3.12 (1.41–6.89; 0.009)	0.39 (0.06–2.57; 0.206)
	Density of Tregs	0.89 (0.75–1.06; 0.867)	1.09 (0.93–1.29; 0.185)	1.26 (0.88–1.81; 0.852)

To further explore the value of the percentage of PD-1⁺ Tregs as a possible biomarker of resistance to PD-1 blockade, we developed an optimized cut-off (using the minimum p-value for association with ORR on nivolumab) to create a binary variable. Based on this cut-off (42.42%), 48 patients (24 of whom were nivolumab-treated) had a high percentage of PD-1⁺ Tregs, and 133 patients (67 of whom were nivolumab-treated) had a low percentage. In the nivolumab arm, patients with high percentage of PD-1⁺ Tregs experienced shorter median PFS (3.19 versus 5.78 months, p=0.021), shorter median OS (18.1 versus 27.7 months, p=0.013) (Figure 1) and marginally significant lower ORR (12.5% versus 31.3%, p=0.059) compared to patients with low percentage of PD-1⁺Tregs. No difference in either PFS or OS was observed between patients with high and low percentage of PD-1⁺Tregs in the everolimus arm (Figure 1).

Figure 1.

Kaplan-Meier curves for PFS (A, C) and OS (B, D) per percentage of PD1⁺ Tregs levels (minimum p-value cutoff) in the nivolumab (nivo) (A, B) and everolimus (evero) (C, D) arm.

The predictive value of the percentage of PD-1⁺ Tregs was then assessed by comparing outcomes on nivolumab versus everolimus according to biomarker levels. A significant interaction between treatment and percentage of PD-1⁺ Tregs (as continuous variable) was observed for OS but not for PFS (2-sided p = 0.067 for OS and 2-sided p = 0.274 for PFS; significance determined as 2-sided p ≤0.10). A similar finding was observed when performing MSKCC-stratified test (Supplementary Table 3). The interaction was such that the higher the PD-1⁺ Tregs, the shorter the OS in patients treated with nivolumab but not in patients treated with everolimus. When the percentage of PD-1⁺ Tregs was analyzed as a binary variable by splitting at the median value, longer OS in the nivolumab arm versus the everolimus arm was observed in the low biomarker group but not in the high biomarker group (Supplementary Figure 4).

Relationship between Tregs and CD8⁺ effector T cells in the ccRCC microenvironment

Our results show that in patients with metastatic ccRCC treated with nivolumab therapy, the expression of PD-1 on intratumoral Tregs is associated with worse clinical outcomes. We have previously demonstrated that the expression of PD-1 on intratumoral CD8⁺ effector T cells negative for TIM-3 and LAG-3 expression (i.e. non-terminally exhausted CD8⁺ effector T cells) is associated with improved response to treatment (4–6). Since there is experimental evidence that Tregs can promote T cell exhaustion (19,20), we hypothesized that in ccRCC tissues, levels of Tregs (either total or PD-1⁺ Tregs) would be negatively correlated with levels of non-terminally exhausted CD8⁺ effector T cells (CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells) and positively correlated with levels of terminally exhausted CD8⁺ effector T cells, defined as CD8⁺ cells co-expressing PD-1 and at least one of the inhibitory receptors TIM-3 and LAG-3 (CD8⁺PD-1⁺TIM-3⁺ and/or LAG-3⁺ cells). In contrast to our hypothesis, however, we found that in the CM-025 cohort, the density of total Tregs correlated poorly both with the percentage or density of terminally exhausted CD8⁺ effector T cells (Spearman rank correlation coefficient of 0.28 and 0.36, respectively) and with the percentage or density of non-terminally CD8⁺ effector T cells (correlation coefficient of −0.11 and 0.22, respectively). Similarly, the percentage of PD-1⁺ Tregs correlated poorly both with the percentage or density of terminally exhausted CD8⁺ effector T cells (correlation coefficient of 0.11 and 0.04, respectively) and with the percentage or density of non-terminally exhausted CD8⁺ effector T cells (correlation coefficient of −0.08 and −0.1, respectively). Overlapping results were obtained when the correlation between levels of Tregs (either total or PD-1⁺ Tregs) and terminally exhausted CD8⁺ effector T cells was assessed by separately analyzing CD8⁺PD-1⁺TIM-3⁺LAG-3⁻, CD8⁺PD-1⁺TIM-3⁻LAG-3⁺, and CD8⁺PD-1⁺TIM-3⁺LAG-3⁺ cells (Supplementary Table 4).

To further explore the link between Treg infiltration and T cell exhaustion in the ccRCC microenvironment, we assessed the spatial relationships between tumor-infiltrating Tregs and tumor-infiltrating CD8⁺ effector T cells in different states of exhaustion. To this end, we re-stained a subset of ccRCC tissues with a 7-plex IF panel, which allowed us to identify all the immune cell populations of interest in the same tissue slide. We then performed a nearest neighbor analysis to determine whether terminally exhausted CD8⁺ effector T cells were localized, on average, closer to Tregs compared to non-terminally exhausted CD8⁺ effector T cells (Figure 2A). The spatial analysis was conducted on total Tregs as well as on the subsets of PD-1⁺ and PD-1⁻ Tregs. We observed that the average distance between total Tregs and terminally exhausted CD8⁺PD-1⁺TIM-3⁺ and/or LAG-3⁺ cells was not significantly different from average distance between total Tregs and non-terminally exhausted CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells (p=0.8). Similar results were obtained when the nearest neighbor analysis was performed on the subsets of PD-1⁺ and PD-1⁻ Tregs (p=0.6 and p=0.8, respectively) (Figure 2B).

Figure 2.

(A) A representative image of a ccRCC tissue analyzed by multiplex IF for expression of CD8, PD-1, CD4, FOXP3, LAG-3 and TIM-3 is shown on the left. An example of the output of the nearest neighbor analysis depicting non-terminally exhausted CD8⁺ effector T cells (red), terminally exhausted CD8⁺ effector T cells (yellow) and Tregs (light blue) is shown on the right. Magenta lines connect each Treg cell to the nearest terminally exhausted CD8⁺ effector T cell and black lines connect each Treg cell to the nearest non-terminally exhausted CD8⁺ effector T cell. (B) Box-plots showing the distribution of average distance between Tregs (total, PD1⁺ and PD1⁻) and terminally exhausted CD8⁺ effector T cells versus the average distance between Tregs (total, PD1⁺ and PD1⁻) and non-terminally exhausted CD8⁺ effector T cells in all the cases analyzed.

Association between an integrated biomarker of PD-1 expression on T cell subsets (PD-1 Treg/CD8 ratio) and clinical outcomes

Taken together our results indicate that PD-1 expression on Tregs and PD-1 expression on non-terminally exhausted CD8⁺ effector T cells are independently associated with opposite clinical outcomes in ccRCC patient treated with nivolumab (but not in patients treated with everolimus). Therefore, we hypothesized that an integrated biomarker model that takes into account the levels of PD-1 expression in each of these two T cell subsets would improve the ability to predict clinical outcomes on nivolumab versus everolimus.

To develop the integrated biomarker, for each patient, we calculated the ratio (hereafter called PD-1 Treg/CD8 ratio) between the percentage of tumor-infiltrating PD-1⁺ Tregs (marker of resistance) and the percentage of tumor-infiltrating CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells (marker of response). In nivolumab-treated patients, the PD-1 Treg/CD8 ratio, measured as a continuous variable, was correlated with worse PFS and OS (HR 1.24; 90% CI: 1.04–1.47; p=0.023 and HR 1.22; 90% CI 1.04–1.44; p=0.022, respectively) and tended to be correlated with lower ORR (OR 0.57; 0.31–1.03; p=0.060). No correlation was observed between the PD-1 Treg/CD8 ratio and either PFS or OS in everolimus-treated patients (Table 2). Similar results were obtained when controlling for MSKCC groups using stratified tests (Supplementary Table 5). The distribution of PD-1 Treg/CD8 ratio in the patient cohort is shown in Supplementary Figure 3.

Table 2:

Association between PD-1 Treg/CD8 ratio (measured as continuous variable) and clinical outcomes.

Treatment Arm	PFS HR (90% CI; p-value)	OS HR (90% CI; p-value)	ORR OR (90% CI; p-value)
Everolimus	0.93 (0.82–1.05; 0.838)	0.91 (0.79–1.05; 0.860)	NA
Nivolumab	1.24 (1.04–1.47; 0.023)	1.22 (1.04–1.449; 0.022)	0.57 (0.31–1.03; 0.060)

The patient population was then dichotomized according to a cut-off (0.57) optimized utilizing the minimum p-value for association with ORR on nivolumab. At this cut-off, 92 patients (51 of whom were nivolumab-treated) had a high PD-1 Treg/CD8 ratio, and 89 patients (40 of whom were nivolumab-treated) had low PD-1 Treg/CD8 ratio. In the nivolumab arm, patients with a high PD-1 Treg/CD8 ratio experienced shorter median PFS (3.48 versus 9.23 months, p<0.001), shorter median OS (18.14 versus 38.21 months, p<0.001) and lower ORR (15.69% versus 40.00%; p=0.009) compared to patients with a low PD-1 Treg/CD8 ratio. The association between a high PD-1 Treg/CD8 ratio and shorter PFS and OS was not observed in patients treated with everolimus (Figure 3).

Figure 3.

Kaplan-Meier curves for PFS (A, C) and OS (B, D) per PD1 Treg/CD8 ratio levels (minimum p-value cutoff) in the nivolumab (A, B) and everolimus (C, D) arm.

We then assessed the predictive value of the PD-1 Treg/CD8 ratio by comparing outcomes on nivolumab versus everolimus according to biomarker levels. Of note, a significant interaction between treatment and the PD-1 Treg/CD8 ratio (as continuous variable) was seen for both PFS and OS (2-sided p = 0.022 for PFS and 2-sided p = 0.028 for OS; significance determined as 2-sided p ≤0.10; Supplementary Table 6). The interaction was such that the higher the PD-1 Treg/CD8 ratio, the shorter the PFS and OS in patients treated with nivolumab but not in patients treated with everolimus. A similar finding was observed by performing MSKCC-stratified test (Supplementary Table 6). When the PD-1 Treg/CD8 ratio was analyzed as a binary variable by splitting at the median value, longer PFS and OS in the nivolumab arm versus the everolimus arm was observed in the low PD-1 Treg/CD8 ratio group but not in the high PD-1 Treg/CD8 ratio group (Supplementary Figure 5).

Lastly, we compared the predictive ability of the various biomarkers (i.e., PD-1 Treg/CD8 ratio, percentage of PD-1⁺ Tregs, percentage of CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells, and density of CD8⁺PD-1⁺TIM-3⁻LAG-3⁻ cells) using various statistical measures including interaction test, discrimination index, and comparison of clinical outcomes by arm. As shown in Table 3 and Supplementary Figure 6, the PD-1 Treg/CD8 ratio performed better than all other biomarkers in predicting both PFS and OS on nivolumab versus everolimus.

Table 3.

Comparison of the predictive value of the various biomarkers

	PFS			OS
Biomarker	Interaction Test p-value	2LogLikelihood Test Statistic*	C-Index**	Interaction Test p-value	2LogLikelihood Test Statistic*	C-Index**
PD-1 Treg/CD8 ratio	0.022	12.276	0.576	0.028	12.976	0.601
% of PD-1+ Tregs	0.274	9.167	0.561	0.067	12.910	0.589
% of CD8+PD-1+ TIM-3−LAG-3− cells	0.076	10.771	0.560	0.172	10.017	0.581
Density of CD8+PD-1+ TIM-3−LAG-3− cells	0.043	11.657	0.568	0.129	11.815	0.588

^*Test statistic for global null hypothesis (all parameters=0) which follows a Chi-squared distributed with 3 degrees of freedom (higher value is better; all models contain the same number of parameters).

^**Higher value is better

Discussion

Biomarkers such as PD-L1 expression in tumor and/or immune cells and tumor mutational burden are currently utilized to select patients for ICI therapy in several tumor types but, unfortunately, they have not proven to be clinically useful in ccRCC. Therefore, the development of clinically significant predictive biomarkers represents a priority for the kidney cancer research field. In this study, we showed that the expression of PD-1 on tumor-infiltrating Tregs is associated with lack of benefit from PD-1, but not everolimus (control), therapy in patients with mccRCC from the randomized CM-025 trial. Moreover, we developed a combined predictive model (PD-1 Treg/CD8 ratio) that integrates levels of PD-1 expression on Tregs and non-terminally exhausted CD8⁺ effector T cells and demonstrated that the combined model has improved predictive value compared to individual biomarkers. Overall, our work supports a scenario in which PD-1 blockade in Tregs and non-terminally exhausted CD8+ effector T cells have opposite effects on tumor growth. In fact, blocking PD-1 on Tregs leads to increased immune suppression and tumor progression, while blocking PD-1 on non-terminally exhausted CD8+ effector T cells leads to increased anti-tumor immunity and tumor regression (Figure 4). According to this scenario, a ccRCC tumor with relatively high PD-1 levels on Tregs and low PD-1 levels on non-terminally exhausted CD8+ effector T cells (high PD-1 Treg/CD8 ratio) is likely to be resistant to PD-1 blockade. In contrast, a ccRCC tumor with relatively low PD-1 levels on Tregs and high PD-1 levels on the non-terminally exhausted CD8⁺ effector T cells (low PD-1 Treg/CD8 ratio) is likely to respond to PD-1 blockade. These findings are consistent with recent data from Kumagai et al, obtained in small patient cohorts of other tumor types (16), and represent an important step toward the development of a clinically useful biomarker for the selection of patients with mccRCC who are most-likely to respond to anti-PD-1 therapy. We recognize that other factors, including somatic genetic alterations also play a role in therapeutic response (21–23), and ultimately, the integration of tumor genetics with microenvironment immunophenotypes may be necessary to better predict response and resistance with immune checkpoint inhibition.

Figure 4.

Schematic representation of the effects of PD-1 blockade on PD1⁺ Tregs and PD1⁺ non-terminally exhausted CD8⁺ effector T cells.

In our dataset, the percentage of PD-1⁺ Tregs (and the PD-1 Treg/CD8 ratio) showed a particularly strong association with OS in patients treated with nivolumab but not in patients treated with everolimus. While the explanation for this result remains under investigation, it should be noted that work from Kamada and colleagues recently suggested that hyperprogressive disease (i.e. rapid acceleration in tumor growth) on ICI treatment, which is observed in a minority of cancer patients treated with these agents, is mediated by PD-1⁺ Tregs (14). Therefore, it is possible that in a subset of mccRCC patients with high percentage of PD-1⁺ tumor-infiltrating Tregs, nivolumab therapy induces particularly high levels of Treg-mediated immunosuppression in the tumor microenvironment, which leads to more aggressive disease, and shorter overall survival. Further investigations are needed to formally test this hypothesis.

Our work has important clinical implications for the development of more effective anti-PD-1 combination therapies for patients with mccRCC, and possibly other tumor types. Taken together with previously published data, our observation that PD-1 expression on Tregs is associated with lack of response to anti-PD-1 monotherapy suggests that combining PD-1 blockade with a regimen that depletes Tregs would overcome the resistance mediated by the inhibition of PD-1 signaling in Tregs and thus lead to improved efficacy. In line with this hypothesis, nivolumab in combination with the anti-CTLA-4 antibody ipilimumab (which is thought to work at least in part by targeting Tregs) is effective for the treatment of mccRCC and represents an FDA-approved first-line option for patients affected by this disease. Of note, novel agents with enhanced ability to target Tregs have been recently developed and have shown promising results in clinical trials. As an example, botensilimab, an antibody that blocks CTLA-4 but also enhances T cell priming and depletes Tregs via its FcγR function, in combination with balstilimab (anti-PD-1) showed impressive anti-tumor activity in diseases where ICI therapy has shown little to no efficacy (i.e. microsatellite stable colorectal cancer) (24). On the basis of our findings, our group is conducting a multicenter phase II study (HCRN GU22–587, NCT05928806, PI: Atkins) comparing the combination of botensilimab and balstilimab (anti-PD-1) to the combination of nivolumab and ipilimumab in patients with treatment naïve mccRCC. Correlative studies of this trial will allow to formally test the hypothesis that improved efficacy of botensilimab /balstilimab over nivolumab/ipilimumab will be particularly evident in patients with high levels of PD-1 expression on intratumoral Tregs.

Tregs exert their immunosuppressive function through several mechanisms, which include promoting T cell exhaustion (19,20). However, the relationship between Tregs and exhausted T cells has not been investigated in depth in human tumors. In this study, we did not find any significant correlation between levels of Tregs (total or PD-1⁺) and levels of the terminally exhausted CD8⁺ effector T cells in ccRCC tumors. Moreover, by performing spatial analysis of the tumor microenvironment we did not observe any enrichment of terminally exhausted CD8⁺ effector T cells in proximity to Tregs (total or PD-1⁺). Although we had not anticipated these results, they are not entirely surprising as many extrinsic and intrinsic factors are known to regulate T cell exhaustion. Further studies are necessary to clarify the extent to which tumor-infiltrating Tregs contribute to the T cell exhaustion observed in ccRCC tumors.

The work presented here has several limitations. First, although we analyzed tissue samples from a randomized clinical trial, this is a retrospective analysis that included only a subset of the patients enrolled in the trial. Second, we mainly worked on primary tumor samples, which are characterized by intratumoral heterogeneity and do not always recapitulate the biology of distant metastases, which are the target of systemic therapy (25). Finally, the CM-025 study enrolled patients that previously failed VEGF-targeted therapies, but the tumor tissues available for analysis were mostly collected before the patients received any systemic treatment. Therefore, these samples are not representative of the changes that might have been induced by the initial therapy and thus may not completely recapitulate the immune composition of the tumor microenvironment at the time of the treatment with nivolumab or everolimus. It should be noted, however, that the concerns listed above are somewhat alleviated by the fact that the association between non-terminally exhausted CD8⁺ effector T cells and improved response to nivolumab treatment that we previously detected in the CM-025 cohort (5) was recently confirmed using a high quality collection of tumor samples from a clinical trial of frontline nivolumab in patients with mccRCC (6). Indeed, the independent validation of our findings suggests that despite the potential limitations, baseline tumor samples from the CM-025 trial represent a very valuable resource for biomarker discovery.

In conclusion, here we provide evidence that PD-1 expression on Tregs is associated with resistance to PD-1 blockade in mccRCC, suggesting that Treg-targeted therapy may synergize with PD-1 inhibition in this disease, and possibly other tumor types. We also show that a model that integrates PD-1 expression on Tregs and non-terminally exhausted CD8⁺ effector T cells displays higher predictive value compared to the individual biomarkers. Validation of these findings in independent clinical trial cohorts of mccRCC patients treated with frontline anti-PD-1 therapy is planned.

Statement of translational relevance.

Anti-PD(L)-1-based therapies are widely utilized for the treatment of metastatic clear cell renal cell carcinoma (mccRCC), but there are no FDA-approved biomarkers to guide patient selection. While our previous work indicates that patients with inflamed ccRCC tumors characterized by high levels of non-terminally exhausted CD8+ effector T cells experience improved clinical outcomes to PD-1 blockade, not all biomarker-positive patients respond to the treatment and mechanisms of resistance remain largely unknown.

By analyzing tissues specimens from a randomized clinical trial of nivolumab versus everolimus (CheckMate 025), we demonstrated that PD-1 expression on Tregs is associated with lack of clinical benefit from PD-1 blockade in mccRCC, suggesting that Treg-targeted therapy may synergize with PD-1 inhibition in this disease, and possibly other malignancies. Moreover, we also showed that a model that integrates PD-1 expression on Tregs and non-terminally exhausted CD8+ effector T cells displays higher predictive value compared to the individual biomarkers.