Differential Expression of PD-L1 between Primary and Metastatic Sites in Clear-Cell Renal Cell Carcinoma

The most common type of renal cell carcinoma (RCC) is clear-cell RCC (ccRCC), which represents >80% of cases and accounts for 2% to 3% of all adult malignant neoplasms (1). Median survival for patients with metastatic disease with approved targeted therapies remains poor and ranges from 8 to 30 months according to prognostic risk groups (2). Therefore, more effective systemic therapies for the treatment of advanced RCC are needed (3). For more than two decades, ccRCC has been recognized as an immunogenic tumor, and cytokine-based immunotherapy can produce durable responses in a small subset of patients (4–7).

Recent studies have demonstrated the role of the programmed death-1 (PD-1) T-cell coreceptor and its ligand PD-L1 (also known as B7-H1) in maintaining an immunosuppressive tumor microenvironment (8). The PD-1/PD-L1 pathway is known to be activated in many tumor types, including lung, ovarian, colorectal, breast, liver, head and neck, kidney, and bladder cancers, and melanoma (9). PD-1 is mainly expressed on tumor-infiltrating lymphocytes, whereas its ligand, PD-L1, is expressed on both hematopoietic cells (B, T, myeloid, and dendritic cells) and tumor cells (10). There is evidence that, similar to epithelial and stromal cells in normal tissues, tumor cells can express PD-L1 on the cell membrane in response to interferon gamma production by activated T cells. Thus, many tumors co-opt the natural physiology of the PD-1 pathway for tissue protection in the face of inflammation to protect themselves from an antitumor immune response. In line with this hypothesis, it has been shown that tumors expressing PD-L1 are able to inhibit antitumoral T-cell immunity by binding PD-1 on T cells (11).

It has been reported that PD-L1 is aberrantly expressed in human ccRCC and that patients with PD-L1–positive tumors display a higher risk of cancer-specific mortality (12–15). Currently, anti–PD-1 and anti–PD-L1 antibodies are actively being investigated in clinical development for metastatic ccRCC (8, 10), and several datasets suggest that primary ccRCC tumors with PD-L1 positivity either on tumor cell membranes or inflammatory cells achieve better response to PD-1/PD-L1–targeting therapies (16–19). Although PD-L1 expression in primary ccRCC tissue increases the likelihood of response to PD-1 pathway inhibition, it fails to identify all responders. Moreover, many patients with PD-L1–positive tumors do not respond to this therapy. Developing biomarkers that reliably predict response will be essential for narrowing the application of PD-1 blockade to those patients most likely to benefit.

Clear-cell RCC is characterized by intratumoral heterogeneity (20). We hypothesized that PD-L1 expression may vary significantly throughout the primary tumors (e.g., high nuclear grade versus low nuclear grade) and/or in the primary tumor versus the metastases and potentially constrain the predictive value of this biomarker. This knowledge is important to determine whether the development of optimal predictive models for PD-1/PD-L1 blockade can be conducted on primary tumor tissue or whether tissue from metastatic sites is likely to be more informative. For this reason, we performed an extensive analysis of PD-L1 expression in a series of primary ccRCCs and corresponding metastases (surgical resections). We assessed PD-L1 expression in both tumor cells and tumor-infiltrating immune cells.

A cohort of 53 primary ccRCC tumors and 76 corresponding metastases from 53 patients, who had undergone surgical tumor resections, were selected from two institutions: Brigham and Women's Hospital and Beth Israel Deaconess Medical Center (Boston, MA). Formalin-fixed paraffin-embedded (FFPE) tissue blocks from primary tumor and corresponding lymph node or distant metastases were retrieved. For each nephrectomy or metastasectomy specimen, all hematoxylin and eosin-stained slides containing tumor were reviewed by expert genitourinary pathologists (S. Signoretti, E.M. Genega, and M. Gupta). To address intratumoral morphologic heterogeneity, the nuclear grade was assessed in all slides using the criteria established by Fuhrman and colleagues (21). For each specimen, both areas of highest nuclear grade, also known as Fuhrman nuclear grade (FNG), and areas of predominant nuclear grade were selected for analysis.

PD-L1 expression was evaluated by immunohistochemistry (IHC) using a mouse monoclonal anti–PD-L1 antibody (405.9A11) developed by Dr. Gordon Freeman (Boston, MA). The assay was validated using FFPE cell line controls known to be either positive or negative for PD-L1 expression by flow cytometry (22).

Four-micron-thick tumor sections were stained with the anti–PD-L1 antibody (final concentration of 3.25 μg/mL) on a Benchmark XT autostainer (Ventana Medical System) with standard antigen retrieval (CC1 buffer, pH8.0, #950-124; Ventana). An UltraView Universal DAB Detection kit (#760-500; Ventana) was used according to the manufacturer's instruction. Counterstaining was performed as part of the automated staining protocol using hematoxylin (#760-2021; Ventana). After staining, slides were then washed in soapy water and distilled water, dehydrated in graded alcohol and xylene, mounted and cover slipped.

CD45 immunostaining was performed on adjacent 4-μm-thick tumor sections. Sections were initially deparaffinized, rehydrated, and heated with a pressure cooker to 125°C for 30 seconds in citrate buffer for antigen retrieval and then incubated with peroxidase (Dako #S2003) and protein blocking reagents (Dako #X0909) each for 5 minutes. Sections were then incubated with anti-CD45 (1:100, Dako, clone 2B11+PD7/26) antibody for 1 hour at room temperature, followed by incubation with the Dako EnVision+ System HRP-labeled polymer anti-mouse (Dako #K4001) for 30 minutes. All sections were developed using the DAB chromogen kit (Dako K3468) for 2 minutes and then lightly counterstained with hematoxylin.

Evaluation of PD-L1 expression in neoplastic cells and tumor-infiltrating mononuclear cells (TIMC) was independently performed by three pathologists (S. Signoretti, M. Gupta, and M. Callea), blinded to clinical data.

Membranous PD-L1 expression in tumor cells was quantified using an H-score (23), which takes into consideration the percentage of positive tumor cells within each staining category (0, negative; 1, weak; 2, moderate; 3, strong). In cases with focal positivity (<1%, positive tumor cells), the H-score was calculated considering the positive tumor cell percentage equal to 1. A case was considered positive when any tumor cell membrane positivity was detected. In addition, in cases with any positivity either in the primary tumor or in the metastases, we recorded PD-L1 status (positive vs. negative) in each nuclear grade component (1–4) present in the lesion.

The extent of TIMCs was evaluated on the basis of the immunoreactivity for CD45, a pan-leukocyte marker (24, 25), and recorded as absent (0), focal (1), mild (2), moderate (3), or marked (4). The percentage of PD-L1–positive TIMCs was determined according to six categories (0% = 0, ≤5% = 1, 6%–25% = 2, 26%–50% = 3, 51%–75% = 4, and >75% = 5). PD-L1 expression in TIMCs was then quantified using an immune cells adjusted score, calculated by multiplying the extent of TIMCs by the “percentage of positive cells” category (26). Any score greater than zero was considered positive.

The primary objective of this study was to characterize PD-L1 expression in primary ccRCC and their corresponding metastases, and to correlate the levels of expression with clinicopathologic features. Patient and tumor characteristics were summarized descriptively. When several samples were available within one primary or multiple metastatic sites, an average was calculated for each case (similar results were obtained when considering a maximum or median value). Proportions of positive PD-L1 expression in matching primary tumors and metastases from an individual case were compared with the exact McNemar test. Median H-score and median immune cells adjusted score in matching primary and metastatic cases were compared with the exact Wilcoxon signed rank test. Comparisons between PD-L1 expression and clinicopathologic features were evaluated using the Fisher exact test. All statistical computations were performed using Stata v.13.1 (StataCorp.), and a P value (two-sided) of <0.05 was considered statistically significant.

We collected tissue samples from 53 primary ccRCCs and 76 matching metastases. In all cases, the metastatic lesions had been removed by surgical excision, providing sufficient and representative tumor tissue for analysis. ccRCCs are characterized by considerable intratumoral morphologic heterogeneity, with areas of low nuclear grade frequently intermixed with areas of high nuclear grade. In order to address the impact of this heterogeneity, for each primary or metastatic lesion, tumor tissue blocks containing both areas of highest nuclear grade, also known as FNG, and areas of predominant nuclear grade were selected for analysis.

Metastatic sites included lung (n = 20), bone (n = 12), lymph node (n = 11), soft tissues (n = 9), adrenal gland (n = 8), pleura (n = 3), brain (n = 2), thyroid (n = 2), and others (n = 9). Although most primary tumors had only one matching metastasis, in 14 cases (26%), two or more metastatic lesions could be retrieved.

Patient characteristics are summarized in Table 1. The median age was 58 years (range, 40–85). Pathologic T stages at diagnosis were T1/T2 in 18 patients, T3/T4 in 32 patients, and unknown in 3 patients. No FNG I or II were identified in the cohort; 35 patients had FNG III and 18 had FNG IV.

Of the 53 cases analyzed, 17 cases (32%) presented PD-L1 tumor cell membrane positivity in the primary tumor and 12 cases (23%) presented PD-L1 tumor cell membrane positivity in the metastases (Table 2 and Fig. 1A and B). The percentage of positive tumor cells ranged between 0% and 40% in primary tumors and 0% and 70% in the metastases.

Discordant tumor cell PD-L1 staining between primary tumors and metastases was detected in 11 of 53 cases [20.8%, 95% confidence interval (CI); 10.8%–34.1%]. Of the 36 cases with primary tumors that did not express PD-L1, 33 cases were also PD-L1–negative in the metastases. Of the 17 cases with primary tumors that expressed PD-L1, only 9 cases also expressed PD-L1 in the metastases (Table 3 and Fig. 1C–F). Among the 11 discordant cases, 6 cases had less than a 3-month time interval between the resection of the primary tumor and the resection of the metastasis (Supplementary Table S1).

It should be noted that several samples were characterized by low percentage (<5%) of PD-L1–positive tumor cells, and only 6 cases (11%) showed ≥5% positive tumor cells in the primary tumor. Similarly, only 8 (15%) cases showed ≥5% positive tumor cells in the metastatic sites. Using the 5% cutoff, we observed discordant tumor cell PD-L1 staining between primary tumors and metastases in 6 of 53 cases (11.3%; 95% CI, 4.3%–23.0%).

In the 20 cases with positive PD-L1 expression in the primary tumors and/or metastases, tumor cell PD-L1 levels (determined by an H-score) were not significantly different in primary tumors compared with the metastatic sites [median H-score, 1.3 (0, 85) vs. 1.5 (0, 170); P = 0.25; Table 2).

All but one primary tumor and all metastases displayed PD-L1–positive TIMCs (range, 5%–75%). PD-L1 expression levels in TIMCs assessed by median immune cells adjusted score were not significantly different in primary tumors and metastases (4 vs. 3; P = 0.82; Table 2).

Among the 14 cases in which more than one metastatic lesion was analyzed, only 1 case (7%) was discordant for tumor cell PD-L1 positivity across the different metastases. Specifically, PD-L1 positivity was observed in a lung lesion but not in a pancreatic lesion. This case also did not present PD-L1 expression on the primary tumor. In the remaining 13 cases, all metastases were PD-L1 negative.

We correlated PD-L1 expression with pathologic features within the cohort of 53 primary tumors (Table 4). We observed that tumor cell PD-L1 positivity was detected in 2 of 18 cases (11.1%) with T stage 1/2, compared with 14 of 32 cases (43.8%) with T stage 3/4, P = 0.03. Furthermore, tumor cell PD-L1 positivity was more frequently detected in primary tumors with FNG IV (n = 12) versus tumors with FNG III (n = 5), P < 0.01.

Pathologic evaluation revealed that in both primary tumors and metastases, PD-L1 positivity was heterogeneous and only present in a subset of tumor cells. Because our analysis was purposely conducted on multiple morphologically different tumor areas that included both the predominant and the highest nuclear grade (i.e., FNG), we further correlated PD-L1 expression with the distinct nuclear grade components detected within each primary or metastatic lesion. We found that PD-L1 expression was strongly associated with areas of nuclear grade 3 or 4 (i.e., high grade; P < 0.001) while areas of nuclear grade 1 or 2 (i.e., low grade) were negative in all but one lesion (Fig. 2 and Supplementary Table S2). It should be noted that within the subset of PD-L1–positive cases, the coexistence of low nuclear grade (mostly PD-L1 negative) and high nuclear grade (PD-L1 positive) areas was observed in 18 of 20 (90%) primary tumors but only in 9 of 21 (43%) metastases. The vast majority of the remaining lesions (2 of 2 primaries and 10 of 12 metastases) were exclusively composed of high-grade tumor cells. Therefore, intratumoral heterogeneity for PD-L1 expression was extensive in primary tumors but more limited in metastases (Supplementary Table S2).

While systemic therapies targeting the VEGF–VEGFR axis and the mammalian target of rapamycin (mTOR) pathway represent major advances in the treatment of patients with metastatic RCC (mRCC), a plateau has been reached in terms of their impact on progression-free survival and overall survival (3). Very encouraging results have been obtained recently with new immunotherapy modalities that target immune checkpoints, including agents blocking the PD-1/PD-L1 pathway. It has been established that interaction of PD-1 with its ligands (PD-L1 and PD-L2) limits T-cell activation, and there is evidence that chronic antigen exposure increases PD-L1 levels in immune cells within the tumor microenvironment, resulting in T-cell “exhaustion” and reduced immune control of tumor progression. Of note, cancer cells can also express PD-L1 and directly contribute to the inhibition of an antitumor immune attack. In this regard, PD-L1 expression has been investigated in several tumor types as both a prognostic biomarker and a potential predictive factor of response to therapeutic antibodies that block the PD-1–PD-L1 axis.

Studies conducted by Thompson and colleagues first demonstrated that PD-L1 expression in RCC is associated with aggressive features, such as higher TNM stage, tumor size, or FNG, and increased risk of cancer-specific mortality (12–15). In these reports, the expression of PD-L1 in either tumor cells or tumor-infiltrating immune cells was found to be an indicator of poor prognosis.

Initial clinical investigations of PD-1- and PD-L1-targeting antibodies in mRCC have raised high expectations and suggested that PD-L1 expression might be a useful biomarker of response to PD-1/PD-L1 inhibition. To date, several distinct clinical trials have shown that responses to PD-L1/PD-1 inhibition are more frequently observed among ccRCC patients whose tumors are positive for PD-L1 expression (16–19, 27, 28). However, it has become increasingly clear that IHC staining for PD-L1 in nephrectomy specimens fails to identify all responders to PD-1/PD-L1 blockade. Indeed, up to 18% of patients with PD-L1–negative tumors have been found to respond to the treatment (19), while many patients with PD-L1–positive tumors fail to respond (18). Although there are several potential explanations for these results, it is possible that the predictive value of PD-L1 expression is negatively affected by tumor heterogeneity. Predictive tissue biomarker research is usually conducted by analyzing the primary tumor because it is easier to obtain. However, given the significant tumor heterogeneity in ccRCC, nephrectomy specimens may not accurately reflect the biology of the metastatic tumors that are the target of the systemic therapy. In line with this hypothesis, we found discordant tumor cell PD-L1 staining between primary tumors and corresponding metastases in a high proportion of cases (∼20%). In contrast, multiple metastases from the same patient presented limited discordance in PD-L1 expression (7%) in the relatively small number of samples that we analyzed (14 cases). Taken together, these data suggest that robust predictive models that include the assessment of PD-L1 expression in ccRCC tumor cells might require the analysis of tissue from metastatic lesions. This possibility should be tested in prospective clinical trials.

Our study also highlights the considerable intratumor heterogeneity of PD-L1 expression in ccRCC. We demonstrate for the first time that PD-L1–positive tumors (especially primary lesions) present considerable morphologic heterogeneity and harbor tumor areas of both low and high nuclear grade, with PD-L1 protein almost exclusively expressed in high nuclear grade areas. These findings have important implications for future predictive biomarker studies and imply that the random selection of tumor blocks for PD-L1 analysis might lead to false-negative results. To avoid this possible bias, we recommend that, in resected lesions characterized by morphologic heterogeneity, high nuclear grade areas should be specifically selected for assessment.

A recent study by Jilaveanu and colleagues explored PD-L1 expression in a cohort of 34 matched pairs of nephrectomy and metastatic tissue samples (29). The authors used an automated quantitative analysis (AQUA) method on a tissue microarray (TMA) consisting of four tissue cores per specimen. Similarly to our current work, they found that the correlation between tumor cell PD-L1 expression in matched primary and metastatic specimens was weak, and the study highlighted PD-L1 staining heterogeneity within one specimen. In contrast to our results, however, the median AQUA score was higher in metastatic sites than in primary specimens. Because our extensive analysis of whole tissue sections from both primary and metastatic tumors reveals that PD-L1 expression is highly heterogeneous and largely restricted to areas with aggressive pathologic features (i.e., high nuclear grade), it is possible that the analysis of only four tissue cores per lesion in a TMA is affected by considerable selection bias.

Several articles have described PD-L1 expression in primary ccRCC, and the reported rate of positivity is highly variable and ranges from 15% to 66% (13–19, 28–30). In the present study, we report a membranous tumor cell PD-L1–positivity rate of 32%, which is higher than the 23.9% rate previously reported by Thompson and colleagues (14). This difference can be ascribed to several factors, including the use of a different anti–PD-L1 antibody, the analysis of a metastatic patient population, the evaluation of multiple tumor blocks per primary tumor, and the fact that in our study a case was considered positive when any tumor cell positivity was detected, while in the study by Thompson and colleagues, cases with <5% tumor staining were considered negative. In our study, we decided to utilize any positivity as the cutoff for the following reasons: (i) the correlation between PD-L1 levels and inhibition of anticancer immunity is currently unknown, and any level of PD-L1 protein detected by IHC might have significant biologic consequences; (ii) the optimal cutoff for PD-L1 expression as a biomarker of response to PD-1/PD-L1 inhibitors still needs to be established, and recent clinical results show that responses can be achieved in patients whose tumors were considered negative using a 5% cutoff; (iii) pathologist-based evaluation is semiquantitative and subjective, and the reproducibility of discerning 1% versus 5% PD-L1–positive tumor staining is questionable.

One major limitation of the PD-L1 staining reports published to date, including ours, is the variability in staining methodologies that utilize antibodies that are not commercially available, and thus prevent a direct comparison of their performance. Standardization of both staining procedures and scoring methods is warranted before PD-L1 can be widely used as a predictive biomarker in the clinic.

Targeting the PD-1/PD-L1 interaction to reinvigorate the immune system is showing promising clinical efficacy in metastatic ccRCC, and the ability to select patients that are more likely to benefit from this therapeutic approach relies on the development of predictive biomarkers, such as PD-L1 expression. We report that discordant expression of PD-L1 between primary tumors and their metastases is detected in approximately 20% of cases, suggesting that accurate assessment of PD-L1 as a predictive biomarker for PD-1 blockade in ccRCC may require the analysis of metastatic lesions. Moreover, we found that PD-L1 staining is almost exclusively observed in the high-grade component of a tumor. This finding should guide pathologists to select appropriate tumor areas for PD-L1 IHC analysis.

L. Albiges is a consultant/advisory board member for Novartis and Pfizer. M.B. Atkins is a consultant/advisory board member for Bristol-Myers Squibb and Merck. F.S. Hodi reports receiving commercial research support from Bristol-Myers Squibb and is a consultant/advisory board member for Bristol-Myers Squibb, Genentech, Merck, and Novartis. D.F. McDermott is a consultant/advisory board member for Bristol-Myers Squibb, Genentech, and Merck. G.J. Freeman has ownership interest (including patents) in Amplimmune, Boehringer-Ingelheim, Bristol-Myers Squibb, EMD Serono, Merck, Novartis, and Roche, and is a consultant/advisory board member for Bristol-Myers Squibb, Eli Lilly, Novartis, and Roche. S. Signoretti is a consultant for Merck, Verastem, and AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Albiges, E.M. Genega, M.B. Atkins, F.S. Hodi, T.K. Choueiri, G.J. Freeman, S. Signoretti

Development of methodology: L. Albiges, E.M. Genega, J. Song, I. Carvo, G.J. Freeman, S. Signoretti

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Callea, L. Albiges, M. Gupta, E.M. Genega, A.P. Fay, J. Song, I. Carvo, R.S. Bhatt, M.B. Atkins, T.K. Choueiri, D.F. McDermott

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Albiges, M. Gupta, S.-C. Cheng, A.P. Fay, M.B. Atkins, F.S. Hodi, T.K. Choueiri, D.F. McDermott, S. Signoretti

Writing, review, and/or revision of the manuscript: L. Albiges, M. Gupta, S.-C. Cheng, E.M. Genega, A.P. Fay, M.B. Atkins, F.S. Hodi, T.K. Choueiri, D.F. McDermott, G.J. Freeman, S. Signoretti

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Callea, L. Albiges, J. Song, I. Carvo, R.S. Bhatt, T.K. Choueiri

Study supervision: T.K. Choueiri, S. Signoretti

This work was supported by the DF/HCC Kidney Cancer SPORE P50 CA101942-01 to S. Signoretti, D.F. McDermott, G.J. Freeman, and T.K. Choueiri. This study was also funded in part by the Trust Family, Michael Brigham, and Loker Pinard Funds for Kidney Cancer Research at Dana-Farber Cancer Institute for T.K. Choueiri.

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.