Advertisement
Translational Cancer Mechanisms and Therapy

Molecular Profiling of Cohorts of Tumor Samples to Guide Clinical Development of Pembrolizumab as Monotherapy

Mark Ayers, Michael Nebozhyn, Razvan Cristescu, Terrill K. McClanahan, Rodolfo Perini, Eric Rubin, Jonathan D. Cheng, David R. Kaufman and Andrey Loboda
DOI: 10.1158/1078-0432.CCR-18-1316 Published March 2019

Abstract

Purpose: Molecular profiling of large databases of human tumor gene expression profiles offers novel opportunities for informing decisions in clinical development programs.

Experimental Design: Gene expression profile of programmed death ligand 1 (PD-L1) was explored in a dataset of 16,000 samples, including approximately 4,000 metastatic tumors, across >25 tumor types prevalent in the United States, looking for new indications for the programmed death 1 (PD-1) inhibitor pembrolizumab. PD-L1 expression was highly concordant with several genomic signatures indicative of immune-inflamed tumor microenvironment. Prevalence of activated immune-inflamed tumors across all tumor types was explored and used to rank tumor types for potential response to pembrolizumab monotherapy.

Results: The analysis yielded 3 tiers of indications in which high levels of PD-L1 and immune-inflamed signatures were found in up to 40% to 60%, 20% to 40%, and 0% to 20% of tumors. Tier 1 contained novel indications known at the time of analysis to be responsive to PD-1 checkpoint blockade in the clinic (such as melanoma and non–small cell lung cancer), as well as indications not studied in the clinic previously, including microsatellite instability–high colorectal, head and neck, bladder, and triple-negative breast cancers. Complementary analysis of an Asian/Pacific cancer dataset (gastric cancer) revealed high prevalence of immune-inflamed tumors in gastric cancer. These data contributed to prioritization of these indications for clinical development of pembrolizumab as monotherapy.

Conclusions: Data highlight the value of molecular profiling in identifying populations with high unmet needs with potentially favorable response characteristics and accelerating development of novel therapies for these patients.

See related commentary by Mansfield and Jen, p. 1443

Translational Relevance

Large multidimensional datasets that include genome-wide gene expression and DNA profiles of thousands of tumor samples enable in silico discovery of gene expression patterns associated with tumor–immune interaction. Programmed death 1 (PD-1) is an immune checkpoint molecule that allows tumors to evade immune surveillance and pembrolizumab, a humanized IgG4κ mAb against PD-1 that directly blocks interactions between PD-1 and its ligands, PD-L1 and PD-L2. The initial clinical activity of pembrolizumab monotherapy in melanoma and non–small cell lung cancer generated great interest in the identification of other pembrolizumab-responsive cancer types. Therefore, a large dataset of gene expression profiles of human tumors was used to identify cancers that exhibit increased PD-L1 gene expression and more general patterns of high immune inflammatory gene expression. The strategy of using molecular profiling enabled rapid expansion of pembrolizumab monotherapy into new indications.

Introduction

Programmed death 1 (PD-1) is an immune checkpoint molecule that allows tumors to evade immune surveillance (1, 2). It is expressed on activated T cells (3, 4) that recognize tumor antigens and invade the tumor microenvironment. Activation of IFNγ signaling by cytolytic T cells transcriptionally activates expression of the PD-1 ligands, PD-L1 and PD-L2, leading to inhibition of effector T-cell functioning (2, 5), a process known as adaptive immune resistance. PD-1 is one of several immune checkpoint molecules expressed on effector T cells (1).

Large multidimensional datasets, such as The Cancer Genome Atlas (TCGA), which include genome-wide gene expression and DNA profiles of thousands of tumor samples enable in silico discovery of gene expression patterns associated with tumor–immune interaction. Newly discovered gene expression signatures can accurately discriminate tumors with high levels of T-cell infiltration, or “hot” tumors, from those with low levels of T-cell infiltration, or “cold” tumors; the presence of T-cell infiltration in the tumor has been associated with good prognosis across multiple tumor types (6–9). Gene signatures characterized by activation of immune-related genes, such as chemokines, were observed in patients with early-stage breast cancer and associated with extended disease-free survival (6). An immune-related 12-chemokine gene signature identified a subset of primary colorectal cancers containing distinct, tertiary, ectopic lymph node–like structures (10). These structures are associated with upregulation of chemokines and have been reported to occur in a variety of tumors types, including tumors of the ovaries, colon, and lungs (10). Generally, the structures are localized to the invasive margin of the tumor, where they recruit lymphocytes and limit cancer progression (11), and are associated with improved patient survival independent of tumor staging or treatment (10). These parallel observations were based on IHC assessment of tumor–immune cell interactions. Another example is the Immunoscore (Integrative Cancer Immunology, Paris, France), an immune stratification system that uses the type, functional orientation (characterized by immune signatures), density, and location of adaptive immune cells within the tumor (i.e., the immune contexture; ref. 6). The Immunoscore ranges from I0 to I4 and has a prognostic role in patients with colorectal cancer; for example, patients with a score of I4 had longer survival compared with patients with a score of I0 (6). Furthermore, patients who had colorectal cancer and a high Immunoscore at diagnosis had early-stage disease, whereas those with a low Immunoscore tended to develop late-stage disease (12). Therefore, immune profiling is prognostic in several malignancies and may help identify patients likely to respond to specific immunotherapies.

Pembrolizumab, a humanized IgG4κ mAb against PD-1 that directly blocks interactions between PD-1 and PD-L1/PD-L2, is approved by the FDA for the treatment of several malignancies (see Supplementary Table 1 for a full list of indications; ref. 13). The clinical development of pembrolizumab monotherapy (hereafter referred to as pembrolizumab, unless stated otherwise) began with the KEYNOTE-001 study (ClinicalTrials.gov identifier, NCT01295827), which evaluated the safety and preliminary clinical activity of pembrolizumab in solid tumors, and the study results showed registrational expansions in metastatic melanoma and non–small cell lung cancer (NSCLC; ref. 14). In patients with metastatic melanoma treated with pembrolizumab, increasing levels of PD-L1 expression were associated with higher objective response rate (ORR), progression-free survival (PFS) rate, and overall survival (15, 16). Similarly, membranous PD-L1 expression in ≥1% of NSCLC tumor cells (tumor proportion score ≥1%) was associated with increased pembrolizumab effectiveness in terms of ORR and PFS (17).

View this table:
Table 1.

Cancer subtypes listed in the Merck and H. Lee Moffitt Cancer Center molecular profiling datasets that were used in the current analysis

The initial clinical activity of pembrolizumab in melanoma and NSCLC in KEYNOTE-001 (14) generated great interest in the identification of other pembrolizumab-responsive cancer types. We hypothesized that other cancers with high expression of PD-L1 and coregulated genes would be most likely to contain subsets of pembrolizumab-responsive patients. Therefore, we used a large dataset of gene expression profiles of human tumors to identify cancers that exhibit increased PD-L1 gene expression and more general patterns of high immune-inflammatory gene expression. These findings steered the prioritization of subsequent clinical trials, leading to the rapid demonstration of clinical benefit beyond melanoma and NSCLC.

Materials and Methods

Clinical specimens

The work flow is depicted in a flow diagram in Supplementary Figure S1. Clinical specimens were provided by a molecular profiling dataset generated by a collaboration between Merck & Co., Inc. and the H. Lee Moffitt Cancer Center. This dataset of 16,000 primary and 4,000 metastatic tumors represents >25 different cancers (Table 1). All samples and data were obtained with appropriate institutional review board approvals. Gene expression data were obtained from primary and metastatic tumors of adult patients treated through the Total Cancer Care initiative, which was created by the H. Lee Moffitt Cancer Center (18). Gene expression data for the Asian/Pacific data-set in the Asian Cancer Research Group collaboration dataset, which contains a large cohort of gastric samples, were also used (19). Written informed consent was obtained from all patients per institutional guidelines and the Declaration of Helsinki.

Sample processing

Tumor tissue was collected and snap frozen within 15 minutes of surgical extirpation, with a typical cold ischemia time of <5 minutes. Frozen sections were then prepared, stained, and submitted for pathologist review. Macrodissection was performed in liquid nitrogen to maintain the frozen tissue and enrich the tumor content, followed by RNA extraction. The specimens were then arrayed on HuRSTA-2a520709 GeneChips (Affymetrix). Samples were processed in parallel in 96-well plates to minimize potential between-sample variation. Reaction purification was achieved using magnetic binding beads for cDNA and Qiagen RNeasy Kits (Qiagen) for cRNA purification. Amplification was carried out using NuGEN 50-ng protocol per manufacturer's instructions (www.nugen.com; NuGEN). GeneChip microarrays (Affymetrix) were loaded with the fragmented target sample/hybridization buffer mix using standard manual techniques and were hybridized for 18 hours at 45°C with vigorous mixing. The unbound portion of the samples was removed, and the remaining sample was stained through the binding of streptavidin-conjugated phycoerythrin to the hybridized target. Finally, the excess label was removed. The washing and staining steps were performed using the Affymetrix GeneChip Fluidics Station 450 per standard protocols.

The arrays were scanned using a GeneChip Scanner 3000 7G system (Affymetrix) with a 48-array autoloader. The scanner maintains the optimal temperature for the arrays before and during scanning. These chips contain >52,000 probe sets, representing >22,000 unique genes. Additional details regarding the platform (e.g., probe set annotation, chip definition files) have been made public and can be assessed as platform GPL10379 at the NCBI GEO website (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GPL10379). The standard RefRMA pipeline was implemented using Affymetrix Power Tools software with default settings. TCGA data were obtained from the TCGA website (https://tcga-data.nci.nih.gov/docs/publications/tcga/?).

Statistical analysis

Pearson correlation with PD-L1 (CD274) expression was calculated for the expression of 18,438 genes present in the transcriptomic characterization of the Merck–Moffitt (N ∼16,000) and TCGA (N ∼10,000) datasets. The reproducibility of correlations of all the genes on the array with PD-L1 profile was validated in an independent dataset (TCGA; Fig. 1).

Figure 1.
Figure 1.

PD-L1 gene coexpression. Pearson correlation with PD-L1 (CD274) expression was calculated for 18,438 genes present in the transcriptomic characterization of the Merck–Moffitt (N ∼16,000) and TCGA (N ∼10,000) datasets. The Pearson correlation across the two datasets was 0.81, indicating strong replication of the biology across the two datasets.

Data availability

Merck & Co., Inc.'s data sharing policy, including restrictions, is available at http://engagezone.msd.com/ds_documentation.php. Requests for access to the clinical study data can be submitted through the EngageZone site or via email to dataaccess{at}merck.com.

Results

Molecular profiling of PD-L1 expression and its biological context

PD-L1 expression profile and its coexpression signature were assessed in Merck–Moffitt (N ∼12,000 primary and N ∼4,000 metastatic) and TCGA (N ∼10,000 primary) datasets. The robustness of PD-L1 gene expression signature across the 2 datasets was explored by comparing the concordance of correlations of other genes with a PD-L1 profile. The correlation of correlations was 0.81 (Fig. 1), which indicated highly significant concordance in the observed correlations of individual genes with PD-L1 between the 2 datasets and established that largely the same common set of genes is most correlated with PD-L1 in both the Merck–Moffitt and TCGA datasets. The top gene correlated with PD-L1 in both datasets was PD-L2 (r = 0.7). Functional annotation via gene set enrichment analysis, which was performed on the top 100 genes that correlated positively with PD-L1 in both datasets, revealed enrichment in gene sets (provided by Ingenuity Inc.) associated with IFN signaling, cytotoxic T cells, and T-cell–active chemokines.

The pattern of PD-L1 expression was also compared with other immune signatures that have been reported in the literature. PD-L1 expression was strongly associated with previously described signatures indicative of T-cell activation, such as the 13-gene signature of the immunologic constant of rejection associated with the Immunoscore (6), cytolytic signature (20), and chemokine signature (ref. 10; minimum c > 0.8; P < 1 × 10–20; Fig. 2). These findings are consistent with the known role of T-cell–derived IFNγ as a driver of PD-L1 transcription (21). Likewise, a distinct gene expression signature of T cell–inflamed tumor microenvironment, which was independently derived through assessment of genes associated with pembrolizumab clinical response and which is being assessed as a potential diagnostic tool for pembrolizumab, is also highly correlated with PD-L1 expression and contains PD-L1 as one of the signature genes (22). These data indicate that a highly coordinated, regulated set of genes is active in a T-cell–inflamed microenvironment. In contrast, other signatures related to tumor cell–intrinsic signaling (23–26), or to distinct types of stromal and myeloid biology (27, 28), were not highly coregulated with PD-L1 (Fig. 2).

Figure 2.
Figure 2.

Comparison of the PD-L1 mRNA expression obtained in this study with four published cytolytic signatures. A particularly strong correlation was found with those reported by Rooney and colleagues and Galon and colleagues. Magenta represents elevated gene expression (relative to the median) and positive correlation; cyan represents decreased gene expression (relative to the median) and negative correlation. EMT, epithelial–mesenchymal transition; GEP, T-cell–inflamed signature; IL, interleukin.

Selection of the threshold

The levels of PD-L1 gene expression and associated cytolytic signatures were highly variable between and within tumor types across the Merck–Moffitt dataset. We hypothesized that this variability could be meaningful, deeply rooted in the causal processes that lead to a cytolytic immune response, and predictive of potential response to anti–PD-1 immunotherapy. Several tumor types were further explored to select an appropriate threshold for identifying the presence of tumors with potentially clinically relevant levels of PD-L1 gene expression. For each profiled sample or indication, the PD-L1 expression cut-off point for positivity was defined as the pan-cancer 75th percentile of PD-L1 expression in the Merck–Moffitt dataset. This pan-cancer cutoff was selected to be consistent with the point at which the distribution of PD-L1 expression in the Merck–Moffitt dataset deviates from a normal distribution (Supplementary Fig. S2), and the cutoff led the percentage of PD-L1–positive tumors to roughly align with response rates reported at the time of the analysis in melanoma, NSCLC, and renal cell carcinoma (RCC). It was also consistent with the cutoffs defined by the Youden index when comparing PD-L1 expression between colorectal microsatellite instability (MSI) and colorectal microsatellite stability (MSS), triple-negative breast cancers (TNBC), and luminal A breast cancers, and a combination of melanoma and NSCLC and remaining solid tumors in the Merck–Moffitt dataset (Supplementary Fig. S3).

Ranking of primary tumors using PD-L1 expression level

In Fig. 3A, the differences in PD-L1 expression are shown across the various tumor types for primary tumor samples included in the Merck–Moffitt dataset; this is reported as the percentage of tumors with PD-L1 positivity above the cutoff, as described in the Results. On the basis of this analysis, tumors could be grouped into 3 tiers: tier 1 (40%–60% of tumors), tier 2 (20%–40% of tumors), and tier 3 (<20% of tumors). The small overlap in confidence intervals across indications suggests that analysis of the Merck–Moffitt dataset is adequately powered to discriminate rankings in primary tumors based on the measurement of PD-L1 mRNA expression. In addition, the rankings of tumor types for metastatic tumor samples included in the Merck–Moffitt dataset were similar to those of primary tumors (Fig. 3B).

Figure 3.
Figure 3.

Percentages of primary (A) and metastatic tumors (B) with high PD-L1 (>75th percentile; with 95% confidence intervals); Merck–Moffitt dataset. Only tumor types with >100 primary tumors available for analysis are shown. A, Three tiers of indications were identified for primary tumors; high levels of PD-L1 expression were found in 40% to 60% (tier 1), 20% to 40% (tier 2), and <20% (tier 3) of tumors, respectively. B, Symbols on the x-axis represent information at the time of analysis and the indications selected for the KEYNOTE-012 study; coloring of symbols indicate the tier to which the primary tumors belong. JHU, Johns Hopkins University;TN, triple negative.

For 3 of the primary tumor sample types included in the Merck–Moffitt dataset, >50% of the samples were above the cutoff for PD-L1 positivity, as described in the Results; for 5 tumor types, ≥25% of samples exhibited high PD-L1 (Fig. 4). The highest proportion of samples above the 75th percentile for PD-L1 positivity was found for primary head and neck cancer (68%), followed by primary squamous cell lung cancer (61%), primary colorectal MSI (59%), and metastatic head and neck cancer (59%). The PD-L1 positivity threshold for primary adenocarcinoma lung cancer was 48%. Other tumors identified with high PD-L1 mRNA expression included primary and metastatic TNBC (40% for both), primary bladder cancer (36%), metastatic clear cell RCC (30%), and metastatic melanoma (40%). In contrast, HER2-positive, luminal A, and luminal B breast cancer subtypes had lower levels of PD-L1 mRNA expression than did TNBC (19%, 6%, and 15%, respectively, for primary and 14%, 4%, and 11%, respectively, for metastatic), as did primary/metastatic endometrial cancer (11%–14% above the positivity threshold).

Figure 4.
Figure 4.

Correlation of PD-L1–high prevalence in primary and metastatic tumors across different cancer types in Merck–Moffitt TCC (P = 1 × 10–5). CRC, colorectal cancer; HRD, homologous recombination deficiency; MSI-H, microsatellite instability–high; TCC, Total Cancer Care; TN, triple negative; UV, ultraviolet radiation.

The concordance of the percentages for primary compared with metastatic samples was significant (R2 = 0.84; P = 1.0 × 10–5; Fig. 4). Although the concordance of PD-L1–positive prevalence between metastatic and primary tumors was very high, our analysis revealed some important differences; for example, RCC and melanoma have relatively higher percentages of PD-L1–positive expression in metastatic tumors than in primary tumors, whereas those for lung cancer are higher in primary tumors than in metastatic tumors (Fig. 4).

Additional molecular profiling of tumors not prevalent in the United States

Some large tumor datasets are biased toward tumor types that are prevalent in the United States and are underpowered for accurate molecular profiling of tumors that occur predominantly in other parts of the world. Gastric cancer is an important example of such a tumor type. Early access to a large cohort of gastric cancer samples from the Asian Cancer Research Group collaboration (19) allowed evaluation of the PD-L1 expression in that tumor type. The overall prevalence of PD-L1–positive gastric tumors based on the pan-cancer 75th percentile in the other tumor types was approximately 25%, which placed gastric cancer in the tier 2 group of indications for anti–PD-1 development (Fig. 3). Specific subsets were highly enriched for PD-L1 expression. As expected, MSI-high (MSI-H) tumors had particularly high levels of PD-L1 expression (Supplementary Fig. S3). In addition, some MSS PD-L1–high tumors are associated with Epstein–Barr virus (EBV) infection, and these were also associated with high levels of PD-L1 expression (Supplementary Fig. S4).

Association of indication ranking by PD-L1 activation with ranking by mutational load

A high level of neoantigens can lead to immune infiltration and subsequent PD-L1 activation (20); high mutation load produced by somatic mutations in a tumor might increase neoantigen expression and elevate PD-L1. This hypothesis was explored by determining the relationship between the percentages of tumors with elevated mutational load and elevated PD-L1 across tumor types, revealing a significant positive association (Fig. 5). Notably, MSI tumors (colon), which have the largest mutational load across all indications, also demonstrated the highest proportion of PD-L1 activation. The association between mutational load and PD-L1 or cytolytic activity is much weaker within indications (Supplementary Table S2) because the two measurements reflect related but largely orthogonal biological processes.

Figure 5.
Figure 5.

Correlation between tumor mutation load and PD-L1 expression in TCGA dataset (https://tcga-data.nci.nih.gov/tcga). Each tumor type is colored according to its primary risk factor (smoking, HRD, MSI, UV, or “other”). CRC, colorectal cancer; ER+, estrogen receptor positive; HRD, homologous recombination deficiency; MSS, microsatellite stable; TN, triple negative; UV, ultraviolet radiation.

Impact of molecular profiling data on clinical trial design

The molecular profiling data shown in the results were taken into consideration in the development of KEYNOTE-012 (ClinicalTrials.gov, NCT01848834), a phase Ib, multicohort trial to evaluate pembrolizumab in patients with TNBC and head and neck, bladder, and gastric cancers that was initiated early in the pembrolizumab clinical development program. Patients were initially selected for PD-L1 positivity by use of IHC; subsequently, an expansion cohort in unselected patients with head and neck cancer was added. Clinical activity in PD-L1–unselected populations was confirmed in all 4 indications in this study; the ORRs and disease control rates were 19% and 44%, respectively, for cohort A TNBC (29), and the ORRs were 20%, 26%, and 22% for head and neck (30), bladder (31), and gastric cancers (32) for cohorts B, C, and D, respectively.

Relative ranking between primary and metastatic tumors was generally consistent. Exceptions were clear cell RCC, which was ranked as tier 3 for primary tumors and tier 2 for metastatic tumors, and gastric cancer, which was ranked as tier 2 for primary tumors and tier 3 for metastatic tumors (Fig. 4). There is evidence of antitumor activity with PD-1/PD-L1 pathway inhibition in patients with metastatic RCC. In the open-label phase III CheckMate-025 trial, PD-1 inhibition with nivolumab as second- or third-line therapy improved the ORR over that achieved with the mTOR inhibitor everolimus (25% vs. 5%) in patients with advanced/metastatic clear cell RCC (33). The combination of nivolumab plus ipilimumab is also associated with improved responses over standard-of-care sunitinib in a large phase III trial of 1,096 patients with treatment-naïve advanced, clear cell RCC (34). Pembrolizumab is under investigation in this patient population as monotherapy (KEYNOTE-427, NCT02853344) and in combination with the tyrosine kinase inhibitor axitinib (KEYOTE-426, NCT02853331). In a phase Ib, nonrandomized dose-finding and dose-expansion trial, the combination of pembrolizumab and axitinib conferred an ORR of 73% in a cohort of 52 patients with previously untreated, advanced RCC who were unselected for PD-L1 expression (KEYNOTE-035, NCT02133742; ref. 35).

Response to pembrolizumab has been evaluated in the tier 3 indications in the basket KEYNOTE-028 trial, mostly in small patient populations with different disease states, including advanced, heavily pretreated, PD-L1–positive, estrogen receptor–positive, HER2-negative breast cancer (n = 25; ref. 36); pretreated, locally advanced/metastatic PD-L1–positive endometrial cancer (n = 24; ref. 37); advanced, PD-L1–positive ovarian cancer (n = 26; ref. 38); and advanced, treatment-resistant, PD-L1–positive colorectal cancer (n = 23; ref. 39). In line with the tier 3 ranking for primary tumors, antitumor activity for these indications seemed modest, with low ORRs of 12.0%, 13.0%, 11.5%, and 4.0%, respectively. The single patient who experienced objective (partial) response among the colorectal cancer cohort had MSI-H disease (39). In a study of patients with mismatch repair–deficient colorectal cancer and non-colorectal cancer tumors and patients with mismatch repair–proficient colorectal cancer, pembrolizumab antitumor activity was found for those with mismatch repair–deficient tumors but not mismatch repair–proficient colorectal cancer (40). Trials being actively pursued in other tier 3 indications include metastatic, castration-resistant prostate cancer (as monotherapy in KEYNOTE-199, NCT02787005; in combination therapies in KEYNOTE-365, NCT02861573), advanced hepatocellular carcinoma (KEYNOTE-224, NCT02702414), previously treated metastatic colorectal cancer (KEYNOTE-164, NCT02460198), and recurrent or metastatic gastric or gastroesophageal junction adenocarcinoma (KEYNOTE-059, NCT02335411).

Discussion

Large genomic datasets provide a unique resource for hypothesis-generating and hypothesis-validating molecular profiling. These sets are becoming essential bioinformatics tools for the development of precision medicine in oncology. In this study, analyses performed across large datasets of genomic profiles were used to steer indication for the anti–PD-1 inhibitor pembrolizumab. This molecular profiling of genomic cohorts allowed linkage of emerging clinical and PD-L1 IHC biomarker data from the pembrolizumab clinical development program with genomic profiles in large datasets based on biomarker prevalence data, biological contextualization using prior knowledge of tumor immune activation signatures, and ranking of tumors based on a gene expression score. This, in turn, enabled mapping of emerging clinical data to the biomarker data, setting a cutoff for positivity, and enabled development of proposals for new indications for clinical study. Although this dataset did not utilize PD-L1 IHC data, IHC scores were moderately correlated with PD-L1 mRNA expression (41).

Analysis of PD-L1 coexpression patterns revealed a highly significant association with a previously developed signature indicative of activation of cytolytic T cells (10). The cytolytic pattern was subsequently further investigated and elucidated by Galon and colleagues (6) and Rooney and colleagues (20), who determined that activation of T cells and natural killer cells is coupled with IFN and chemokine signaling. Moreover, a recently published analysis of mRNA profiles of baseline tumor samples from pembrolizumab-treated patients on a NanoString nCounter platform (NanoString Technologies, Inc.,) yielded an 18-gene, T-cell–inflamed immune signature that predicted response to pembrolizumab across several solid tumors. The predictive value of this signature, which contained IFN-responsive genes related to antigen presentation, chemokine expression, cytotoxic activity, and adaptive immune resistance, compared favorably with PD-L1 IHC in PD-L1–unselected populations (22). The finding that PD-L1 expression is positively associated with the cytolytic process is consistent with the role of PD-L1 as part of a negative feedback loop that is activated in the presence of cytolytic activity. Thus, activation of PD-L1 is a biomarker of robust cytolytic activity suppressed by a tumor-induced microenvironment. A strong correlation between the cause (cytolytic activity) and the negative feedback (activation of PD-L1 and PD-L2) suggests a state of equilibrium between tumor and immune system that, in some tumors, can be shifted by anti–PD-1 therapy toward antitumor activity.

Large datasets, such as those of Merck–Moffitt and TCGA, are well powered to accurately identify all the genes that play a role in cytolytic activation and concomitant negative feedback. The correlation of cytolytic genes with PD-L1 expression in these two large datasets and the almost perfect overlap between distinct large gene sets demonstrates a robust association between the cytolytic machinery and tumor PD-L1 overexpression. These datasets are also powered to accurately assess the distribution of PD-L1 activation scores for each tumor type and identify statistically significant differences in the scores among the tumors, particularly for metastatic tumors. The Merck–Moffitt dataset included profiling of a large number of metastatic tumors (N ∼4,000) compared with the TCGA dataset, which notably lacks profiling of tumors in the metastatic setting in most indications. A similar ranking of indications was obtained in both datasets for primary tumors; however, notable exceptions in the ranking of primary versus metastatic tumors in the Merck–Moffitt dataset are RCC and melanoma, which have a significantly higher prevalence of PD-L1 positivity in metastatic lesions.

To further substantiate the biological rationale underlying indication selection, we demonstrated that ranking of tumors by mutational burden is consistent with ranking based on PD-L1 activation (Fig. 5). Mutational burden is one of the potential causal events underlying the cytolytic immune response and PD-L1 activation. Clinically validated indications at the time of our analyses, such as melanoma and NSCLC, and our candidate indications from tier 1 that were prioritized for clinical development, such as head and neck, bladder, and MSI-H colon cancers, not only have high levels of PD-L1 activation, but also have high rates of mutational burden. In gastric tumors, the observed PD-L1 activation was caused by not only mutational burden (MSI), but also EBV infection. The presence of viral infections, such as human papillomavirus and EBV, may explain PD-L1 activation in cervical tumors and in subsets of head and neck tumors.

The strong relationship between mutational rate and elevated PD-L1 expression is true from the point of view of a pan-cancer analysis but not when the relationship is studied in individual indications (except for MSI cancers; Supplementary Table S2). This is potentially because of the heterogeneity of causal events for cytolytic immune response and tumor escape mechanisms, which decouple the presence of neoantigens from the presence of cytolytic infiltrate and PD-L1 activation over the long-term evolution of the tumor.

Molecular profiling also guided the clinical development of pembrolizumab in tier 3 lower ranked tumor types. It was predicted that the likelihood of observing any meaningful clinical monotherapy activity in the PD-L1–unselected population from those indications would be low; therefore, a biomarker-based enrichment strategy was used. This led to the initiation of an ongoing, multicohort, nonrandomized phase Ib basket clinical trial (KEYNOTE-028, NCT02054806) of pembrolizumab monotherapy (10 mg/kg dosed every 2 weeks) in more than 450 patients across 20 different tumor types. The basket trial design allowed for the clinical evaluation of tumors from different tumor types in a single study. In KEYNOTE-028, patients were enrolled whose tumors expressed elevated levels of PD-L1, but not as high as the cutoff selected to separate tier 1 tumors from tier 3 tumors. The exact levels of PD-L1 expression necessary to achieve substantial response rates among tier 3 tumor types remain to be determined.

Pembrolizumab may be active as monotherapy for the biomarker-selected patients in tier 3 indications, and its use in conjunction with other therapies in these indications may be even more effective. Indeed, pembrolizumab in combination with pemetrexed and carboplatin is already approved by the FDA as a first-line option for metastatic nonsquamous NSCLC (13), and multiple clinical trials are under way in which combination therapy with pembrolizumab and a broad range of agents are being tested for initial safety and efficacy across indications, including those in tier 3. In this rapidly evolving field, many hypotheses regarding combination activity will be clinically tested, ranging from specific immune-based combination mechanisms to broadly orthogonal approaches with standard chemotherapies and newer targeted therapies. The results of such trials must be evaluated to understand the molecular basis for successful and unsuccessful combination therapies. In combination with the results of ongoing clinical trials, a deeper understanding of tumor molecular subtypes and characteristics of immune infiltration will potentially offer insight into mechanisms of resistance to pembrolizumab and approaches to overcoming that resistance.

As a testimony of the validity of our analytic approach and the value of molecular databases to drive clinical development, pembrolizumab has been approved in 3 of the 4 indications prioritized for rapid clinical development as a result of the effort described in this article (head and neck, urothelial carcinoma, and gastric cancer; Supplementary Table 1).

In conclusion, the strategy of using molecular profiling of PD-L1 expression in a large dataset of gene expression profiles of human tumors has enabled rapid expansion of pembrolizumab development into new indications and has been validated by the significant levels of clinical activity subsequently reported for those indications.

Disclosure of Potential Conflicts of Interest

M. Ayers holds ownership interest (including patents) in Merck. T.K. McClanahan holds ownership interest (including patents) in Merck. R.F. Perini holds ownership interest (including patents) in Merck. J. Cheng holds ownership interest (including patents) in Merck. No potential conflicts of interest were disclosed by the other authors.

Authors' Contributions

Conception and design: M. Ayers, M. Nebozhyn, R. Cristescu, T.K. McClanahan, J. Cheng, A. Loboda

Development of methodology: M. Ayers, M. Nebozhyn, R. Cristescu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Nebozhyn, J. Cheng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Ayers, M. Nebozhyn, R. Cristescu, R.F. Perini, J. Cheng, D.R. Kaufman, A. Loboda

Writing, review, and/or revision of the manuscript: M. Ayers, M. Nebozhyn, R. Cristescu, T.K. McClanahan, R.F. Perini, E.H. Rubin, J. Cheng, D.R. Kaufman, A. Loboda

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Nebozhyn, R. Cristescu

Study supervision: M. Ayers, J. Cheng

Acknowledgments

The authors thank Heather Hirsch, Hongyue Dai, Theresa Zhang, and Rob Iannone (former employees of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ) for their contributions to the development of this manuscript. Medical writing and/or editorial assistance was provided by Matthew Grzywacz, PhD, of the ApotheCom pembrolizumab team. This assistance was funded by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ. Funding for this research was provided by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ.Funding for this research was also provided by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ.

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.

Footnotes

  • Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

  • Prior presentation: The data herein were presented in part at the American Association for Cancer Research Annual Meeting 2015; April 18–22, 2015.

  • Received April 27, 2018.
  • Revision received June 29, 2018.
  • Accepted August 16, 2018.
  • Published first November 15, 2018.

References

  1. 1.↵
    1. Pardoll DM
    . The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer 2012;12:252–64.
    OpenUrlCrossRefPubMed
  2. 2.↵
    1. Hirano F,
    2. Kaneko K,
    3. Tamura H,
    4. Dong H,
    5. Wang S,
    6. Ichikawa M,
    7. et al.
    Blockade of B7-H1 and PD-1 by monoclonal antibodies potentiates cancer therapeutic immunity. Cancer Res 2005;65:1089–96.
    OpenUrlAbstract/FREE Full Text
  3. 3.↵
    1. Agata Y,
    2. Kawasaki A,
    3. Nishimura H,
    4. Ishida Y,
    5. Tsubata T,
    6. Yagita H,
    7. et al.
    Expression of the PD-1 antigen on the surface of stimulated mouse T and B lymphocytes. Int Immunol 1996;8:765–72.
    OpenUrlCrossRefPubMed
  4. 4.↵
    1. Keir ME,
    2. Butte MJ,
    3. Freeman GJ,
    4. Sharpe AH
    . PD-1 and its ligands in tolerance and immunity. Ann Rev Immunol 2008;26:677–704.
    OpenUrlCrossRefPubMed
  5. 5.↵
    1. Latchman Y,
    2. Wood CR,
    3. Chernova T,
    4. Chaudhary D,
    5. Borde M,
    6. Chernova I,
    7. et al.
    PD-L2 is a second ligand for PD-1 and inhibits T cell activation. Nat Immunol 2001;2:261–8.
    OpenUrlCrossRefPubMed
  6. 6.↵
    1. Galon J,
    2. Angell HK,
    3. Bedognetti D,
    4. Marincola FM
    . The continuum of cancer immunosurveillance: prognostic, predictive, and mechanistic signatures. Immunity 2013;39:11–26.
    OpenUrlCrossRefPubMed
  7. 7.↵
    1. Schumacher K,
    2. Haensch W,
    3. Roefzaad C,
    4. Schlag PM
    . Prognostic significance of activated CD8(+) T cell infiltrations within esophageal carcinomas. Cancer Res 2001;61:3932–6.
    OpenUrlAbstract/FREE Full Text
  8. 8.↵
    1. Sato E,
    2. Olson SH,
    3. Ahn J,
    4. Bundy B,
    5. Nishikawa H,
    6. Qian F,
    7. et al.
    Intraepithelial CD8+ tumor-infiltrating lymphocytes and a high CD8+/regulatory T cell ratio are associated with favorable prognosis in ovarian cancer. Proc Natl Acad Sci U S A 2005;102:18538–43.
    OpenUrlAbstract/FREE Full Text
  9. 9.↵
    1. Pagès F,
    2. Berger A,
    3. Camus M,
    4. Sanchez-Cabo F,
    5. Costes A,
    6. Molidor R,
    7. et al.
    Effector memory T cells, early metastasis, and survival in colorectal cancer. N Engl J Med 2005;353:2654–66.
    OpenUrlCrossRefPubMed
  10. 10.↵
    1. Coppola D,
    2. Nebozhyn M,
    3. Khalil F,
    4. Dai H,
    5. Yeatman T,
    6. Loboda A,
    7. et al.
    Unique ectopic lymph node-like structures present in human primary colorectal carcinoma are identified by immune gene array profiling. Am J Pathol 2011;179:37–45.
    OpenUrlCrossRefPubMed
  11. 11.↵
    1. Sautes-Fridman C,
    2. Lawand M,
    3. Giraldo NA,
    4. Kaplon H,
    5. Germain C,
    6. Fridman WH,
    7. et al.
    Tertiary lymphoid structures in cancers: prognostic value regulation, and manipulation for therapeutic intervention. Front Immunol 2016;7:407.
    OpenUrl
  12. 12.↵
    1. Mlecnik B,
    2. Tosolini M,
    3. Kirilovsky A,
    4. Berger A,
    5. Bindea G,
    6. Meatchi T,
    7. et al.
    Histopathologic-based prognostic factors of colorectal cancers are associated with the state of the local immune reaction. J Clin Oncol 2011;29:610–8.
    OpenUrlAbstract/FREE Full Text
  13. 13.↵
    Merck Sharp & Dohme Corp. KEYTRUDA® (pembrolizumab) for injection, for intravenous use. Whitehouse Station (NJ): Merck Sharp & Dohme Corp.; 2018.
  14. 14.↵
    1. Kang SP,
    2. Gergich K,
    3. Lubiniecki GM,
    4. de Alwis DP,
    5. Chen C,
    6. Tice MAB,
    7. et al.
    Pembrolizumab KEYNOTE-001: an adaptive study leading to accelerated approval for two indications and a companion diagnostic. Ann Oncol 2017;28:1388–98.
    OpenUrl
  15. 15.↵
    1. Kefford R,
    2. Ribas A,
    3. Hamid MZ
    . Clinical efficacy and correlation with tumor PD-L1 expression in patients (pts) with melanoma (MEL) treated with the anti-PD-1 monoclonal antibody MK-3475. J Clin Oncol 2014;32:3005.
    OpenUrlAbstract/FREE Full Text
  16. 16.↵
    1. Daud AI,
    2. Wolchok JD,
    3. Robert C,
    4. Hwu WJ,
    5. Weber JS,
    6. Ribas A,
    7. et al.
    Programmed death-ligand 1 expression and response to the anti-programmed death 1 antibody pembrolizumab in melanoma. J Clin Oncol 2016;34:4102–9.
    OpenUrl
  17. 17.↵
    1. Garon EB,
    2. Rizvi NA,
    3. Hui R,
    4. Leighl N,
    5. Balmanoukian AS,
    6. Eder JP,
    7. et al.
    Pembrolizumab for the treatment of non–small-cell lung cancer. N Engl J Med 2015;372:2018–28.
    OpenUrlCrossRefPubMed
  18. 18.↵
    1. Fenstermacher DA,
    2. Wenham RM,
    3. Rollison DE,
    4. Dalton WS
    . Implementing personalized medicine in a cancer center. Cancer J 2011;17:528–36.
    OpenUrlCrossRefPubMed
  19. 19.↵
    1. Cristescu R,
    2. Lee J,
    3. Nebozhyn M,
    4. Kim KM,
    5. Ting JC,
    6. Wong SS,
    7. et al.
    Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med 2015;21:449–56.
    OpenUrlCrossRefPubMed
  20. 20.↵
    1. Rooney MS,
    2. Shukla SA,
    3. Wu CJ,
    4. Getz G,
    5. Hacohen N
    . Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell 2015;160:48–61.
    OpenUrlCrossRefPubMed
  21. 21.↵
    1. Garcia-Diaz A,
    2. Shin DS,
    3. Moreno BH,
    4. Saco J,
    5. Escuin-Ordinas H,
    6. Rodriguez GA,
    7. et al.
    Interferon receptor signaling pathways regulating PD-L1 and PD-L2 expression. Cell Rep 2017;19:1189–201.
    OpenUrl
  22. 22.↵
    1. Ayers M,
    2. Lunceford J,
    3. Nebozhyn M,
    4. Murphy E,
    5. Loboda A,
    6. Kaufman DR,
    7. et al.
    IFN-gamma-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 2017;127:2930–40.
    OpenUrlCrossRefPubMed
  23. 23.↵
    1. Loboda A,
    2. Nebozhyn M,
    3. Klinghoffer R,
    4. Frazier J,
    5. Chastain M,
    6. Arthur W,
    7. et al.
    A gene expression signature of RAS pathway dependence predicts response to PI3K and RAS pathway inhibitors and expands the population of RAS pathway activated tumors. BMC Med Genomics 2010;3:26.
    OpenUrlCrossRefPubMed
  24. 24.↵
    1. Loboda A,
    2. Nebozhyn MV,
    3. Watters JW,
    4. Buser CA,
    5. Shaw PM,
    6. Huang PS,
    7. et al.
    EMT is the dominant program in human colon cancer. BMC Med Genomics 2011;4:9.
    OpenUrlCrossRefPubMed
  25. 25.↵
    1. van't Veer LJ,
    2. Dai H,
    3. van de Vijver MJ,
    4. He YD,
    5. Hart AA,
    6. Mao M,
    7. et al.
    Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002;415:530–6.
    OpenUrlCrossRefPubMed
  26. 26.↵
    1. Loboda A,
    2. Nebozhyn M,
    3. Cheng C,
    4. Vessey R,
    5. Huang P,
    6. Dai H,
    7. et al.
    Biomarker discovery: identification of a growth factor gene signature. Clin Pharmacol Ther 2009;86:92–6.
    OpenUrlCrossRefPubMed
  27. 27.↵
    1. Yoshihara K,
    2. Shahmoradgoli M,
    3. Martinez E,
    4. Vegesna R,
    5. Kim H,
    6. Torres-Garcia W,
    7. et al.
    Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 2013;4:2612.
    OpenUrlCrossRefPubMed
  28. 28.↵
    1. Casneuf T,
    2. Axel AE,
    3. King P,
    4. Alvarez JD,
    5. Werbeck JL,
    6. Verhulst T,
    7. et al.
    Interleukin-6 is a potential therapeutic target in interleukin-6 dependent, estrogen receptor-alpha-positive breast cancer. Breast Cancer 2016;8:13–27.
    OpenUrl
  29. 29.↵
    1. Nanda R,
    2. Chow LQ,
    3. Dees EC,
    4. Berger R,
    5. Gupta S,
    6. Geva R,
    7. et al.
    Pembrolizumab in patients with advanced triple-negative breast cancer: phase Ib KEYNOTE-012 study. J Clin Oncol 2016;34:2460–7.
    OpenUrlAbstract/FREE Full Text
  30. 30.↵
    1. Chow LQ,
    2. Haddad R,
    3. Gupta S,
    4. Mahipal A,
    5. Mehra R,
    6. Tahara M,
    7. et al.
    Antitumor activity of pembrolizumab in biomarker-unselected patients with recurrent and/or metastatic head and neck squamous cell carcinoma: results from the phase Ib KEYNOTE-012 expansion cohort. J Clin Oncol 2016;34:3838–45.
    OpenUrl
  31. 31.↵
    1. Plimack ER,
    2. Bellmunt J,
    3. Gupta S,
    4. Berger R,
    5. Chow LQ,
    6. Juco J,
    7. et al.
    Safety and activity of pembrolizumab in patients with locally advanced or metastatic urothelial cancer (KEYNOTE-012): a non-randomised, open-label, phase 1b study. Lancet Oncol 2017;18:212–20.
    OpenUrl
  32. 32.↵
    1. Muro K,
    2. Chung HC,
    3. Shankaran V,
    4. Geva R,
    5. Catenacci D,
    6. Gupta S,
    7. et al.
    Pembrolizumab for patients with PD-L1-positive advanced gastric cancer (KEYNOTE-012): a multicentre, open-label, phase 1b trial. Lancet 2016;17:717–26.
    OpenUrl
  33. 33.↵
    1. Motzer RJ,
    2. Escudier B,
    3. McDermott DF,
    4. George S,
    5. Hammers HJ,
    6. Srinivas S,
    7. et al.
    Nivolumab versus everolimus in advanced renal-cell carcinoma. N Engl J Med 2015;373:1803–13.
    OpenUrlCrossRefPubMed
  34. 34.↵
    1. Motzer RJ,
    2. Tannir NM,
    3. McDermott DF,
    4. Aren Frontera O,
    5. Melichar B,
    6. Choueiri TK,
    7. et al.
    Nivolumab plus ipilimumab versus sunitinib in advanced renal-cell carcinoma. N Engl J Med 2018;378:1277–90.
    OpenUrl
  35. 35.↵
    1. Atkins MB,
    2. Plimack ER,
    3. Puzanov I,
    4. Fishman MN,
    5. McDermott DF,
    6. Cho DC,
    7. et al.
    Axitinib in combination with pembrolizumab in patients with advanced renal cell cancer: a non-randomised, open-label, dose-finding, and dose-expansion phase 1b trial. Lancet Oncol 2018;19:405–15.
    OpenUrl
  36. 36.↵
    1. Rugo HS,
    2. Delord JP,
    3. Im SA,
    4. Ott PA,
    5. Piha-Paul SA,
    6. Bedard PL,
    7. et al.
    Safety and antitumor activity of pembrolizumab in patients with estrogen receptor-positive/human epidermal growth factor receptor 2-negative advanced breast cancer. Clin Cancer Res 2018;24:2804–11.
    OpenUrlAbstract/FREE Full Text
  37. 37.↵
    1. Ott PA,
    2. Bang YJ,
    3. Berton-Rigaud D,
    4. Elez E,
    5. Pishvaian MJ,
    6. Rugo HS,
    7. et al.
    Safety and antitumor activity of pembrolizumab in advanced programmed death ligand 1-positive endometrial cancer: results from the KEYNOTE-028 study. J Clin Oncol 2017;35:2535–41.
    OpenUrl
  38. 38.↵
    1. Varga A,
    2. Piha-Paul SA,
    3. Ott PA,
    4. Mehnert JM,
    5. Berton-Rigaud D,
    6. Morosky A,
    7. et al.
    Pembrolizumab in patients (pts) with PD-L1–positive (PD-L1+) advanced ovarian cancer: updated analysis of KEYNOTE-028. J Clin Oncol 2017;35:5513.
    OpenUrl
  39. 39.↵
    1. O'Neil BH,
    2. Wallmark JM,
    3. Lorente D,
    4. Elez E,
    5. Raimbourg J,
    6. Gomez-Roca C,
    7. et al.
    Safety and antitumor activity of the anti-PD-1 antibody pembrolizumab in patients with advanced colorectal carcinoma. PLoS One 2017;12:e0189848.
    OpenUrlCrossRefPubMed
  40. 40.↵
    1. Le DT,
    2. Uram JN,
    3. Wang H,
    4. Bartlett BR,
    5. Kemberling H,
    6. Eyring AD,
    7. et al.
    PD-1 blockade in tumors with mismatch-repair deficiency. N Engl J Med 2015;372:2509–20.
    OpenUrlCrossRefPubMed
  41. 41.↵
    1. Liu X,
    2. Lee J,
    3. Kim KM,
    4. Kim ST,
    5. Park SH,
    6. Kang WK,
    7. et al.
    Comprehensive investigation of programmed death receptor ligand 1 (PD-L1) expression and associated molecular features in gastric cancer patients. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14–18; Chicago, IL. Philadelphia (PA): AACR; 2018. Abstract nr 4215.
Back to top
View this article with LENS

Open full page PDF
Article Alerts
Email Article
Citation Tools
Molecular Profiling of Cohorts of Tumor Samples to Guide Clinical Development of Pembrolizumab as Monotherapy
Mark Ayers, Michael Nebozhyn, Razvan Cristescu, Terrill K. McClanahan, Rodolfo Perini, Eric Rubin, Jonathan D. Cheng, David R. Kaufman and Andrey Loboda
Clin Cancer Res March 1 2019 (25) (5) 1564-1573; DOI: 10.1158/1078-0432.CCR-18-1316
del.icio.us logo Digg logo Reddit logo Twitter logo CiteULike logo Facebook logo Google logo Mendeley logo

Jump to section

Advertisement