Programmed Cell Death Ligand-1 (PD-L1) and CD8 Expression Profiling Identify an Immunologic Subtype of Pancreatic Ductal Adenocarcinomas with Favorable Survival

Pancreatic adenocarcinoma (PDAC) is the fourth most common cause of cancer-related death in the United States, and the incidence of PDAC is rising worldwide (1, 2). Most cases of PDAC are metastatic disease at the time of diagnosis, and the median survival for these patients is less than one year (2). This highlights the continued need for new and effective therapies in PDAC. Systemic chemotherapy is the mainstay of treatment for PDAC. Although novel immune-checkpoint inhibitors targeting programmed cell death protein 1 (PD-1) or its ligand (PD-L1) or cytotoxic T lymphocyte–associated protein 4 (CTLA-4) have activity in many different cancers, they have thus far failed to show meaningful clinical benefit in unselected patients with PDAC. The objective response rate to immune-checkpoint therapy in unselected patients with PDAC has been approximately 5% in prior clinical trials (3–5). Based on its poor response rate to current systemic immunotherapies, PDAC is largely considered to be a poorly immunogenic cancer type. To optimally identify patients who may have intact antitumor immune responses and/or benefit from immunotherapeutic strategies, an improved immunologic characterization of PDACs is needed.

Despite its resistance to systemic immunotherapies, an integrated genomic analysis has identified that a subset of PDACs (the “immunogenic” subset) have upregulated pathways involved in acquired immune suppression with enhanced signaling through the PD-1 and CTLA-4 pathways (6). Within this subset of PDAC, patients with high macrophage or T-cell coinhibition scores had shorter overall survival (OS). PD-L1 expression in the tumor microenvironment has also been identified as a poor prognostic factor for survival in other independent cohorts of patients with PDAC (7–12). This suggests the possibility that although immune-checkpoint inhibitors may be insufficient to produce clinical responses in unselected populations of PDAC, there may be subsets of PDAC where acquired immune suppression through the PD-L1 pathway is, nonetheless, an important factor for PDAC disease progression.

Subtyping of the immune microenvironment in PDAC offers the opportunity for therapeutic development through personalization of systemic immunotherapies. Mismatch-repair deficiency (MMRd) is the only biomarker currently used to identify cases of PDAC that are likely to respond to immune-checkpoint inhibitors (13–15). Although PD-L1 expression in tumors is one of the most established biomarkers of active immunoregulation, it is less informative without contextualization. Although PD-L1 expression has been interrogated in multiple cohorts of PDAC, only a few studies have investigated the importance of PD-L1 expression in combination with CD8 expression and other adaptive immune-resistance pathways that may contribute to PDAC immunosuppression (8, 9, 16–18). Using clinical survival and RNA expression data from The Cancer Genome Atlas (TCGA), we investigated the relationship between PD-L1 expression and survival in patients with pancreatic cancer in combination with tumor-infiltrating immune subsets, and identified additional immune checkpoints that are upregulated in PDAC and may be implicated in tumor progression. We also sought to confirm the significance of these relationships using a complementary, IHC method in an independent patient cohort. Our findings provide grounds for immune subtyping of PDAC and infer opportunities for therapeutic development of immunotherapy in PDAC.

RNA sequencing (RNA-seq) data Level 3 RSEM normalized data for 152 pancreatic adenocarcinoma (PAAD) patients (19) from TCGA were accessed from the Broad Institute TCGA GDAC Firehose (https://ezid.cdlib.org/id/doi:10.7908/C11G0KM9) and log₂-transformed. Mutation annotation files (MutSig 2.0) for 146 pancreatic adenocarcinoma patients were downloaded from the Broad Institute TCGA GDAC Firehose's analysis_2015_04_02, (http://gdac.broadinstitute.org/). As a measure of mutational burden, we used the number of nonsynonimus mutations per sample that were log₁₀-transformed for the further analysis. Corresponding clinical data were obtained from Liu and colleagues (20). Data analysis was performed using R/Bioconductor software (version 3.5.0) with build-in packages and custom routines.

This research was performed in accordance with the Declaration of Helsinki. The evaluation of archived pathology specimens described in this article was approved by the Institutional Review Board at Johns Hopkins (NA-00080582), which does not require individual written patient consent. Cases of PDAC collected from patients who had undergone a pancreatic resection with curative intent from 2008 to 2012 were identified using the Johns Hopkins Pancreas Database, a repository of patient clinical data. We excluded cases that were lost to follow-up or for whom long-term survival data were unknown. To specifically enrich for differences in tumor biology, we subsequently selected patients who died as a result of PDAC recurrence in less than 500 days from (short-term survivors, n = 18) or recurred with PDAC but remained alive for at least 1,000 days (long-term survivors, n = 15).

The clinical diagnosis for each case was confirmed by a pathologist (R. Anders) through review of the corresponding hematoxylin and eosin slides. One representative formalin-fixed paraffin-embedded block from each tumor specimen was selected. Consecutive 5-μm-thick sections were cut from each block and mounted on glass slides. Automated staining was performed on the Leica Bond RX (Leica Biosystems) for CD8 and IDO1. Slides were baked and dewaxed online, followed by antigen retrieval, followed by blocking for nonspecific binding using Protein block (X090930, Agilent Technologies). Detection was performed using the Bond Polymer Refine Kit (DS9800, Leica Biosystems). Samples were counterstained and dehydrated through a graded-alcohol series into Histoclear II (508990150, Thermo Fisher Scientific) and coverslipped using Ecomount (5082832, Biocare Medical). Manual staining was performed for PD-L1 staining as the protocol previously described (21). Additional staining reagent information for CD8, IDO1, and PD-L1 is listed in Supplementary Table S1.

CD8 density was determined using a quantitative image analysis approach. Whole slides were scanned at 20× objective equivalent (0.49 μm/pixel) using a digital slide scanner (Nanozoomer, Hamamatsu). Image analysis (HALO Indica Labs) was used to determine the density (# of cells/surface area analyzed) of CD8-expressing lymphocytes within the tumor compartment. Scoring of tumor PD-L1 and IDO1 expression was performed manually by a trained pathologist (R. Anders), who was blinded to the clinical features of the case, utilizing a cutoff value of 1% of tumor cells with positive staining. Representative PD-L1 and IDO1 IHC images are shown in Supplementary Fig. S1.

To estimate a threshold for overexpression of immune-checkpoint markers, we used the Memorial Sloan Kettering Cancer Center cBioPortal (22). We found that approximately 10% (range, 5%–17%; Supplementary Fig. S2) of TCGA PDACs had expression higher than one standard deviation above the mean of the reference population used to define the cBioPortal z-score, and used this 10% cutoff to define checkpoint positivity. We defined high and low lymphocyte infiltration as values above and below median expression in tumor samples, and similarly defined high and low CD8⁺ T-cell density as above and below the median for the Johns Hopkins University samples. Data analysis was performed using R/Bioconductor software (22) with build-in packages and custom routines. Heat maps, box-and-whisker plots, and scatter plots were created using the gplots package and build-in R graphic functions.

For the correlation of overexpression with 5-year survival, we split samples to high versus low using the top 10% for each marker. Kaplan–Meier survival curves were made using the survfit function from the survival package, and hazard ratios and P values from log-rank tests were reported. To test the association of continuous expression with 5-year survival, we fitted the univariate Cox proportional hazards regression model as implemented in the coxph function from the R survival package (Supplementary Table S2).

For the Th1/IFNγ gene signature, we combined genes from the published Th1 signature from Gentleman and colleagues (22) and genes from IFNγ signaling pathway from the Reactome database (http://www.reactome.org; Supplementary Table S3). Pearson correlation coefficients were reported for correlation of expression of PD-L1 (CD274) and IDO1 with CD8A, CD68, GZMB, NOS2, and the Th1/IFNγ gene signature, and the Spearman correlation coefficients were reported for mutational load analyses.

To define candidate immune checkpoints, including PD-L1, that may induce tumor immunosuppression in the tumor microenvironment in PDAC and provide prognostic insight, we correlated tumor expression of immune-modulating pathways with 5-year survival curves using RNA expression data from TCGA. Many PDAC tumors upregulated only a few, if any, adaptive immune pathways, but upregulation of select adaptive immune-resistance pathways, including PD-L1, was observed in a subset of PDAC tumors. The expression of immune checkpoints in PDAC showed limited clustering (Fig. 1; Supplementary Table S4). Upregulation of PD-L1 (CD274) frequently occurred with upregulation of IDO1, upregulation of CTLA4 frequently co-occurred with upregulation of TIGIT and TNFRSF18, and LAG3 upregulation frequently co-occurred with upregulation of IDO1 and CD40.

When analyzing the 5-year OS data, tumor expression of PD-L1 (CD274) was associated with poor survival in PDAC [P = 0.001, hazard ratio for survival (HR) = 2.55 (CI, 1.15–5.64)], consistent with prior reports (Fig. 2A; refs. 8–12). Expression of indoleamine 2, 3-dioxygenase (IDO1) was also associated with a poor OS [P = 0.008, HR = 2.24 (0.96–5.2)]. However, no association between the expression of multiple other immune checkpoints, e.g., LAG3, TIGIT, and TIM-3 (HAVCR2), in the tumor microenvironment and survival in PDAC specimens was seen (Fig. 2B). Expression of CTLA-4 in the tumor microenvironment was marginally correlated with increased survival. Across many tumor types, a high density of tumor-infiltrating lymphocytes is a favorable prognostic indicator (23, 24). Based on mRNA expression from the TCGA data, however, we did not find any significant relationship between expression of any canonical immune subset markers within the tumor, including CD8A, CD4, CD19, or CD68, and survival (all P > 0.05).

Upon identifying that PD-L1 (CD274) and IDO1 expression conferred a poor prognosis in PDAC, we subsequently investigated factors that are known to contribute to their expression. Although most pancreatic cancers have a low tumor mutational burden, a small percentage of pancreatic cancers have MMRd, which results in a failure to repair errors in base pair mismatches in tumor DNA and a high tumor mutational burden (25–28). Anti–PD-1 therapy has shown clinical activity in a small subset of human cancers with MMRd, including select cases of PDAC (14, 15). We hypothesized that a high tumor mutational burden might lead to an increased density of neoantigen-specific CD8⁺ lymphocytes in the tumor microenvironment and, thus, upregulation of adaptive immune pathways by the tumor cells such as PD-L1 (CD274) or IDO1. Therefore, we investigated whether the expression of PD-L1 (CD274) or IDO1 was associated with tumor mutational burden in PDAC. We did not observe any relationship between the tumor mutational burden and either PD-L1 or IDO1 expression in TCGA (Supplementary Fig. S3A and S3B). However, only two samples in this cohort had MMRd with a mutation in MLH1, MSH2, MSH6, or PMS2. We also examined the relationship between these markers and tumor stage, hypothesizing that adaptive immune resistance might increase with increasing tumor stage. Most patients (n = 131 out of 152) in TCGA had stage 2 disease, limiting the power of this analysis, but we did not observe any signal for increased expression of PD-L1 (CD274) or IDO1 with increased tumor stage (Supplementary Fig. S4).

CD8⁺ T cells in the tumor microenvironment can produce interferon-gamma (IFNγ), leading to upregulation of adaptive immune-resistance pathways, including the PD-1/PD-L1 axis and IDO1 (29, 30). Therefore, we examined whether expression of CD8A or other immune subsets in the tumor microenvironment was associated with expression of PD-L1 (CD274) or IDO1 (Fig. 3). Consistent with our hypothesis, a positive relationship between CD8A and PD-L1 (CD274) was seen (r = 0.39, P = 5.09 × 10^–7), as well as between CD8A and IDO1 (r = 0.59, P = 6.3 × 10^–16), and expression of IDO1 and PD-L1 (CD274) was also positively correlated with one another (r = 0.54, P = 1.03 × 10^–12; Fig. 3A–C). PD-L1 (CD274) can also be expressed by monocyte-lineage cells, so we examined the relationship between PD-L1 (CD274) and CD68, a marker of macrophages and other monocyte-lineage cells. A positive association between CD68 and PD-L1 (CD274) expression in the microenvironment was also observed (r = 0.33, P = 3.32 × 10^–5; Fig. 3D). The association between PD-L1/IDO1 expression and CD8A/CD68 expression extended to granzyme B (GZMB), a marker of CD8⁺ T-cell cytolytic activity and NOS2 (also known as iNOS), a marker for the M1 macrophage phenotype (Fig. 3E–I).

We also examined the relationship between PD-L1 (CD274) and IDO1 expression and a combined Th1/IFNγ gene signature (Fig. 4). A Th1 bias is characterized by the secretion of IFNγ and is primarily responsible for activating an antitumor immune response. If PD-L1 (CD274) and IDO1 arise as a result of a cytotoxic T-cell antitumor response in the tumor immune microenvironment in PDAC, we hypothesized that expression of both molecules would be associated with a Th1/IFNG gene signature. Consistent with this hypothesis, we found that expression of both PD-L1 (CD274) and IDO1 was similarly associated with the Th1/IFNG gene signature.

Although CD8A expression and the Th1/IFNG gene signature were associated with PD-L1 (CD274) expression, some patients had elevated CD8A expression in the absence of PD-1/PD-L1 expression. Because the different expression patterns of PD-L1 (CD274) and CD8A may implicate biologically and possibly clinically different behavior, we then looked at the survival of patients with elevated or normal PD-L1 expression, stratified by CD8A expression (Fig. 5A). Although CD8A expression alone did not correlate with survival in PDAC (Supplementary Table S2), patients with the best survival over five years had increased CD8A infiltration without expression of PD-L1, whereas patients with PD-L1 expression without increased CD8A infiltration had the worst survival. Similar results were obtained when the cohort was stratified by IDO1 and CD8A expression, with the high CD8A expression and low IDO1 expression group showing the highest survival (Fig. 5B). These data suggest that the combined analysis of CD8A and PD-L1 (or IDO1) expression yields different subtypes of PDACs; specifically, our results suggested that CD8A expression in the absence of PD-L1 or IDO1 expression is associated with improved survival in PDAC.

To extend our observations based on mRNA expression data from the TCGA, we sought to further confirm the relationship between PD-L1/IDO1 expression and poor patient survival, as well as improved survival in the subtype with high CD8 expression but low PD-L1/IDO1 expression by analyzing the protein expression in an independent cohort of PDAC from Johns Hopkins University (n = 33). Banked PDAC tumor samples were analyzed from patients who had undergone a resection for PDAC with curative intent but died as a result of PDAC recurrence in 500 days or less (short-term survivors) or recurred with PDAC but remained alive for at least 1,000 days (long-term survivors). Most patients with PDAC recur after resection, and the selective use of patients with either poor or long survival after disease recurrence allowed us to investigate specific differences in disease biology that might contribute to the differences in survival for these patients. Samples were stained for PD-L1, IDO1, and CD8 by IHC. CD8 expression was quantified using digital image analysis to determine the density of CD8⁺ cells. Expression of PD-L1 and IDO1 on tumor cells was quantified by a pathologist (R. Anders) who was blinded to the clinical history of the patients. The clinicopathologic characteristics of the 33 PDAC samples analyzed are shown in Table 1.

Of the samples stained, 11 (33.3%) had positive PD-L1 expression and 7 (21.2%) had positive IDO1 expression on tumor cells. No significant association between PD-L1 status or IDO1 expression and any clinical tumor features, including pathologic stage, histologic grade, resection margin, venous or perineural invasion (P > 0.05 for all comparisons), was seen. Consistent with TCGA data, patients with positive PD-L1 expression by IHC had worse survival than patients without detectable PD-L1 expression [P = 0.01, HR = 2.61 (1.03–6.61); Fig. 6A]. A trend toward worse survival in the IDO expression group was also observed but did not meet the threshold for statistical significance (P = 0.068; Fig. 6B). An overlap of IDO1 and PD-L1 expression occurred in this cohort, with 6/7 IDO1⁺ samples also staining positive for PD-L1. A larger separation of survival curves was seen when the six PD-L1⁺IDO1⁺ samples were compared with samples without expression of either marker [n = 21; P = 0.000382, HR = 4.51 (CI: 0.94–21.53); Fig. 6C].

We also sought to confirm our findings from TCGA data showing that higher CD8A expression in the absence of PD-L1 (CD274) expression was associated with improved survival in PDAC. Given the small number of PD-L1⁺ cases, we separated the cohort into three groups: PD-L1⁺, PD-L1⁻/CD8^low, and PD-L1⁻/CD8^high. Again, consistent with our prior TCGA findings, the group of patients with high CD8 expression in the absence of PD-L1 expression (PD-L1⁻/CD8^high) had the best survival, whereas PD-L1⁺ cases and PD-L1⁻ cases with low CD8 expression had similar survival (Fig. 6D). Stratified analysis using IDO1 and CD8 expression only showed a trend toward improved prognosis for the IDO1⁻/CD8^high subtype (P = 0.11; Fig. 6E).

In this study, we aimed to characterize a subtype of PDAC based on the expression of immune-modulating pathways in combination with CD8A expression, and explore their clinical relevance by correlation with survival. Using mRNA expression data from TCGA and protein expression data via IHC from an independent cohort, we have shown the following: (i) high PD-L1 expression is associated with poor survival in PDAC; (ii) PD-L1 expression in PDAC is associated with CD8A expression, IDO1 expression, and a Th1/IFNγ gene signature; and (iii) although CD8A density alone was not predictive of improved survival in PDAC, PD-L1⁻/CD8^high expression identified a subtype with favorable survival PDAC. Our findings further strengthen prior observations (6) that despite being widely described as a poorly immunogenic tumor type, the endogenous antitumor immune response is a critical factor for PDAC survival. Specifically, our data suggest that the coassessment of PD-L1 expression and CD8 expression density may provide a better assessment of the immunologic state of PDACs and, in turn, the patient survival than expression of PD-L1 alone. Our findings also infer potential opportunities for development of immunotherapeutic strategies based on immuno-subtyping of PDACs.

PD-L1 expression has been interrogated in multiple PDAC cohorts. Overall, prior studies based on select cohorts (8, 9, 17, 18) are consistent with our observation, including a meta-analysis of such reports (16), demonstrating that PD-L1 expression is associated with poor prognosis in PDAC. However, these studies have not contextualized the PD-L1 expression by CD8 expression and corresponding survival outcomes. Contextualization of PD-L1 expression is not trivial. In unique clinical contexts, such as diffuse large B-cell lymphomas, PD-L1 expression can be overexpressed on tumor cells as a result of genetic events (31), but such genetic events are not thought to occur in PDAC. More often, PD-L1 expression on tumor cells can be induced by IFNγ as a consequence of an antitumor T-cell response, for which CD8⁺ T cells constitute the major effector arm of antitumor immunity (32). Accordingly, the correlation between PD-L1 expression and CD8 expression, as well as a Th1/IFNγ gene signature, in our study suggests that PD-L1 expression in PDAC arises as a mechanism of adaptive immune resistance to an antitumor T-cell response. These findings, accompanied by the different survival outcomes in the PD-L1/CD8 expression subtypes, underscore the importance of considering CD8 expression when studying the PD-L1 expression in this tumor type.

Immune-checkpoint inhibitors have thus far failed to demonstrate meaningful clinical benefit in PDAC. Thus, an improved understanding of the PDAC tumor microenvironment, thereby identifying subsets of patients who may derive benefit from these therapies, is needed to improve upon current therapies. The association of PD-L1 expression and poor survival suggests that PD-L1 may be a driver of disease progression in a subset of patients with PDAC. Despite a lack of compelling radiographic responses to PD-1 pathway inhibitors in unselected populations, our data suggest that there may a subset of PD-L1⁺/CD8^high PDAC patients for whom blocking the PD-1 pathway may reinvigorate CD8⁺ T-cell responses and lead to improved survival. Our results are consistent with a recent integrated genomic analysis of PDAC that identified the existence of an “immunogenic” disease subset (6), but their analysis did not stratify PDAC based on PD-L1 or CD8 alone. Referring to the immunologic classification paradigm described by Teng and colleagues (33), based on the clinical experience with melanoma patients, “Type I” cancers are those with high presence of tumor-infiltrating lymphocytes and PD-L1 expression. Type I is the subtype that primarily responds to immune-checkpoint inhibitor (ICI) therapies but constitutes only a minority of patients with PDAC in our study. This provides a possible explanation as to why PDACs in aggregate do not seem to respond to ICIs and again proposes the importance of refining patient selection for immunotherapy trials. This hypothesis requires clinical validation in prospective clinical trials. For patients with lower T-cell infiltration without and with PD-L1 expression, i.e., “Type II” and “Type III,” respectively, it is likely that PD-1 pathway inhibitors will need to be used in combination with other forms of immunotherapy such as vaccines to first induce a T-cell response and a compensatory upregulation of the PD-1 pathway (34). Such combinatorial approaches to recruit immune cells and enhance immune recognition may effectively convert these types into an immunotherapy-responsive state. This is an especially critical challenge for PDACs given that most patients have relatively low CD8⁺ T-cell infiltration based on our analysis.

Our study recognized IDO1 expression as a correlate of poor survival in PDACs consistent with previous observations (35, 36), but the clinical evidence that targeting IDO1 can improve outcomes in human cancers is lacking thus far (37). Another finding was that there was very little clustering of immune checkpoints in PDAC, with a subset of patients highly expressing at least one adaptive immune-resistance pathway. This suggests that immunotherapies could eventually be paired on the basis of immune biomarkers in the tumor microenvironment, but the feasibility of a personalized approach to immunotherapy in PDAC requires prospective investigation.

Strengths of our study include the use of two different disease cohorts with well-documented clinical outcomes and the use of sensitive antibodies to stain clinical samples in parallel. Limitations include the retrospective nature of our analyses and selection biases inherent in our cohort, such as a tertiary care practice referral bias. We demonstrated longer survival for PD-L1⁻/CD8^high patients and poor survival for PD-L1⁺ patients with PDAC. Although PDAC is widely regarded as a poorly immunogenic cancer type, these findings support the notion that an active antitumor immune response is a feature of long-term survival in PDAC and infer opportunities for the development of novel immunotherapies based on data-supported biomarkers to refine patient selection.

W.J. Ho has ownership interest (including stock, patents, etc.) in Rodeo Therapeutics. R. Anders has received honoraria from the speakers bureau of Bristol-Myers Squibb, Merck, and AstraZeneca. E.M. Jaffee reports receiving a commercial research grant from Bristol-Myers Squibb, Aduro Biotech, and Amgen, has ownership interest (including stock, patents, etc.) in Aduro Biotech, and is a consultant/advisory board member for CStone, Dragonfly, Genocea, and Adaptive Biotechnologies. M. Yarchoan reports receiving a commercial research grant from Bristol-Myers Squibb, Exelixis, and Merck & Co, and is a consultant/advisory board member for Eisai and Exelixis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Q. Zhu, A. De Jesus-Acosta, N.S. Azad, D.A. Laheru, R. Anders, E.M. Jaffee, M. Yarchoan

Development of methodology: L. Danilova, Q. Zhu, N.S. Azad, D.A. Laheru, R. Anders, E.M. Jaffee, M. Yarchoan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Danilova, T. Vithayathil, A. De Jesus-Acosta, D.A. Laheru, R. Anders, M. Yarchoan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Danilova, W.J. Ho, Q. Zhu, D.A. Laheru, E.J. Fertig, R. Anders, E.M. Jaffee, M. Yarchoan

Writing, review, and/or revision of the manuscript: L. Danilova, W.J. Ho, Q. Zhu, A. De Jesus-Acosta, N.S. Azad, D.A. Laheru, E.J. Fertig, R. Anders, E.M. Jaffee, M. Yarchoan

Study supervision: D.A. Laheru, E.M. Jaffee, M. Yarchoan

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.