Programmed cell death ligand 1 cut-point is associated with reduced disease specific survival in resected pancreatic ductal adenocarcinoma

A tissue microarray was constructed using duplicate 0.6 mm cores from the epithelial component of all available, resected, pathologically confirmed pancreatic ductal adenocarcinomas derived from the archives of the Vancouver Coastal Health Region between 1995 and 2014. All patients received primary surgery with a subset receiving adjuvant chemotherapy with a pyrimidine nucleoside analog. Cores for the tissue microarray were obtained from areas of tumor as determined by routine microscopy on hematoxylin and eosin-stained sections. Cases were excluded if they lacked clinical follow-up data or if clinico-pathologic variables were lacking.

Immunohistochemistry was performed on 4-μm-thick formalin-fixed paraffin-embedded sections of tissue microarrays. PD-L1 immunohistochemistry was performed at the Deely Research Centre at the British Columbia Cancer Agency using the Intellipath FLX autostainer (Biocare) platform. Mismatch repair, CD3 and CD8 immunohistochemistry was performed in the clinical laboratory of Vancouver General Hospital using the Ventana Discovery XT and the Ventana Benchmark XT automated system (Ventana Medical Systems, Tucson, AZ).

For PD-L1, slides were incubated with the clone SP142 (Spring Bioscience, Pleasanton, USA) at 1/100 dilution in Da Vinci Green diluent at room temperature for 30 min. Slides were then washed and incubated with Mach2 Rabbit-HRP polymer for 30 min at room temperature and detected with IP DAB chromogen for 5 min. Nuclei were counterstained with a 1/10 dilution of CAT hematoxylin then slides were again washed, air dried and coverslipped with Ecomount. The antibody clone was selected based on its strong concordance to three other PD-L1 clones and RNA in situ hybridization (ISH) in NSCLC [ 19].

For MMR stains, slides were incubated with MLH1 (mouse monoclonal antibody, 1:50 dilution, cat#: NCL-L- MLH1, clone ID:ES05; Leica Microsystems, New- castle, UK), MSH2 (mouse monoclonal antibody, 1:1000 dilution, cat#: 286 M-16, clone ID:G219–1129; Cellmarque, Rocklin, CA), MSH6 (rabbit monoclonal antibody, 1:200 dilution, cat#: CLAC-0047, clone ID: EP49; Cedarlane Corporation, Burlington, ON, Canada), and PMS2 (rabbit monoclonal antibody, 1:20 dilution, cat#: CLAC-0049, clone ID:EP51; Cedarlane Corporation) for 32 min at room temperature. For the slides to be stained for PMS2 were additionally prepped with the Epitomics DAB prep kit before antibody incubation.

Antibodies were detected using the Ventana DABMap kit, counterstained with hematoxylin and treated with a proprietary bluing agent (Ventana). Positive and negative controls are performed as part of the routine clinical quality assurance; in addition to the external quality control program (Canadian Immunohistochemistry Quality Control (cIQc), a provider of proficiency testing for Canadian clinical laboratories).

PD-L1 status was assessed independently by two anatomical pathologists (BSS and DG) and consensus achieved on all cases. Positivity was evaluated by H-Score, a combination of staining intensity and percentage of tumor cell staining. Staining intensity was scored as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong) based on membranous localization and each score multiplied by the percentage of cells (0% - 100%) staining. Therefore, H-scores ranged from 0 to 300. To account for potential intra-tumoral heterogeneity, the mean of both cores was used to generate the score for each case.

Mismatch repair (MMR) was quantified as per Riazy et al. [ 20]. Briefly, protein expression for MLH1, MSH2, MSH6, and PMS2 was considered intact (normal) if any percentage of definite positive nuclear staining of the malignant cells was detected on either TMA core. In cases where one or more mismatch repair proteins were interpreted as negative staining, examination utilizing immunohistochemistry on whole sections was performed. Each protein was considered lost (abnormal) if there was complete loss of nuclear staining in the tumor cells and if there was a positive internal control (intact nuclear staining of stromal elements such as inflammatory cells and/or endothelial cells) on whole section. Cases showing a complete absence of nuclear staining pattern of both tumor cells and stromal elements were deemed uninterpretable and thus excluded from the study. Cases that demonstrated loss of any MMR marker on the TMA were subjected to confirmatory whole slide section staining and were scored independently by two pathologists (BSS and DFS), who were blinded to clinical characteristics and patient outcomes. Divergent assessments were reconciled by consensus conference. A case was labeled as mismatch repair deficient (MMRd) if any of the four mismatch repair proteins was completely absent on immunohistochemistry. Cases were classified as mismatch repair proficient (MMRp) if all four proteins stained positive to some degree.

Individual tumor infiltrating lymphocytes were counted and typed in the epithelial and stromal compartments using clinically validated IHC stains for CD3 and CD8. Scoring was performed independently by two anatomical pathologists (BSS and MA-K) and consensus achieved on all cases. To account for potential intra-tumoral heterogeneity, the average of both cores were used to generate the final score for CD8+ and CD3+ tumor infiltrating lymphocytes.

Standard treatment, clinical and pathologic parameters were collected from the British Columbia Cancer Agency which included: age at surgery, sex, adjuvant chemotherapy agents used, lymphovascular invasion, perineural invasion, pathologic primary tumor (pT) stage, and pathologic regional lymph-node status (pN). The primary outcome measure was defined to be disease-specific survival, where survival time was calculated as the difference between the date of last follow-up and the date of surgery, expressed in years. Patients were censored if they were alive at last follow-up or had died from a cause other than their pancreatic malignancy. Deaths attributable to treatment related toxicities or inter-current diseases were considered censored observations for this analysis.

To determine if H-Score or the percent of positive cells for PD-L1 expression yield differential prognostic ability, each scoring method was subjected to an omnibus assessment utilizing the Cox-Proportional Hazards Model to determine if the expression of PD-L1 was a significant prognostic marker in the context of the clinico-pathologic variables outlined previously with the exception of pT-Stage due the fact that the vast majority of the cases in this cohort are pT3. The proportionality assumption for each variable was assessed through examination of Cox-Snell residuals and continuous variables were assessed for linearity. The PD-L1 scoring methodology with the smallest P-value was determined to have the strongest prognostic effect. Parametric survival analysis was used in order to further elucidate the gradient dependent effect of PD-L1 expression on disease specific survival (DSS) for the scoring methodology with the greatest prognostic effect. This procedure modelled the disease specific survival data with 5 different distributions which included: weibull, log-normal, exponential, frechet, and log-logistic. The best distribution to be used for parametric survival analysis was determined by selecting the one with the lowest Bayesian Information Criterion from the model fits. This analysis produces a quantile plot with logSurvival Time plotted against PD-L1 expression which illustrates the gradient dependent relationship between disease specific survival and PD-L1 expression. Based on these findings, a series of three cut-points were created starting with an H-Score or percentage positive cells of 1- as these identify the equivalent cases. Subsequent cut-points were set at increments of 5 which correspond to the increments used for the assessment of percent positive cells. The resultant groups were subjected to univariable survival analysis to quantify differences in disease specific survival using the Kaplan-Meier method. A multivariable approach to disease specific survival, using the Cox Proportional Hazards Model, was used to determine if survival differences between PD-L1 expression categories were independent of adjuvant chemotherapy. Assessment for heterogeneity of clinico-pathologic variables was performed with the following statistical approaches: continuous variables were examined using the Wilcoxon Rank-Sum Test, categorical comparisons were computed using Fisher’s exact test. A P-value of <0.05 was considered as statistically significant for all analyses. All analyses were computed with JMP v13.1 (SAS Institute, Cary, NC).