Thomas Arnason

Background: At diagnosis, ~80% of pancreatic ductal adenocarcinomas (PDAC) have metastasized. Relapse is thus common even among patients who undergo surgical resection, the only curative option. PDAC progresses rapidly, and existing immunotherapies have been ineffective. We hypothesized that natural killer (NK) cell immunotherapies could be effective against PDAC because they recognize conserved and heterogeneous features associated with cellular stress and transformation and can seek out metastases distal to the primary tumor site. Here, we aim to define the key features of NK cells as effective agents for PDAC immunotherapy. Methods: We used The Cancer Genome Atlas Program (TCGA) PDAC Firehose data and flow cytometry to predict and measure the most common activating or inhibitory ligands available on PDAC for NK cell activation. To ascertain how the tumor might alter expression of these ligands during treatment, inflammation or immune pressure, we measured expression of NK ligands at rest, or after exposure to immune cells or inflammation. To test and rank the functional importance of these dynamic ligands in the recognition, killing and control of PDAC we used co-culture, antibody-blocking and an NK-competent humanized mouse model. Results: Leveraging the known sequential acquisition of mutations as a surrogate for disease progression, we observed a progressive loss of transcript expression for activating NK cell ligands and chemoattractants. Exposure of PDAC to NK cells or interferon-γ, an inflammatory stimulus, drove dynamic changes in expression of both activating and inhibitory ligands. In vitro co-culture assays revealed a redundancy in the activating receptors engaged in NK:PDAC interactions, but that human leukocyte antigens (HLA)-killer immunoglobulin-like receptors (KIR) signaling dominantly interrupted anti-PDAC activity. In NK-competent humanized mice, adoptively transferred, unselected, unmodified NK cells slowed tumor growth in a dose-dependent manner, but NK cells selected to avoid HLA I-driven inhibition were the most competent effectors for PDAC control. Conclusions: Although there is redundancy among activating ligand:receptor pairs for recognizing PDAC tumors, interactions between KIR and HLA define the extent to which antitumor activity can proceed. During tumor progression and in response to immunotherapy, NK:tumor interactions drive upregulation of HLA I molecules. Thus, educated NK cells from HLA I-disparate donors may be the more effective allogeneic NK immunotherapy for PDAC.

Surgical resection combined with chemotherapy offers the best chance of survival in pancreatic ductal adenocarcinoma (PDAC), but many will experience recurrence and early mortality. We examined soluble AXL (sAXL), a blood protein, for its ability to predict 6-month mortality after resection and compared it to CA19-9. Fifty-four patients with PDAC who underwent tumour resection were analyzed to assess biomarker performance and identify optimal cut-off levels. The cut-off for sAXL was 40.26 ng/mL (sensitivity 0.729; specificity 0.643), while it 253.3 U/mL for CA19-9 (sensitivity 0.591; specificity 0.621). Patients with sAXL > 40.26 ng/mL had a non-significant trend toward worse survival (log-rank p = 0.088). Univariate Cox regression revealed that high tumour grade (3 + 4) and positive resection margin significantly predicted early mortality. Multivariate Cox regression showed that sAXL > 40.26 ng/mL remained associated with 6-month mortality (hazard ratio 2.42, bootstrap 95% CI 1.15–5.65, p = 0.020), independent of high tumour grade (hazard ratio 4.02, bootstrap 95% CI 1.68–13.2, p = 0.002). These findings suggest that a preoperative blood test (sAXL) has utility for predicting futile surgery beyond the current standard, CA19-9, and can be incorporated into larger models to assist in risk stratification and follow-up planning.

Background: Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers, and has limited options for treatment. Low immune infiltration, desmoplastic stroma, and poor tumour immunogenicity are all expected to contribute to PDAC’s rapid progression and limited response to existing immunotherapies. PDAC tumours are mosaics of different sub-tumour microenvironments, including some that do contain immune cells. We hypothesized that increasing frequency of lymphocyte:tumour interactions would correlate with lengthened survival for patients with PDAC. Methods: Using multiplex immunofluorescence, digital pathology, and computational analyses, we profiled the spatial distribution, co-localization, and neighbourhood architecture of immune cells in tumours from 73 patients with PDAC. Results: Higher densities of CD3+CD8– (CD4+) T cells were associated with improved five-year overall survival, particularly when enriched at the tumour-stroma boundary (i.e. peritumoural). CD3+CD8-T cells, CD8+ T cells (CD3+CD8+), and B cells (CD20+) frequently co-infiltrated and co-localized, forming distinct immune neighbourhoods indicative of organized adaptive immunity. We defined three common immune neighbourhoods within PDAC (1) Macrophage-dominant; (2) T cell dominant; and (3) Disorganized. With greater tumour and peri-tumoural area represented by the T cell dominant neighbourhood, overall survival was increased. The other neighbourhoods were not significantly associated with outcomes. Conclusion: Immune cells self-assemble in predictable patterns in PDAC. The presence of T cell dominant neighbourhoods, which we interpret as supporting ongoing immune activation, best predict lengthened survival. Spatially organized immune interactions may serve as prognostic indicators and inform future studies for immunotherapies in PDAC.

Gastric cancer is common globally and has a generally poor prognosis with a low 5-year survival rate. Targeted therapies and immunotherapies have improved the treatment landscape, providing more options for efficacious treatment. The use of these therapies requires predictive biomarker testing to identify patients who can benefit from their use. New therapies on the horizon, such as CLDN18.2 monoclonal antibody therapy, require laboratories to implement new biomarker tests. A multidisciplinary pan-Canadian expert working group was convened to develop guidance for pathologists and oncologists on the implementation of CLDN18.2 IHC testing for gastric and gastroesophageal junction (G/GEJ) adenocarcinoma in Canada, as well as general recommendations to optimize predictive biomarker testing in G/GEJ adenocarcinoma. The expert working group recommendations highlight the importance of reflex testing for HER2, MMR and/or MSI, CLDN18, and PD-L1 in all patients at first diagnosis of G/GEJ adenocarcinoma. Testing for NTRK fusions may also be included in reflex testing or requested by the treating clinician when third-line therapy is being considered. The expert working group also made recommendations for pre-analytic, analytic, and post-analytic considerations for predictive biomarker testing in G/GEJ adenocarcinoma. Implementation of these recommendations will provide medical oncologists with accurate, timely biomarker results to use for treatment decision-making.

Background/Objectives: One study of pancreatic ductal adenocarcinoma has found expression of glypican-3 (GPC3) and cytokeratin-19 (CK19) determined by immunohistochemistry to be associated with higher stage and grade disease, with a more adverse prognosis. The reported 44% rate of GPC3 expression in pancreatic cancer raises the important possibility that targeted immunotherapies currently in development for hepatocellular carcinoma may also prove useful for GPC3-expressing pancreatic cancers. The present study aims to determine if a similar expression pattern of these markers and stage/grade/prognostic associations is present in our Canadian patient population. Methods: Patients with a pancreatic surgical resection for adenocarcinoma or neuroendocrine tumor (NET) were identified from pathology records over a 5-year period. Immunohistochemistry for GPC3 and CK19 was performed on archived tumor tissue and the proportion of positive cells and intensity of staining were recorded. Grade, stage, and overall survival were compared in patients with NETs that were CK19-positive versus -negative. Results: All 72 pancreatic adenocarcinomas and 20 NETs tested were negative for GPC3, apart from a single case of pancreatic adenocarcinoma. All 72 adenocarcinomas were positive for CK19 expression. Half of the NETs were positive for CK19. There was no correlation between CK19 expression in NETs and tumor grade, lymph node metastasis, distant metastasis, or overall survival. Conclusions: We are skeptical of the reported prognostic value of GPC3 and CK19 in pancreatic adenocarcinomas. CK19 as a prognostic marker in NETs has potential for further study. The results with our protocol for GPC3 immunohistochemistry suggest that pancreatic cancer may be a less promising target for GPC3-targeted immunotherapies than previously thought.

Genomic profiling of pancreatic cancer using small core biopsies has taken an increasingly prominent role in precision medicine. However, if not appropriately preserved, nucleic acids (NA) from pancreatic tissues are known to be susceptible to degradation due to high intrinsic levels of nucleases. PAXgene fixation (PreAnalytix, Switzerland) represents a novel formalin-free tissue preservation method. We sought to compare the NA and histomorphological preservation of pancreatic cancer tissues preserved with PAXgene-fixed paraffin-embedding (PFPE) and formalin-fixed paraffin-embedding (FFPE). Tissues from 19 patients were obtained prospectively from pancreaticoduodenectomy specimens and evaluated by four gastrointestinal pathologists. The extracted NA were quantified by Nanodrop and Qubit and assessed for quality by qPCR, targeted next-generation sequencing (NGS) assay, and RNA-sequencing. Our results demonstrated that, when assessed blindly for morphological quality, the four pathologists deemed the PFPE slides adequate for diagnostic purposes. PFPE tissues enable greater yields of less fragmented and more amplifiable DNA. PFPE tissues demonstrated significantly improved quality control (QC) metrics in a targeted NGS assay including Median Absolute Pair-wise Difference (MAPD) scores. Our results support the use of PAXgene fixative for the processing of specimens from pancreatic cancers with the potential benefits of improved yields for more amplifiable DNA in low-yield biopsy specimens and its ideal use for amplicon-based NGS assays.

Background: High grade serous cancer (HGSC) remains a highly fatal malignancy with less than 50% of patients surviving 5 years after diagnosis. Despite its high mutational burden, HGSC is relatively refractory to checkpoint immunotherapy, suggesting that additional features of the cancer and its interactions with the immune system remain to be understood. Natural killer (NK) cells may contribute to HGSC control, but the role(s) of this population or its subsets in this disease are poorly understood. Methods: We used a TMA containing duplicate treatment-naïve tumors from 1145 patients with HGSC and a custom staining panel to simultaneously measure macrophages, T cells and NK cells, separating NK cells based on CD16a expression. Using pathologist-validated digital pathology, machine learning, computational analysis and Pearson’s correlations, we quantitated infiltrating immune cell density, co-infiltration and co-localization with spatial resolution to tumor region. We compared the prognostic value of innate, general, and adaptive immune cell “neighborhoods” to define characteristics of HGSC tumors predictive for progression-free survival and used flow cytometry to define additional features of the CD16a ^dim NK cell subset. Results: NK cells were observed in >95% of tumor cores. Intrastromal localization of CD16a ^low and CD16a ^high NK cells was associated with shorter and longer progression-free survival, respectively. CD16a ^high NK cells most frequently co-localized with T cells and macrophages; their proximity was termed an “adaptive” neighborhood. We find that tumors with more area represented by adaptive immune cell neighborhoods corresponded to superior progression free survival. In contrast, CD16a ^low NK cells did not co-infiltrate with other immune cell types, and expressed the ectonucleotidases, CD39 and CD73, which have been previously associated with poor prognosis in patients with HGSC. Conclusions: Progression-free survival for patients with HGSC may be predicted by the subset of NK cells within the tumor infiltrate (i.e. CD16a ^high vs. CD16a ^low ). NK cell subtypes were associated predictable co-infiltrating and co-localizing leukocyte subsets, suggesting that their presence and activity may influence, or be influenced by the tumor microenvironment. Our data suggest that immunotherapeutic strategies for HGSC should consider the constitution of NK cell subsets and may benefit from mobilizing and activating CD16 ^high NK cells.

Patients with hereditary diffuse gastric cancer often undergo prophylactic gastrectomy to minimize cancer risk. Because intramucosal poorly cohesive carcinomas in this setting are typically not grossly visible, many pathologists assess the entire gastrectomy specimen microscopically. With 150 or more slides per case, this is a major time burden for pathologists. This study utilizes deep learning methods to analyze digitized slides and detect regions of carcinoma. Prophylactic gastrectomy specimens from seven patients with germline CDH1 mutations were analyzed (five for training/validation and two for testing, with a total of 133 tumor foci). All hematoxylin and eosin slides containing cancer foci were digitally scanned, and patches of size 256×256 pixels were randomly extracted from regions of cancer as well as from regions of normal background tissue, resulting in 15,851 images for training/validation and 970 images for testing. A model with DenseNet-169 architecture was trained for 150 epochs, then evaluated on images from the test set. External validation was conducted on 814 images scanned at an outside institution. On individual patches, the trained model achieved a receiver operating characteristic (ROC) area under the curve (AUC) of 0.9986. This enabled it to maintain a sensitivity of 90% with a false-positive rate of less than 0.1%. On the external validation dataset, the model achieved a similar ROC AUC of 0.9984. On whole slide images, the network detected 100% of tumor foci and correctly eliminated an average of 99.9% of the non-cancer slide area from consideration. Overall, our model shows encouraging progress towards computer-assisted diagnosis of hereditary diffuse gastric cancer.