Genetics and Therapeutic Responses to Tumor-Infiltrating Lymphocyte Therapy of Pancreatic Cancer Patient-Derived Xenograft Models

Abstract

Background and aims: Pancreatic cancer is the seventh leading cause of cancer-related deaths worldwide. Checkpoint immunotherapy has not yet shown encouraging results in pancreatic cancer possibly because of a poor immunogenicity and/or an immune suppressive microenvironment. The aim of this study was to develop patient-derived xenograft (PDX) models, compare their genetics to the original biopsies, and assess if autologous tumor-infiltrating lymphocytes (TILs) would have antitumoral activity in pancreatic cancer.

Methods: We subcutaneously transplanted tumors from 29 patients into NOG mice to generate PDX models. We established TIL cultures and injected them into PDX mice. We analyzed histology and genetics of biopsies and PDX tumors.

Results: Tumor growths were confirmed in 11 of 29 transplantations. The PDX tumors histologically resembled their original biopsies, but because stromal cells in the PDX model tumors were from mouse, their gene expression differed from the original biopsies. Immune checkpoint ligands other than programmed death ligand-1 (PD-L1) were expressed in pancreatic cancers, but PD-L1 was rarely expressed. When it was expressed, it correlated with tumor take in PDX models. One of the 3 tumors that expressed PD-L1 was an adenosquamous cancer, and another had a mismatch repair deficiency. TILs were expanded from 6 tumors and were injected into NOG or human interleukin-2 transgenic-NOG mice carrying PDX tumors. Regression of tumors could be verified in human interleukin-2 transgenic-NOG mice in 3 of the 6 PDX models treated with autologous TILs, including the adenosquamous PDX model.

Conclusion: PDX models of pancreatic cancer can be used to learn more about tumor characteristics and biomarkers and to evaluate responses to adoptive cell therapy and combination therapies. The major benefit of the model is that modifications of T cells can be tested in an autologous humanized mouse model to gain preclinical data to support the initiation of a clinical trial.

Keywords: Pancreatic Cancer; Patient-Derived Xenografts; Transcriptomic classification; Tumor-Infiltrating Lymphocytes.

Figure 1

Establishment of pancreatic cancer PDX models for cellular immunotherapy studies. (A) Schematic presentation of the project. Tumors are used for generation of PDX models, DNA/RNA for sequencing, immunohistochemistry (IHC), and TILs for later immunotherapy. TIL therapy is monitored using caliper measurements of tumors growing in PDX mice. (B) Tumor growth curves of tumors from 11 patients. (C) Representative IHC of pancreatic cancer in patients and in 2 passages in mice. Staining was performed with H&E or antibodies against human cytokeratins, HLA-ABC, and Ki67. Another patient sample is seen in Figure A1. (D) Survival of patients with PDAC in the cohort divided into whether tumors formed (n = 8) or not (n = 13). H&E, hematoxylin and eosin; IL2, interleukin 2.

Figure 2

Somatic genomic alterations and transcriptomic classification. (A) Mutations in genes that have either previously found to be significantly mutated in pancreatic ductal adenocarcinoma, which are listed in the COSMIC Cancer Gene Census (CGC) or which have been described as pan-cancer drivers. For genes in the latter 2 categories, only those mutated in at least 2 independent samples are shown. (B) Copy number alterations, shown as color-coded in proportion to log₂ ratios of tumor relative to normal. Red indicates gains and, blue indicates losses. (C) Predicted cancer or cell types for each sample based on k-nearest neighbor classification (k-NN; k = 6²⁸) on the basis of Spearman correlation coefficients calculated with respect to all coding genes, using either a pan-cancer data set from TCGA as reference or the Cancer Cell Line Encyclopedia (CCLE, Table A1). The colors indicate the proportion among the top 10 most strongly correlated samples that are the same cancer type as the final prediction, as a measure of agreement. (D) As in (C), but showing the proportions of each tumor or cell type that occurred among the top 10 samples in the ranked list used for classification of a given tumor. (E) Classification of pancreatic cancer subtypes based on metadata from a previous TCGA study, using k-NN classification with k = 15 (found to be optimal in leave-one-out cross-validation on the TCGA-PAAD cohort; Figure A3A).

Figure 3

RNA-based immune profiling. (A) Inference of the proportions of immune cell types, as well as cancer-associated fibroblasts and endothelial cells present in samples, based on deconvolution of bulk RNA-seq data of patient biopsies using EPIC. (B) As in (A), but excluding uncharacterized cells (likely to be cancer cells). (C) Expression levels of genes involved in interactions with immune cells. (D) Expression levels of genes encoding immune checkpoint proteins. A list of genes describing HLA genes and immune checkpoint ligands are present in Table A2.

Figure 4

Pancreatic cancer PDX tumors (n = 3–5) growing in NOG mice or in hIL2-NOG mice and treated with autologous TILs. (A) Cumulative tumor growth curves from caliper measurements. Individual growth curves and controls where tumors are grown in hIL2-NOG without TILs are shown in Figure A4. (B) Immunohistochemistry analysis of TIL infiltration (CD3 and CD8) and expression of the immune checkpoint ligand PD-L1. Shown are a representative image of minimum 3 mice per PDX model. Higher magnification images of representative sections of all 6 PDX models are in Figures A6–A11. H&E, hematoxylin and eosin;