Precision Medicine in Pancreatic Cancer: Patient-Derived Organoid Pharmacotyping Is a Predictive Biomarker of Clinical Treatment Response

Abstract

Purpose: Patient-derived organoids (PDO) are a promising technology to support precision medicine initiatives for patients with pancreatic ductal adenocarcinoma (PDAC). PDOs may improve clinical next-generation sequencing (NGS) and enable rapid ex vivo chemotherapeutic screening (pharmacotyping).

Experimental design: PDOs were derived from tissues obtained during surgical resection and endoscopic biopsies and studied with NGS and pharmacotyping. PDO-specific pharmacotype is assessed prospectively as a predictive biomarker of clinical therapeutic response by leveraging data from a randomized controlled clinical trial.

Results: Clinical sequencing pipelines often fail to detect PDAC-associated somatic mutations in surgical specimens that demonstrate a good pathologic response to previously administered chemotherapy. Sequencing the PDOs derived from these surgical specimens, after biomass expansion, improves the detection of somatic mutations and enables quantification of copy number variants. The detection of clinically relevant mutations and structural variants is improved following PDO biomass expansion. On clinical trial, PDOs were derived from biopsies of treatment-naïve patients prior to treatment with FOLFIRINOX (FFX). Ex vivo PDO pharmacotyping with FFX components predicted clinical therapeutic response in these patients with borderline resectable or locally advanced PDAC treated in a neoadjuvant or induction paradigm. PDO pharmacotypes suggesting sensitivity to FFX components were associated with longitudinal declines of tumor marker, carbohydrate-antigen 19-9 (CA-19-9), and favorable RECIST imaging response.

Conclusions: PDOs established from tissues obtained from patients previously receiving cytotoxic chemotherapies can be accomplished in a clinically certified laboratory. Sequencing PDOs following biomass expansion improves clinical sequencing quality. High in vitro sensitivity to standard-of-care chemotherapeutics predicts good clinical response to systemic chemotherapy in PDAC. See related commentary by Zhang et al., p. 3176.

Figure 1.

Frequency (%) of the wild type gene variant reported as a result of the clinical NGS panel of the primary pancreatic ductal adenocarcinoma in a registry cohort. Specimen type (biopsy or resection) was not significantly associated with the frequency of the wild type gene reported (A), whereas the wild type findings were more commonly reported after pretreatment by chemo- (B) or radiotherapy (C), and good response to pretreatment in the surgical pathology report (D).

Figure 2.

Mutated gene variants detected in three independent datasets by CLIA-approved next generation sequencing of primary tumors and patient-derived organoids (PDO) established from the same primary tumor. In discovery set (A), whole exome sequencing of the PDO was compared to CLIA solid tumor panel (STP) of the primary tumor. In validation set 1 (B), whole-exome sequencing (WES) was performed on both primary tumor and organoid, and in CLIA validation set 2 (C), the CLIA-approved NGS-STP methods were performed for both primary tumor and organoid. The detection of variant allele frequencies (VAF, %) for typical pancreatic cancer driver mutations (KRAS, TP53, SMAD4 and CDKN2A) are substantially higher in PDOs than in primary tumors in discovery dataset (D) and validation datasets 1 (E) and 2 (F). Relevant pancreatic cancer mutations detected by WES of the primary tumor and the corresponding organoids established from the same tumor in the validation set 1 (G). A homozygous copy number loss of the chromosomal area of the CDKN2A gene in the organoid compared to primary tumor. Though demonstrating a slight decrease in CNV, clinically-relevant bioinformatics pipelines would fail to call CDKN2A loss in the primary tumor.

Figure 3.

Pharmacotyping-derived population distribution of PDOs (left axis label) showing half maximal inhibitory concentration (IC50) and 95% confidence interval in dose-response testing against 5-fluorouracil (5-FU (A). A population distribution of mean PDO IC50 values presented as a violin plot for Irinotecan (B), 5-FU (C), Oxaliplatin (D), Gemcitabine (E) and Paclitaxel (F). Putative clinically-relevant cohorts are categorized in thirds as sensitive (blue), intermediate (orange) and resistant (red) to the chemotherapeutic described.

Figure 4.

Synergy estimates of Gemcitabine and Paclitaxel combination in organoid ex vivo pharmacotyping of case 010 presented in 2D plot (A) and 3D surface plot (B). The red areas reflect synergy of the axis concentrations by ZIP synergy score. Despite the red peak, the result shows practically no synergistic effect of the selected drug combination. Synergy tensor cubes showing the viability-based inhibition after ex vivo pharmacotyping by the combinations of the three drugs (5-FU, Irinotecan and Oxaliplatin) for a more sensitive culture 001 (C) and for a more resistant culture 041 (D). ZIP synergy scores (E-F) after combination pharmacotyping of 5-FU, Irinotecan and Oxaliplatin. Negative ZIP scores throughout reflect the lack of synergy of the drugs in the combination.

Figure 5.

A waterfall plot showing patients’ CA19–9 responses to FOLFIRINOX as percentage change during the time period of neoadjuvant therapeutic administration (A). Only patients clearly producing plasma CA19–9 were included, whereas one patient with a CA19–9 baseline level of 3 U/mL was excluded. Sensitivity of the patients’ organoid in single agent pharmacotyping to each three compounds of FOLFIRINOX are presented by colors of the bar: dark blue = sensitive to all three drugs, light blue = sensitive to two drugs, orange = sensitive to one drug, and red = sensitive to none of the three drugs. Waterfall plot presenting the tumor volume change in computed tomography imaging (RECIST response) during the period of neoadjuvant therapy (B).