A central research portal for mining pancreatic clinical and molecular datasets and accessing biobanked samples

Abstract

Objectives: We present Pancreas Genome Phenome Atlas (PGPA) as a resource for the mining and analysis of pancreatic -omics datasets, and demonstrate the biological interpretations possible due to this dynamic analytics hub accommodating an extensive range of publicly available datasets.

Methods: Clinical and molecular datasets from four primary sources are included (The Cancer Genome Atlas, International Cancer Genome Consortium, Cancer Cell Line Encyclopaedia, Genomics Evidence Neoplasia Information Exchange), which form the foundation of -omics profiling of pancreatic malignancies and related lesions (n = 7760 specimens). Several user-friendly analytical tools to integrate and explore molecular data derived from these primary specimens and cell lines are available. Crucially, PGPA is positioned as the data access point for Pancreatic Cancer Research Fund Tissue Bank - the only national pancreatic cancer biobank in the UK. This will pioneer a new era of biobanking to promote collaborative studies and effective sharing of multi-modal molecular, histopathology and imaging data (>125 000 specimens from >3980 cases and controls; >2700 radiology images, and >2630 digitised H&Es from 401 donors) to accelerate validation of in silico findings in patient-derived material.

Results: We demonstrate the practical utility of PGPA by investigating somatic variants associated with established transcriptomic subtypes and disease prognosis: several patient-specific variants are clinically actionable and may be leveraged for precision medicine.

Conclusions: This places PGPA at the analytical forefront of pancreatic biomarker-based research, providing the user community with a distinct resource to facilitate hypothesis-testing on public data, validate novel research findings, and access curated, high-quality patient tissues for translational research.

Keywords: Biomarkers; Genomics; Pancreas biobank; Transcriptomics; Translational.

Graphical abstract

Fig. 1

Summary of available tissue types. (A) The proportion of >3 400 unique organ site tissues available for study in the UK national Pancreatic Cancer Research Fund Tissue Bank, with the breakdown of controls (B) and pancreas (C) highlighted. (D) Distribution of >125 000 PCRFTB specimens by type, across all >3980 patients. Details are updated weekly. Additionally, radiological imaging is available for 2702 patients with malignant, pre-malignant and benign pancreatic diagnoses, and >2630 digitised H&E images from 401 donors. Samples can be applied for here. (E) Geographical locations of PCRFTB patient recruitment sites. *=pancreatic juice, CTC, bile, organoids, cancer-associated fibroblasts. Map created with mapchart.net.

Fig. 2

Advanced filtering options and clinical summaries for publicly available PDAC datasets. (A) Available data can be filtered according to various patient-related factors and tumour characteristics, including the stratification and analysis of cohorts according to KRAS and TP53 mutational status and established transcriptomic, immune (TCGA, ICGC), genomic (ICGC) or histologically-derived AI (PacPaint) subtypes. (B) Dynamic bar charts allow multiple covariates to be viewed in relation to each other: e.g. survival trends in PDAC (TCGA) and neuroendocrine (ICGC PAEN-AU). (C) Filtered attributes can also be visualized as summary thumbnail figures for each study cohort; distribution of tumour grade in 185 TCGA PDAC samples is shown as an example. Data underpinning visual outputs can also be downloaded as .csv or .xls files, for offline analysis.

Fig. 3

Transcriptomic stratification in PDAC reveals subtype-specific somatic variants. Oncoplots* of the top 25 most frequently mutated genes for consensus (A)n = 27 classical-type and (B)n = 16 basal-type PDAC cases (TCGA). (C) Overlap^ between the somatically mutated genes associated with best/worst prognosis subtypes across TCGA and ICGC PACA-AU cohorts combined. Genes highlighted in bold contain Tier 1 predicted oncogenic driver variants that have associated pharmacological inhibitors or chemotherapies (see Supplementary Table S1). *Mutated genes are ranked in order of the total number of mutations in each given gene (where genes may have >1 mutation present; black ‘multi-hit’), while the percentage to the right of each bar reflects the proportion of samples altered in the cohort. ^Created in Venny 2.1.

Fig. 4

Differentially expressed genes between classical/progenitor and basal-like/QM/squamous TCGA PDAC tumours. Box plots showing the trends of (A) TP53 and (B) MUC16 mRNA expression levels across all patients in each filtered TCGA PDAC group; best prognosis (n = 27; classical/progenitor; left) and worst prognosis (n = 16; basal/squamous/QM; right). C. Kaplan-Meier curve showing elevated MUC16 expression significantly associated with lower patient survival over 3 years, from n = 402 PACA-AU PDAC patients with expression and outcome data (logrank p = 0.011; hazard ratio (HR)=2.23). D. No association between MUC16 mRNA expression levels and outcome were observed in n = 65 neuroendocrine carcinomas (PAEN-AU).

Fig. 5

Frequently altered genes and biological pathways amongst n=756 KRAS wild-type PDAC tumours from GENIE. (A) Oncoplot showing the top 10 most frequently mutated genes in KRAS wild-type PDAC tumours (confirmed somatic missense mutations filtered out; insertions or duplications may still be present). (B) Alluvial plot showing gene targets harbouring any variants with therapeutic biomarker potential in ≥5 % of patients, as identified by the Cancer Genome Interpreter and based on data from OncoKB, CIVic (Clinical Interpretation of Variants in Cancer) and the Cancer Biomarkers database. (C) Altered biological pathways amongst KRAS wild-type PDAC tumours include MAPK and p53 signalling, as derived from the KEGG pathway database. The proportion of genes mutated in each pathway (left) and the proportion of all KRAS wild-type patients affected (x-axis) are given.