Pancreatic ductal adenocarcinoma remains one of the deadliest cancers, with a five-year survival rate below 10%. The disease is notoriously difficult to treat because it is surrounded by a thick, fibrous barrier (desmoplastic stroma) that physically blocks drug delivery and immune cells. Additionally, pancreatic cancer cells themselves have evolved multiple resistance mechanisms to chemotherapy and immunotherapy.
One underexplored vulnerability is the TRAIL death receptor pathway. Cancer cells can normally be killed when immune cells signal them to self-destruct through receptors on their surface, but pancreatic cancer cells often suppress or bypass this pathway. The TNFRSF10A gene encodes one of these receptors, called TRAILR1, and this study investigated whether it could be targeted to trigger cancer cell death.
Researchers integrated data from six independent pancreatic cancer patient cohorts downloaded from the Gene Expression Omnibus database. Using Bayesian robust inference and random forest machine learning, they identified differentially expressed genes comparing tumor versus normal pancreatic tissue. This multi-cohort approach ensures findings are reproducible across different patient populations.
To understand where TNFRSF10A is active within the tumor, spatial transcriptomics data was analyzed using the 10X Genomics Visium platform, which maps gene expression to specific locations within tissue sections. Complementary proteomics data from CPTAC was used to confirm whether gene expression translated to actual protein abundance. A competing endogenous RNA network analysis then mapped the microRNA regulators controlling TNFRSF10A expression.
TNFRSF10A was identified as the most significantly upregulated gene in pancreatic cancer across all six patient cohorts. Spatial transcriptomics revealed that this elevation was not uniform throughout the tumor but was specifically concentrated in malignant epithelial cells at the tumor's invasive leading edge, the region closest to surrounding stromal tissue.
This spatial enrichment in the most invasive tumor compartment is scientifically significant. The invasive front is where cancer cells are most actively spreading and are most exposed to pro-inflammatory signals from the stroma. High TRAILR1 expression in this dangerous zone suggests it may be part of the signaling that drives invasion, making it a rational therapeutic target.
Using SELFormer, a transformer-based deep learning model for molecular property prediction, combined with traditional QSAR modeling, the team screened FDA-approved drugs and natural compounds for their ability to modulate TRAILR1. This virtual screening approach tested millions of molecular interactions computationally before any lab experiments.
Three candidates emerged: temsirolimus (an mTOR inhibitor approved for kidney cancer), ergotamine (a migraine drug), and capivasertib (an AKT inhibitor in clinical trials). All three underwent 300-nanosecond molecular dynamics simulations to test whether they would maintain stable binding to the TRAILR1 protein over time, confirming favorable binding energetics with minimal structural drift.
Because all three identified drug candidates are already FDA-approved or in clinical trials, their safety profiles in humans are well-established. This dramatically shortens the path to clinical testing compared to developing entirely new compounds. Temsirolimus in particular has mechanistic evidence supporting its role in enhancing death receptor-mediated cancer cell killing.
This study demonstrates a powerful workflow for pancreatic cancer drug discovery: integrate multi-omics data to find a biologically validated target, use AI to predict which drugs will engage that target, and confirm with molecular simulations. This approach could be applied to identify additional druggable targets in pancreatic cancer and other treatment-resistant malignancies.