# Multi-scale systems toxicology defines a KLF5-centered adverse outcome pathway linking DEHP exposure to pancreatic cancer progression and signaling programs relevant to therapy tolerance

## TL;DR

This study explores how DEHP exposure may lead to pancreatic cancer by identifying a KLF5-centered pathway that influences tumor progression and therapy resistance.

## Contribution

The paper introduces a multi-scale systems toxicology approach to define a novel adverse outcome pathway linking DEHP exposure to pancreatic cancer.

## Key findings

- A six-gene PDAC signature was identified with high accuracy (AUC = 0.946) using machine learning.
- KLF5 expression was modestly associated with pancreatic cancer risk (OR = 1.188, p = 0.046) via Mendelian randomization.
- DEHP exposure upregulated KLF5 and MMP7, promoting cancer cell migration and invasion.

## Abstract

Di(2-ethylhexyl) phthalate (DEHP) is a ubiquitous plasticizer implicated in pancreatic carcinogenesis, yet the molecular initiating events and adverse outcome pathways (AOPs) linking exposure to disease mechanisms remain poorly resolved. In this study, we integrated a multi-scale systems toxicology framework—combining heterogeneous ensemble machine learning, Mendelian randomization (MR), molecular docking and molecular dynamics (MD), single-cell transcriptomics, and in vitro assays—to delineate a candidate mechanistic trajectory. A Tabular Prior-Data Fitted Network (TabPFN)-enhanced ensemble identified a six-gene pancreatic ductal adenocarcinoma (PDAC) signature (AUC = 0.946). Within this signature, MR provided suggestive evidence for a modest association between genetically predicted Krüppel-like factor 5 (KLF5) expression and pancreatic cancer risk (OR = 1.188, p = 0.046). Functional enrichment of DEHP–PDAC intersection targets highlighted pro-survival signaling modules, including PI3K–AKT- and MAPK-related pathways. Structure-based analyses supported the biophysical plausibility of a non-covalent DEHP–KLF5 interaction (−6.4 kcal/mol), and 100-ns MD simulations indicated a persistent binding mode with conformational accommodation. Single-cell analysis localized KLF5 predominantly to malignant ductal cells and, together with CellChat inference, was consistent with malignant cell-derived TGF-β- and MIF-related signaling. Network-based virtual knockout further suggested that KLF5 may contribute to sustaining a TGF-β/MMP7-linked matrix remodeling program. Consistently, DEHP exposure upregulated KLF5 and MMP7 and enhanced migration and invasion of PANC-1 cells. Collectively, these findings support a working AOP model linking DEHP-responsive KLF5-centered activity to extracellular matrix (ECM) remodeling and immunomodulatory communication, providing a mechanistic rationale that aligns with signaling programs characteristic of therapy-tolerant tumor niches, particularly PI3K–AKT-coupled survival signaling and TGF-β-linked stromal remodeling. Although these implications are hypothesis-generating, they highlight a potential avenue for future drug-response investigations.

Diagram illustrating the DEHP-KLF5 axis linking environmental exposure to the progression of pancreatic ductal adenocarcinoma (PDAC). It includes steps such as DEHP exposure, ensemble machine learning and TaPFN, Mendelian randomization, molecular docking, and scRNA sequencing. KLF5 causal driving, network interactions, and in vitro validation are shown, highlighting stromal fibrosis and immune exclusion processes leading to PDAC progression.