Systems toxicology prioritization of HSD11B1 as a candidate PFOA-related metabolic homeostasis node in adrenocortical carcinoma

Abstract

This study aimed to construct a hypothesis-generating evidence framework linking PFOA-related molecular target annotations to biological alterations in adrenocortical carcinoma (ACC), with emphasis on HSD11B1 as a downregulated adrenal metabolic homeostasis node and an in silico prioritized candidate PFOA-related protein. PFOA-related targets were obtained from SwissTargetPrediction and CTD and intersected with ACC-associated genes from GeneCards. GO and Reactome enrichment analyses were performed. Using a combined TCGA-ACC/GTEx expression matrix, eight machine-learning models-GLM, SVM, Random Forest, GBDT, XGBoost, Naive Bayes, Kernel SVM, and LASSO-were evaluated with stratified five-fold cross-validation. A matched negative-control sensitivity analysis compared the 25-gene candidate panel with 21 housekeeping-gene panels of the same size. Core genes were identified by integrating SHAP/feature-importance rankings and cross-model selection frequency, but model AUCs were treated as dataset-level rather than disease-specific performance. HSD11B1 expression was examined in TCGA/GTEx and three GEO cohorts. Single-cell analyses used 20 ACC 10x samples from Zenodo record 10534062/GSE252108. Molecular docking of PFOA with HSD11B1 structures 3PDJ, 3OQ1, and 3BZU was performed, and the predicted poses were assessed relative to the catalytic pocket. Forty-four PFOA-related targets and 2,272 ACC-associated genes yielded 25 intersecting genes. The primary models produced high five-fold cross-validated AUCs (0.9785–0.9961); however, in the matched sensitivity analysis, housekeeping panels also produced high AUCs in most model families (model-specific median 0.789–0.999), and the candidate panel exceeded the housekeeping median in only one of eight model families. These values therefore indicate source-confounded dataset separability, not ACC-specific discrimination. IGF2 and PTGER2 were selected by all models and HSD11B1 by seven. HSD11B1 was downregulated in the combined TCGA/GTEx comparison and showed the same direction across three separately analyzed GEO cohorts. Single-cell analysis retained 92,097 cells and showed low, sparse HSD11B1 expression in malignant adrenal cells. Molecular docking generated top-ranked PFOA-HSD11B1 poses outside the catalytic pocket; these poses do not support an effect on cortisone-cortisol turnover. HSD11B1 may represent a downregulated adrenal differentiation and steroid/glucocorticoid homeostasis node rather than a classical ACC oncogene. Because disease status was aligned with TCGA versus GTEx source and PFOA exposure was not measured, the machine-learning and exposure-related results are hypothesis-generating. The non-catalytic docking poses do not establish binding, enzyme inhibition, or altered glucocorticoid activation.