by Christos Evangelou, MSc, PhD – Medical Writer and Editor 

Adrenocortical carcinoma (ACC), a rare endocrine malignancy affecting fewer than two people per million annually, has a poor prognosis. Mitotane is currently the only FDA-approved first-line therapy, and there is limited understanding of which patients might respond to emerging immunotherapies. In a recent study, researchers developed an AI-driven scoring system that integrates tumor metabolism, immune infiltration, and digital pathology to predict treatment response and prognosis in patients with ACC. 

The study was published in npj Precision Oncology. 

Study Rationale 

The research team, led by Dr. Zhiyu Liu from Dalian Medical University, aimed to address two clinical problems in ACC care. 

“Previous studies focused on a single modality — genomics, pathology, or transcriptomics — without integrating steroid metabolism, immune microenvironment, and tumor morphology,” Dr. Liu explained in an interview with Pathology News. “Two major unmet needs motivated our work: first, a lack of integrative biomarkers to complement the Weiss score for better risk stratification; second, the absence of reliable predictors for immunotherapy response in ACC, traditionally considered an immune-cold tumor.” 

Dr. Liu further explained that the Weiss scoring system relies on morphological features such as mitotic rate, necrosis, and vascular invasion. The researchers hypothesized that combining genomic data with AI analysis of whole-slide pathology images could identify mechanisms that link metabolism and immunity and that could influence response to treatment. 

A Multi-Modal Approach to Molecular Classification

The research team analyzed data from 142 patients with ACC from The Cancer Genome Atlas (TCGA) and multiple Gene Expression Omnibus (GEO) datasets. Their analysis integrated genomic profiling, digital pathology, and single-cell RNA sequencing data. The researchers first performed gene set enrichment analysis on 24 microenvironmental cell subpopulations, classifying patients into three tumor immune microenvironment (TIME) clusters with distinct immune infiltration patterns and survival outcomes. 

From differential gene expression analysis between these clusters, they identified 18 genes associated with both immune function and steroid metabolism. Unsupervised clustering of these genes revealed two molecular subtypes, which the team used to develop the Steroid-related Immune Score (SIS) through principal component analysis. Patients were then stratified into high-SIS and low-SIS groups based on an optimal cutoff value. 

To validate this classification, the researchers employed two deep learning models, ResNet50 and Vision Transformer-B16, to analyze whole-slide pathology images from 55 patients in the TCGA database. ResNet50 reached an area under the curve of 0.82 and 71% accuracy in five-fold cross-validation. The team validated their best-performing model in an independent cohort of 20 patients from two Chinese hospitals, confirming the generalizability of the model. 

SIS Captures Biological Tumor Characteristics Beyond Conventional Pathological Assessment

The team conducted Class Activation Mapping (CAM) analysis to visualize which regions of pathology slides drive AI predictions. They found that high-SIS tumors were characterized by prominent lymphocyte infiltration, particularly CD8+ T cells, which are typically lacking in ACC tumors. CAM analysis of data from an external validation cohort confirmed that high-SIS tumors had lymphocytic infiltration, whereas low-SIS patients had minimal immune cell infiltration 

“We were surprised to find that a subset of ACC cases, traditionally considered immunologically ‘cold,’ exhibited high SIS values accompanied by robust lymphocyte infiltration,” Dr. Liu noted. “This finding challenges the conventional view that cortisol-secreting ACCs are uniformly immunosuppressed and instead highlights a metabolically driven immune diversity within the disease.” 

DHCR7 Is A Strong Prognostic Marker

The researchers identified DHCR7, an enzyme involved in cholesterol biosynthesis, as the metabolic gene most strongly associated with SIS. Pan-cancer analysis showed that DHCR7 expression levels are higher in ACC than in other cancers, and elevated expression correlated with worse survival outcomes. Additionally, DHCR7 knockdown in ACC cells (SW-13 and NCI-H295R) increased sensitivity to mitotane, with IC50 values decreasing significantly after gene silencing. 

“High DHCR7 expression was strongly associated with poor prognosis and reduced treatment responsiveness, suggesting that metabolic activity can profoundly influence immune phenotypes and therapeutic vulnerability,” Dr. Liu said. “This points to DHCR7 and related pathways as potential metabolic targets in ACC.” 

Tailoring Treatment to Molecular Subtype

Patients with high-SIS tumors demonstrated features associated with immunotherapy responsiveness, including elevated expression of immune checkpoint molecules (PD-1, PD-L1, PD-L2, CTLA4), higher immunophenoscore, and enrichment in immune-related pathways. Tumor Immune Dysfunction and Exclusion (TIDE) analysis showed that over 70% of patients with high-SIS tumors would likely respond to immune checkpoint blockade, and more than 65% of those with low-SIS tumors would be resistant. 

In addition, patients with low-SIS tumors exhibited characteristics of active steroid biosynthesis: 78.7% had abnormal hormone secretion, 57.4% had elevated cortisol, and 66% showed high adrenal cortical differentiation index scores. Gene set enrichment analysis confirmed enrichment in steroid biosynthesis and cholesterol homeostasis pathways in this subgroup. 

Drug sensitivity prediction using the Genomics of Drug Sensitivity in Cancer database revealed that high-SIS patients were significantly more responsive to PI3K/Akt/mTOR pathway inhibitors (90% sensitivity). In contrast, low-SIS patients showed greater sensitivity to PLK1 inhibitors and various agents targeting steroid hormone synthesis. 

Single-Cell Insights Into Immune Escape

Single-cell RNA sequencing of a low-SIS tumor demonstrated that cancer cells expressed 79 exosome-associated genes enriched in hormone synthesis, metabolic reprogramming, immune evasion, and drug resistance pathways. Analysis of cell-cell communication showed that tumor-derived exosomes interact extensively with T cells and macrophages through ligand-receptor pairs, including VCAM1-ITGA4 and TGFB1-TGFBR2. 

Dr. Liu explained that the VCAM1-ITGA4 axis represents a novel immune escape mechanism in ACC. Tumor-secreted VCAM1 binding to T cell surface receptor ITGA4 may prevent T cells from adhering to tumor cells and compete with fibronectin (FN1) for ITGA4 binding, further suppressing T cell function. 

Additionally, the analysis identified ligand-receptor pairs (AXL, HGF, PDGFB, PDGFRB) in macrophages and mesenchymal cells that may contribute to resistance against EGFR tyrosine kinase inhibitors, which have shown limited efficacy in ACC despite frequent EGFR overexpression. 

An Unexpected Connection to Neuropsychiatry

The researchers found that ACC shares metabolic features with schizophrenia, particularly in calcium and iron homeostasis. Immune quantitative trait loci (immunQTLs) analysis revealed that calcium-related genetic variants affecting regulatory T cells were most strongly associated with survival in ACC. Single-cell data showed that tumor-specific genes were enriched in calcium ion binding and iron ion binding pathways, both of which are implicated in schizophrenia. 

This led to the prediction that certain dopamine receptor antagonists used in schizophrenia treatment might have therapeutic potential in low-SIS ACC patients when combined with mitotane. 

“We hope to conduct targeted analyses of these shared ‘neuro-metabolic’ pathways to explore their impact on immune regulation in ACC,” Dr. Liu said, noting plans for experimental validation and potential clinical trials. 

Limitations and Future Directions

Study limitations include the relatively small size of the external validation cohort and the limited follow-up duration. In addition, the single-cell analysis included only one patient, limiting the generalizability of the immune escape mechanisms identified. 

Dr. Liu noted that upon further validation in prospective multicenter studies, “SIS could complement the Weiss score in borderline or ambiguous cases, adding a quantitative immune-metabolic dimension to histopathologic assessment.” He added, “High-SIS tumors are more likely to respond to immunotherapy, whereas low-SIS tumors may benefit more from hormone-suppressive or metabolic interventions.” 

The team is also pursuing experimental validation of DHCR7 as a therapeutic target and planning to integrate SIS into early-phase clinical trials as a stratification biomarker. 

“Although ACC remains a rare malignancy, its extreme heterogeneity and metabolic complexity make it a fertile ground for precision medicine,” Dr. Liu concluded. “We hope SIS will serve as a bridge between morphology, metabolism, and immunity, enabling clinicians to make more informed, individualized treatment decisions.” 

The study received financial support from the Scientific Research Project of the Ministry of Education of Liaoning Province, the Interdisciplinary Research Cooperation Project Team Funding of Dalian Medical University, and the Joint Foundation of the Dalian Institute of Chemical Physics, Chinese Academy of Sciences, and the Second Hospital of Dalian Medical University. 

References 

1. Hao W, Yao L, Wang Y, et al. Multi-modal characterization of metabolic and immune gene clusters in adrenocortical carcinoma treatment. NPJ Precis Oncol. 2025;9(1):314. Published 2025 Sep 18. doi:10.1038/s41698-025-01092-4.