Sexually dimorphic radiogenomic models identify distinct imaging and biological pathways that are prognostic of overall survival in glioblastoma

Niha Beig, Salendra Singh, Kaustav Bera, Prateek Prasanna, Gagandeep Singh, Jonathan Chen, Anas Saeed Bamashmos, Addison Barnett, Kyle Hunter, Volodymyr Statsevych, Virginia Hill, Vinay Varadan, Anant Madabhushi, Manmeet S. Ahluwalia, Pallavi TiwariNeuro-Oncology 2020 · DOI

Abstract

Sexually dimorphic radiogenomic models could help risk-stratify GBM patients for personalized treatment decisions.

Introduction

Glioblastoma (GBM) is the most aggressive primary brain tumor, characterized by poor prognosis and significant variability in patient outcomes. While epidemiological data consistently shows that females survive longer than males, most prognostic models are "all-comers" models that treat sex as a simple covariate or ignore it entirely. This study addresses this gap by developing the first "sex-controlled" radiogenomic models, investigating whether sex-specific imaging phenotypes can better predict overall survival (OS) and reveal the distinct biological pathways driving these outcomes in men versus women.

Methodology

The researchers employed a multi-scale radiogenomic approach to link macroscopic MRI features with microscopic gene expression data.

Overview of the radiogenomic pipeline, including data acquisition, feature extraction, and survival analysis.

The methodology involved three primary stages:

  1. Radiomics: 2,850 quantitative features were extracted from three MRI protocols (Gd-T1w, T2w, and FLAIR) across three tumor subcompartments: the necrotic core, the enhancing tumor, and peritumoral edema.
  2. Model Development: Using LASSO Cox regression, the team built three distinct Radiomic Risk Score (RRS) models: male-specific ($m$-RRS), female-specific ($f$-RRS), and a combined "all-comers" model ($all$-RRS).
  3. Radiogenomics: To provide biological context, the RRS was linked to transcriptomic (RNA-seq) data. Gene Ontology (GO) and single-sample Gene Set Enrichment Analysis (ssGSEA) were used to identify enriched biological processes and correlations between MRI textures and gene expression.

Datasets and Cohort Selection

The study utilized a large multi-institutional dataset of 313 GBM patients (196 male, 117 female) curated from four major sources: The Cancer Imaging Archive (TCIA), Ivy Glioblastoma Atlas Project (Ivy GAP), Clinical Proteomic Tumor Analysis Consortium (CPTAC), and the Cleveland Clinic Foundation (CCF).

Flowchart illustrating the inclusion/exclusion criteria and the division of patients into training and independent test cohorts for both radiomic and radiogenomic analyses.

For the radiogenomic analysis, a subset of 147 patients with available gene expression data was used to validate the biological underpinnings of the radiomic features.

Key Results and Biological Insights

A critical finding of the study was the failure of the "all-comers" model. When a model trained on a mixed-sex population was validated on sex-specific cohorts, it failed to remain prognostic. This suggests that "averaging" the two sexes masks the unique features that drive survival in each group.

Male-Specific High-Risk Biology

In males, "high-risk" (poor survival) was associated with Laws energy features—which capture textural heterogeneity like spots and ripples—within the enhancing tumor and edema. These imaging markers correlated strongly with aggressive biological processes, specifically angiogenesis and cell adhesion. The $m$-RRS was highly significant in both training ($P < 0.00001$) and independent validation ($P = 0.0028, HR = 1.73$).

Female-Specific Low-Risk Biology

In females, "low-risk" (better survival) was linked to Laws energy features (detecting levels and edges) in the necrotic core. These features were significantly associated with immune-related signaling pathways, such as IL-6 signaling and T-cell cytotoxicity. The $f$-RRS was also prognostic in validation ($P = 0.046, HR = 0.607$).

Clinical Utility and Limitations

Combining these sex-specific radiomic scores with traditional clinical factors (age, resection extent, MGMT status) significantly improved survival prediction accuracy, reaching a C-index of 0.88 for males and 0.69 for females. This suggests that MRI features can serve as non-invasive proxies for underlying tumor microvasculature and immune environments.

However, the study is limited by its retrospective nature and the relatively small female sub-cohort for genomic analysis ($n=53$). Future work should focus on prospective, multi-site trials to validate these "sex-controlled" markers, potentially allowing for the selection of patients for sex-specific clinical trials, such as testing immunotherapies specifically in female cohorts where immune pathways are dominant.

Summary updated

Abstract

Sexually dimorphic radiogenomic models could help risk-stratify GBM patients for personalized treatment decisions.
BibTeX
@inproceedings{beig2020sexually,
  author = {Niha Beig and Salendra Singh and Kaustav Bera and Prateek Prasanna and Gagandeep Singh and Jonathan Chen and Anas Saeed Bamashmos and Addison Barnett and Kyle Hunter and Volodymyr Statsevych and Virginia Hill and Vinay Varadan and Anant Madabhushi and Manmeet S. Ahluwalia and Pallavi Tiwari},
  title = {{Sexually dimorphic radiogenomic models identify distinct imaging and biological pathways that are prognostic of overall survival in glioblastoma}},
  booktitle = {Neuro-Oncology},
  year = {2020},
  doi = {10.1093/neuonc/noaa231},
}
Figure