Kannan, Kasthuri S.

Abstract High-grade gliomas (HGG) have significant spatial heterogeneity, complicating therapeutic approaches. While multiparametric MRI (mpMRI) enables noninvasive assessment of this heterogeneity, most predictive models lack biological interpretability when used to classify or predict outcomes. Here, we developed an unsupervised framework that identifies novel mpMRI feature patterns through dimensional reduction, linking them to clinically relevant molecular signatures. We analyzed 18 advanced mpMRI features from 73 stereotactic biopsies across 20 HGG patients. Imaging data were normalized, followed by dimensional reduction using an iterative UMAP across varying parameters. Density-based clustering was used to generate stable image-derived clusters through consensus co-clustering. RNA-seq data from matched biopsies were analyzed using DESeq2 and GSEA to identify differentially expressed genes and enriched pathways. We then constructed a network linking specific imaging features to these transcriptional signatures. Validation was performed in an independent cohort, allowing us to extend clusters across full brain mpMRI. Unsupervised clustering of mpMRI features produced nine stable image-based clusters. These clusters were not fully explained by contrast enhancement, metabolic class, or tumor purity, shown by intraclass correlation (ICC) analyses, suggesting they capture novel biological variation. Transcriptomic analysis showed each cluster was associated with distinct gene expression patterns and enriched biological pathways, including neuronal differentiation, immune chemotaxis, and chromatin remodeling. Key imaging features (sRCBV, MTT, K2)—combined in cluster-specific patterns—enabled construction of an interpretable imaging–transcriptomic network. This was validated in an independent cohort. We then applied to full brain mpMRI, mapping spatially distinct regions to biological trends and patient outcomes. Addressing a research gap in imaging HGG, we present an interpretable network that connects specific patterns of mpMRI to clinically meaningful molecular signatures. This approach advances our understanding of tumor heterogeneity and supports future translational work in precision patient care.

Abstract BACKGROUND Glioblastoma (GBM) is a molecularly and phenotypically heterogeneous tumor, with regional invasion patterns driving treatment resistance and universal recurrence. In infiltrative regions, GBM cells adopt at least two predominant invasive phenotypes: Neuronal-enriched (NEU), with synaptic and neurodevelopment programs, and Glycolytic/Plurimetabolic (GPM), with heightened immune and metabolic activity. These subtypes differ in therapeutic vulnerability and microenvironmental interactions; features not captured by conventional contrast-enhanced (T1+C) MRI due to minimal blood-brain barrier (BBB) disruption. Our group sought to develop predictive models to distinguish variability of NEU/GPM subtypes within infiltrative GBM, utilizing graph convolutional networks (GCN), advanced multiparametric MRI (mpMRI), and spatially-matched image-localized biopsies with transcriptomic profiling. METHODS We analyzed 18 mpMRI features —including conventional, diffusion-weighted, and Dynamic Susceptibility Contrast (DSC) perfusion MRI— of 101 image-localized biopsies from MRI non-enhancing infiltrative margins of 64 GBM patient tumors. Features were hierarchically clustered per modality; each cluster was represented as a node in a Neo4j graph. Edges connected nodes sharing overlapping samples. Nodes were annotated with NEU or GPM subtype proportion. GCNs were trained to classify subtype-enriched nodes using graph topology and node features. RESULTS GCNs achieved high predictive performance on held-out test data (AUC = 0.95 for NEU, 0.88 for GPM; 70:30 train-test split) with similar 5-fold cross-validation performance and generalized well to an independent cohort (AUC > 0.85, n = 45). Importantly, performance was robust without relying on T1+C imaging, enabling prediction beyond BBB-disrupted regions. GCNs outperformed traditional classifiers (logistic regression, random forest). CONCLUSION Our results demonstrate that GCN-modeling of mpMRI features enables non-invasive prediction of invasive GBM subtypes at the tumor margin. Unlike convolutional neural networks (CNN), which rely on spatial continuity, GCNs leverage relational structure between heterogeneous imaging features, making them ideal for modeling sparse, non-grid data. This may support non-invasive regional vulnerability detection, surgical targeting, and improved response assessment.

Abstract Introduction: High-grade gliomas (HGG) are aggressive brain tumors with heterogeneous cell populations, including invasive regions driving disease progression and recurrence. Our previous work integrated molecular profiling with spatially matched advanced multiparametric MRI in a cohort of HGG patients with multi-regional sampling, uncovering distinct biological signatures linked to individual advanced MRI phenotypes. Yet, the relationships between MRI phenotypes—and how this potentially informs underlying biology—remains unexplored. Here, we leverage an unsupervised approach to dimensional reduction for a broad array of MRI features and compare regional transcriptomic signatures in a cohort of image-localized HGG samples. Methods: From our existing data, we selected 20 HGG patients with 73 multi-regional tumor biopsies, each including 18 advanced MRI parameter calculations. With an iterative approach, data was projected using uniform manifold approximation (UMAP) at multiple resolutions and clustered, producing high-confidence stable clusters. We used random forest modeling to assess cluster associations with previously identified signatures. Cluster-specific transcriptomic profiles were detected by DESeq2 and subsequent gene-set enrichment analysis. We further evaluated Pearson correlations and statistical fit with MRI image parameters by performing pathway analysis on genes with significant correlations by image feature, across clusters. Results: 8 clusters emerged from our analyses. Most advanced MRI features exhibited statistically significant differences across clusters, forming unique patterns of feature behavior. Notably, while these image-based clusters did not correlate well with metabolic profiles, contrast enhancement, or genomics, they did demonstrate distinct and HGG-relevant transcriptomic profiles, with shared and divergent signatures. Across clusters, we also identified several correlations between specific MRI parameters and gene pathway trends. These highlighted combined patterns of multiple MRI features, offering a spectrum of regional phenotypes imperceptible with single feature analysis. We generalize these results to an image-based decision matrix to detect biologic signatures for validation. Conclusion: Our integrative analysis of MRI mapping stratifies imaged-based patterns to non-invasively assess molecular signatures within HGG. This approach will help elucidate MRI biomarkers that may predict glioma behavior and support future individualized treatment approaches. Citation Format: Matthew W. Kenaston, Matthew Flick, Kasthuri Kannan, Michele Ceccarelli, Nhan Tran, Leland S. Hu. Integrative analysis of regional MRI phenotypes in high-grade glioma: Dimensional reduction of multi-omics imaging and molecular profiles [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5776.

Abstract BACKGROUND Standard of care surgery for high grade glioma (HGG) employs Magnetic Resonance Imaging to identify the contrast enhancing (CE) core for resection, leaving a non-enhancing (NE) invasive rim of tumor which contributes to recurrence. Multi-parametric MRI (mpMRI) has recently been used to map regional biological heterogeneity of neuronal and immune signatures in HGG, and continues to offer a promising strategy for non-invasive surveillance of tumor progression. Here, we present a graph network analysis relating mpMRI to biological signatures in the NE tumor rim and show the utility of communities of imaging features for discriminating tumor phenotypes. METHODS 53 mpMRI features across 101 spatially-localized RNA-sequenced biopsy samples from 64 patients with IDH-wt HGG were represented as nodes within a Neo4j graph network. Modularity optimization was used to identify communities of densely connected nodes. Communities of nodes representing high proportions of NE samples (one-tailed test) were kept for further analysis. Gene set enrichment analysis (GSEA) on samples within selected imaging communities was analyzed for biological process enrichment using clusterProfiler. Single sample GSEA (ssGSEA) using ssGSEA2 was used to annotate neuronal and immune cell states across samples within selected communities. mpMRI feature selection within each community identified the most relevant imaging features for isolating samples classified into neuronal and immune cell states. RESULTS 11 imaging communities gathered a high proportion of NE samples. Three NE communities showed significant enrichment of neuronal signatures, while one showed significant enrichment of immune pathways. One neuronal-predominant community showed unique upregulation of 14 pathways related to neuronal migration/astrocyte projection. CONCLUSION Network imaging communities can identify clusters of NE glioma samples showing contrasting neuronal and immune signatures, including distinct upregulation of migration and projection pathways. Mapping these imaging feature communities to specific biological processes implicated in glioma progression may aid treatment planning and prognostication for glioma patients.

Abstract BACKGROUND High-grade glioma (HGG) is the most common and aggressive form of primary brain cancer. Standard of care surgery employs Magnetic Resonance Imaging to identify the contrast enhancing (CE) core for resection, leaving a non-enhancing (NE) invasive rim of tumor which contributes to recurrence. The hypoxic tumor microenvironment is thought to drive invasion into healthy brain, and understanding cellular responses to hypoxia may provide insight into combatting recurrence. Here, we apply a graph network to integrate regionally heterogeneous data from a unique dataset of spatially-localized CE and NE biopsies from a large cohort of patients with HGG to identify gene networks regulating hypoxia response and invasion. METHODS RNA-sequencing of 1159 differentially expressed genes across 159 IDH-wt biopsies (74 patients) from NE/CE regions were represented as nodes in a Neo4j graph network. Candidate genes for network random walks occurred exclusively within NE communities detected by modularity optimization and were localized by cell-type using single-cell RNA-sequencing (scRNAseq). CSF2RA was selected as source (walk length = 5; 10,000 walks/source node). Frequently occurring genes (one-tailed test) in walks were ranked by correlation with CSF2RA expression in NE samples and analyzed by gene set enrichment analysis for biological processes (clusterProfiler). CSF2RA and hypoxia gene expression were visualized with DoHeatmap in composite NE/CE scRNAseq profiles. RESULTS CSF2RA, an immune receptor, was upregulated (logFC = 0.789) in NE samples, scRNAseq myeloid populations, and Glycolytic/Plurimetabolic glioma subtype. CSF2RA random walk genes (n=46) enriched metabolic/proliferative signatures. CSF2RA expression correlated negatively with hypoxia genes globally and regionally (NE/CE), while NE/CE scRNAseq revealed distinct cellular expression patterns of CSF2RA and hypoxia response genes. CONCLUSION CSF2RA expression may prime NE glioma cells for invasion by altering hypoxia response. The genetic network bridging CSF2RA and hypoxia response may be a targetable mechanism of tumor recurrence.

Abstract BACKGROUND High-grade glioma (HGG) is the most common and aggressive form of primary brain cancer. Standard of care surgery employs Magnetic Resonance Imaging to identify the contrast enhancing (CE) core for resection, leaving a non-enhancing (NE) invasive rim of tumor which contributes to recurrence. Recent multi-omics analyses identified predominant neuronal and immune signatures within NE tumor. Here, we apply a graph network to integrate regionally heterogeneous data from a unique dataset of spatially-localized CE and NE biopsies from a large cohort of patients with HGG. METHODS RNA-sequencing of 1159 differentially expressed genes across 159 IDH-wt biopsies (74 patients) from NE/CE regions were represented as nodes in a Neo4j network. Candidate genes for network random walks occurred exclusively within NE communities detected by modularity optimization and localized by cell-type using single-cell RNA-sequencing (scRNAseq). CSF2RA and RBFOX1 were selected as sources (walk length = 5; 10,000 walks/source node). Frequently occurring genes (one-tailed test) in walks were ranked by correlation with CSF2RA/RBFOX1 expression in NE samples and analyzed by gene set enrichment analysis for biological processes (clusterProfiler). Random walk genes were visualized (DoHeatmap) against composite NE/CE scRNAseq profiles. RESULTS CSF2RA and RBFOX1 were upregulated in NE samples (logFC = 0.789, 1.85). Upregulation of CSF2RA, an immune receptor, in the Glycolytic/Plurimetabolic and RBFOX1, a neurodevelopmental regulator, in the Neuronal glioma subtypes were observed in NE samples (ANOVA p < 0.05). CSF2RA walk genes (n=46) enriched metabolic/proliferative signatures, while RBFOX1 walk genes (n=201) showed neurodevelopmental upregulation. scRNAseq localized CSF2RA to monocytes/macrophages while RBFOX1 was expressed in distinct astrocyte/neuroepithelial populations. CONCLUSION CSF2RA and RBFOX1’s gene networks modulate immune and neurodevelopmental processes influencing glioma progression in NE tumor. Understanding the region-specific roles of these networks may offer strategies for targeting cells in the invasive glioma rim.

Abstract BACKGROUND High-grade glioma (HGG) is the most common and aggressive form of primary brain cancer with poor prognosis. Standard of care surgery employs Magnetic Resonance Imaging to identify the contrast enhancing (CE) core for resection, typically leaving behind a non-enhancing (NE) invasive rim of tumor which invariably contributes to recurrence. Recent multi-omics analyses identified predominant neuronal and immune signatures within NE tumor, which could provide key insights for combatting recurrence. We present the novel application of graph network analyses to integrate multi-omic data from a unique dataset of multi-regional spatially-localized CE and NE biopsy samples from a large cohort of patients with HGG. MATERIAL AND METHODS 159 IDH-wildtype tumor biopsy samples from NE and CE regions of 74 glioma patients were profiled by RNA sequencing. Differentially expressed genes (DEGs) across NE and CE samples (n = 1159) were determined using the EdgeR package, and represented as nodes within a Neo4j graph network. Samples were hierarchically clustered for each gene (method = median), and edges were generated between gene nodes sharing ≥ 1 sample. Modularity optimization was used to identify communities of densely connected nodes. Genes occurring exclusively within NE communities were localized by cell type using external single-cell RNA sequencing datasets and identified as source node candidates for NE network random walks. These were performed with CSF2RA nodes as source (walk length = 5; 10,000 walks/source node). Random walks traversing samples with significantly elevated CSF2RA expression (one-tailed test) were kept for analysis. Genes occurring at high frequency (one-tailed test) within random walks were ranked by correlation with CSF2RA expression in NE samples and analyzed by gene set enrichment analysis (GSEA) for biological processes using clusterProfiler. RESULTS CSF2RA, an immune signaling receptor gene, was upregulated (logFC = 0.789) in NE samples and uniquely occurred within NE communities. We show its differential expression within the immune-enriched Glycolytic/Plurimetabolic (GPM) glioma subtype and confirm its enrichment in myeloid populations across single cell HGG datasets. NE network random walks identified 46 DEGs associated with CSF2RA. GSEA revealed upregulation of proliferative and metabolic signatures and downregulation of signaling pathways within CSF2RA’s network. CONCLUSION CSF2RA’s genetic network in NE tumor represents a potential mechanism for immunologic facilitation of glioma progression within this region. Given the expression of CSF2 ligand by glioma cells, which plays a crucial role in recruiting myeloid cells during peripheral cancer progression, modulating the expression of key genes in the CSF2RA network may be a promising approach for perturbing the crosstalk signals between immune and tumor cells in the NE region.