Integrating morphometric similarity networks with imaging transcriptomics

Neurodegenerative diseases represent a heterogeneous group of progressive neurological disorders whose prevalence and socioeconomic burden are expected to rise with increasing global life expectancy. While traditional neuroimaging approaches are capable of detecting macroscopic structural alterations, they are limited in resolving the underlying microscale cellular and molecular mechanisms. This work proposes and validates a computational pipeline integrating Morphometric Similarity Networks (MSNs) with imaging transcriptomics to bridge whole-brain structural topology and gene expression signatures in neurodegenerative disease cohorts. As a case study, it focuses on Parkinson’s disease (PD), which is increasingly recognized as a brain-wide structural network disorder rather than a focal dopaminergic pathology. After an exploratory mass univariate analysis, individualized MSNs were constructed from T1-weighted MRI data for 212 subjects (106 PD patients, 106 healthy controls) from the Parkinson’s Progression Markers Initiative (PPMI), using five cortical morphometric features extracted via FreeSurfer and parcellated across the 68-region Desikan-Killiany atlas. Pairwise structural similarity was quantified through multivariate Euclidean distance, and node strength was derived as the primary topological metric. Group-level differences were assessed using a General Linear Model controlling for age, sex, estimated total intracranial volume and scanner model, with Benjamini-Hochberg FDR correction applied across all regions. The resulting spatial phenotype map of beta coefficients was cross-referenced with Allen Human Brain Atlas microarray gene expression data via Partial Least Squares regression, with spatial autocorrelation addressed through spin-test permutations using the Váša algorithm. Finally, gene set enrichment analyses were performed. Mass univariate and node strength analyses revealed subtle but directionally consistent structural alterations in PD, confirming the pipeline’s sensitivity to detect low-effect signals. Hub decoupling emerged as the dominant topological reorganization pattern, affecting 32.4% of regions and aligning with established hub vulnerability hypotheses. Although global PLS permutation testing did not reach significance, transcriptomic analysis identified LRRK2 as a significant canonical PD gene correlate, while Gene Set Enrichment Analysis consistently implicated neuroinflammatory pathways, glial cell signatures and excitatory neuronal enrichment as molecular correlates of the observed structural reorganization. This confirmed coherent integration between structural and transcriptomic layers. Secondary application of the full pipeline to a frontotemporal dementia cohort confirmed the framework’s generalizability, yielding more pronounced results consistent with the greater cortical involvement characteristic of FTD. This work delivers a modern, scalable and reproducible computational framework, providing a foundation for future integrative investigations of structural connectome alterations and their molecular underpinnings across neurodegenerative conditions.