Voxel-based texture similarity networks reveal abnormal brain network topology linked to molecular signatures in amyotrophic lateral sclerosis

While amyotrophic lateral sclerosis (ALS) is characterized by disrupted brain network function, the abnormal patterns of structural similarity networks and their biological underpinnings remain poorly understood. This study employed voxel-based texture similarity network (vTSN) analysis to identify complex brain network disruptions in ALS, aiming to advance the pathophysiological understanding of the disease. Individual vTSNs were constructed based on structural T1-weighted magnetic resonance imaging data from 113 patients with ALS and 144 healthy controls. We first explored the vTSN disruption patterns in ALS patients and compared their sensitivity with two other popular structural similarity networks: morphometric similarity networks (MSN) and morphometric inverse divergence (MIND). We then linked vTSN abnormalities with molecular signatures. Finally, we explored the potential of vTSN in predicting the clinical traits and states of ALS patients. ALS patients demonstrated bidirectional patterns of both increased and decreased connectivity in vTSN ( P < 0.05, Bonferroni correction), particularly involving sensorimotor network (SMN), default mode network (DMN), frontoparietal network (FPN) and subcortical network (SN). vTSN showed the highest sensitivity in detecting abnormal connectivity ( N = 137) compared to the two well-known approaches, MSN ( N = 0) and MIND ( N = 30) ( P < 0.05, Bonferroni correction), and revealed unique disruption patterns ( r < 0.02, P > 0.05). Notably, vTSN alterations were linked to ALS risk genes; representative genes CHMP2B and SOD1 showed distinct developmental profiles and significant regional correlations. Furthermore, vTSN changes mirrored neuroreceptor distributions (notably M1, 5-HT6, and MOR), with M1 showing coupling in the SN and FPN, linking network decay to neurochemical imbalances. Finally, the network-based vTSN model achieved robust diagnostic power (AUC = 0.757), which was statistically equivalent to edge-level or whole-brain models (DeLong’s test, P > 0.05). vTSN exhibits excellent performance for detecting network abnormality in ALS and reveals potential molecular signatures of the pathophysiology of disease. These findings underscore the potential of vTSN-derived network measure as a biologically meaningful biomarker for ALS.