Unraveling multi-scale neuroimaging biomarkers and molecular foundations for schizophrenia: A combined multivariate pattern analysis and transcriptome-neuroimaging association study

Abstract

Aims: Schizophrenia is characterized by alterations in resting-state spontaneous brain activity; however, it remains uncertain whether variations at diverse spatial scales are capable of effectively distinguishing patients from healthy controls. Additionally, the genetic underpinnings of these alterations remain poorly elucidated. We aimed to address these questions in this study to gain better understanding of brain alterations and their underlying genetic factors in schizophrenia.

Methods: A cohort of 103 individuals with diagnosed schizophrenia and 110 healthy controls underwent resting-state functional MRI scans. Spontaneous brain activity was assessed using the regional homogeneity (ReHo) metric at four spatial scales: voxel-level (Scale 1) and regional-level (Scales 2-4: 272, 53, 17 regions, respectively). For each spatial scale, multivariate pattern analysis was performed to classify schizophrenia patients from healthy controls, and a transcriptome-neuroimaging association analysis was performed to establish connections between gene expression data and ReHo alterations in schizophrenia.

Results: The ReHo metrics at all spatial scales effectively discriminated schizophrenia from healthy controls. Scale 2 showed the highest classification accuracy at 84.6%, followed by Scale 1 (83.1%) and Scale 3 (78.5%), while Scale 4 exhibited the lowest accuracy (74.2%). Furthermore, the transcriptome-neuroimaging association analysis showed that there were not only shared but also unique enriched biological processes across the four spatial scales. These related biological processes were mainly linked to immune responses, inflammation, synaptic signaling, ion channels, cellular development, myelination, and transporter activity.

Conclusions: This study highlights the potential of multi-scale ReHo as a valuable neuroimaging biomarker in the diagnosis of schizophrenia. By elucidating the complex molecular basis underlying the ReHo alterations of this disorder, this study not only enhances our understanding of its pathophysiology, but also pave the way for future advancements in genetic diagnosis and treatment of schizophrenia.

Keywords: gene expression; multiscale analysis; multivariate pattern analysis; regional homogeneity; schizophrenia.

FIGURE 1

The flowchart of the spatial‐scale analysis to acquire the information of classifying schizophrenia from HCs and to characterize the involved gene expression profiles correlated with ReHo difference in schizophrenia at each spatial scale.

FIGURE 2

Pearson's correlation between MEs of gene modules and t‐statistics values of case–control ReHo differences in cortical and subcortical regions of schizophrenia at Scales 1–3. (A–C) Showed the correlation coefficients between MEs of gene modules and ReHo differences in Scale 1 (voxel level), Scale 2 (272 cerebral regions), and Scale 3 (53 regions), respectively. The color bar represents the correlation coefficients. At Scale 4 (17 regions), the limited number of cerebral regions prevented the implementation of WGCNA analysis, leading to underpowered analysis and inaccurate network modules.

FIGURE 3

Gene enrichment of genes significantly correlated with ReHo alterations in schizophrenia in Scale 1. (A) Significant GO items of biological processes; (B) Significant gene ontology (GO) items of molecular function; (C) Significant GO items of cellular components. The x‐axis represented the p value of enrichment for each GO item (y‐axis). The size of each sphere indicated the number of genes overlapped with each GO item, and the color of each sphere indicated the significance level of enrichment, as shown in the color bar.