Demographic characteristics and two EOS subtypes

We used a combination of multiple cortical features and transcriptional data to establish the connections between gene expression and alterations in the MSN of EOS subtypes compared to HC (Fig. 1). The demographic analysis demonstrated no statistically significant differences in sex, age, TIV, or Euler number between EOS patients (mean age: 14.60 ± 1.95) and HC (mean age: 14.58 ± 2.46) (Table 1). The consistency of clustering assignments across different resolutions (2–10 clusters) was assessed using the ARI, which is known for its insensitivity to the number of clusters (K) [12]. The highest reproducibility was observed with a two-cluster solution (K = 2), yielding an ARI value of 0.62 (Additional file 1: Fig. S1). At resolutions of K = 3 to 10, the ARIs were lower than those at K = 2. Consequently, 53 EOS patients were assigned to EOS1, and 47 were assigned to EOS2. Based on these convergent findings, a subsequent analysis focused on the two EOS subtypes (EOS1 and EOS2). Furthermore, both EOS subtypes exhibited decreased verbal (VIQ), performance (PIQ), and full-scale intelligence quotient (FSIQ) scores compared with HC. The EOS1 displayed lower VIQ and FSIQ scores compared to EOS2 (Additional file 1: Fig. S2). Additionally, EOS1 and EOS2 exhibited no significant difference in sex, age, TIV, and PANSS scores.

Fig. 1

The workflow of this study. a. MSN construction. The 308 × 308 matrix for MSN was calculated across multiple macrostructural features (surface area, cortical thickness, gray matter volume, Gaussian curvature, and mean curvature). The MSN strength was obtained by the average weighted correlation coefficients between a given region and all other regions. b. Classification. The HYDRA method was used to identify EOS subtypes within the MSN strength. c. EOS-related gene analysis. PLS regression was then used to identify imaging transcriptomic associations. The link between brain-wide gene expression and morphometric changes in different EOS subtypes was evaluated through PLS weight mapping, genetic commonalities evaluation with psychiatric disorders, functional enrichment of PLS weighted genes, and cellular transcriptional signature assessment

Table 1 Clinical and demographic characteristicsSubtyping EOS-related changes in MSN strength

The temporal and frontal lobes displayed high MSN strength, whereas the somatosensory and occipital cortices exhibited low MSN strength (Fig. 2a). The global MSN strength of EOS1 was reduced compared with HC, whereas the global MSN strength of EOS2 did not differ significantly from HC (Additional file 1: Fig. S3). Specific cortical regions exhibited significant differences in the MSN strength between EOS1 and HC (Additional file 2: Table S2, Fig. 2b). Significantly decreased MSN strength in EOS1 was found in the temporal cortex, precentral gyrus, paracentral gyrus, superior frontal gyrus, and insula. Conversely, significantly increased MSN strength in EOS1 was observed in the lateral occipital gyrus, medial frontal gyrus, left entorhinal cortex, superior parietal gyrus, cuneus, postcentral gyrus, and lingual gyrus. Moreover, we found two cortical regions with significant MSN strength differences between EOS2 and HC (Additional file 2: Table S2, Fig. 2c). Significantly decreased MSN strength in the EOS2 subtype was observed in the right paracentral gyrus (part 3), whereas significantly increased MSN strength was observed in the right superior frontal gyrus (part 2) in the EOS2 subtype. We next explored the differences in MSN strength between EOS1 and EOS2 subtypes. Compared to the EOS2 subtype, the EOS1 subtype exhibited decreased MSN strength in the superior frontal gyrus, middle frontal gyrus, insula, and anterior central gyrus, and increased MSN strength in the lateral occipital cortex, lingual gyrus, and cuneus (Additional file 1: Fig. S4).

Fig. 2

Subtyping EOS-related regional changes in MSN strength. a. The MSN strength of EOS subtypes and HC. b-c. Case–control comparison of MSN strength for EOS1 and EOS2. d. Scatterplot of the control MSN strength and case–control t-map

Abnormal MSN strength was observed in functional Yeo 7 networks [38] (Additional file 1: Fig. S5a) and von Economo atlas (Additional file 1: Fig. S5b) [39] between EOS subtypes and HC. For the Yeo 7 functional networks, the EOS1 subtype displayed significantly increased MSN strength in the visual network and decreased MSN strength in the somatomotor, ventral attention, and default mode networks, whereas the EOS2 subtype displayed significantly increased MSN strength in the somatomotor network. Concerning the von Economo atlas, EOS1 exhibited significantly increased MSN strength in the secondary and primary sensory networks and decreased MSN strength in the primary motor, association, and insular networks. whereas EOS2 displayed significantly increased MSN strength in the primary motor network.

We used a quadratic non-linear model to construct the global MSN strength and the MSN strength of the functional network to explore the developmental trajectories of EOS subtypes from childhood to adolescence. The reason for selecting this approach is because age exerts non-linear patterns on the brain [40]. Thus, the default network of HC demonstrated a trend of an initial increase followed by a gradual decline from childhood to adolescence. The limbic and default network of EOS1 exhibited a rapid increase, followed by a gradual decline. Furthermore, the global strength and ventral attention network of EOS2 displayed a pronounced and consistent decline (Additional file 1: Fig. S6).

The MSN strength of HC and the case–control t-map of EOS1 exhibited a negative and spatial correlation (r(308) = -0.96, pspin < 0.0001) (Fig. 2d), suggesting that more connected regions exhibited larger case–control differences [14]. In addition, 56% of the positive MSN strength in HC and the negative t-values in the EOS1 were in decoupling, and 37% of the negative MSN strength in HC and the positive t-values in the EOS1 displayed dedifferentiation. However, EOS2 displayed the opposite pattern. The MSN strength of HC and the case–control t-map of EOS2 were positively and spatially correlated (r(308) = 0.70, pspin < 0.0001) (Fig. 2d). We found that 30% of negative MSN strength in HC and negative t-values in the EOS2 exhibited hyperdifferentiation, and 44% of positive MSN strength in HC and positive t-values in the EOS2 displayed hypercoupling. Spearman’s correlation analysis, conducted to assess the association between abnormal MSN strength and symptoms, revealed no significant correlations following Bonferroni’s correction (Additional file 2: Table S3). Additionally, we explored the spatial correlation between case–control t-maps of EOS subtypes and correlation coefficient maps of MSN strength and PANSS. In EOS1 patients, the case–control t-map displayed a significant spatial and positive correlation with statistical maps of MSN strength and PANSS positive and total scores but not with negative scores. In EOS2 patients, the case–control t-map exhibited significant spatial and negative correlation with statistical maps in PANSS positive and total scores, whereas positive correlation with statistical maps in PANSS negative scores (Additional file 1: Fig. S7).

In addition, we investigated the spatial correlation of case–control t-maps of the MSN strength between EOS subtypes and adult-SCZ in Morgan et al.’s study. EOS1 exhibited a significant spatial and positive correlation, whereas EOS2 displayed a significant spatial and negative correlation with case–control t-map of adult-SCZ (Additional file 1: Fig. S8). The positive/negative correlation suggested that EOS1 could be in a “classical SCZ” state, whereas EOS2 could fall in the “non-classical SCZ” state. The highest reproducibility of adult-SCZ was observed with a two-cluster solution (K = 2), yielding an ARI value of 0.52 (Additional file 1: Fig. S1) for subtype classification. The spatial correlation analysis revealed significant spatial and positive correlations of case–control t-map in both EOS1/2 and corresponding SCZ1/2 subtypes (Additional file 1: Fig. S9a-c). The positive correlations suggested that adult-SCZ could manifest two different subtypes similar to the EOS states. Moreover, type-I diseases displayed shared abnormal regions, encompassing increased regions in the superior parietal gyrus, lingual gyrus, cuneus, and postcentral gyrus, and reduced regions in the temporal gyrus, superior frontal gyrus, precentral gyrus, and insula (Additional file 1: Fig. S9d). However, no common abnormal regions were observed in type-II diseases.

Transcriptional patterns related to regional changes in MSN strength

PLS regression was performed to uncover the gene expression patterns based on the distinct anatomical distributions of case–control t-maps in MSN strength. Therefore, the PLS1 of two subtypes effectively explained 39% and 25% of the variations in the macrostructural differences for EOS1 and EOS2 patients, respectively (pperm < 0.0001) (Additional file 1: Fig. S10). The distribution of the PLS1 score-weighted map demonstrated an anterior–posterior gradient of gene expression in EOS1, whereas an inverse distribution in EOS2 cohorts was performed (Fig. 3a-b, Additional file 2: Table S4). This gradient reflected variations in the transcriptional architecture of the brain cortex and distinct expressional differences in EOS subtypes, which was also manifested in the EOS-related regional changes observed in the MSN strength map. Irrespective of EOS1 or EOS2 patients, the PLS1 scores exhibited a significant spatial correlation with the case–control t-value maps in MSN strength (EOS1, Spearman’s r = 0.57, pspin < 0.0001; EOS2, Spearman’s r = 0.52, pspin < 0.0001; Fig. 3c). Univariate one-sample Z tests were used to identified PLS1 weighted genes to represent transcriptional signatures for subsequent analysis, including 1,651 PLS1 + genes (Z > 5) and 1,743 PLS1- genes (Z < -5) for EOS1, and 79 PLS1 + and 318 PLS1- genes for EOS2 (all pFDR < 0.0001) (Additional file 2: Table S5).

Fig. 3

Transcriptional expression patterns related to differences in MSN strength. a. The distribution of differences in MSN strength and PLS1 scores in the left hemisphere of the EOS1 subtype. b. The distribution of differences in MSN strength and PLS1 scores in the left hemisphere of the EOS2 subtype. c. Scatterplots showing the significant spatial correlation between PLS1 scores and the case–control t-value maps of MSN strength in both EOS subtypes; EOS1, Spearman’s r = 0.57, pspin < 0.0001; EOS2, Spearman’s r = 0.52, pspin < 0.0001. d-e. The expression of SCZ-related genes from ISH datasets was positively or negatively associated with regional changes in MSN, including 6 positive genes (i.e., CIT, ARC, GRIN2A, PVALB, GABRB2, and PPP3CC) and 8 negative genes (i.e., HTR2C, GRM3, RGS4, DTNBP1, SYN2, GRIK4, TAC1, and CNR1). All r values were determined by Spearman’s correlation analysis, and p values were obtained from spatial correlation tests and adjusted with FDR correction

Next, we investigated the relationships between the published SCZ-related gene expression and corresponding regional changes in MSN strength. We acquired 77 SCZ-related genes by screening the keyword “schizophrenia” in the “Gene List” project from the ISH data in the AHBA database. In total, we screened 69 SCZ-related genes overlapping with 15,631 background genes and performed a spatial correlation analysis with case–control t-maps in MSN (Additional file 2: Table S6). In EOS1 cohorts, 14 SCZ-related genes exhibited a significant association (absolute(r) > 0.35, FDR correction, pspin < 0.05) with homologous case–control t-values, including six positive genes (i.e., CIT, ARC, GRIN2A, PVALB, GABRB2, and PPP3CC) and eight negative genes (i.e., HTR2C, GRM3, RGS4, DTNBP1, SYN2, GRIK4, TAC1, and CNR1). However, only one SCZ-related gene (TAC1) exhibited a significant positive correlation in EOS2 cohorts, and two genes (GRM7 and CIT) displayed a negative correlation with corresponding case–control t-maps (Fig. 3d). The expression of the top positively (or negatively) weighted genes was consistent with (or in contrast to) the distribution of variant regional changes in MSN strength, including CIT (Spearman’s r = 0.52, FDR correction, pspin = 0.006), CNR1 (Spearman’s r = -0.47, FDR correction pspin = 0.005), TAC1 (Spearman’s r = 0.29, FDR correction pspin = 0.068), and GRM7 (Spearman’s r = -0.36, FDR correction pspin = 0.028) (Fig. 3e).

Potential relationship between EOS-related changes in MSN strength and transcriptional dysregulation of other mental disorders

To gain deeper insights into the potential relationship between regional changes and gene dysregulation in mental disorders, we performed the correlation analysis of PLS1- weighted genes and published disease-related DGEs (Additional file 2: Table S7). In EOS1 patients, we first obtained the DGEs of six mental disorders from Gandal’s study and further identified overlapping upregulated (log-2 [fold change] > 0) genes, including 2 for MDD, 106 for ASD, 127 for adult SCZ, 35 for BD, 41 for AAD, and 418 for IBD. The PLS1- gene weights displayed positive correlations with ASD-related and adult-SCZ-related DGE values, as confirmed by permutation tests: ASD (rs (106) = 0.29, adjusted pperm = 0.006) and adult-SCZ (rs(127) = 0.24, adjusted pperm = 0.007) (Fig. 4a-b). However, no significant correlation was found with other mental disorders, including BD (rs(35) = -0.10, adjusted pperm = 0.709), AAD (rs(41) = 0.16, adjusted pperm = 0.171) and IBD (rs(418) = 0.03, adjusted pperm = 0.306) (Fig. 4c-e). In contrast, in EOS2 cohorts, PLS1- gene weights were only positively related to IBD-related DGEs values (rs(70) = 0.33, adjusted pperm = 0.011), with few overlapping genes in the other five disorders (Fig. 4f, Additional file 1: Fig. S11). These results indicated that regional changes in two EOS subtypes exhibited distinct gene expression patterns of psychiatric disorders.

Fig. 4

Correlation analysis between PLS1 weighted gene expressions of changes in MSN strength and transcriptional dysregulation of various mental disorders. a-b. In EOS1 patients, PLS1- weights exhibited significant positive associations with upregulated differential gene expression (DGE) in autism spectrum disorder (ASD) (rs (106) = 0.29, adjusted pperm = 0.006) and adult schizophrenia (rs(127) = 0.24, adjusted pperm = 0.007). c-e. There was no significant correlation with DGE in other mental disorders in EOS1 patients, including bipolar disorder (BD) (rs(35) = -0.10, adjusted pperm = 0.709), alcohol abuse disorder (AAD) (rs(41) = 0.16, adjusted pperm = 0.171) and inflammatory bowel disease (IBD) (rs(418) = 0.03, adjusted pperm = 0.306). f. In EOS2 patients, PLS1- weights exhibited significant positive associations with upregulated DGE in IBD (rs(70) = 0.33, adjusted pperm = 0.011). All r values were determined by Spearman’s correlation analysis, and p values were obtained from permutation tests and adjusted with FDR correction

Moreover, the MAGMA gene-set enrichment analysis identified characteristic biological processes for these mental disorders, including energy metabolism-related processes (“cAMP catabolic process”, “electron transport”, “cytochrome c to oxygen”, and “thyroid hormone catabolic process”) in ASD, and pathways of amino-acid metabolism (“fatty acid homeostasis”, “aspartate metabolic process” and “aspartate biosynthetic process”) and neuronal signals (“cerebral cortex GABAergic interneuron fate commitment”) in adult-SCZ (Additional file 1: Fig. S12a-b). Additionally, the GWAS genes of BD were primarily enriched in neuronal interaction-related processes, including “regulation of dendrite development” and “beta-adrenergic receptor kinase activity”, whereas AAD was significantly enriched in Ion signal regulation related pathways, such as “nickel ion binding”, “localization within membrane”, and “regulation of potassium ion transport” (Additional file 1: Fig. S12c-d). In contrast, MDD was significantly associated with neuroreceptor signaling, including “ganglioside metabolic process” and “negative regulation of dopamine receptor signaling pathway”, whereas IBD was enriched in processes related to cellular development and immune activation, such as “endothelial cell development”, “JAK-STAT cascade involved in growth hormone signaling pathway” and “positive regulation of interleukin-1 beta production” (Additional file 1: Fig. S12e-f).

Functional enrichment of genes correlated with regional changes in MSN strength

To further elucidate the functional characteristics of genes correlated with regional changes in MSN, we aligned GO and KEGG enrichment analyses with PLS1 weighted gene lists (Additional file 2: Table S8). For EOS1 patients, the column diagram displayed the top 10 significant BP terms, including “regulation of cellular metabolic process” and “regulation of RNA metabolic process” for PLS1 + genes and “synaptic signaling”, “cell–cell signaling”, “localization” and “nervous system development” for PLS- genes (Fig. 5a). In EOS2 cohorts, PLS1 + gene lists were only enriched in five BP, such as “polyamine metabolic process” and “polyamine biosynthetic process”, whereas PLS- genes exhibited similar enrichment in cellular and RNA metabolic processes as the PLS + genes in EOS1 patients (Fig. 5b). In addition, the KEGG analysis validated the enrichment of common pathways between EOS1 PLS1 + genes and EOS2 PLS1- genes, especially in the “Herpes simplex virus 1 infection” pathway, whereas no significant enrichment for PLS1 + genes were observed in EOS2 (Fig. 5c). The EOS1 PLS1 + genes were uniquely enriched in the “MAPK signaling pathway” and “calcium signaling pathway”, suggesting their involvement encephalic regional abnormities in EOS1 patients. In contrast, the EOS1 PLS1- genes were significantly enriched in synaptic signaling-related functions, such as serotonergic synapse, phagosome, gap junction, neurotrophin signaling pathway, and others (Fig. 5d).

Fig. 5

Functional enrichment of PLS1 weighted genes related to regional changes in MSN. a-b. Top 10 biological process (BP) terms of PLS1 + (Z > 5, pFDR < 0.05) and PLS1- (Z <  − 5, pFDR < 0.05) gene enrichment in the two EOS subtypes. c-d. KEGG pathway enrichment of PLS + and PLS- genes in the two EOS subtypes. The large circle nodes represent terms of the pathway, and small circle nodes represent related genes. The size of large nodes represents the -log10 (adjusted p values) of pathways, and lines indicate the relationships between genes and pathways

Transcriptional signature assessment for canonical brain cell types and specific developmental stages in EOS

We next explored the transcriptional signatures at the cellular level along with the regional changes in MSN strength. We adopted an indirect method to match PLS1 weighted genes into seven canonical brain cell types, including endothelial cells, astrocytes, OPCs, microglia, oligodendrocytes, and excitatory and inhibitory neurons. The ssGSEA scores were used to summarize the gene expression of cells, displaying their distribution in different brain regions (Fig. 6a, Additional file 2: Table S9). In EOS1 patients, the PLS1 weighted gene list was significantly related to inhibitory neurons (number = 25, FDR-corrected adjusted pperm = 0.021, FDR-corrected) and excitatory neurons (number = 22, FDR-corrected adjusted pperm = 0.006) for PLS1 + genes and astrocytes (number = 28, FDR-corrected adjusted pperm = 0.019) for PLS1- genes (Fig. 6b-d, Additional file 2: Table S10). Only a few overlapping genes were present in EOS2 cohorts between cells and the PLS1 weighted gene list without statistical significance. Based on cell-specific genes, the enrichment analysis revealed that MSN strength changes in EOS individuals were significantly enriched in BPs associated with signaling transport, as well as pathways of “calcium signaling”, “neuronal system” and “tyrosine kinases receptor for neuronal cells” (Fig. 6c). Astrocyte-specific genes were enriched in cellular response-related processes, including “cellular response to chemical stress”, “response to glucocorticoid”, “regulation of epithelial cell proliferation”, “cell–cell adhesion” and “Transport of small molecules” (Fig. 6e).

Fig. 6

Cellular specific transcriptional signatures in accordance with MSN changes in EOS subtypes. a. The distribution of regional gene expression maps of seven brain cells using ssGSEA scores. b. The number of overlapping genes with PLS1 + weighted genes for each cell type, including inhibitory neurons (number = 25, adjusted pperm = 0.021); excitatory neurons (number = 22, adjusted pperm = 0.006); endothelial cells (number = 25, adjusted pperm = 1); microglia (number = 8, adjusted pperm = 0.825); astrocytes (number = 8, adjusted pperm = 1); oligodendrocytes (number = 2, adjusted pperm = 1); oligodendrocyte precursors (OPCs) (number = 2, adjusted pperm = 1). c. Gene ontology and pathway terms enriched for PLS1 + weighted genes for different cell types. d. The number of overlapping genes with PLS1- weighted genes for each cell type, including inhibitory neurons (number = 17, adjusted pperm = 0.9243); excitatory neurons (number = 9, adjusted pperm = 1); endothelial cells (number = 30, adjusted pperm = 0.9243); microglia (number = 4, adjusted pperm = 1); astrocytes (number = 28, adjusted pperm = 0.019); oligodendrocytes (number = 2, adjusted pperm = 1); and oligodendrocyte precursors (OPCs) (number = 6, adjusted pperm = 0.9243). e. Gene ontology and pathway terms enriched for PLS1- weighted genes for different cell types. All p values were obtained from permutation tests and adjusted with FDR correction

In addition, we evaluated the link between PLS weighted genes and developmental time windows across different brain regions in distinct EOS subtypes by developmental gene expression enrichment analysis, spanning from early fetal (EF) to young adulthood (YA) stages. The analysis revealed the PLS1 ± gene lists were predominantly expressed in brain regions from developmental terminal stages, particularly across specific areas including the cerebellum and thalamus for PLS1 + genes, and striatum, hippocampus, cortex, and amygdala for PLS1- genes in EOS1 patients (Additional file 1: Fig. S13 a-b). In contrast, no significant enrichment was present in any stages for PLS1 + genes in EOS2 patients, whereas PLS1- genes were primarily enriched in the early developmental stages in the cortex and amygdala areas (Additional file 1: Fig. S13 c-d). These findings identified specific cell types and distinct impressionable developmental stages, with distinct gene expression patterns according to the regional changes in MSN strength, providing maps of known specific cell types and developmental trajectories associated with EOS pathology.