Biological subtyping of autism via cross-species fMRI

Abstract

It is frequently assumed that the phenotypic heterogeneity in autism spectrum disorder reflects underlying pathobiological variation. However, direct evidence in support of this hypothesis is lacking. Here, we leverage cross-species functional neuroimaging to examine whether variability in brain functional connectivity reflects distinct biological mechanisms. We find that fMRI connectivity alterations in 20 distinct mouse models of autism (n=549 individual mice) can be clustered into two prominent hypo- and hyperconnectivity subtypes. We show that these connectivity profiles are linked to distinct signaling pathways, with hypoconnectivity being associated with synaptic dysfunction, and hyperconnectivity reflecting transcriptional and immune-related alterations. Extending these findings to humans, we identify analogous hypo- and hyperconnectivity subtypes in a large, multicenter resting state fMRI dataset of n=940 autistic and n=1036 neurotypical individuals. Remarkably, hypo- and hyperconnectivity autism subtypes are replicable across independent cohorts (accounting for 25.1% of all autism data), exhibit distinct functional network architecture, are behaviorally dissociable, and recapitulate synaptic and immune mechanisms identified in corresponding mouse subtypes. Our cross-species investigation, thus, decodes the heterogeneity of fMRI connectivity in autism into distinct pathway-specific etiologies, offering a new empirical framework for targeted subtyping of autism.

Keywords: autism; connectomics; cross-species; hyperconnectivity; hypoconnectivity; immune regulation; mouse; subtyping; synaptic signaling; transcriptional regulation.

Figure 1.. fMRI connectivity in 20 autism mouse models clusters into dominant hypo- and hyperconnectivity subtypes.

a) Ridgeline plot quantifications of whole-brain fMRI connectivity strength differences (mutant vs. control) for the 20 autism-relevant mouse models examined. Ridgelines show voxel distribution of fMRI connectivity strength indexed by Cohen’s d color-coded by connectivity difference (blue: lower connectivity in mouse mutants vs. control, i.e., hypoconnectivity; red: higher connectivity in mouse mutants vs. control, i.e., hyperconnectivity). b) Top-view brain maps showing fMRI connectivity differences (with Cohen’s d > 0.5) in the 20 autism-related mouse models relative to wild-type littermates. Blue/light blue indicates hypoconnectivity, red/yellow indicates hyperconnectivity in mutants. c) Heatmap of Cohen’s d values across brain regions. Hierarchical clustering revealed two distinct connectivity subtypes: a hyperconnectivity cluster (n=9 mouse models) and a hypoconnectivity cluster (n=11 mouse models). Hypo, hypoconnectivity; Hyper, hyperconnectivity. d) Coronal brain views highlighting voxels with consistent hypo- or hyperconnectivity in the two fMRI dysconnectivity subtypes. Color intensity reflects the number of autism-related mouse models showing fMRI dysconnectivity (blue, hypoconnectivity, top panel; red/yellow, hyperconnectivity, bottom panel) in each subtype (threshold: Cohen’s d > 0.8). Amy, amygdala; BF, basal forebrain; Cing, anterior cingulate; CPu, caudoputamen; HPC, Hippocampus; Motor ctx, motor cortex; Hypt, hypothalamus; Ins, Insula; mPFC, medial prefrontal cortex; MB, mid brain; NAc, nucleus accumbens.

Figure 2.. Distinct signaling pathways underlie fMRI connectivity subtypes in rodents.

a) Illustrative schematic of gene enrichment analysis used to link autism relevant pathways to rodent hypo- and hyperconnectivity subtypes. Autism-risk genes and immune factors (i.e., IL-6) modelled in mouse lines associated with either subtype (listed in blue and red typeface, respectively) were used as seed genes to generate mouse line-specific protein-protein interactomes. Within each subtype, these interactomes were then concatenated, and upon removal of shared genes, we generated two subtype-specific gene interactomes. Filled circles represent individual genes, gray links indicate gene interactions. Dashed lines delineate individual interactomes and solid lines outline the two concatenated interactomes. b) Heatmap displaying significant enrichment for autism-relevant pathways in the two interactomes. c) Heatmap displaying significant enrichment for modules of genes differentially expressed in autism for each of the two interactomes. The odds ratio for hypoconnectivity subtype are shown in the left column (blue coloring); those for the hyperconnectivity subtype in the right column (red coloring). Thick cell borders indicate that enrichment is significant at q(FDR) < 0.05. We report the list of genes belonging to each interactome and the pathways we probed in Supplementary Table 2.

Figure 3.. Cross-species identification of autism-related dysconnectivity subtypes.

Schematic illustration of the workflow we used to identify hypo- and hyperconnectivity subtypes across species. a) We applied data-driven hierarchical clustering analyses to our rodent database to identify dominant hypo- and hyperconnectivity subtypes across autism mouse models relevant to autism. For each subtype we generated a dysconnectivity prior mask, consisting of a set of anatomical regions exhibiting hypo- or hyperconnectivity across models. b) Gene enrichment analyses were used to uncover molecular pathways associated with hypo- vs. hyperconnectivity in rodent autism models (Supplementary Figure 1). c) In humans, we computed fMRI dysconnectivity by comparing resting state fMRI data of autistic vs. neurotypical (NT) individuals. Leveraging the cross-species translatability of fMRI, we used a region-wise approach to quantify fMRI dysconnectivity in individuals with autism relative to NTs. Specifically, we selected fMRI scans exhibiting hypo- or hyperconnectivity in human brain regions corresponding to the dysconnectivity priors identified in rodents (Supplementary Figure 2). d) Finally, we performed gene enrichment analyses to investigate whether brain-decoded genes from each map were enriched for molecular ontologies or gene modules known to be associated with autism.

Figure 4.. Replicable hypo- and hyperconnectivity subtypes can be identified in autism.

a) Sample size distribution by data collection (ASD; light gray, NT; dark gray); ASD total n=940; NT total n=1036, n=38 data collections across 23 sites (Table 2). b) Hypoconnectivity subtype maps in discovery (top subpanel: n=55, 7.4%) and replication (bottom subpanel: n=19, 9.7%) datasets. Blue indicates regions exhibiting fMRI hypoconnectivity compared to NTs (Cohen’s d≤0.8). c) Hyperconnectivity subtype maps in discovery (top subpanel: n=124, 16.7%) and replication (bottom subpanel: n=38, 19.4%) datasets. Red/yellow indicates fMRI hyperconnectivity (Cohen’s d≥0.8). d) Distribution of autistic participants included in the hypo- or hyperconnectivity subtypes in the aggregated (discovery plus replication) autism dataset; Blue: hypoconnectivity subtype, n=74, across n=28 data collections; red: hyperconnectivity, n=162 across n=38 data collections), as well as those not subtyped (light gray n=704, across all 38 collections). e) Connectograms showing atypical fMRI network structure in hypo- (left) and hyperconnectivity (right) subtypes (upon regression of mean fMRI connectivity across 414 parcellation units) ^,. Link thickness is proportional to the number of between-network edges displaying a significant difference in ASD vs. NT (red: increased, blue: decreased connectivity; t > 3.1, NBS corrected at p < 0.05). f) Radar plot showing the percentage of overlap (range: 0–50%) between subtype mean regressed connectivity difference maps and 12 neuro-cognitive ontology probability maps . g) Autism subdomain severity scores (top subpanel: social affect (SA); bottom subpanel: restricted repetitive behaviors (RRB)) based on Autism Diagnostic Observation Schedule (ADOS; see Supplementary Methods) for each subtype. SA: hypoconnectivity subtype n=33, mean=6.1±2.2; hyperconnectivity, n=84, mean=7.0±1.8; t₍₁₁₅₎=2.37, p_uncorr=0.019, p_(FDR)=0.030. RRB: hypoconnectivity subtype n=34, mean=6.9±2.8; hyperconnectivity, n=85, mean=7.5±2.2; t₍₁₁₇₎=1.20, p_uncorr=0.23, p_(FDR)=0.23. Error bars: SEM; *p < 0.05; ns: non-significant; CSS: calibrated severity score. DMN, default mode network; Dorsal Atten, dorsal attentional network; Limbic, limbic network; Salience, salience network; SomatMot, somatomotor network; TempPar, temporoparietal network; Visual, visual network. NBS, network-based statistics.

Figure 5.. Hypo- and hyperconnectivity subtypes recapitulate synaptic and immune pathways modeled in mice.

a) Venn diagrams showing enrichment between brain-decoded genes (i.e., genes spatially correlated with dysconnectivity patterns) and differentially expressed genes in autism (Gandal et al., 2022) for both subtypes. Blue/red areas report the number (N) of brain-decoded genes for each subtype. Grey: differentially expressed genes; overlap (i.e. enriched genes) is reported with corresponding odds ratios. ***p(FDR) < 0.001. b) Heatmap of enrichments between brain-decoded genes and autism-dysregulated pathways or c) modules of genes differentially expressed in autism . Left columns: hypoconnectivity (blue); right: hyperconnectivity (red). Color intensity indicates enrichment significance (−log p-value). Thick borders mark significant enrichments (q(FDR) < 0.05). d) Left: Scatterplot representation of odds ratio (OR) of the molecular ontologies associated with mouse hypoconnectivity (x-axis), versus OR of the same molecular ontologies decoded in autism hypoconnectivity map (y-axis). Right: the same plot is also reported for ORs of genes differentially expressed in autism . e) Left: Scatterplot representation of odds ratio (OR) of the molecular ontologies of mouse hyperconnectivity (x-axis), versus OR of the same molecular ontologies decoded in autism hyperconnectivity map (y-axis). Right: the same plot is also reported for ORs of genes differentially expressed in autism . Pathways significantly enriched at q(FDR) < 0.05 in both mouse models and autism are highlighted with blue (hypoconnectivity) or red (hyperconnectivity) shading. We report the list of brain decoded genes and the pathways we probed in Supplementary Table 3.