Multiscale neural gradients reflect transdiagnostic effects of major psychiatric conditions on cortical morphology

The study addresses whether common psychiatric conditions share a cortex-wide pattern of morphological alterations and how these transdiagnostic effects relate to multiscale brain organization. Motivated by high comorbidity and shared risk factors across disorders, prior ENIGMA studies have shown widespread cortical thickness changes within individual diagnoses. The authors hypothesize that a shared dimension of cortical morphology alterations exists across autism spectrum disorder (ASD), attention deficit/hyperactivity disorder (ADHD), major depressive disorder (MDD), obsessive-compulsive disorder (OCD), bipolar disorder (BD), and schizophrenia (SCZ), and that this dimension aligns with sensory-fugal gradients of cortical microstructure/cytoarchitecture, intrinsic functional connectivity, and neurotransmitter receptor/transporter distributions. The purpose is to contextualize transdiagnostic morphological effects across multiple neural scales to gain mechanistic insight into regional vulnerability in psychiatric conditions.

Prior work has demonstrated reproducible structural brain alterations in psychiatric disorders, often using multisite meta-analytic frameworks like ENIGMA. Transdiagnostic analyses indicated shared anatomical compromises and linked them to spatial patterns of gene expression (e.g., CA1 pyramidal genes implicated in cortical thickness regulation). The structural model posits that regional laminar differentiation and connectivity influence plasticity and disease susceptibility, with paralimbic cortices being particularly vulnerable. Data-driven gradients from in vivo MRI and postmortem histology reveal a sensory-fugal axis differentiating highly myelinated, well-laminated sensory/motor cortices from less differentiated, sparsely myelinated heteromodal/paralimbic regions; functional gradients diverge partly, emphasizing heteromodal networks. Neurotransmitter mapping (PET/SPECT) has cataloged cortex-wide distributions of dopaminergic and serotonergic systems, both implicated in psychiatric pathophysiology (e.g., dopamine in psychosis and ADHD; serotonin in mood disorders). These literatures motivate aligning shared structural effects with gradients of microstructure/function and neurotransmitter topographies.

- Data sources: Meta-analytic cortical thickness case-control effect size maps from six ENIGMA Working Groups (ASD, ADHD, MDD, OCD, BD, SCZ), totaling 28,546 participants (12,876 patients; 15,670 controls) across 145 cohorts. ENIGMA standardized T1-weighted MRI processing (FreeSurfer) yielded thickness for 68 Desikan–Killiany regions; effects adjusted for age, age^2, sex, and site, combined via inverse variance-weighted random-effects models. - Shared dimension estimation: Principal component analysis (PCA) was applied to concatenated cross-disorder effect size maps to derive a shared disease dimension (first principal component). Robustness checks included mean effect across disorders and PCA on Cohen’s d maps from the ENIGMA Toolbox, spatial similarity to individual disorder maps, and leave-one-condition-out analyses. Spin-tests (n=1000) accounted for spatial autocorrelation; multiple comparisons controlled via FDR. - Multiscale contextualization: 1) In vivo gradients: From Human Connectome Project (HCP) data (n=207 unrelated adults), computed (a) microstructural gradient from T1w/T2w-derived intracortical myelin profile similarities, and (b) functional connectivity gradient from resting-state fMRI parcel-wise correlations, using diffusion map embedding (BrainSpace; α=0.5, t=0). Correlated the shared dimension with each gradient. 2) Postmortem cytoarchitecture: From BigBrain 3D histology, sampled intracortical intensity profiles across 18 equivolumetric surfaces and computed moments (mean, SD, skewness, kurtosis) and externopyramidization; correlated these with the shared dimension. Replicated moments/externopyramidization from in vivo T1w/T2w profiles (HCP) as sensitivity analysis. 3) Neurotransmitter systems: Mapped PET/SPECT-derived cortical distributions (JuSpace) for ten systems (FDOPA, GABAa, DAT, NAT, SERT, D1, D2, 5‑HT1a, 5‑HT1b, 5‑HT2a) to parcellation; correlated each with the shared dimension. Spin-tests (n=1000) with FDR across maps. - Additional analyses: Examined associations for other PCs and a surface area-based shared dimension (PCA on surface area Cohen’s d for five conditions), repeating multiscale correlations. - Predictive modeling: LASSO regression with five-fold nested cross-validation (100 repetitions) to predict the shared dimension from concatenated multiscale features (gradients, cytoarchitectural metrics, neurotransmitter maps). Evaluated performance by correlation (r), mean absolute error (MAE), and permutation tests (n=1000). Recorded feature selection frequencies. - Parcellation: Desikan–Killiany (68 regions); BigBrain lacked data for banks of STS, frontal pole, temporal pole (3 regions/hemisphere) in cytoarchitecture analyses.

- Shared dimension: The first principal component of cortical thickness alterations explained 55.7% of cross-disorder variance. It followed a sensory-to-paralimbic axis, with positive scores in sensory/motor regions and negative scores in transmodal/paralimbic cortices. Paralimbic regions showed the strongest mean atrophy across disorders, primary motor cortex the least. - Robustness: The shared component resembled PCA on ENIGMA Toolbox Cohen’s d maps (r=0.552, P_spin<0.001). Spatial similarity was strongest with SCZ and BD, followed by MDD, ADHD, ASD, and OCD (all P_spin-FDR<0.001). Leave-one-condition-out analyses yielded highly similar shared maps (r>0.9, P_spin-FDR<0.001). Shared maps modestly related to cortical expansion/reconfiguration (e.g., r=-0.256, P_spin=0.036 for the shared dimension). - Gradients: Significant negative association with the microstructural gradient (r=-0.400, P_spin-FDR=0.042), and a weaker trend with the principal functional gradient (r=-0.247, P_spin-FDR=0.090), indicating greater transdiagnostic susceptibility in less myelinated, less laminated paralimbic regions. - Cytoarchitecture: Significant positive associations of the shared dimension with BigBrain-derived laminar profile skewness (r=0.400, P_spin-FDR=0.015) and externopyramidization (r=0.472, P_spin-FDR=0.015), consistent with heightened alterations in less laminarly differentiated cortex. Results replicated using in vivo T1w/T2w-derived proxies. - Neurotransmitter systems: Significant correlations with PET/SPECT maps—D2 receptor (r=0.280, P_spin-FDR=0.035, positive), 5‑HT1b receptor (r=0.349, P_spin-FDR=0.025, positive), dopamine transporter DAT (r=-0.240, P_spin-FDR=0.041, negative), and 5‑HT1a receptor (r=-0.307, P_spin-FDR=0.033, negative)—implicating dopaminergic and serotonergic architectures in regional vulnerability. - Other shared dimensions: The second PC showed stronger, inverted associations with the functional gradient, suggesting a sensory–heteromodal axis distinct from the first sensory–paralimbic axis. A surface area-based shared dimension differed from thickness (r=-0.030, P_spin-FDR=0.446) and related to SERT (r=0.274, P_spin-FDR=0.029), 5‑HT2a (r=-0.297, P_spin-FDR=0.011), and GABAa (r=-0.325, P_spin-FDR=0.005). - Prediction: Multiscale features predicted the shared dimension reliably (r=0.518±0.044; MAE=0.828±0.039; P_perm<0.001). Most frequently selected predictors included cytoarchitectural skewness and externopyramidization, D2 and 5‑HT1b receptor maps, and the microstructural gradient.

The shared transdiagnostic cortical thickness alteration pattern aligns with a sensory–paralimbic organizational gradient, indicating that paralimbic regions with low laminar differentiation and reduced myelination are particularly susceptible across ASD, ADHD, MDD, OCD, BD, and SCZ. This supports the structural model linking cortical microarchitecture to plasticity and vulnerability. The stronger association with microstructural (vs. functional) gradients suggests that disease-related susceptibility is more tightly anchored to cytoarchitectural context than to placement along sensory–heteromodal functional hierarchies. Associations with dopaminergic (D2, DAT) and serotonergic (5‑HT1a, 5‑HT1b) distributions further implicate neurotransmitter systems as modulators of regional vulnerability. Supervised learning demonstrated that combinations of microstructural features and transmitter maps best explain the transdiagnostic pattern, underscoring multiscale determinants rather than a single factor. Overall, findings provide a unifying framework embedding shared psychiatric effects within cortical microstructure and molecular architectures, coherent with evidence for shared cellular and genetic risk factors across disorders.

This work identifies a robust, cortex-wide shared dimension of morphological alterations across six major psychiatric conditions that follows a sensory–paralimbic gradient, with paralimbic regions showing greatest susceptibility. By integrating in vivo microstructural and functional gradients, postmortem cytoarchitectonics, and neurotransmitter maps, the study demonstrates that transdiagnostic vulnerability is anchored in cortical microarchitecture and related to serotonergic/dopaminergic systems. Machine learning confirms the synergistic explanatory power of multiscale features. These insights may inform transdiagnostic neurobiological models and guide development of diagnostic and therapeutic strategies that cut across traditional nosological boundaries. Future research should incorporate subject-level multimodal measurements within the same cohorts, employ higher-resolution/functional parcellations, and rigorously address comorbidities and medication effects to refine mechanistic interpretations.

- Comorbidities and medication: Heterogeneous cohorts across sites limited adjustment for co-occurring diagnoses and treatment effects, which may influence shared patterns. - Cross-cohort multimodality: Microstructural, cytoarchitectural, and neurotransmitter features were derived from independent datasets, precluding direct subject-level mechanistic inferences. - Parcellation granularity: Analyses were constrained to the Desikan–Killiany atlas; limited areal specificity and alignment with functional topographies. Vertex-level or functionally defined parcellations could improve sensitivity. - Histology reference: BigBrain represents a single older donor, potentially limiting generalizability of cytoarchitectural associations. - PET/SPECT constraints: Tracer specificity, spatial resolution, and partial volume effects may affect neurotransmitter map accuracy; broader tracer panels and higher-resolution autoradiography would strengthen conclusions. - Modality differences: Surface area-based shared effects diverged from thickness-based patterns, indicating modality-specific sensitivities. Low mean effects in sensory/motor regions do not exclude functional relevance that may be undetected by current structural analyses.