Background:Major depressive disorder (MDD) is heterogeneous in clinical presentation and treatment response. The COORDINATE-MDD consortium identified two magnetic resonance imaging (MRI)-derived neuroanatomical profiles: dimension 1 (D1), with relatively preserved gray and white matter, and dimension 2 (D2), showing widespread reductions aligned with immunometabolic profile. Profiles were associated with distinct responses to selective serotonin reuptake inhibitor (SSRI) antidepressant and placebo (PLA). In this study, we examined electrophysiological correlates of the neuroanatomical profiles and their relationship to treatment outcome. Methods:Baseline resting-state, eyes-closed electroencephalography (EEG) was acquired from 237 medication-free participants with MDD who were in a current depressive episode (155 women; mean age [SD] = 37.47 [13.36] years) from CAN-BIND (Canadian Biomarker Integration Network in Depression) (SSRI) and EMBARC (Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care) (SSRI or PLA). EEG features included spectral power, frontal alpha asymmetry (FAA), multiscale sample entropy, and intersite phase clustering. Effects of profile (D1 and D2) and clinical outcome (responder, nonresponder; defined as ≥50% symptom improvement) were examined with age, sex, and site as covariates. Results:No significant electrophysiological differences were observed after covariate adjustment. However, among participants who subsequently responded to treatment, D1 showed greater baseline alpha power in frontal and central regions and lower relative delta posteriorly compared with D2. In PLA-treated responders, D2 showed spectral slowing, elevated low-frequency power, reduced gamma, and coarse-scale entropy compared with D1. Baseline FAA was lower in responders than nonresponders, independent of the neuroanatomical profile. Conclusions:EEG differences between MRI-defined neuroanatomical profiles emerged in relation to clinical outcome. D1 was associated with electrophysiological patterns consistent with flexible, globally regulated cortical dynamics in SSRI responders, whereas D2 showed a distinct pattern in PLA responders, indicating partially separable neural mechanisms underlying pharmacological and PLA treatment effects.

Background Major depressive disorder (MDD) is heterogeneous in clinical presentation and treatment response. COORDINATE-MDD consortium identified two MRI-derived neuroanatomical profiles: Dimension 1 (D1), with relatively preserved grey and white matter, and Dimension 2 (D2), showing widespread reductions aligned with immuno-metabolic profile. Profiles were associated with distinct responses to SSRI antidepressant and placebo. This study examined electrophysiological correlates and relationship to treatment outcome. Methods Baseline resting-state, eyes-closed EEG was acquired from 237 medication-free MDD participants in current depressive episode (155 women; mean age 37.47 ± 13.36 years) from CAN-BIND (SSRI) and EMBARC (SSRI or placebo). EEG features included spectral power, frontal alpha asymmetry (FAA), multiscale sample entropy and inter-site phase clustering. Effects of profile (D1, D2) and clinical outcome (responder, non-responder; defined as ≥ 50% symptom improvement) were examined with age, sex and site as covariates. Results No significant electrophysiological differences were observed after covariate adjustment. However, among participants who subsequently responded to treatment, D1 showed greater baseline alpha power in frontal and central regions and lower relative delta posteriorly compared with D2. In placebo-treated responders, D2 showed spectral slowing, elevated low-frequency power, reduced gamma and coarse-scale entropy relative to D1. Baseline FAA was lower in responders than non-responders, independent of neuroanatomical profile. Conclusions EEG differences between MRI-defined neuroanatomical profiles emerge in relation to clinical outcome. D1 is associated with electrophysiological patterns consistent with flexible, globally regulated cortical dynamics in SSRI responders, whereas D2 shows distinct pattern in placebo responders, indicating partially separable neural mechanisms underlying pharmacological and placebo treatment effects. Plain Language Summary : Depression is a common condition, but people differ in their symptoms, underlying biology, and response to treatment. This makes it difficult to predict which treatments will be most effective for each individual.Previous research from the COORDINATE-MDD consortium identified two brain-based groups of depression using MRI scans. One group showed relatively preserved brain structure and better response to antidepressant medication, while the other showed more widespread changes and similar improvement with medication or placebo, suggesting different underlying mechanisms.In this study, we examined whether these groups also differ in brain activity measured using EEG before treatment. No clear differences were seen across all participants. However, among those who later improved, distinct patterns emerged: one group showed activity consistent with more flexible and coordinated brain function and greater response to medication, while the other showed slower activity and reduced complexity linked to placebo response.These findings suggest that different biological mechanisms underlie treatment response, supporting more personalised approaches to care.

Abstract Background Major depressive disorder (MDD) is clinically heterogeneous, hindering identification of reproducible biomarkers. Using a semi-supervised machine learning approach, HYDRA, we previously identified two neuroanatomical dimensions from structural MRI in medication-free MDD from COORDINATE-MDD consortium. These dimensions (D1, D2) showed differential responses to selective serotonin reuptake inhibitor (SSRI) antidepressants and placebo. External replication in UK Biobank linked D2, characterized by widespread subtle neuroanatomical reductions, to an immuno-metabolic profile. Here, we examined whether these dimensions are detectable early in the course of illness. Methods We applied the pre-trained model to structural MRI data from the multisite PRONIA cohort, comprising individuals with recent-onset depression (ROD; n = 377; mean age 25.8 years, SD 6.0; 51.3% female) and healthy controls (n = 267; mean age 25.5 years, SD 6.4; 61.0% female). Participants were assigned to clusters (C1, C2) corresponding to the previously identified dimensions (D1, D2). Clusters were compared on clinical symptom profiles, peripheral inflammatory markers, and in a subset (n = 107), proteomic ageing indices. Results Two neuroanatomical clusters were identified in PRONIA. C1 (n = 265) showed higher negative symptom severity and elevated interleukin-2 levels. C2 (n = 140) was associated with higher residual proteomic age. Overall depressive symptom severity did not differ significantly between clusters. Conclusions Neuroanatomical dimensions of MDD are reproducible and detectable at illness onset. Associations with negative symptom severity, inflammatory signalling, and proteomic ageing suggest these dimensions capture biologically meaningful heterogeneity early in depression. These findings support a biologically informed framework for stratified treatment approaches in MDD.

Intrinsic timescale is a commonly used measure of spontaneous neural dynamics that quantifies the temporal window of processing of neuronal populations. Intrinsic timescale displays a hierarchical cortical organization across multiple species and imaging modalities, with shorter timescales in sensorimotor cortex compared to association cortex. However, less is known about how intrinsic timescale evolves during human brain development and whether its cortical maturation patterns generalize to independent developmental samples. Here we estimate the intrinsic timescale in two independent datasets of youth (HCPD: n=565; HBN: n=729; age range 8-22 years) and investigate its neurodevelopmental patterns. We find that developmental changes in the intrinsic timescale follow a hierarchical pattern that recapitulates an axis spanning sensorimotor to association cortices (S-A axis). Our analysis of an independent healthy young adult dataset (HCPYA: n=973, age range 22-37 years) underscores the specificity of these developmental findings, suggesting that the intrinsic timescale develops along the S-A axis in youth and stabilizes in adulthood. Together, these results reveal convergence between major axes of cortical organization and development, highlighting intrinsic timescale as a principled marker of hierarchical brain maturation in youth.

Background:Major depressive disorder (MDD) is associated with altered brain structure and evidence of accelerated brain aging. However, previous studies have been limited by clinical samples with mixed medication status and multiple mood states, modest sample sizes, small percentage of MDD individuals older than 65 years of age, and/or reliance on summary-level data. Methods:Harmonized T1-weighted MRI from MDD (n = 645), all medication-free and in a current depressive episode, and matched healthy controls (n = 645), segmented into 145 regional volumes, from 11 sites in COORDINATE-MDD consortium. Brain age gap (BAG) was estimated using gradient boosting regression with nested cross-validation. Group differences in BAG (and age-corrected BAG [cBAG]) were examined across age strata. Regional contributions were evaluated using Shapley Additive exPlanations. Results:MDD was associated with significantly elevated cBAG compared with healthy controls (mean difference + 2.01 years). Age-stratified analyses showed no differences before mid-30s, with progressively larger gaps thereafter, reaching +6.85 years in MDD aged 55 and older. cBAG differed across neuroanatomical phenotypes associated with differential antidepressant response, cognitive impairment, increased adverse life events, increased self-harm and suicide attempts, and a pro-atherogenic metabolic profile. Key contributing regions included lateral and medial prefrontal regions, middle temporal gyrus, putamen, supplementary motor cortex, central operculum, and cerebellum. Conclusions:Accelerated structural brain aging in MDD is age-dependent and is most pronounced in a neuroanatomical phenotype associated with worse key clinical outcomes. The findings support neuroprogression models of MDD while demonstrating that cBAG is not a uniform feature of MDD and seem to be more strongly expressed in a specifically clinically vulnerable disease phenotype.

Spermatogonial stem cells (SSCs) hold great promise for treating male infertility, but their clinical translation is impeded by the lack of optimal conditions to maintain their undifferentiated state in vitro. In this study, we focused on epigenetic regulators upregulated during differentiation as potential targets. Through a small-molecule screen targeting such conserved regulators, we found that PRMT5 inhibition suppressed mouse SSC differentiation and enhanced their proliferation in a GDNF-deficient, differentiation-prone microenvironment in vitro. Using SSC transplantation assays, we confirmed that EPZ015666-treated SSCs retained their spermatogonial identity. This differentiation-inhibitory effect was reversible upon EPZ015666 withdrawal, allowing restoration of normal spermatogenesis. Notably, EPZ015666 also inhibited differentiation and promoted the proliferation of human and non-human primate spermatogonia in vitro. Mechanistically, EPZ015666 exerted this effect by inhibiting the enzymatic active site of PRMT5. These findings suggest that PRMT5 inhibition could provide a novel strategy for culturing human SSCs in vitro.

Medical vision-language pretraining (VLP) from paired CT images and radiology reports enables scalable representation learning, but most existing methods align either whole scans with entire reports or local image regions with text fragments. These formulations underuse a key property of radiology reports: findings are organized around anatomical structures, with abnormalities described by organs, disease concepts, locations, and severity-related attributes. We propose OKA-CT, an organ-hierarchical knowledge-augmented framework for CT-report VLP. OKA-CT first converts free-text reports into organ-conditioned knowledge using radiology report parsing and LLM-assisted semantic structuring. The extracted hierarchy is used across two learning stages. Stage 1 injects anatomy-grounded evidence into the CT visual representation through fine-grained organ-conditioned supervision, while Stage 2 uses organ-specific report evidence to guide structured report-CT contrastive learning, where hierarchy-derived semantic soft targets treat non-paired cases with shared organ-level findings as weak semantic positives rather than uniform negatives. A lightweight query-based global branch further aggregates disease-relevant volumetric evidence for whole-scan representation. On CT-RATE and RAD-ChestCT datasets, OKA-CT achieves zero-shot abnormality diagnosis AUROCs of 84.9 and 72.2, outperforming prior CT VLP baselines. Retrieval and patch-occlusion analyses further show improved report-image alignment and stronger sensitivity to disease-associated anatomical regions.

Cynomolgus monkey blastoids are promising models for investigating early primate embryogenesis, yet previous generation systems have suffered from low efficiency and the presence of undefined cell clusters, limiting their broader application due to high costs, time consumption, and ethical concerns. In this study, we established an optimized protocol for efficiently inducing cynomolgus monkey blastoids by systematically modifying key parameters based on established human blastoid protocols. The resulting cyBlastoids consistently exhibited an induction efficiency exceeding 70%, with morphology and cell lineage allocation closely resembling those of natural blastocysts. Immunofluorescence staining and transcriptomic analyses revealed that the ICM-like and TE-like cells within cyBlastoids shared high molecular similarity to their counterparts in natural embryos and previously reported cynomolgus monkey blastoids. Extended in vitro culture further demonstrated that cyBlastoids displayed typical morphological and lineage characteristics of post-implantation embryos. Our stable and efficient protocol provides a robust alternative for primate embryo research, facilitating reproducible studies on early embryonic development in non-human primates.

Brain age prediction has been widely utilized to assess functional connectivity (FC) development, but conventional global brain age indices are limited in capturing spatial heterogeneity across the cortex. This study introduces a regional brain development index to characterize fine-grained FC maturation across cortical regions. We examined its spatial variability and stratified individuals into subtypes with distinct region-wise FC developmental patterns. Using data from the Philadelphia Neurodevelopmental Cohort (ages 8-23 years), we identified three distinct subtypes and found that individuals with advanced FC developmental pattern aligning with the sensorimotor-association axis exhibited superior cognitive performance. Robustness was confirmed through replication in the Human Connectome Project Development cohort. Further analyses revealed associations between FC development and gene expression linked to neural differentiation, synaptogenesis, and myelination. These findings suggest that spatial heterogeneity in FC development reflects cortical microstructure and hierarchical organization, underscoring its critical role in neurocognitive maturation during youth. This study applies machine learning to fMRI data to map developmental variations in functional connectivity, uncovering heterogeneity across individuals and cortical regions that predicts neurocognitive maturation in youth.

Human embryos undergo pivotal morphogenetic remodelling shortly after implantation. The understanding of this crucial stage is severely impeded by the scarcity of embryonic samples and ethical constraints. Pluripotent stem cells with the competence for somatic and germline differentiation serve as in vitro models of epiblast. In this study, we established human formative pluripotent stem cell-like cells (hfPSC-LCs) from naïve human embryonic stem cells (hESCs), conventional hESCs, human induced pluripotent stem cells (hiPSCs), as well as human blastocysts using the three-dimensional (3D) Matrigel culture system. Similar to pre-gastrula stage epiblast, hfPSC-LCs self-organise into self-renewing colonies with an apical lumen and exhibit several hallmarks of formative pluripotency, consistent with the properties observed in mouse fPSCs. Notably, single cells of hfPSC-LCs could differentiate into amnion-like precursor cells (hALPCs) which are transcriptionally and morphologically similar to the bona fide amnion. Meanwhile, hfPSC-LCs directly respond to primordial germ cell (PGC) induction signals, generating PGC-like cells (PGCLCs) either as single-cell aggregates or intact colonies, with an efficiency of approximately 50%. Chromatin accessibility analysis revealed that the differentiation capacity of hfPSC-LCs for gametes and amnion lineages might correlate with the accessible chromatin architecture of PGC and amnion associated genes. Loss of 3D-Matrigel niche disrupts formative pluripotency in both mouse and human, manifesting as downregulated formative markers and compromised differentiation capacity. Collectively, our findings establish hfPSC-LCs as a 3D model for investigating formative pluripotency of humans, thereby probably addressing a critical gap in the understanding of human pluripotency transitions.

ABSTRACT The next frontier in cognitive neuromodulation is defined by personalized and adaptive protocols, necessitating approaches tailored to individual functional neuroanatomy and brain‐state fluctuations. Here, we introduce an adaptive neuromodulation framework that integrates individualized network targeting with real‐time decoding of brain states to precisely target working memory functional networks. Using concurrent transcranial magnetic stimulation (TMS) and functional magnetic resonance imaging (fMRI), we first mapped participant‐specific networks and identified personalized targets. A real‐time decoder then tracked stimulation‐evoked neural dynamics to empirically determine the optimal frequency (i.e., the best‐performing within a tested set of 5, 10, and 20 Hz) and a corresponding suboptimal frequency for each individual. In a multi‐session crossover study, only the optimal‐frequency stimulation significantly improved working memory, with the decoder's output predicting behavioral gains. A key finding is the substantial inter‐individual variability in the optimal frequency, providing evidence against the notion of a universal “best” frequency. Our results demonstrate that cognitive enhancement is governed by the precise interaction between stimulation target and frequency. This work provides a causal demonstration of personalized, network‐based neuromodulation and offers proof of concept for a generalizable, biomarker‐driven framework, representing a step toward advancing cognitive therapeutics. Trial Registration: This study is registered at ClinicalTrials.gov (identifier: NCT04402294).

Deep learning models for neuroimaging have largely been developed for individual tasks, limiting knowledge transfer across applications. Here we introduce GenFAR, a modular deep learning framework that learns general, clinically informed features from brain MRIs. We trained this modular architecture on 49,246 individuals across 11 cohorts, using 17 diverse classification and regression tasks spanning cognition, clinical, diagnosis, demographics, and biomarkers. This yields aggregated, focused feature sets that capture rich, clinically- and biologically-relevant brain representations. We developed a sequential learning approach where tasks progressively build on previously learned representations. Through an analysis of 5,000 task sequences, we identified an optimal sequence length of six tasks and introduced a Donor Score metric to quantify each task's contribution to downstream performance. This analysis revealed five consistently strong donor tasks (Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia) that formed the base of our sequential model. We demonstrated the utility of our learned representation, in various tasks beyond those included in the training set, to serve as the foundation for specialized secondary predictors. We further showed that using the learned feature representation can substantially increase the sample efficiency of secondary deep learning training tasks and models, as well as improve their accuracy.

Purpose:Interpretability is highly desirable for oncologic outcome prediction, as it increases the level of transparency and trustworthiness of the model. This model characteristic is particularly relevant in the setting of modest sample size. Existing work has focused on Shapley Additive Explanations to provide post-hoc explanations on black-box models. These models are not intrinsically interpretable. In this study, we investigated the applicability of an intrinsically interpretable glass-box model, Explainable Boosting Machine (EBM), for hypothesis generation from an early-stage lung cancer data set. Methods and Materials:We applied EBM to a stripped data set on postradiation therapy lung cancer recurrence, aiming to extract as much information as possible by using pristine EBM configurations. We compared the key features ranked by EBM with those identified through univariate statistical analysis. Additionally, we benchmarked its performance against logistic regression and random forest models, while also evaluating the hypotheses generated by EBM. This study was approved by an institutional review board at the University of Pennsylvania. Results:EBM identified primary tumor size and body mass index as the most prognostic features, aligning with the results of the univariate analysis. Its interpretability provides safeguards against misinterpretation; the model revealed potential age-related bias in this single-arm data set and possible confounding interactions between race and body mass index. EBM yielded competitive performances and more interpretable insights compared with logistic regression and random forest but was not immune from generalizability challenges arising from limited data. Conclusions:The modest performance prevents EBM from being used as a clinical decision support tool, when applied to limited data. However, its interpretable, glass-box nature makes it useful for hypothesis generation.

Interpreting brain-behavior relationships through the lens of anatomical parcellations or functional networks is commonplace in human brain mapping. However, statistical approaches for testing whether brain-behavior associations are stronger (i.e., enriched) within a region of interest remain underdeveloped. Here, we propose a permutation-based approach for network enrichment testing using ordinal dominance curves (NETDOM). In simulation studies, we demonstrate that NETDOM properly controls the type I error rate-unlike other prominent enrichment methods-while exhibiting increased statistical power when enrichment occurs in a subset of in-network locations. Using data from two large-scale neurodevelopmental cohorts, we illustrate that NETDOM effectively detects enriched associations between structural and functional brain measures and neurocognitive performance.

Neurodegenerative diseases exhibit substantial heterogeneity, complicating both diagnosis and prognosis. Identifying clinically meaningful subtypes is crucial for understanding disease mechanisms and can also improve diagnostic precision and prognostic accuracy. Existing subtyping approaches primarily rely on unsupervised learning of patient data for capturing inter-individual variability, often failing to uncover subtypes that are informative for diagnosis or prognosis. To address this limitation, we propose a novel mixture-of-experts (MoE) framework that integrates predictive modeling with subtype identification. Unlike traditional subtyping methods, our approach learns a router to assign individuals to specialized expert networks, each corresponding to a distinct subtype, to improve predictive accuracy. This MoE framework ensures that the discovered subtypes are not only statistically distinct but also clinically informative. We evaluate the framework on a real-world dataset of mild cognitive impairment (MCI) subjects and a semi-simulated dataset, demonstrating superior performance for predicting MCI subjects progression to Alzheimers disease while identifying distinct clinically meaningful MCI subtypes. Code is available at https://github.com/Kateridge/MoESubtyping

Characterizing brain dynamic functional connectivity (dFC) patterns from functional Magnetic Resonance Imaging (fMRI) data is of paramount importance in imaging neuroscience and medicine. Recently, graph neural network (GNN) models, combined with transformers or recurrent neural networks (RNNs), have shown great potential for modeling the dFC patterns. However, these methods face challenges in characterizing the modularity organization of brain networks and capturing varying dFC state patterns. To address these limitations, we propose dFCExpert, a novel method designed to learn robust representations of dFC patterns from fMRI data with modularity experts and state experts. Specifically, the modularity experts optimize multiple experts to characterize the brain modularity organization during graph feature learning process by combining GNN and mixture of experts (MoE), with each expert focusing on brain network nodes within the same functional network module. The state experts aggregate temporal dFC features into a set of distinct connectivity states using a soft prototype clustering method, providing insight into how these states support diverse brain functions and vary across brain conditions. Experiments on three large-scale fMRI datasets have demonstrated the superiority of our method over existing alternatives. The learned dFC representations not only enhance interpretability but also hold promise for advancing our understanding of brain function across a range of conditions, including brain development, sex differences, and Autism Spectrum Disorder. Our implementation is publicly available at https://github.com/MLDataAnalytics/dFCExperts.

Addiction is a chronically relapsing disease characterized by drug intoxication, craving, bingeing, and withdrawal with loss of control. An expanding body of literature has suggested that non-substance addictions, such as internet addiction and pathological gambling, exhibit shared clinical, phenomenological, and biological features with substance-based addictions. With the development of imaging technology in the past three decades, neuroimaging studies have yielded critical insights into the neurobiological implications of substance and non-substance addictions, revealing neurochemical and functional alterations inherent in the brains of individuals afflicted by these diverse forms of addiction. Imaging techniques have increasingly taken the forefront in elucidating the neuronal underpinnings of addiction as they are poised to guide future research toward developing therapeutic interventions for addiction, particularly for non-substance addiction, which constitutes a growing proportion of addiction disorders. However, the current literature lacks an overview of comparing different types of addictions. This review is set to explore the differences and similarities in neural correlates underlying substance and non-substance addiction based on neuroimaging studies. Specifically, the objectives of this review are (1) to summarize structural brain changes in substance and non-substance addiction and (2) to provide a focused review of commonalities and differences in neural correlates of reward processing, cue-reactivity, and inhibitory control in individuals with substance and non-substance addiction.

We present a preliminary analysis of a GAN-based normative modeling technique for capturing individual-level deviations in brain measures, addressing heterogeneity in neurological disorders. By leveraging self-supervised training on pseudo-synthetically simulated patient data, our method detects disease-related effects without the need for large, disease-specific datasets. We demonstrate the versatility of this approach by applying it to structural MRI and resting-state fMRI data, identifying neuroanatomical and functional connectivity deviations in Alzheimer's disease (AD) and Traumatic Brain Injury (TBI). This model's ability to accurately capture disease-related abnormalities in brain measures highlights its potential as a powerful tool for personalized diagnosis and the study of brain disorders, opening new avenues for research.