A cross-species study published in Nature Neuroscience has sorted brain connectivity patterns from 20 genetic mouse models of autism and resting-state brain scans from 940 people with autism into two distinct subtypes, one dominated by hypoconnectivity and the other by hyperconnectivity. The research drew on roughly 1,000 neurotypical controls for comparison and pulled human imaging data from 38 separate data collections. The finding challenges the long-standing practice of treating autism as a single neurological profile and raises a pointed question: if these two subtypes respond differently to targeted therapies, clinical trials that lump all autistic participants together may be missing real treatment effects.
Two connectivity subtypes across species and why they change the research agenda
For decades, researchers have struggled to reconcile the wide genetic diversity behind autism with any unified brain signature. The new study addresses that gap by applying the same resting-state functional MRI technique to both engineered mouse lines and a large human cohort. On the animal side, the team clustered fMRI connectivity alterations across genetic mouse models and found that the alterations fell into two groups: one in which connections between brain regions were weaker than typical, and another in which they were stronger. When the researchers turned to human data, they identified analogous subtypes using scans aggregated from 940 individuals with autism.
The practical consequence is immediate for anyone designing or enrolling in autism intervention research. If a sensory-integration therapy, for example, works by strengthening weak neural connections, it could help the hypoconnectivity subtype while doing little, or even proving counterproductive, for people whose brains already show excessive connectivity. Conversely, a treatment aimed at dampening overactive networks might benefit the hyperconnected subtype but risk blunting compensatory pathways in those with weaker connectivity.
A prospective pediatric trial that stratifies participants by subtype and tracks social-communication scores, adaptive behavior, and sensory outcomes over time would be the clearest test of whether these biological groupings predict different treatment responses. Such a design would also allow researchers to ask whether subtype membership shifts with development or remains stable across childhood and adolescence. No subtype-stratified clinical trial has been reported yet, but the framework now gives investigators a measurable way to move beyond one-size-fits-all protocols.
The subtyping approach could also reshape how basic science is organized. Instead of evaluating each autism-linked gene in isolation, labs can ask whether a new mouse model aligns more closely with the hypoconnected or hyperconnected pattern and then prioritize experiments that probe mechanisms shared across that cluster. This could streamline the path from gene discovery to circuit-level explanations, and eventually to targeted pharmacological or behavioral interventions.
How 38 data collections and earlier mouse mapping built the evidence base
The human side of the analysis rests on two major open-access repositories. The first, known as ABIDE I, is a consortium that aggregates and openly shares resting-state fMRI data along with phenotypic information. Its initial release contained more than a thousand scans with 539 individuals diagnosed with autism and 573 typical controls, ranging in age from 7 to 64. The second collection, ABIDE II, expanded the pool and brought the total number of contributing sites and datasets to dozens of imaging centers. By drawing from both repositories, the 2026 study assembled a human sample large enough to test whether the mouse-derived subtypes held up across different scanners, clinical sites, and demographic groups.
The mouse work also did not start from scratch. An earlier study mapped resting-state fMRI connectivity across 16 autism mouse models and reported heterogeneous, model-specific connectivity alterations that nonetheless clustered into a limited number of patterns. That effort supplied both the conceptual framework and the analytic methods that the newer paper extended to 20 genetic lines. Expanding from 16 to 20 models strengthened the case that the two-subtype pattern is not an artifact of any single gene mutation but instead reflects a broader organizational principle in how autism-linked genetic changes reshape brain wiring.
Bridging these animal and human datasets required careful alignment. Resting-state fMRI measures spontaneous fluctuations in blood oxygen levels across brain regions while a subject lies still, providing an indirect index of functional connectivity. In mice, the scans are performed under light anesthesia to minimize motion while preserving intrinsic network dynamics; in humans, participants simply rest with eyes open or closed. Despite these procedural differences, the statistical clustering methods applied to both species produced matching subtype structures, a convergence the authors treat as evidence that the subtypes reflect real biological variation rather than measurement noise or site-specific artifacts.
Another strength of the cross-species strategy is that it allows mechanistic hypotheses generated in mice to be tested against human data. If a particular neurotransmitter system or developmental window appears critical for the hypoconnected subtype in animal work, researchers can look for converging evidence in human genetics, pharmacology, or longitudinal imaging. Conversely, patterns emerging from the large human datasets can guide which mouse lines to prioritize for follow-up experiments, creating a feedback loop between clinical and preclinical research.
Gaps in the subtyping framework and what to watch next
Several questions remain open. The study has not released the full list of all 20 mouse genotypes alongside their individual connectivity matrices, so independent teams cannot yet replicate the clustering at the single-model level using only published data. Without that transparency, it is difficult to know how sensitive the subtype assignments are to analytic choices such as thresholding, motion correction, or the specific brain atlas used to define regions of interest.
The human cohort, while large, inherits the limitations of its source repositories. ABIDE I, for instance, spans ages 7 to 64, and the 2026 paper does not break out how many of the 940 participants fall within the pediatric range most relevant to early intervention research. ABIDE II adds additional sites and participants but also introduces more variability in diagnostic instruments and comorbidity assessments. Without per-participant demographic tables, it is difficult to assess whether the subtypes are equally stable across age groups, sexes, and IQ levels, or whether certain clinical profiles are overrepresented in one connectivity pattern.
Scanner harmonization is another concern. The 38 data collections feeding the human analysis came from different MRI machines at different institutions, each with its own acquisition parameters and preprocessing pipelines. Multi-site fMRI studies routinely apply statistical corrections for scanner differences, but the specific harmonization steps used for the final 940-person aggregate have not been detailed in public summaries of the paper. If residual site effects correlate with subtype labels, they could artificially inflate or obscure true biological differences.
There are also conceptual questions about how connectivity-based subtypes relate to the lived experience of autism. The current analysis focuses on resting-state networks, which capture baseline communication among brain regions but do not directly measure how individuals process language, social cues, or sensory input in real time. It remains unclear whether the hypoconnected and hyperconnected groups differ systematically in traits such as social motivation, repetitive behaviors, anxiety, or sensory sensitivities, or whether they instead represent different neural routes to similar outward behaviors.
For families and clinicians, the most concrete development to track is whether any research group launches a prospective trial that assigns participants to subtype-matched interventions. The two-subtype framework suggests that a therapy’s efficacy may depend not only on age and symptom profile but also on underlying network architecture. A future study might, for example, pair the hypoconnected subgroup with intensive social-skills training designed to strengthen specific circuits, while offering the hyperconnected subgroup interventions that emphasize regulation and reduction of overload, then compare outcomes across arms.
In the nearer term, imaging centers could begin exploring whether subtype labels can be estimated reliably at the individual level using clinically feasible scan times. If robust classifiers can be developed, subtype assignment might one day become part of research-grade assessment batteries, sitting alongside behavioral evaluations and genetic testing. That prospect is still distant, and significant work remains to validate the stability and clinical relevance of the two connectivity patterns. But by demonstrating that autism’s neural diversity can be organized into reproducible subtypes across species, the new study offers a roadmap for more targeted, biologically informed approaches to both research and care.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
