A study published in Nature Neuroscience has split autism into two distinct biological types based on how the brain wires its connections, drawing on resting-state fMRI scans from nearly 1,000 autistic participants and cross-species validation with 20 mouse models. One subtype features hypoconnectivity tied to synaptic gene pathways; the other shows hyperconnectivity linked to immune-related biological enrichment. Roughly 25% of participants fell cleanly into one of the two groups, a finding that challenges the longstanding practice of treating autism as a single diagnostic spectrum.
Why two biological subtypes change the treatment equation
Autism diagnoses currently rely on behavioral observation, not biology. Clinicians assess social communication, restricted interests, and repetitive behaviors, then assign a single label that covers an enormous range of severity and underlying causes. That approach has left drug development largely stuck: interventions that help some individuals fail others, and clinical trials routinely produce mixed results because they pool biologically different people into the same group.
The new subtyping framework offers a testable alternative. If hypoconnectivity tracks with synaptic pathway disruptions while hyperconnectivity tracks with immune pathway disruptions, then matching a person’s wiring profile to a targeted therapy could sharpen treatment selection. A reasonable prediction follows: people assigned to the immune-linked hyperconnectivity subtype should respond better to anti-inflammatory interventions than to compounds targeting synaptic signaling, and the opposite should hold for the synaptic hypoconnectivity subtype. No trial has tested that prediction yet, but the biological separation reported in the study provides the first mechanistic rationale for designing one.
In principle, subtype-based stratification could also help explain why some past autism drug trials have produced modest average benefits but large individual differences. If participants with immune-driven hyperconnectivity and those with synaptic hypoconnectivity were mixed together, any treatment that worked well for only one group would look weak overall. Reanalyzing existing trial data through the lens of connectivity profiles, if archived brain scans are available, could be one of the fastest ways to test whether the new framework has practical value.
How fMRI scans and mouse models produced the split
The research team analyzed resting-state functional MRI data aggregated through the Autism Brain Imaging Data Exchange, a large multisite dataset that pools brain scans and phenotypic information from laboratories worldwide. The ABIDE II release alone provides a roughly 1,000-person rsfMRI sample, giving the researchers enough statistical power to detect reproducible connectivity patterns rather than one-off signals.
From those scans, the team identified two groups whose brain networks diverged in opposite directions. One group showed weaker-than-typical connections between regions, a pattern called hypoconnectivity. The other showed stronger-than-typical connections, or hyperconnectivity. When the researchers mapped those connectivity profiles onto gene expression databases, the hypoconnectivity subtype aligned with enrichment in synaptic-related biological pathways, while the hyperconnectivity subtype aligned with immune-related pathway enrichments. The two profiles were not simply opposite ends of a single dial; they pointed to different cellular and molecular mechanisms.
The cross-species arm of the study reinforced that conclusion. Researchers examined 20 mouse models carrying different autism-associated genetic mutations and found that the same two connectivity patterns appeared in the animals. That replication across species strengthens the case that the subtypes reflect real biological divisions rather than artifacts of scanner noise or demographic variation in the human sample. Still, only about 25% of participants mapped neatly onto one subtype or the other, leaving the majority in an intermediate or mixed zone that the current framework does not yet classify.
Methodologically, the work also underscores the value of large, harmonized imaging datasets. Resting-state fMRI is notoriously sensitive to motion, scanner differences, and preprocessing choices. By drawing on a broad, multisite resource and applying consistent analytic pipelines, the authors could show that the hypoconnective and hyperconnective signatures persisted across different cohorts. That robustness will be essential if other groups are to reproduce the findings and refine the boundaries of each subtype.
Genetic heterogeneity explains why a single spectrum never fit
Separate genomic work has already shown that autism’s genetic architecture is far from uniform. Large-scale exome sequencing has implicated both developmental and functional changes in autism neurobiology, identifying dozens of high-confidence risk genes that affect different cell types and brain circuits. Whole-genome sequencing studies have expanded beyond protein-coding regions to estimate how noncoding and regulatory variants contribute to autism risk. Together, these findings paint a picture of a condition driven by many different genetic entry points, which helps explain why a single behavioral label has never captured the underlying biology.
The connectivity subtypes reported in the Nature Neuroscience paper sit on top of that genetic diversity. Synaptic gene disruptions and immune pathway disruptions represent two broad categories of biological consequence, each potentially fed by different sets of risk variants. Sorting people by the downstream wiring effect, rather than by any single gene, could offer a practical way to group individuals for clinical purposes even when their exact mutations differ. In that sense, the hypoconnective and hyperconnective patterns act as convergent endpoints for many heterogeneous genetic and environmental influences.
At the same time, the partial classification rate serves as a reminder that two subtypes will not be enough to explain every case. Some individuals may show mixtures of synaptic and immune signatures in different brain networks, while others may follow entirely different biological routes that do not strongly affect resting-state connectivity at all. Future work that combines genomics, imaging, and detailed clinical phenotyping may reveal a more nuanced map in which the current subtypes are two prominent peaks within a broader landscape.
Open questions before subtype-guided treatment trials
Several gaps stand between the current findings and clinical use. The study does not report longitudinal outcome data, so there is no direct evidence yet that subtype assignment predicts who benefits from which therapy. The 25% classification rate means three-quarters of autistic individuals remain unclassified under this scheme, and it is unclear whether they represent additional subtypes, transitional biology, or simply noisier data. The mouse model results, while consistent with the human findings, lack published detail on exact connectivity metrics or pathway enrichment scores beyond summary statements.
Phenotypic and comorbidity data from the ABIDE datasets have not been publicly linked to the two subtypes, so researchers cannot yet say whether one subtype correlates with, for example, higher rates of epilepsy, gastrointestinal issues, or intellectual disability. And the whole-genome sequencing literature has not broken down which specific noncoding variants feed into each connectivity profile, leaving a gap between broad pathway labels and precise molecular mechanisms. Bridging that gap will require new studies that collect imaging, genomic, and clinical data in the same individuals, ideally across development.
There are also practical hurdles. Resting-state fMRI is expensive, time-consuming, and not yet standardized for routine clinical use. Motion artifacts pose particular challenges for autistic children, who are often the focus of early-intervention efforts. Developing shorter, more tolerable scan protocols and validating less costly proxies for connectivity patterns-such as EEG-based measures or blood biomarkers tied to synaptic or immune activity-will be crucial if subtype-guided treatment is ever to reach typical clinics.
Ethical considerations will need equal attention. Labeling a person as belonging to an “immune” or “synaptic” subtype could influence how families, clinicians, and insurers think about prognosis and support, even before strong evidence links those labels to outcomes. Researchers and advocates will have to ensure that biological subtyping complements, rather than replaces, attention to individual needs and preferences. For now, the most responsible stance is to treat the new framework as a research tool that sharpens hypotheses, not as a diagnostic or prognostic label.
Despite these caveats, the work marks a notable shift. Instead of asking whether autism as a whole is associated with over- or underconnectivity, the field can now ask which autistic brains show which pattern, why, and with what consequences. As more datasets integrate imaging with genetics and detailed clinical profiles, the hope is that connectivity-based subtypes will evolve from statistical clusters into clinically meaningful categories that guide targeted, evidence-based care.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
