Researchers have identified at least two biologically distinct subtypes of autism, each defined by opposite patterns of brain connectivity and tied to different molecular pathways. One subtype features widespread hypoconnectivity linked to synaptic gene enrichment, while the other shows hyperconnectivity associated with immune-related pathways. The findings, drawn from cross-species analyses and large multi-site imaging datasets, challenge the long-standing clinical practice of treating autism as a single behavioral spectrum and open the door to subtype-specific treatment strategies.
Two connectivity subtypes rewrite the autism diagnosis playbook
Current diagnostic criteria for autism spectrum disorder rely on behavioral observation, grouping individuals with vastly different neurobiological profiles under one label. That approach has frustrated drug development for decades because clinical trials pool participants whose brains may be wired in fundamentally opposing ways. A treatment that helps one group could, in theory, do nothing for or even harm another.
The new research, published in a recent journal report, changes that calculus. By analyzing resting-state functional MRI data, the study team found that one subtype is characterized by reduced connectivity across brain regions, with gene expression profiles enriched for synaptic signaling pathways. The other subtype shows the reverse: elevated connectivity tied to immune and microglial gene expression. These are not subtle statistical artifacts. They represent reproducible, opposing signatures that held up across independent cohorts.
The practical consequence is direct. If a person’s autism is driven primarily by synaptic overgrowth through the mTOR pathway, an mTOR inhibitor might reduce their hyperconnectivity and ease social difficulties. If another person’s autism stems from immune-mediated disruption, a compound that modulates microglial activity could be more effective. Testing this idea, where baseline connectivity predicts which individuals respond better to mTOR inhibition versus microglial-modulating drugs over an eight-week trial, is a logical next step that the subtype framework now makes possible. No such trial has been conducted yet, but the biological rationale is now far more specific than anything the field had before.
The work also reframes long-running debates about whether autism should be split into multiple conditions or kept as a single spectrum diagnosis. Rather than drawing boundaries based on surface behavior, the connectivity-based subtypes emerge from quantitative brain measures and are backed by molecular signatures. That provides a more objective foundation for stratifying participants in future research and, eventually, in clinical practice.
Cross-species data and federal repositories anchor the evidence
The strength of these findings rests on the datasets and methods behind them. The researchers drew on resting-state fMRI data shared through the Autism Brain Imaging Data Exchange II, a multi-site initiative that pools brain scans from research centers around the world. ABIDE II was specifically designed to accelerate connectome studies in autism by making raw imaging data freely available, which allows independent teams to replicate and challenge results using the same scans.
Data harmonization, the process of making scans collected on different machines and protocols comparable, relied on standards maintained by the National Database for Autism Research, a federal platform run by the U.S. Department of Health and Human Services. NDAR provides common data elements that let researchers merge heterogeneous datasets without introducing systematic bias from scanner differences or demographic imbalances. That infrastructure is essential for teasing apart true biological subtypes from site-specific quirks.
Separate mechanistic work adds biological plausibility. Studies using mouse models with mTOR-related synaptic pathology have shown that this specific pathway produces large-scale functional hyperconnectivity resembling one of the two human subtypes, according to research published in a communication on mTOR-driven networks. On the other side, research on the 16p11.2 microdeletion, a well-known autism-associated genetic variant, demonstrated impaired prefrontal functional connectivity in both mice and humans. That cross-species consistency matters because it links defined genetic lesions to specific connectivity phenotypes rather than relying on behavioral classification alone.
By integrating these strands-human imaging, animal genetics, and transcriptomic data-the subtype framework moves beyond correlation. The hypoconnected subtype aligns with disruptions in cortical circuits seen when key neurodevelopmental genes are perturbed, while the hyperconnected subtype echoes patterns observed when immune signaling and microglial pruning go awry. Although the causal chain in people remains to be fully mapped, the convergence across methods and species lowers the risk that the two subtypes are mere statistical flukes.
Gaps between brain scans and the clinic remain wide
Several questions stand between these imaging findings and any change in how autism is diagnosed or treated. No study has yet matched individual-level clinical symptom scores and treatment histories to each connectivity subtype within the same group of participants. Without that link, it is unclear whether the two subtypes predict different behavioral profiles, different responses to existing therapies, or different developmental trajectories from childhood through adulthood.
Longitudinal data are also missing. The current evidence captures brain connectivity at a single time point. Whether a person classified as hypoconnected at age eight remains in that subtype at age eighteen is unknown. If subtype membership shifts over development, treatment strategies built around baseline connectivity would need repeated reassessment. Developmental plasticity, hormonal changes, and environmental factors such as intensive behavioral therapy could all reshape connectivity in ways that blur or sharpen subtype distinctions over time.
The most pressing gap is experimental. No human trial has directly tested whether targeting immune versus synaptic pathways shifts connectivity in the predicted direction for each subtype. Animal models show that specific genetic lesions produce specific connectivity patterns, but translating that into a clinical protocol requires confirming that the same molecular logic holds in people receiving a drug rather than carrying a defined mutation. Regulators and ethics boards will also want reassurance that modulating brain-wide connectivity does not introduce new cognitive or emotional side effects.
Scanner and population confounds also deserve scrutiny. Multi-site datasets like ABIDE II are powerful because of their size, but differences in MRI hardware, acquisition parameters, and the demographic makeup of each site can create false patterns. Full correction for these confounds, with transparent reporting of subtype prevalence at each site, has not been documented in enough detail to rule out technical artifacts entirely. Future analyses will need to show that the hypoconnected and hyperconnected groups appear consistently across scanners, age bands, and sex distributions, rather than clustering in a handful of sites.
Another open issue is how many subtypes truly exist. The current work emphasizes a primary split into two opposing connectivity profiles, but finer-grained clustering might reveal additional groups nested within each category. For clinicians, however, more subtypes are not automatically better. The field will have to balance biological nuance with practical usability, ensuring that any proposed subtype scheme can be implemented with standard imaging protocols and interpreted in ways that meaningfully guide care.
From lab finding to stratified care
The next development to watch is whether any research group designs a prospective clinical trial that stratifies autistic participants by baseline connectivity subtype before randomization. In such a study, individuals could be assigned to interventions that either target synaptic signaling or modulate immune activity, with resting-state fMRI serving as both a selection tool and an outcome measure. If connectivity shifts in the expected direction and symptoms improve preferentially within the matched subtype, that would provide the strongest test yet of the framework’s clinical value.
Even short of drug trials, connectivity-informed stratification could reshape basic research. Studies of social cognition, sensory processing, or language in autism might yield clearer results if participants are grouped by brain subtype rather than by a single diagnostic label. That, in turn, could help explain why some behavioral interventions work well for certain autistic people but not others, and could support more personalized recommendations for families navigating therapy options.
For now, the two connectivity subtypes should be viewed as a promising but provisional map of autism’s biological diversity. They underscore that the same behavioral diagnosis can arise from very different neural and molecular routes, and they hint at a future in which autism research and care are organized around objective brain measures rather than one-size-fits-all categories. Turning that promise into practice will require careful replication, rigorous control of confounds, and, ultimately, trials that test whether tailoring treatment to connectivity patterns can improve real-world outcomes.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
