A team of researchers from three institutions has identified two biologically distinct subtypes of autism, each defined by opposite patterns of brain-wide communication. One subtype is marked by reduced connectivity across most brain regions, while the other shows unusually strong connections. The findings, drawn from mouse models and large human brain-imaging datasets, challenge the long-standing practice of diagnosing autism as a single behavioral condition and raise the possibility that children in each subtype may respond differently to the same therapies.
Why two connectivity subtypes change the clinical picture
Autism diagnoses today rely entirely on observed behavior. Clinicians assess social communication, repetitive actions, and sensory responses, then assign a single label. That approach groups together individuals whose brains may be wired in fundamentally different ways. The new research, published in the journal Nature Neuroscience, directly tests whether those differences can be sorted into reproducible biological categories. The answer, based on resting-state functional MRI data from both mice and humans, is that at least two such categories exist.
The practical stakes are immediate. If a child’s brain shows widespread hypoconnectivity, meaning reduced signaling between distant regions, the biological drivers differ from those of a child whose brain is hyperconnected. According to the study, the hypoconnectivity pattern is tied to disruptions in synaptic genes, pointing toward a distinct molecular mechanism that affects how neurons communicate across large-scale networks. Hyperconnectivity, by contrast, appears linked to other pathways that may involve altered regulation of long-range circuits, although these mechanisms are not yet characterized with the same precision.
That distinction matters for treatment design. A reasonable hypothesis, though not yet tested in clinical trials, is that children assigned to the hyperconnectivity subtype could show faster gains in social-communication skills after targeted cognitive training than those in the hypoconnectivity subtype, independent of overall autism severity. The logic is straightforward: if hyperconnected brains already have strong signal pathways, structured training may redirect that activity more efficiently than in brains where the signal infrastructure itself is diminished. Conversely, individuals with hypoconnectivity might need interventions that first bolster basic network strength, potentially through pharmacological strategies that enhance synaptic function, before higher-level training can be fully effective.
At the same time, the authors emphasize that no clinical outcome data are yet tied directly to these connectivity-defined subtypes. For now, the biological split offers a framework for designing future trials rather than a ready-made tool for clinicians. Any move toward subtype-guided treatment will require carefully controlled studies that assign participants based on brain connectivity and then track their responses to specific interventions over time.
Cross-species fMRI data and the ABIDE datasets
The research team, a collaboration between the Istituto Italiano di Tecnologia, the Child Mind Institute, and the University of Trento, built its case by working across species. Earlier mouse-model work mapped resting-state functional connectivity across numerous autism-related genetic models, revealing that different mutations produce diverging whole-brain signatures rather than a single uniform pattern. These experiments, conducted under tightly controlled conditions, allowed the investigators to see how particular genetic disruptions reshape long-range brain communication.
The next step was to test whether those mouse-derived signatures also appear in humans. To do this, the team turned to the Autism Brain Imaging Data Exchange, a large open-access repository that aggregates resting-state fMRI scans from multiple research sites. The first wave of data, known as ABIDE I, was released in 2012, and a second wave described in a 2017 data report substantially expanded the available pool of scans from individuals with autism and neurotypical controls. Together, these datasets provide one of the largest resources for studying intrinsic brain connectivity in autism.
By applying the connectivity patterns first observed in mice to these human datasets, the researchers confirmed two reproducible subtypes: one defined by predominantly whole-brain hypoconnectivity and the other by widespread hyperconnectivity. Individuals with autism in the ABIDE cohorts could be assigned to one of these patterns based on how strongly their resting-state networks resembled the mouse-derived templates. Importantly, both subtypes were present across independent cohorts, suggesting that the patterns are not artifacts of any single scanner, site, or analysis pipeline.
The cross-species design is what separates this work from earlier attempts to subtype autism using brain imaging alone. Previous human-only studies often struggled with small samples, inconsistent preprocessing methods, and limited power to detect robust patterns. Starting from controlled genetic mouse models, where the biological cause is known, and then matching those patterns to human data provides a stronger anchor for interpreting the connectivity differences. The authors describe in detail, via the full methods section, how they validated subtype assignments across independent human cohorts and tested the stability of their clustering procedures.
Gaps between subtype biology and treatment response
Several questions remain open. The most pressing is whether these two connectivity subtypes predict different responses to existing interventions. Behavioral therapies, speech and language programs, and pharmacological treatments are currently prescribed without reference to brain connectivity profiles. If the hypoconnectivity subtype, tied to synaptic gene disruption, responds better to interventions that strengthen neural signaling, while the hyperconnectivity subtype benefits more from approaches that refine or redirect excess connectivity, treatment selection could become far more precise. But no trial has yet tested that possibility, and the study does not provide outcome data linking subtype to therapy success.
Another unresolved issue is how these connectivity patterns evolve across development. Exact participant counts and age distributions from the aggregated human sample used in the Nature Neuroscience paper are not specified in the available summaries, leaving open whether the two subtypes are equally common in children, adolescents, and adults. Autism presents differently across age groups and sexes, and it is unclear whether the hypoconnectivity and hyperconnectivity patterns remain stable over time or whether individuals might shift between them as their brains mature.
Sex differences are another potential source of variation. Autism is more frequently diagnosed in males, but the underlying reasons remain debated. Without detailed breakdowns of how many males and females fall into each connectivity subtype, researchers cannot yet say whether one pattern is more prevalent in a particular sex or whether the biological mechanisms differ in subtle ways. Future analyses that stratify ABIDE and other large datasets by sex, age, and IQ could help clarify whether the same two subtypes capture the full range of variability.
There are also methodological questions. Resting-state fMRI, while powerful for mapping large-scale networks, is sensitive to motion, scanner differences, and preprocessing choices. The cross-cohort validation reported by the authors suggests that the hypoconnectivity and hyperconnectivity patterns are robust, but the field still lacks standardized pipelines that would make subtype assignments directly comparable across laboratories and clinics. Before these patterns can inform individual-level decisions, researchers will need to show that a given person’s subtype classification remains consistent across different scanning sessions and analysis methods.
Despite these gaps, the work represents a concrete step toward a biologically grounded taxonomy of autism. By demonstrating that at least two distinct patterns of brain-wide connectivity recur across species and datasets, the study moves beyond the idea of autism as a single, undifferentiated condition. Instead, it points to a future in which diagnoses might incorporate both behavioral assessments and objective measures of brain network organization.
For families and clinicians, that shift will not happen overnight. Translating connectivity subtypes into treatment guidelines will require new trials that stratify participants based on brain patterns, careful monitoring of outcomes, and replication across independent samples. Yet the framework introduced by this research offers a roadmap: identify reproducible biological signatures, link them to underlying molecular pathways, and then test whether they predict who benefits most from which intervention. If that roadmap holds, the discovery of hypoconnectivity and hyperconnectivity subtypes may mark the beginning of more personalized, biologically informed care for people on the autism spectrum.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
