Autism Is Not One Thing — And Now We Can See It in the Brain

TL;DR

• Researchers at the Italian Institute of Technology and the Child Mind Institute in New York analysed brain connectivity in 20 mouse models of autism and in autistic children.
• They identified two distinct subtypes: one characterised by reduced connectivity between brain regions (hypoconnectivity), the other by increased connectivity (hyperconnectivity).
• The hypoconnected subtype was linked to synaptic pathways — the machinery neurons use to communicate.
• The hyperconnected subtype was associated with immune-related pathways and gene regulation changes.
• Both subtypes were found in mice and humans, giving the findings cross-species validation.
• This is not a clinical diagnostic tool yet, but it provides a framework for precision psychiatry in autism.

What Happened

For years, brain imaging studies of autism have produced a mess of contradictory results. Some found reduced functional connectivity between brain regions. Others found increased connectivity. Still others found complex patterns that didn't fit either category. The field treated this variability as noise — a reproducibility problem, or something to be averaged away.

A team led by Alessandro Gozzi at the Italian Institute of Technology's Center for Neuroscience and Cognitive Systems decided to test a different hypothesis: what if the variability is the signal?

Their study, published 6 June in Nature Neuroscience, used resting-state fMRI to measure functional connectivity — the way different brain regions' activity fluctuates together — across 20 different mouse models of autism. Each model carries a different genetic or biological perturbation linked to autism. The question was simple: do these models cluster into distinct connectivity patterns, and if so, can those patterns be linked to specific biological mechanisms?

The answer was yes. Two patterns emerged. One group of mouse models showed broadly reduced connectivity between brain regions. Another showed broadly increased connectivity. When the researchers dug into the biology underlying each pattern, they found that the hypoconnected subtype was associated with synaptic pathways — alterations in how neurons form and maintain connections. The hyperconnected subtype was linked to immune-related pathways and changes in gene regulation.

Then came the critical step. Working with Adriana Di Martino and colleagues at the Child Mind Institute, Gozzi's team looked for the same patterns in brain scans from autistic and non-autistic children. They found them. The two subtypes — hypoconnected and hyperconnected — were present in humans, with the same biological associations.

What It Actually Means

This study changes how we should think about autism at a biological level.

The current diagnostic framework treats autism as a single spectrum — a collection of behavioural and developmental differences that vary in severity but share a common label. That framework has value for clinical practice and support access, but it has been a blunt instrument for research. Two children can receive the same diagnosis and have fundamentally different biology driving their experiences.

Gozzi's work provides the first cross-species evidence that at least some of autism's heterogeneity can be parsed into biologically meaningful subtypes. The hypoconnected subtype points to problems at the synapse — the physical gaps between neurons where signals are transmitted. The hyperconnected subtype points to immune and gene-regulation mechanisms that may produce a different form of circuit dysfunction entirely.

"Two individuals may both receive an autism diagnosis, and may even show overlapping behavioural features, but the brain and molecular mechanisms contributing to their condition could be quite different," Gozzi said. "That distinction matters if we want to move toward more precise and personalized interventions."

The Hype Check

This is not a diagnostic tool. Gozzi is explicit about that. You cannot walk into a clinic, get an fMRI, and receive a subtype classification that guides treatment. The study is foundational — it establishes that subtypes exist and that they are biologically meaningful. Translating that into clinical practice will require years of additional research, including studies that link the subtypes to specific cognitive, sensory, and behavioural profiles.

The study also identified two subtypes, but Gozzi believes there may be more. "If we can build richer and larger mouse imaging datasets, including more models and more biological perturbations, we may be able to partition this space more finely — perhaps identifying three, four, or more mechanistically distinct subtypes," he said.

The finding is a beginning, not an endpoint.

The Stakeholder Landscape

Autistic people and their families are the most important audience, and the one most likely to have complicated reactions. On one hand, the idea of biologically distinct subtypes could eventually lead to more personalised support — interventions matched to the underlying biology rather than the behavioural surface. On the other hand, the history of autism research includes repeated attempts to carve autistic people into "types" that sometimes served to exclude or stigmatise. The researchers' framing — that this is about understanding diversity, not ranking it — matters enormously.

Clinicians and diagnosticians gain a framework for thinking about why two patients with the same diagnosis can respond so differently to the same interventions. The hypoconnected and hyperconnected subtypes may eventually map onto different treatment responses.

Neuroscience researchers gain a methodological template. The cross-species approach — using mouse models as a "biological Rosetta Stone" to interpret human imaging data — is a powerful innovation that could be applied to other neurodevelopmental and psychiatric conditions.

Pharmaceutical researchers gain targets. If the hyperconnected subtype is immune-mediated, anti-inflammatory or immunomodulatory approaches might be more relevant for those individuals. If the hypoconnected subtype is synaptic, interventions targeting synaptic function might be more appropriate. This is speculative, but it is the kind of speculation that drives drug development.

Cross-Layer Implications

The study is a methodological breakthrough as much as a biological one. Cross-species fMRI — using mouse models to validate and interpret human brain imaging — has been talked about for years but rarely executed at this scale. The fact that Gozzi's team analysed 20 different mouse models and then confirmed the findings in human children gives the results a level of confidence that single-species studies cannot match.

There is also a quiet implication for how we think about psychiatric classification more broadly. If autism — a condition defined entirely by behaviour — can be split into biologically distinct subtypes, the same may be true for depression, anxiety, schizophrenia, and ADHD. The DSM's symptom-based categories may eventually give way to biology-based ones. This study is a small but significant step in that direction.

What This Means for Parents

If you are parenting an autistic child, here is what this study means — and what it does not:

What it means:

• The variability you see among autistic children — including your own child's unique profile of strengths and challenges — may reflect real biological differences, not just "severity" on a single dimension.
• The idea that different interventions might work better for different subtypes is now a testable hypothesis. It is not yet a clinical reality, but it is a direction of travel.
• The study reinforces that autism is not one thing. If a particular therapy or approach did not work for your child, that does not mean nothing will. It may mean the approach was not matched to the underlying biology.

What it does not mean:

• You cannot get your child's brain scanned to determine their subtype. This is research-stage science, not clinical practice.
• The subtypes are not "better" or "worse." They are different biological pathways to overlapping behavioural profiles.
• This study does not challenge the value of behavioural and developmental support. It suggests that in the future, those supports might be better targeted.

The Uncertainty Ledger

• Only two subtypes were identified. Gozzi and colleagues believe there are likely more. The current study is a proof of concept, not a complete taxonomy.
• The link to behaviour is not yet mapped. The team is planning studies that combine brain imaging with "deep phenotyping" — detailed data on cognition, sensory symptoms, development, and clinical history. Until those studies are done, we do not know how the subtypes relate to the real-world diversity of autistic experience.
• The mouse-to-human translation is strong but not perfect. Mouse models capture specific genetic and biological perturbations, but human autism is far more heterogeneous. The fact that the subtypes appeared in both species is encouraging, but replication in larger and more diverse human samples is essential.
• The physiological mechanisms are still being worked out. Gozzi's team is now investigating what the fMRI signals mean in terms of neuronal activity, circuit dynamics, and the balance between excitation and inhibition in the brain.
• Clinical application is years away. Even if the subtypes are validated and linked to behaviour, developing subtype-specific interventions will require clinical trials that do not yet exist.

Bottom Line

For years, brain imaging studies of autism produced results that seemed to contradict each other. This study shows that the contradictions were not noise — they were evidence that autism is more than one thing at the biological level. Two distinct subtypes, one synaptic and one immune-related, appear in both mice and humans. The finding does not change clinical practice today, but it changes the research agenda tomorrow. Instead of asking whether the autistic brain is "more connected" or "less connected," we can begin to ask which circuit-level subtype is present, what biology it reflects, and whether different subtypes respond differently to support. That is the difference between treating autism as a single condition and treating it as what it has always been: a spectrum, all the way down to the biology.

Sources:

• Pagani, M. et al. "Autism subtypes identified using cross-species functional connectivity analyses." Nature Neuroscience (2026). DOI: 10.1038/s41593-026-02287-z [Tier 1]
• Medical Xpress / Ingrid Fadelli. "Autism may have two distinct subtypes based on brain connectivity patterns." 6 June 2026. [Tier 2]
• Direct quotes from Alessandro Gozzi via Medical Xpress interview. [Tier 2]
• Child Mind Institute / Italian Institute of Technology (CNCS@UNITN) — institutional sources. [Tier 1]