Five Sensory Profiles, Five Wiring Patterns: Mapping the Mechanism of Sensory Reactivity in Autism
- Among 146 autistic adults and youth, sensory questionnaire scores sorted cleanly into five distinct sensory processing profiles rather than one uniform autistic sensory signature.
- Each profile mapped onto a measurably different pattern of brain network organisation, derived from resting functional connectivity across 232 cortical and subcortical regions.
- Machine learning told seven of the ten possible profile pairs apart using only graph measures of how efficiently brain regions exchanged information, at a strict statistical threshold.
- The somatomotor network, orbitofrontal and prefrontal cortex, posterior parietal cortex and subcortical regions carried the connectivity differences that predicted which sensory profile a person belonged to.
Sensory reactivity is one of the most pervasive yet most under-modelled features of autism, reported in up to 87 percent of autistic people and folded into the diagnostic criteria since 2013. Clinically it is also one of the most consequential: hypersensitivity drives avoidance, withdrawal and the exhausting work of getting through ordinary environments. Yet the field has long treated "sensory differences" as a single block, which obscures the fact that two autistic clients can have opposite sensory worlds. This study, drawn from the Province of Ontario Neurodevelopmental Disorders (POND) Network, takes a different stance: it asks whether sensory variation is organised, and whether that organisation has a neural signature.
The authors first used k-means clustering on Short Sensory Profile scores from 146 autistic participants and recovered five sensory phenotypes with distinct behavioural presentations. This in itself reframes the clinical picture: sensory reactivity is not a dial running from low to high, but a small set of qualitatively different configurations. The mechanistic question is whether these behavioural clusters correspond to anything in the brain, or whether they are merely an artefact of how questionnaires carve up self-report.
To answer that, the team built functional connectivity matrices from resting-state fMRI across 200 cortical and 32 subcortical regions, then computed graph-theoretical measures – betweenness centrality, strength, local efficiency, clustering coefficient – that index how information moves through a network rather than how active any single region is. Machine learning models trained on these measures distinguished 7 of the 10 phenotype pairs at p < .005. The discriminating signal concentrated in the somatomotor network, orbitofrontal and prefrontal cortex, posterior parietal cortex and subcortical structures – a circuit that spans raw sensory intake, integration and the executive weighting of what matters.
The reason this belongs to the mechanism conversation, not the diagnosis conversation, is that the model never tried to separate autistic from non-autistic brains. It separated autistic people from each other, by the texture of their sensory experience. That is a shift from categorical labelling toward what might be called sensory subtyping: an account of how the reactivity is generated, with implications for who will tolerate which environment and which accommodation.
For the practitioner, the takeaway is conceptual before it is practical. The familiar request to "reduce sensory load" assumes a shared sensory profile that this work suggests does not exist. If sensory reactivity comes in five flavours with five connectivity signatures, then sensory accommodations, occupational therapy goals and even the framing of a client's daily fatigue should be profile-specific. The data also caution against reading a client's sensory style off their autism diagnosis alone.
What "sensory phenotype" actually means here
A phenotype, in this study, is a reproducible cluster of sensory-questionnaire responses – patterns of seeking, avoiding, over- and under-responding – that recur across people. The novelty is grounding those behavioural clusters in network connectivity, moving the construct from a description of behaviour toward a candidate explanation of it.
Why connectivity, not activation
The analysis deliberately used graph measures of information exchange rather than where the brain "lights up". Sensory reactivity is increasingly understood as a problem of integration and gain control across distributed circuits, so a network-topology lens is better matched to the construct than a single-region activation map.
The model did not sort autistic brains from neurotypical ones – it sorted autistic people from one another by the wiring of their sensory worlds.
This is a cross-sectional, correlational design, so the connectivity differences cannot be read as causes of the sensory profiles. The sample mixes ages and is drawn from a single Ontario research network, and three of the ten phenotype pairs were not reliably separable, leaving the boundaries between adjacent profiles uncertain.