When Two Brains Fail to Sync: A Neural Marker of Social Difficulty in Autistic Preschoolers
- In 54 children (28 autistic, 26 non-autistic, pre-school age), autistic children showed significantly lower behavioural synchrony with an adult partner across both intentional imitation tasks and a spontaneous hand-clapping task (University of Jinan, China; fNIRS hyperscanning).
- Interpersonal neural synchrony (INS) — the moment-to-moment coupling of brain activity between child and adult — was reduced in autistic dyads specifically in the right temporoparietal junction (r-TPJ), left inferior frontal gyrus (l-IFG) and inferior parietal lobule (IPL): the core social-cognitive circuit.
- Single-region activation told a weaker story than the *between-brain* coupling: during the spontaneous task there were no group differences in activation across all 20 fNIRS channels, yet INS still separated the groups.
- A GaussianNB classifier trained on task-related INS values discriminated autistic from non-autistic children with accuracy rising from 55.6% (delayed imitation) to 74.1% in the spontaneous clapping condition; SHAP-identified driver channels overlapped exactly with the statistically significant INS channels.
For decades we have described autistic social difficulty from the outside — eye contact, reciprocity, the texture of an interaction that does not quite land. This study moves the locus of measurement inside, to the live coupling of two brains during a real exchange. That matters because synchrony is not a static trait an individual carries; it is something that happens, or fails to happen, between people.
What the data shows
The design separates two kinds of synchrony that clinicians intuitively distinguish but rarely measure apart. Intentional synchrony — deliberately copying or matching another person — was probed with delayed and synchronous imitation. Spontaneous synchrony — the automatic entrainment of rhythmic hand-clapping — was probed without instruction to match. Autistic children underperformed on both, but the neural signature differed by condition. In the intentional imitation tasks, reduced INS appeared in l-IFG, IPL and r-TPJ; in the spontaneous task, the deficit concentrated in the IPL and r-TPJ even though no single brain region showed altered activation.
That dissociation is the clinically interesting part. Activation asks "is this region working?" Interpersonal synchrony asks "are these two people's regions working together?" The spontaneous condition produced normal-looking activation yet abnormal coupling — a reminder that a child can recruit the right machinery and still fail to lock onto a partner. The classifier results reinforce this: accuracy was highest (74.1%) precisely in the spontaneous, unscripted task, where social demand is least cued and most ecologically valid.
For your practice
This is a mechanism paper, not a tool you will run on Monday — fNIRS hyperscanning lives in research labs. But the conceptual yield is immediately usable. When you assess a young autistic child, resist scoring the child alone. The unit of dysfunction here is the dyad. A child who "imitates on request" (relatively preserved intentional pathways) may still struggle profoundly with the unscripted, rhythmic give-and-take that real play demands. That predicts where to aim intervention: less on drilling discrete imitation, more on co-regulated, rhythm-based, low-instruction interaction — music, movement, turn-taking games where the goal is the shared beat, not a correct copy.
It also reframes parent coaching. If synchrony is a property of the pair, then the adult's own regulation, timing and responsiveness are part of the neural circuit, not background. Coaching a caregiver to slow down, match the child's tempo and let synchrony emerge is, on this evidence, intervening on the mechanism itself.
Autistic social difficulty here is not a fault inside one brain but a failure of two brains to couple — which makes the dyad, not the child, the proper target of assessment and intervention.
Small sample (N=54) from a single Chinese centre, pre-school age only, and classifier accuracies in the intentional conditions hover near chance (55–57%) — the discriminative signal is real but modest and not yet a diagnostic instrument.