Autism Severity Tracked Frontoparietal-DMN Connectivity Across ADHD and Autism

TL;DR: 166 children, two different chart labels, one consistent brain pattern. Stronger resting-state coupling between the left middle frontal gyrus and the posterior cingulate cortex tracked clinician-rated autism severity — even in children whose primary diagnosis was ADHD. The same pipeline found no comparable signal for ADHD severity itself.

Source: Molecular Psychiatry (2026) | Segura et al.

Autism and ADHD spend a lot of time presented as neighboring diagnoses with overlapping symptoms — and a lot of imaging research treats that overlap like contamination to be controlled for. This paper does the opposite. It treats the overlap as the point. And what surfaces is unexpected: the cleanest cross-diagnostic brain signal in this cohort was not ADHD’s. It was autism’s.

Why the Diagnostic Label Was Not the Most Interesting Variable

The standard developmental-psychiatry design is one diagnosis vs. another, or one diagnosis vs. controls, with average brain differences as the readout. That design works until the symptoms stop respecting the labels — and autism and ADHD are exactly that case. They co-occur often, share genetic liability, and bleed into each other in clinic.

The team recruited 166 verbal children ages 6 to 12, including 63 with autism and 103 with ADHD-without-autism, then deeply phenotyped every child for both autism and ADHD symptoms regardless of which diagnosis they had walked in with. The data hid no overlap. They were structured to expose it.

One methodological choice deserves attention. The autism signal in this dataset belonged to the clinician-observed ADOS-2 calibrated severity score, not to a parent questionnaire. Different informants do not produce equivalent ratings, and in this sample, the observation-based autism measure carried information the parent-report measures did not.

One Left-Hemisphere Bridge, Two Diagnoses, One Direction

The result narrowed to a single connection between two well-known networks. The first node sat in the left middle frontal gyrus, part of the frontoparietal control network — the system that flexibly directs attention. The second sat in the posterior cingulate cortex and precuneus, a hub of the default mode network involved in internally directed thought and social cognition.

Stronger coupling between these nodes corresponded to more severe autism symptoms — across the whole sample, not just within the autism diagnostic group. The relationship held after statistically controlling for ADHD ratings. That is a stronger claim than “autism diagnosis correlates with circuit X.” It says the circuit follows the symptom dimension itself, regardless of the chart label that came with the child.

Network segregation analysis added depth. As autism severity rose, the default mode network became less cleanly separated from the frontoparietal, dorsal attention, and visual networks. The boundaries that normally keep these systems partly distinct looked leakier. That fits a pattern autism imaging has been circling for years — not a single broken region, but altered coordination between systems that should stay differentiated.

The ADHD Null Result Is Almost As Important

The pipeline that surfaced the autism signal was tested against multiple ADHD measures: clinician-rated inattentive symptoms, hyperactive symptoms, parent questionnaires. None of them produced a statistically significant whole-brain relationship after correction.

That does not mean ADHD has no functional-connectivity biology. It means that in this tightly phenotyped, motion-controlled cohort, the strongest transdiagnostic brain–behavior signal belonged to autism severity. The shared biology may not be evenly distributed across the two diagnoses, even though the clinical overlap is heavy in both directions.

The asymmetry is harder to dismiss when you remember the cohort makeup. 38 of 103 ADHD-without-autism children — 37% — still crossed the ADOS-2 autism-spectrum threshold. The study was not comparing tidy, isolated syndromes. It was comparing children whose symptoms bled into each other in real ways. That is exactly what makes the autism-specific connectivity result hard to explain as a labeling artifact.

The Gene Enrichment Move Anchors the Imaging Result

Most fMRI papers stop after the connectivity map. This one took one more step. Using in-silico methods, the authors asked whether the spatial pattern of the autism-associated connectivity differences overlapped with known gene-expression gradients in the human brain.

The enriched genes had three properties worth noting. They overlapped with gene sets carrying elevated variant burden in both autism and ADHD. They were enriched for neuron projection biology — the cellular machinery needed for long-range communication, exactly the kind of process you would expect to be relevant to between-network coordination. And the spatial fit lined up with the imaging result rather than pointing to a biologically unrelated system.

This is in-silico work, not direct molecular measurement from the children, so it cannot establish causality. What it can do is constrain the conversation. The connectivity pattern was not floating in statistical space. It mapped onto genomic territory that already makes biological sense for neurodevelopmental circuit formation.

What This Pushes Biomarker Research Toward

The practical implication is more about strategy than diagnosis. Biomarker work built around categorical case-control comparisons can flatten exactly the signal this paper just surfaced. If autism severity tracks the same circuit phenotype in two different diagnostic groups, then comparing “autism” against “ADHD” averages over the variation that actually contains the brain signal.

The cohort tells the same story clinically. Many children diagnosed with ADHD without autism still carried meaningful autism-spectrum burden on direct observation. Rigid diagnostic silos have a way of erasing those children’s actual phenotypes — and with them, the chance to find the biology that matches.

The study has the limits any reasonable reader expects. The sample was verbal children with IQ above 65; the resting-state scans were relatively short. This is a mechanistic clue, not a deployable clinical test. But it is a useful clue — and the cleanest sentence in the paper might be the implicit one: the diagnosis crossed over, and the connectome kept following autism severity anyway. That is the kind of result transdiagnostic imaging has been promising for a decade.