On the largest protein map ever built for autism — and the question it makes unavoidable
On 27 August, a UCSF-led team published in Science the largest map of molecular interactions ever assembled for autism. Alison Singer, president of the Autism Science Foundation, called it a watershed moment. She is right. I want to say why I think she is right, and why the reason may be slightly different from the one making headlines.
What the study did
The team took 100 high-confidence autism risk genes and, instead of studying the genes, studied the proteins those genes encode. Using affinity purification–mass spectrometry, they mapped more than 1,800 protein–protein interactions, roughly 87% of them previously unreported. They then analysed 54 patient-derived mutations to see how particular variants changed those interactions, and used AlphaFold structural predictions to locate where in a protein interface the disruption occurs.
The findings are specific. The 100 proteins and their interaction partners form a highly interconnected network that mirrors autism risk-gene expression patterns in the developing human brain. That network is enriched for additional autism risk genes but not for the schizophrenia risk set used in the comparison — evidence that the map is capturing something more particular than a generic neurodevelopmental network. Functional testing of selected interactions showed effects on progenitor cells, cortical neurogenesis, and neural excitability. In one striking case, separate mutations in FOXP1 and FOXP2 converged on disrupting the same FOXP1–FOXP4 interaction, and the damage turned out to run through a pathogenic gain of function in the partner protein rather than a simple loss of FOXP1.
The paper’s own summary is the sentence worth sitting with: genetically heterogeneous risk genes and distinct mutations converge onto shared complexes and rewire protein interactions.
The headline and the result are not saying the same thing
The UCSF release led with Krogan’s statement that the science demonstrates autism is written in our genes. But the paper’s decisive advance begins where that sentence stops.
Krogan is candid that gene discovery had hit a wall. Hundreds of genes, dozens of mutations, and very few therapeutic leads. His own metaphor is that genes and mutations are a parts list, and what was missing was the wiring diagram. The breakthrough came from moving up a level — from the parts to how the parts are organized.
This is not less genetic. It is what genetics looks like once it becomes systems biology. The genes supplied the entry points; the finding is that their protein products do not act as isolated parts. Many different genetic starting points arrive at a small number of shared functional consequences, and the shared consequence is the thing worth treating. Matthew State says as much: convergence of multiple mutations on the same process is what tells you where the target is.
Singer’s own framing caught this. She noted that the shared pathways could yield targets relevant to much larger groups of autistic people, including people with no identified genetic cause.
That does not prove autism is one biological entity, and it does not rule out mechanistically meaningful subgroups. What it does show is why visible difference cannot, by itself, establish biological separation. Distinct genetic routes can converge on shared molecular organization, and apparently similar outcomes can arise through different routes. Architecture has to be measured, not inferred from phenotype.
The same discipline runs in the other direction, and it should be said plainly. This map is built on 100 high-confidence risk genes, mostly rare high-impact variants, in a population the researchers identify as most directly relevant to profound autism — around 30% of diagnoses. Extending the convergence story to people without an identified genetic cause is a hypothesis the map makes testable; it is not a result the paper delivers, and State has said as much himself. The atlas is also model-based: selected findings were tested in human forebrain organoids, not measured in a developing human brain.
The question the map makes unavoidable
Here is what a wiring diagram gives you: it tells you what is perturbed. Here is what it cannot tell you: why two people carrying the same perturbation do not live the same life.
That is not hypothetical. It is the ordinary experience of every clinician and every family. Among people carrying the same identified variant, developmental and functional trajectories diverge profoundly. One child speaks and another does not. One person’s difficulties ease through adolescence while another’s intensify. One adult sustains work for a decade and then, after an illness or a bereavement or a move, cannot.
The DNA sequence did not change. The state of the system did.
Between a perturbed protein interaction and a person’s life course lies a multiscale regulatory architecture spanning cells, circuits, body systems, and environments. It includes how the system regulates itself across development, what regulatory cost to the person is being paid to sustain observable function, and what happens when cumulative demand exceeds available capacity. That is where sleep, sensory environment, illness, transitions, support, and timing do their work. It is also where much of the care currently capable of changing an ordinary day operates — not by repairing a protein interface, but by treating physiological burdens, improving sleep and communication access, reducing destabilizing demand, supporting recovery, and improving the fit between a person and their environment.
None of this is a criticism of the study. The map is a prerequisite. You cannot ask a well-formed question about regulation without knowing what is being regulated, and until this week we largely did not. But the map raises its own next question, and it is empirical: what governs the passage from a rewired network to an individual trajectory?
Elements of that question are being studied. They are rarely studied as a single longitudinal, multiscale problem — how molecular perturbation, developmental timing, physiological state, cumulative demand, and environmental conditions interact to produce divergent trajectories. On the same day this paper appeared, QBI announced $46 million to advance the molecular programme. There is no comparable investment in the middle layer.
For families
The honest thing is to say both parts, in the same breath.
This is real progress. Krogan says three programmes are already underway to develop potential therapies from the work. That means therapeutic development has begun. It does not mean a treatment is near. The map warrants hope, not a timetable.
And the endpoint cannot be molecular correction for its own sake. It has to be a better ordinary day: less pain, fewer seizures, steadier sleep and health, stronger communication access, more autonomy, more sustainable participation.
The distance between a druggable target and those outcomes is measured in the middle layer — the regulatory architecture connecting molecules to cells, cells to bodies, and bodies to lives. We are building a powerful science at the molecular end of that gap. We still lack an integrated science of the middle.
If this map is going to become medicine, the next decade has to study not only the molecular target, but the architecture connecting that target to the life.
Lori Hogenkamp is founder and director of the Center for Adaptive Stress. She writes on regulatory architecture in neurodevelopment and health at ndstress.org.
Sources
- Autism mutations rewire protein interaction networks to drive neurodevelopmental pathology. Science, 27 August 2026. DOI: 10.1126/science.ady4523. Co-first authors Belinda Wang, Rasika Vartak, and Kelsey Hennick; co-corresponding authors Kirsten Obernier, Tomasz J. Nowakowski, and A. Jeremy Willsey; senior authors Matthew W. State and Nevan J. Krogan.
- UCSF news release, “Autism Decoded: New Science Is Opening Paths to Better Treatments,” 27 August 2026 — study scope, 54 patient-derived mutations, FOXP1–FOXP4 finding, profound autism figure, State and Krogan quotes.
- UCSF news release, “UCSF QBI Is Awarded $46 Million to Advance Autism Research,” 27 August 2026.
- CNN, “Largest ever ‘map’ of autism may hold clues for new targeted therapies,” 27 August 2026 — Singer’s emailed comments; Krogan’s parts-list metaphor and the three therapeutic programmes.
Leave a Reply