The Simons Foundation Autism Research Initiative (SFARI) held its annual investigator meeting on March 16-18, 2026, in New York City. Simons Foundation Executive Vice President of Autism & Neuroscience Kelsey Martin welcomed over 215 attendees at the Fashion Institute of Technology. Martin highlighted SFARI’s many activities, resources and grant opportunities while emphasizing SFARI’s commitment to basic science research on autism spectrum disorder (ASD) and related neurodevelopmental disorders (NDDs). The three-day meeting comprised five scientific sessions and featured a keynote lecture by Roy Grinker of George Washington University.

While many studies have sought neurobiological differences in the cerebral cortex to explain ASD, Mustafa Sahin of Boston Children’s Hospital argued in his introduction to the session that the cerebellum may have an overlooked role. Beyond its motor functions, the cerebellum also mediates nonmotor activities, including executive function, language and working memory. Alterations in cerebellar volume and white matter have been found in ASD, and isolated cerebellar injury at birth raises ASD risk by 36 times¹. In mouse models of tuberous sclerosis complex, Sahin and colleagues have found that targeted loss of the TSC1 gene in Purkinje cells of the cerebellum resulted in ASD-like behaviors².

Mari Sepp of the University of Tartu presented her work on cerebellum evolution, which could highlight potential vulnerabilities relevant to ASD. Using single-nucleus transcriptomics and chromatin accessibility profiling, Sepp has inventoried cerebellar cell types in humans, non-human primates, mice and opossums. She found many conserved features, but also a human-specific expansion of early-born Purkinje cells and numerous gene expression differences³. Sequence-based models trained on these data and applied across 200 mammalian species further revealed human cis-regulatory elements associated with gene expression divergence, including the gain of transcription factor THRB expression in human progenitor cells⁴. Preliminary results indicate that ASD risk genes are highly expressed in the developing cerebellum. Compared with genes predominantly associated with developmental delay, ASD-predominant risk genes show expression in more differentiated cell states.

Though not a core symptom, motor deficits are highly prevalent in individuals with ASD and correlate with social difficulties. Stephanie Rudolph of Albert Einstein College of Medicine hypothesized that motor, cognitive and social dysfunction in ASD may share common circuit disruptions in the cerebellum. To examine this, she has developed a behavioral pipeline to track the emergence of quantifiable motor skills, social communication and cerebellar-dependent learning in mice throughout life. In studies of cerebellar circuit development, Rudolph found that transient oxytocin receptor expression drives maturation of specific Purkinje cell populations involved in valence and vestibular processing. Disruption of these transient neuromodulatory events may therefore have long-lasting effects on circuit function and behavior.

Kevin Bender of the University of California, San Francisco has been studying the ASD-related SCN2A gene, which encodes a sodium channel subunit. While his previous work has found roles for SCN2A in the cortex⁵, SCN2A is also highly expressed in the cerebellum, where it is essential for plasticity⁶. In mice missing a copy of Scn2a, Bender described not only a lack of cerebellar plasticity, but also alterations in the vestibulo-ocular reflex (VOR), which relies on this plasticity. The same VOR abnormalities were found in children with loss-of-function SCN2A mutations, which suggests VOR can provide a quantifiable and translatable biomarker that may extend to a range of NDDs.

Adam Hantman of the University of North Carolina at Chapel Hill presented work on the neural underpinnings of skilled movements and their relevance to ASD. His experiments use Neuropixels probes to monitor activity in multiple brain regions, including the cerebellum, in mice while they make reach and grab movements. Cross-correlation analysis of spiking patterns reveal interregional coordination, including before movement; coordination occurred across all pairs of areas, but only in subsets of cells, and it was also accompanied by increases in brain oscillation coherence. Hantman suggested the cerebellum acts as a master coordinator of interregional coupling. Consistent with this, preliminary results in a mouse model of Angelman syndrome showed increased movement precision when Ube3a gene expression was selectively restored to neurons in the cerebellum.

The next session explored different scales of brain activity measurement. The high spatial and temporal resolution possible in animal models has allowed Gabrielle Pouchelon of Cold Spring Harbor Laboratory to detect transient thalamocortical circuits in mice that provide a pivotal step for circuit maturation⁷. This transient connectivity may play a role in ASD because it persists in fragile X mouse models and because diminishing it chemogenetically normalizes circuit function⁸. To detect transient dynamics brain-wide, Pouchelon is developing functional ultrasound (fUS) imaging methods in neonatal mice.

Jessica Cardin of Yale University presented work exploring brain network dynamics in multiple mouse models of ASD using mesoscopic imaging. This technique involves wide-field imaging of fluorescent signals⁹ and can detect state-dependent activity across the mouse cortex¹⁰. In mouse models of ASD developed through CRISPR-mediated knockdown or overexpression of Mecp2, the gene related to Rett syndrome, Cardin reported altered functional connectivity across cortical regions. The mice also showed expected alterations in gene expression as well as deficits in a social task. Cardin also described how the whole imaging system could be reduced to fit onto the head of awake, behaving mice.

Michael Lombardo of the Istituto Italiano di Tecnologia examined how fractal and periodic characteristics of EEG data could be used to infer mechanisms that control excitation-inhibition (E/I) imbalance and to parse heterogeneity in autism. Using computational and animal modeling, Lombardo and colleagues showed that fractal components measured by the Hurst exponent (H) track with input excitability while γ oscillations track with the ratio of E/I synaptic conductances. Both electrophysiological measures (i.e. H, γ) were also complementary in helping to accurately predict underlying spiking activity. Using those same electrophysiological measures in autistic children, Lombardo and colleagues then showed that male autism children cluster into two E/I neurosubtypes. One subtype is characterized by increased excitability, while the other is characterized by decreased excitability. Atypical EEG excitability patterns in these subtypes were differentially associated with behavior. Excitability tracks with language, cognition and motor behavior in the more excitable subtype, but tracks with neuropsychiatric comorbidities (e.g., ADHD, anxiety, sleep issues) in the less excitable subtype.

Sahin led a panel discussion with the day’s speakers, beginning with a question about designing cross-species studies to make them more impactful to ASD. Ideas included studies of conserved, reflexive behaviors, which might be more likely to translate to humans. Quantifiable motor behaviors were also highlighted. Another discussion centered on where idiopathic ASD fit amid the bulk of scientific results based on rare, monogenic models: The monogenic work may outline the extremes of possible neural mechanisms, with idiopathic ASD falling somewhere within. Another comment highlighted the variability within animal models, despite their genetic homogeneity, and even within an animal. This variability itself could be informative.

This session addressed the challenge of tracing how genotype translates into phenotype, including an understanding of pleiotropy, in which a single genetic variant influences multiple independent traits. Hyejung Won of the University of North Carolina at Chapel Hill addressed pleiotropy in a study of common variants found in genome-wide association studies (GWAS) of several psychiatric disorders, including ASD. Using a massively parallel reporter assay (MPRA) to test thousands of single nucleotide polymorphisms (SNPs), she has identified those that modulate expression of a reporter gene¹¹. While some SNPs were associated with a single disorder, others were associated with three or more disorders spanning multiple domains, making them pleiotropic. Mapping these variants to their target genes suggested that pleiotropy reflects gene function: Target genes were expressed across many cell types for a protracted time during development, and their protein products exhibited greater network connectivity than nonpleiotropic proteins.

The next two talks presented genetic evidence for ASD subtypes based on common variants. Varun Warrier of the University of Cambridge has found two polygenic profiles of ASD that differ by age of diagnosis¹². In longitudinal cohorts of child development, Warrier found that age of diagnosis is associated with two distinct trajectories: one with difficulties early on, and another with difficulties that appear later in childhood or adolescence. A GWAS on two large ASD cohorts found that age of diagnosis is a heritable trait. Modeling genetic covariance among ASD GWAS, including ASD stratified by age at diagnosis, identified two genetic factors: one characterized by earlier diagnosis, and the other with later diagnosis. These factors differed in the associations with other traits, such as high genetic correlation for ADHD and depression in the later-diagnosed group.

Taking a different approach, Lauren Weiss of the University of California, San Francisco presented evidence for two ASD subtypes, based on principal components analysis of autosomal SNP data. While analyzing SPARK data, her group detected two clusters: a “major” one associated with an earlier age of diagnosis and a higher male-to-female ratio, and a “minor” one associated with a later age of diagnosis and reduced bias in male-to-female cases. The clusters were observed in multiple ancestries, and in another cohort with different ascertainment. The minor cluster was associated with polygenic risk for other mental health conditions, including major depressive disorder (MDD).

Natalie Sauerwald of the Flatiron Institute presented her evidence for ASD subtypes based on phenotypes¹³. Looking at 239 different phenotypic features from 5,000 children in the SPARK cohort, mixture modeling separated individuals into one of four categories: a social-behavior group enriched for repetitive behavior and limited social communication; a mixed ASD with developmental delay group; a broadly affected group; and a moderately challenged group. Genetic analysis of these groupings showed differences in common variants, de novo rare variants and de novo inherited variants, which suggests the phenotypic groupings may arise from different biological mechanisms.

To help educate parents about clinical genetic testing, Elise Robinson of Massachusetts General Hospital and the Broad Institute presented a tool developed in her lab called MINERVA (Mapping Individualized Estimates of Rare Variants in Autism). Although clinical genetic testing is recommended for children with ASD, less than 15 percent of children in the U.S. receive it. To help close this gap, the MINERVA website offers a tool that parents can use to obtain the probability of their child receiving a clinical genetic diagnosis based on simple variables like age of walking, sex and presence of seizures. MINERVA also contains education modules to help users understand clinical genetic testing results and interpretation.

A panel discussion moderated by Jonathan Sebat of the University of California, San Diego followed with the morning’s speakers and Thomas Bourgeron of Institut Pasteur. The discussants agreed that Warrier and Weiss had likely detected the same genetic subtypes, but that Sauerwald’s categories were probably within their early diagnosis groups. Others suggested additional information that could help refine genotype-phenotype subtypes, including variants of medium effect size and sensory phenotype information. Though differing ascertainment may contribute to ASD heterogeneity, discussants affirmed the benefits of studying real-world cohorts versus tightly phenotyped ones.

The next session focused on points of biological convergence emerging from ASD’s heterogeneity. Nael Nadif Kasri of Radboud University Medical Center presented work that has looked for syndrome-specific signatures of electrical activity from excitatory neurons derived from iPSCs from people with NDDs. When grown on microelectrode arrays to record field potentials in vitro, these human neurons can show distinctive outputs such as those discovered for mutations related to Kleefstra syndrome¹⁴. Looking at neurons developed from iPSC lines of 16 different NDDs, Kasri’s group has found four neural activity phenotypes, with some NDDs exhibiting only one phenotype and others showing multiple phenotypes. Transcriptomic data obtained from the cells after recording showed some convergence on activity phenotypes, but not always: In some cases, similar transcriptomes resulted in different network activity. The data show that transcriptomic impact alone does not reliably predict functional consequences and multimodal data is needed to understand when genetic perturbations become circuit-relevant cellular phenotypes.

André Sousa of the University of Wisconsin—Madison presented work that has identified neuronal maturation as a process affected by ASD risk genes¹⁵. Using single nucleus-multiomics and Patch-seq techniques on excitatory neurons of prefrontal cortex from rhesus macaque brain in late prenatal development, his lab has been able to track transcriptomic changes alongside electrophysiological maturation in the same cells. This highlighted RAPGEF4, a gene whose expression was highly correlated with maturation of resting membrane potential and inward sodium current. When RAPGEF4 expression was blocked in both macaque and human brain slices, neurons took on an immature morphology and were more immature for resting membrane potential and inward sodium current, as predicted. Similarly, loss of CHD8, a chromatin remodeler and ASD-related gene, led to immature cell phenotypes. Transcriptional characterization of neurons in which CHD8 was suppressed found that RAPGEF4 was downregulated, and that increasing RAPGEF4 in a CHD8 suppressed background could rescue the immature phenotype of CHD8 suppressed neurons.

Neville Sanjana of the New York Genome Center and New York University presented recently published results demonstrating the feasibility and power of multiomic pooled CRISPR screens to interrogate hundreds of genes in pediatric brain tumor cells while simultaneously obtaining single cell-level information about open chromatin and gene expression (MultiPerturb-seq)¹⁶. Deploying this system to knock down about 100 high confidence ASD-related genes involved in chromatin and transcriptional regulation in neurons derived from human induced pluripotent stem cells has allowed a systematic study of changes to transcriptional patterns. Some are shared among different ASD gene manipulation, while others differ. These CRISPR-mediated knock downs were chemically induced and temporary, allowing for gene re-activity. For the majority of genes targeted with the CRISPR knock down, they found that at least half of the changes to transcription could be reversed after gene re-activation, which lays the foundation for future gene therapies to reverse the effects of ASD-related gene loss.

In her talk, Anne West of Duke University offered a cell biological model of convergence among ASD genes involving chromatin architecture and consequences for neuronal maturation. In mouse cerebellar granule cell neurons, she described how KDM6B, an ASD-related gene and chromatin regulator, was required to induce expression of synaptic genes important for neuron maturation through histone demethylation¹⁷. West also showed how methylation patterns could compartmentalize chromatin, with condensed regions bringing certain genes in close proximity, thus coordinating their expression. She suggested that mutations in other ASD- and NDD-related genes may disrupt this chromatin organization, resulting in derailed gene expression programs of neural maturation.

Roy Grinker of George Washington University gave the keynote talk in which he shared his thoughts on autism as a diagnostic entity from his perspective as an anthropologist and a parent of a daughter with ASD. With personal anecdotes and professional insights, he illustrated how disease entities are products of history and culture. For example, when his daughter received a Pervasive Developmental Disorder-Not Otherwise Specified (PDD-NOS) diagnosis as a toddler in the 1990s, the category of autism wasn’t “legible” in the way it is today: Pediatricians didn’t screen for it, psychiatrists did not encounter it, and the general public was not aware of it. Similarly, ascertainment affects prevalence, as Grinker illustrated in his own study of ASD in South Korea at a time when ASD was not widely recognized there¹⁸.

That ASD is heterogeneous and shape-shifting through history doesn’t make it less real, however. Though this complexity may seem to interfere with the reductionism that the scientific method requires, Grinker urged scientists to keep in mind the social and cultural contributions to a diagnostic category. This means treating the Diagnostic Statistical Manual (DSM) as a living document and not shying away from talking about ASD’s complexity, even when public discourse rewards simplistic answers.

Genvieve Konopka of the University of California, Los Angeles presented preliminary analysis of differences in white matter tracts in ASD based on tissue from Autism BrainNet, a postmortem brain tissue resource. While transcriptomic studies often focus on gray matter, Konopka’s previous work has found human-specific differences in oligodendrocytes, the myelin-producing cells that ensheath axon fibers¹⁹. Single-nucleus multiomics on the tissue from six white matter regions in case-matched samples yielded data from 700,000 nuclei, mostly from oligodendrocytes. This revealed a decrease in mature oligodendrocytes in ASD, and differentially expressed genes in oligodendrocytes and microglia compared to controls, which showed some overlap with ASD risk genes.

Tomasz Nowakowski of the University of California, San Francisco presented work looking beyond the cortex for a molecular signature of ASD. In a mouse model of 16p11.2 deletion syndrome, Nowakowski found evidence of alterations in the number of D1 medium spiny neurons in the striatum. This cell type also showed differentially expressed genes in human postmortem samples in those with a genetic diagnosis of ASD²⁰. ASD brains showed a reduced signature of a coexpression network depending on EGR1, a transcription factor. Nowakowski also examined language region Brodmann area 22. Molecular profiling showed transcriptional differences in types of interneurons and excitatory neurons in those with a genetic diagnosis of ASD, with accentuated effects in nonverbal individuals.

Studying human cortical tissue from prenatal stages to adolescence, Li Wang of Stanford University has identified some mechanisms involved in cell-type-specification²¹. Through single cell-multiomics combined with knowledge of motif sequences, Wang has pinpointed 582 “eRegulons” that show cell-type-specific transcription factor expression. This has revealed key bifurcation points for cell subtype specification during development. Wang also showed preliminary efforts to obtain proteome data from single cells. With mass spectrometry technology increasing its specificity and output, Wang has been able to identify 1,000 proteins per cell in 100 cells a day.

Omer Bayraktar of the Wellcome Sanger Institute has been precisely mapping ASD gene expression using spatial transcriptomics, which allows an unbiased and quantitative brain-wide view of gene expression in situ. His team has made an atlas based on 250 ASD genes in midgestation human brain, 65 cell type markers and 10 million cells (stageatlas.org)²². Using a method to identify programs of gene coexpression across cells, Bayraktar found five different ASD-relevant gene programs. Each program localized to five different brain regions, including the thalamus. Most risk genes converged in the thalamus, with elevated expression in excitatory neurons, which suggests the thalamus may be a hub of ASD pathology.

Michael Gandal of the University of Pennsylvania moderated a panel discussion with the day’s speakers, beginning with a question on whether ASD’s phenotypic heterogeneity could ever be reconciled with a single, common signature in the brain. Speakers endorsed the idea that there will be some commonalities, but that the heterogeneity was so rich that ASD will probably involve more than one circuit. Another comment emphasized the need to distinguish between molecular changes that are causal or compensatory, or cell-autonomous or not, as a way to understand ASD’s heterogeneity, and the necessity of animal models to resolve this. Another idea was the need to establish a brain “phenotype” reference to understand what the brain endpoint of ASD is, which could provide a useful benchmark for researchers.

Martin closed the meeting by emphasizing the diversity of ASD, which requires diverse research approaches. She hoped the SFARI investigators would leave the meeting inspired to embrace ASD’s complexity, and the different scientific perspectives and approaches it will take to ultimately understand it.

1. Wang S.S.-H. et al. Neuron 83, 518–532 (2014) PubMed
2. Tsai P.T. et al. Nature 488, 647–651 (2012) PubMed
3. Sepp M. et al. Nature 625, 788–796 (2024) PubMed
4. Sarropoulos I. et al. Science 391, eadw9154 (2026) PubMed
5. Spratt P.W.E. et al. Neuron 103, 673–685 (2019) PubMed
6. Wang C. et al. Neuron 112, 1444–1455 (2024) PubMed
7. Dwivedi D. et al. Nat. Commun. 15, 5421 (2024) PubMed
8. Dumontier D. et al. bioRxiv (2024) Preprint
9. Cardin J.A. et al. Neuron 108, 33–43 (2020) PubMed
10. Lohani S. et al. Nat. Neurosci. 25, 1706–1713 (2022) PubMed
11. Lee S. et al. Cell 188, 1409–1424 (2025) PubMed
12. Zhang X. et al. Nature 646, 1146–1155 (2025) PubMed
13. Litman A. et al. Nat. Genet. 57, 1611–1619 (2025) PubMed
14. Frega M. et al. Nat. Commun. 10, 4928 (2019) PubMed
15. Gao Y. et al. Neuron 113, 2490–2507 (2025) PubMed
16. Yan R.E. et al. Nat. Biotechnol. 43, 1628–1634 (2025) PubMed
17. Ramesh V. et al. eLife 12, e86273 (2023) PubMed
18. Kim Y.S. et al. Am. J. Psychiatry 168, 904–912 (2011) PubMed
19. Caglayan E. et al. Nature 620, 145–153 (2023) PubMed
20. Yuan G. et al. bioRxiv (2025) Preprint
21. Wang L. et al. Nature 647, 169–178 (2025) PubMed
22. Aivazidis A. et al. bioRxiv (2025) Preprint