Developmental convergence and divergence in human stem cell models of autism

Autism spectrum disorder (ASD) is a common neurodevelopmental disorder (NDD), with a childhood prevalence of close to 2%¹⁸. The last decade of genetic studies has yielded hundreds of risk genes, consistent with extraordinary aetiological heterogeneity^1,2,3,4. More than 100 high-confidence mutations have been associated with ASD in genetic studies^{2,5,6,7,8,9,10,11,12,13}. These rare, usually de novo mutations with large effect sizes are expected to account for 10–15% of ASD cases^19,20, whereas common genetic variation is predicted to explain at least 50% of genetic risk^21,22,23. Overall, ASD shows a complex genetic architecture, with a substantial component derived from a collection of distinct, rare disorders with overlapping clinical features.

Despite its genetic heterogeneity, postmortem transcriptome analysis has revealed consistent changes in most individuals with idiopathic ASD, as well as individuals with a specific syndromic form of ASD, (dup)15q11–13 (refs. ^14,15,16,17). However, the mechanisms by which distinct mutations can lead to convergent molecular pathology, and whether convergence occurs across rare forms of ASD remains unknown. Understanding how these processes develop is complicated by the fact that the expression of most ASD risk genes peaks during fetal development, yet gene expression studies in human brain are conducted after this critical developmental window has ended^{16,24,25,26,27}. Several lines of evidence, including genetic²⁸, genomic^29,30, neuroimaging³¹ and neuropathology³², indicate that early neurodevelopment has an essential role in the development of ASD.

The advent of stem cell-based in vitro systems enables high-fidelity modelling of human brain development in NDDs^{33,34,35,36,37,38,39,40,41}. Most studies have investigated relatively small numbers of lines from individuals with idiopathic ASD⁴², or focused on individual mutations, which have demonstrated the utility of human induced pluripotent stem (hiPS) cell-based systems to study the impact of ASD genetic risk on neurodevelopment^43,44,45,46. In addition, recent work has highlighted the power of studying many genes in parallel, identifying evidence for molecular convergence using CRISPR-based perturbations of ASD risk genes on control genetic backgrounds^47,48,49. However, studies of lines derived from affected individuals are a major gap in the field.

Here we profile a large cohort of hiPS cell lines, starting from 96 hiPS cell lines ascertained from individuals with 8 different mutations associated with ASD and 11 individuals with idiopathic ASD and 30 lines derived from 25 matched control participants. From each of these lines, we derived neural organoids using a guided differentiation approach to create human cortical organoids (hCOs)^50,51. Using orthogonal analytic approaches, we find evidence for convergence during early neuronal differentiation in those with genetically defined forms, but no significant shared signal across the idiopathic cases. We identify and characterize a downregulated, chromatin and transcriptional network that contains several ASD risk genes, including members of the SWI–SNF complex. This network is predicted to drive the observed changes in downstream gene expression associated with these ASD susceptibility mutations. We use CRISPRi to validate the effects of many putative network drivers on downstream gene expression, which includes downregulation of important neurodevelopmental pathways including many ASD susceptibility genes.

We reprogrammed somatic cells (Methods) to generate hiPS cells from a cohort of individuals with eight different mutations associated with ASD: (1) 22q11.2 deletion, a 1.5–3-megabase (Mb) deletion that leads to a constellation of variably present symptoms including heart defects, craniofacial features, intellectual disability, ASD (roughly 20%) and psychosis (roughly 25%)^52,53 (n = 18)⁴⁴; (2) 22q13.3 deletion, known as Phelan–McDermid syndrome, a deletion spanning 130 kilobases (kb) to 9 Mb that includes SHANK3, among other genes and presents with developmental delay, hypotonia and impaired social interactions⁵⁴ (n = 4); (3) 15q13.3 deletion, a 1.5-Mb deletion that presents variably with intellectual disability and epilepsy⁵⁵ (n = 3); (4) 16p11.2 deletion, a roughly 600-kb deletion presenting variably with intellectual disability, motor impairments, communication deficits and ASD⁵⁶ (n = 4); (5) 16p11.2 duplication, which can share intellectual disability and ASD phenotypes with the reciprocal deletion⁵⁷ (n = 4); (6) Timothy syndrome, which is characterized by variants within the CACNA1C gene and presents with syndactyly, prolonged QT interval, ASD and intellectual disability^58,59 (n = 2)^45,60,61; (7) PCDH19-related disorder, which is associated with epilepsy and can also include intellectual disability and ASD^62,63 (n = 2) and (8) the SHANK3 R522W mutation (n = 1), a point mutation associated with neurodevelopmental risk⁶⁴; mutations in SHANK3 share behavioural phenotypes with the larger 22q13.3 deletion^61,65,66. We also profiled individuals with idiopathic ASD with no known pathogenic variants (n = 11) and unaffected individuals (n = 25) (Fig. 1a and Supplementary Table 1) for a total of 74 individuals. For several individuals we used several hiPS cell lines to assess reproducibility, amounting to a total of 96 hiPS cell lines (Fig. 1a,b and Extended Data Fig. 1).