Functional impact of genetic background on variable expressivity in neurodevelopmental disorders

Abstract

Disease-associated variants can lead to variable phenotypic outcomes in neurodevelopmental disorders, but the biological mechanisms underlying this variability remain poorly understood. Here, we develop a framework to investigate this phenomenon using the 16p12.1 deletion as a paradigm of variable expressivity. Using induced pluripotent stem cell models from affected families and CRISPR-edited lines with the 16p12.1 deletion, we find that the deletion and rare variants in the genetic background jointly influence chromatin accessibility and expression of neurodevelopmental genes. Cellular analyses identify family-specific phenotypes, including altered inhibitory neuron production and neural progenitor cell proliferation, which correlate with head-size variation. CRISPR activation of individual 16p12.1 genes variably rescue these defects by modulating key developmental signaling pathways. Integrative analyses further identify regulatory hubs, including transcription factors FOXG1 and JUN, as mediators of these effects. Our study provides a functional framework for investigating how individual genetic architectures contribute to phenotypic variability in neurodevelopmental disorders.

Genetic variants can lead to variable outcomes in neurodevelopmental disorders. Here, the authors show that the 16p12.1 deletion and genetic background jointly shape chromatin regulation and neurodevelopmental phenotypes, explaining variable disease expressivity.

Introduction

In contrast to Mendelian disorders with straightforward genotype to phenotype relationships, variants associated with complex disorders often lead to variable clinical features^1,2. This phenomenon of variable expressivity is particularly true for rare copy number variants (CNVs), such as deletions and duplications within chromosomal regions 1q21.1, 15q13.3, and 22q11.2, which have been associated with different neurodevelopmental outcomes and have also been identified in population controls^3–5. Among these CNVs, the approximately 500-kbp 16p12.1 deletion, encompassing eight genes, represents an ideal model to study variable expressivity of disease-associated variants for several reasons. First, the deletion is associated with a wide range of phenotypic outcomes, including intellectual disability/developmental delay, autism, congenital anomalies, and epilepsy in affected children, as well as psychiatric features such as schizophrenia, depression, and anxiety in adolescents and adults, with varying levels of severity^6–12. Second, unlike syndromic CNVs such as 17p11.2 deletion in Smith-Magenis syndrome or 5q35 deletion in Sotos syndrome³, that mostly occur de novo, the 16p12.1 deletion is inherited in over 90% of affected individuals from parents with milder cognitive or neuropsychiatric features¹³. The high inheritance rate allows us to assess phenotypic outcomes in multiple carriers from the same family. Third, specific patterns of rare variants in the genetic background (“secondary variants”) correlated with distinct phenotypic trajectories associated with the deletion¹⁴. Secondary modifier variants have also been reported to contribute to phenotypic variability in both monogenic and complex neurodevelopmental disorders, including in individuals affected by 7q11.23 duplication and 16p11.2 duplication^2,3,15,16. However, the molecular basis by which these CNVs confer disease susceptibility and interact with secondary variants to produce variable outcomes remains poorly understood.

Previous studies have shown the utility of induced pluripotent stem cells (iPSCs) to recapitulate complex biological processes in vitro, enabling investigation into the molecular etiology of neurodevelopmental disorders^17–19. However, these studies have typically focused on aggregate analysis of subjects with the same primary variant, such as the 16p11.2 and 22q11.2 deletions, or the same phenotype, such as autism and schizophrenia, without investigating variability among these subjects^20–22. Thus, the effects of disease-associated variants within the context of an individual’s genetic background remain unexplored. Here, we developed a framework to dissect the mechanistic basis of variable expressivity in iPSC models of the 16p12.1 deletion (Fig. 1). By integrating multi-omics profiling, CRISPR/dCas9-mediated gene activation, and family-based analyses, we demonstrate that phenotypic trajectories associated with the 16p12.1 deletion are shaped by the combinatorial effects of the deletion and secondary variants through the modulation of key signaling pathways including TGF-β and PI3K-AKT. We further identify regulatory hubs that mediate these effects, including transcription factors (TFs) FOXG1 and JUN. Our study emphasizes a conceptual shift from a focus on single causal variants to a system-level understanding of individual genetic architecture, advancing more effective precision medicine strategies.

Fig. 1. Overview of the integrated framework using 16p12.1 deletion iPSC models.

a Experimental design. WGS was performed on DNA isolated from blood or iPSCs derived from healthy donors and individuals in the three families. RNA-seq was performed across all four differentiation stages and ATAC-seq was performed at iPSC and NPC stages. Information on correlation between replicates and lines can be found in Supplemental Data 1. b Representative images of neural-converted cells stained with different markers for distinct cell stages (scale bars, 20 μm). c Pedigrees of three families with clinical phenotypes. Filled squares, 16p12.1 deletion carrier. More detailed phenotypic information is provided in Supplementary Data 1. No color fill indicates that clinical information was unavailable. d Overview of experimental strategy used to investigate how the 16p12.1 deletion and secondary variants contribute to phenotypic variability. Cartoons in (a) and (d) were created in BioRender. Girirajan, S. (2026) https://BioRender.com/1zw6tq0.

Results

Generation and characterization of iPSC models for 16p12.1 deletion

To model the effects of the deletion on neural development in an isogenic setting, we created a 465-kbp 16p12.1 deletion in the HD_01 line, derived from a healthy donor, using a CRISPR/Cas-9-based strategy (Supplementary Fig. 1a–e). To study the effects of the 16p12.1 deletion under different family-specific genetic backgrounds, we reprogrammed iPSC lines from peripheral blood mononuclear cells derived from three families (Fig. 1a–d, Supplementary Data 1). As controls, we used iPSC lines (HD_01 and HD_02; obtained from NINDS repository) from unrelated healthy male and female donors, as well as an HD_01 line transfected with an empty vector (EV) as a CRISPR control. These iPSC lines were then differentiated to neural progenitor cells (NPC) and immature (iMN) and mature neurons (MN) using dual-SMAD inhibition¹⁸ (see Methods, Fig. 1b). RNA sequencing was performed at all four cell stages, while ATAC-seq was performed for iPSC and NPC lines (Fig. 1a). We confirmed reduced expression of 16p12.1 deletion genes in deletion cell lines, including UQCRC2, POLR3E, MOSMO, and CDR2, as well as reduced ATAC-seq signals across the region (Supplementary Fig. 2a–b, Supplementary Data 2). The expression of expected gene markers were robust for each differentiation cell state, such as expression of SOX2, POU5F1, and FUT4 for iPSC; SOX2, HES1, NES, and EMX2 for NPC; and MAP2, DLG4, NCAM1, and RBFOX3 for neurons (Supplementary Fig. 2c), and ATAC-seq peaks revealed differences in chromatin accessibility near cell-type-specific genetic markers such as POU5F1, SOX2, and HES1, confirming the expected cell identities across all lines (Supplementary Fig. 2d).

Combined effects on gene expression related to neurodevelopment

We compared the CRISPR-edited and patient 16p12.1 deletion lines to patient nondeletion, CRISPR control, and healthy donor (HD) lines at each cell stage and found that the most significantly differentially expressed genes (DEGs) were those within the deleted region (Supplementary Fig. 2a, Supplementary Data 2). This suggested that a combined analysis may obscure the downstream effects of the deletion due to variability between samples, likely due to differences in their genetic background (Supplementary Fig. 2e). To isolate the direct impact of the deletion, we compared the CRISPR deletion line to its isogenic control (i.e., isogenic setting). This approach is commonly used in iPSC models of disease²³ to control for the effects of the genetic background. DEGs identified in NPC, iMN, and MN were enriched for neurodevelopmental and psychiatric disorder genes^24–29 (Supplementary Fig. 3a–b). To assess the extent to which the effects of genetic background on gene expression are controlled in the isogenic setting, we performed whole genome sequencing to identify all variants shared between the CRISPR deletion and its isogenic control (see Methods). In this context, altered expression of genes (i.e., DEGs and altered isoforms) overlapping with genetic-background variants would suggest interactive effects between the variants and the deletion. We first tested if altered gene expression identified from isogenic comparisons were enriched for expression QTLs (eQTLs) and splicing QTLs (sQTLs), respectively, from the PsychENCODE and GTEx databases^30,31 (see Methods). Both eQTLs and sQTLs were depleted within the DEGs and altered isoforms after accounting for predicted direction of QTL effect, indicating limited interactions between common variants and the deletion (Supplementary Fig. 3c, Supplementary Data 2). We next identified 5,526 rare variants that were shared between the CRISPR deletion and its isogenic control (referred to as secondary variants). After removing genes potentially affected by QTLs, we found that 76/5,526 (1%), 935/5,526 (17%), 960/5,526 (17%), and 357/5,526 (6%) of the secondary variants overlapped with the DEGs and isoforms with altered usages identified in the iPSC, NPC, iMN, and MN lines, respectively (Fig. 2a, Supplementary Data 2). A subset of these DEGs that overlap secondary variants, including SHANK1, DLX5 and BBS4, showed extreme expression in the CRISPR deletion line compared to other lines ( | z-score | ≥2)^32,33 (Supplementary Fig. 3d, Supplementary Data 2). Over-representation analysis of DEGs and altered isoforms with secondary variants revealed enrichment for terms associated with neurodevelopment, neuronal function, and signaling pathways, such as axon guidance, synaptic transmission, and PI3K-AKT signaling pathway (Fig. 2b, Supplementary Data 2). These results suggest that secondary variants, even in a control genetic background, may alter gene expression when combined with the 16p12.1 deletion³⁴. This also supports the idea that secondary variants could exert latent effects that become unmasked in the context of a disease-associated variant^35,36.

Fig. 2. Evaluating the impact of secondary variants on chromatin accessibility and gene expression in the context of 16p12.1 deletion.

a Models (left) of interactive effects between the 16p12.1 deletion and secondary variants. Forest plots (right) with dots representing estimated odds ratio (OR) along with 95% confidence intervals (CI) show the overlap of secondary variants with DEGs (left, padj<0.05), isoforms with altered usages (center, q < 0.05), and differential peaks (Diffpeaks) obtained from ATAC-seq (right, padj<0.01) in the isogenic setting (CRISPR deletion compared to its isogenic control). **Benjamini-Hochberg FDR < 0.05, mRNA: 0.16, 7.71 × 10⁻⁴², 6.12 × 10⁻²², 1.36 × 10⁻¹¹, isoform usage: 0.0098, 9.84 × 10⁻⁰⁷, 4.12 × 10⁻³³, 8.36 × 10⁻¹¹ for iPSC, NPC, iMN and MN separately, chromatin accessibility: 0.40, 3.63 × 10^-05 for iPSC and NPC separately, two-sided Fisher’s exact test (see Supplementary Data 2 and Supplementary Data 3). b Selected pathways enriched for DEGs (padj<0.05) and isoforms with altered usage (q < 0.05) overlapping secondary variants in the isogenic setting. Note that DEGs, isoforms with altered usages, and Diffpeaks which overlap QTLs with concordant direction of effects and those overlapping with rare variants not shared between the CRISPR deletion and HD_01 were excluded (see Supplementary Data 2 and Supplementary Data 3. c Representative motifs of TF binding sites enriched within ATAC-seq Diffpeaks that intersect secondary variants in the isogenic lines. See Supplementary Data 3. d Examples of secondary variants altering chromatin accessibility and differential expression (transcripts per million, TPM) of the corresponding target gene in the isogenic (mean values) and familial settings (median values) in NPCs are shown. ATAC-seq peaks were visualized using Integrative Genome Viewer (IGV). SNV, single nucleotide variant; Del, 16p12.1 deletion lines; Nondel, nondeletion lines; var, variant. SLC29 A3 also showed outlier expression ( | z-score | ≥2) in the isogenic setting. Two-way ANOVA tests using TPM showed significant interactive effects (p < 0.05) for the analyzed familial samples. We used n = 3 independent experiments for all RNA-seq and ATAC-seq assays. Diffpeak locations for SLC29A3, SMYD3, and PGF are, respectively, chr10: 71,341,518–71,344,099, chr1: 245,822,000-245,825,000, chr14: 74,956,000–74,958,000. Data for isogenic and familial samples are provided in Supplementary Data 2 and Supplementary Data 5, respectively. Cartoons in (a) and (d) were created in BioRender. Girirajan, S. (2026) https://BioRender.com/1zw6tq0. Source data are provided as a Source Data file.

We also evaluated the combined effect of the deletion and genetic background on chromatin accessibility (Supplementary Fig. 3c, Supplementary Data 3). While chromatin accessibility QTLs (caQTLs) were not enriched at ATAC-seq defined differentially accessible regions (“Diffpeaks”), we identified five and 105 secondary variants at these regions in iPSCs and NPCs, respectively (Fig. 2a, Supplementary Data 3). Diffpeaks intersecting secondary variants were associated with differential expression of target genes such as NRXN3, SOX6, and DPP10. Furthermore, motif enrichment analysis revealed binding motifs for multiple transcription factors (TFs) within these Diffpeaks, including MYB and TCF4, which are associated with neurodevelopmental pathways³⁷ (Fig. 2c, Supplementary Data 3).

We next assessed potential interactive effects using iPSC models derived from families with variable clinical features (Fig. 1c). The DEGs identified in deletion compared to nondeletion lines, as well as between proband and carrier parent lines in the same family, were enriched for genes associated with neurodevelopmental and psychiatric disorders; although enrichment patterns varied across families (Supplementary Fig. 4a–d, Supplementary Data 4). When we analyzed each family separately, we identified genes such as RBFOX1, SRRM1, and WRD70 that showed significant changes in gene expression when both the deletion and the secondary variants were present (Supplementary Fig. 4e, Supplementary Data 5). In both isogenic and family lines with the deletion, secondary variants were also associated with significant changes in chromatin accessibility and altered expression of target genes, such as SLC29A3, SMYD3, and PGF (Fig. 2d). These results suggest that secondary variants in combination with the deletion lead to nonadditive changes in gene expression within neurodevelopmental pathways.

Variable neural phenotypes across genetic backgrounds

Beyond transcriptomic changes, we also evaluated the combined effects of the deletion and secondary variants on neural phenotypes. At the NPC stage, we found no differences in NESTIN and SOX2 (neural progenitor cell markers) levels between the deletion and nondeletion lines. However, we observed a decrease in PAX6 (a dorsal telencephalic marker) (Supplementary Fig. 5a–b) and a highly variable increase in NKX2.1 (a ventral telencephalic marker) in the deletion lines compared to nondeletion lines (Fig. 3a), suggesting altered neuronal lineage commitment during differentiation. To further investigate this, we compared marker expression in each family’s deletion lines to HD lines and, when available, nondeletion lines from the same family. We found a marked increase in NKX2.1-positive cells specifically in GL_007 deletion lines and the CRISPR deletion line (Fig. 3a–b, Supplementary Fig. 6a). In addition to NKX2.1, which is critical for the development of inhibitory neurons, we examined VGAT, the vesicular GABA transporter, in mature neurons expressing NeuN and MAP2 (Supplementary Fig. 5C). We found opposing trends across families, with a substantial increase in VGAT intensity in the deletion lines of GL_007 and a variable decrease in the deletion lines of GL_077 and GL_079 compared to controls (Fig. 3c–d, Supplementary Fig. 6b–c). We also assessed other genes involved in inhibitory neuron differentiation and migration, including DLX1, DLX2, LHX6, and LHX8, and found them to be highly expressed across NPCs, iMNs, and MNs in both the CRISPR deletion and GL_007 deletion lines compared to controls (Supplementary Fig. 5d). Although NKX2.1 was elevated in the CRISPR deletion line compared to its isogenic control, the absence of a significant difference in VGAT intensity suggests that this alteration may not persist into the MN stage or may be too subtle to be detected with this assay. Additionally, glutamatergic neuron production did not appear to be affected, as expression levels of glutamatergic markers (SLC17A6, SLC17A7, GRIN1, GRIN2B, and GLS) were not consistently altered across the deletion lines, and mature neurons showed no overt change in VGLUT1 expression (Supplementary Fig. 5c). These results suggest that, in specific genetic backgrounds, the deletion contributes to altered inhibitory neuron production, consistent with findings in other iPSC models of neurodevelopmental disorders³⁸.

Fig. 3. Assessing neural phenotypes across genetic backgrounds.

a Violin plot (left) shows the percentage of NKX2.1-positive cells at the NPC stage in all deletion compared to nondeletion lines, **p = 0.0075. Bar plot (right) shows the percentage pooled by family and deletion status, **p = 0.0015; ***p = 5.02 × 10^-05. b Representative images of NPCs stained for NKX2.1. Scale bar, 20 μm. c Violin plot (left) shows fluorescence intensity for VGAT normalized by MAP2 at the MN stage in all deletion compared to nondeletion lines, ns = 0.41. Bar plot (right) shows fluorescence intensity pooled by family and deletion status, ns: CRISPR del/con, 0.61; GL_077 c/nc, 0.10; GL_079 c/nc, 0.22; GL_079 c/HD, 0.57; *p = 0.019; ***p = 0.00039. d Representative images of neurons stained for VGAT and MAP2. Scale bar, 20 μm. e Violin plot (left) shows percentage of Ki-67-positive cells at the NPC stage in all deletion compared to nondeletion lines, ns = 0.65. Bar plot (right) shows the percentage pooled by family and deletion status, ns: CRISPR del/con, 0.49; GL_077 c/nc, 0.94; GL_007 c/HD, 0.10; ***p = 0.00014 and 0.00018 for GL_079 c/nc and c/HD separately. f Representative images of Ki-67-positive NPCs. Scale bar, 70 μm. g Bubble plot showing selected pathways (q < 0.05) enriched by GSEA from DEGs (padj < 0.05) annotated in the Reactome, WP, and KEGG databases, “isogenic” for CRISPR lines. NES, normalized enrichment score (see Supplementary Data 4). For (a), (c) and (e), embedded box plots in violin plots indicate the interquartile range (IQR) and median (horizontal line) values. Data presented in bar plots are mean ± s.e.m. Significance is determined with a linear mixed effects model treating image, replicate, and line as random effects when multiple lines are grouped for comparison (see Supplementary Data 10 for all statistical test results). “n/a” indicates that stats tests were not performed due to high zero inflation. n = 3 independent experiments for each line, except for RNA-seq and downstream analyses where n = 2 for iMNs and MNs of P1C_077, and MNs of HD_01. n = 6 images were quantified per experiment. “c” represents deletion lines and “nc” represents nondeletion lines, for their respective families (see Supplementary Data 10 for lines used in each group). “HD” represents HD_01 and HD_02. Source data are provided as a Source Data file.

Aberrant neural proliferation and apoptosis, which alter the proportions of distinct brain cell types, represent another convergent phenotype in neurodevelopmental disorders³⁹. We evaluated NPC proliferation by measuring Ki-67-positive cells as well as EdU-positive cells at 0, 24, and 48 chase hours and found no differences in proliferation rates between the deletion and nondeletion lines (Fig. 3e,f, Supplementary Fig. 6d–f). Similarly, TUNEL assays revealed no differences in apoptosis between the two groups (Supplementary Fig. 6g–h). However, when we assessed individual families, we observed opposite trends in the deletion probands from GL_079 and GL_007 compared to HD controls: both proband and carrier mother (P2C_079 and MC_079) from GL_079 showed increased proliferation, while in family GL_007, proband P1C_007 showed decreased proliferation and increased apoptosis (Fig. 3e–f, Supplementary Fig. 6d–h). Proband P2C_079 also showed decreased apoptosis, but only when compared to nondeletion lines from the same family (Supplementary Fig. 6g–h). Notably, these results correspond with the head-size phenotypes in P2C_079 and MC_079 from family GL_079 (macrocephaly) and P1C_007 from family GL_007 (microcephaly) (Fig. 1c, Supplementary Data 1).

We also found evidence of altered developmental timing in deletion lines, a phenomenon also reported in iPSC models of autism and schizophrenia^22,40. A significant increase in TUBB3-positive cells (a marker for newborn neurons) at the NPC stage was detected in the CRISPR deletion and GL_007 deletion lines compared to controls (Supplementary Fig. 6i–j). To further investigate this trend, we performed weighted gene co-expression network analysis (WGCNA) on RNA-seq data from NPC, iMN, and MN stages to identify gene modules correlated with specific stages of differentiation and examine changes in these modules within each family (Supplementary Fig. 7a, Supplementary Data 6). These gene modules were enriched for differentiation stage-specific Gene Ontology (GO) terms such as cell division and chemical synaptic transmission (Supplementary Fig. 7b, Supplementary Data 6). The CRISPR deletion and GL_007 deletion lines exhibited significantly lower eigengene scores for the NPC-associated module (blue) at the NPC stage and significantly higher eigengene scores for the mature neuron-associated module (turquoise) at both the NPC and iMN stages (Supplementary Fig. 7c, Supplementary Data 6). These results suggest premature differentiation of NPCs in these lines. Furthermore, gene set enrichment analysis (GSEA) comparing deletion lines to HD or familial nondeletion lines across cell stages revealed significant enrichment for GABAergic neuron and DNA replication-related terms, as well as for signaling pathways such as PI3K-AKT, TGF-β, and Wnt (Fig. 3g, Supplementary Fig. 7d, Supplementary Data 4). While we were not able to disentangle the effects of the 16p12.1 deletion alone, these findings suggest that the deletion contributes to alterations in neurogenesis and associated pathways in a genetic background–specific manner.

CRISPR activation of individual 16p12.1 genes reverses transcriptomic and neural defects

We previously found that 16p12.1 genes show limited connectivity to each other within a brain-specific co-expression network, especially when compared to genes within the more penetrant 16p11.2 deletion^13,41,42. Therefore, we sought to investigate the role of individual 16p12.1 genes in driving distinct functional changes across genetic backgrounds. We designed single-guide RNAs (sgRNAs) targeted to the promoters of MOSMO, POLR3E, and UQCRC2 and used CRISPR-mediated transcriptional activation (dCas9-VP64 and MS2-P65-HSF1-mediated CRISPRa⁴³) to increase their expression in the proband lines (see Methods, Fig. 4a). We selected these genes as they showed significantly reduced expression in the deletion samples across neural differentiation cell states (Supplementary Fig. 2a). Previous studies have also implicated these genes in cellular functions such as Sonic hedgehog (Shh) signaling, immune response, and mitochondrial homeostasis^44–46. We confirmed sustained overexpression of the targeted genes during neural conversion to the NPC stage, the cell state in which most cellular effects were assessed in this study (Supplementary Fig. 8a). However, neurosphere formation failed for CRISPRa of MOSMO in P2C_079. This could be potentially due to a dosage-sensitive role of MOSMO in the genetic background of P2C_079 in regulating Shh signaling^47,48, which is known to influence neurosphere formation⁴⁹.

Fig. 4. CRISPR-mediated transcriptional activation of 16p12.1 genes across probands.

a Schematic of the CRISPRa approach (using dCas9-VP64 and MS2-P65-HSF1) in probands from the three families. b UpSet plot shows overlaps among DEGs (log₂|FC | ≥ 0.5, padj<0.05) for each comparison (CRISPRa sgRNAs of deletion genes vs. empty sgRNA-MS2 vector) in the three proband lines. c Quantification and representative images of NKX2.1-positive cells with CRISPRa of 16p12.1 genes in NPCs from P1C_007. **p = 0.0067; ***p = 0.00017; ns = 1. Scale bar, 20 μm. d Quantification and representative images of Ki-67-positive cells with CRISPR activation of 16p12.1 genes in NPCs from P1C_007. ****p = 5.23 × 10⁻⁰⁷; ***p = 0.00065; ns=1. Scale bar, 70 μm. e Quantification and representative images of Ki-67-positive cells with CRISPR activation of 16p12.1 genes in NPCs from P2C_079. Scale bar, 70 μm. ****p = 9.53 × 10⁻⁰⁷; ns=0.225. f Select pathways (q < 0.05) enriched by GSEA across different sets of DEGs (padj<0.05) annotated in the Reactome, WP, and KEGG databases. GSEA was performed for DEGs from deletion NPCs versus HD controls and DEGs from CRISPRa versus EV controls for each proband. NES, normalized enrichment score; M, pathways derived from CRISPRa_MOSMO DEGs; P, pathways derived from CRISPRa_POLR3E DEGs; U, pathways derived from CRISPRa_UQCRC2 DEGs. White color fill indicates absence of terms in GSEA results. g Heatmap of normalized gene expression in selected pathways (colored in (f)) across CRISPRa lines, clustered using Ward D2. Genes were selected from GSEA pathway databases. Color scale represents z-scores calculated from TPM for each group. EV, empty sgRNA-MS2 vector; M, CRISPRa of MOSMO; P, CRISPRa of POLR3E; U, CRISPRa of UQCRC2. HD refers to HD_01 and HD_02. Data presented in bar plots are mean±s.e.m. One-way ANOVA followed by Dunnett’s post hoc tests were used for (c), (d), and (e). see Supplementary Data 10 for all statistical test results. n = 3 independent experiments for each line including RNA-seq and downstream analyses. For (c), (d), and (e), n = 6 images per experiment were quantified (see Supplementary Data 10). Cartoons in (a) were created in BioRender. Girirajan, S. (2026) https://BioRender.com/1zw6tq0. Source data are provided as a Source Data file.

Using RNA-seq in NPCs, we identified DEGs (i.e., CRISPRa DEGs) comparing each CRISPRa line to its corresponding EV control. DEGs were primarily shared across different CRISPRa lines in the same proband, rather than across different probands when the same 16p12.1 gene was activated (Fig. 4b, Supplementary Data 7). Furthermore, CRISPRa of each of these genes did not change the expression of other 16p12.1 genes (Supplementary Fig. 8b, Supplementary Data 7). These findings suggest that CRISPRa-mediated restoration of 16p12.1 genes leads to outcomes modulated by the genetic background and that these deletion genes are not strongly interconnected, consistent with previous observations⁴².

We next investigated whether CRISPRa of individual 16p12.1 genes could reverse transcriptomic changes initially observed in probands with the deletion. By comparing CRISPRa DEGs to those from probands compared to HD controls, we identified 4,858 genes with opposing directions of gene expression change (i.e., reversed DEGs) (Supplementary Fig. 8c, Supplementary Data 7). For example, activating POLR3E reversed expression of 16 genes across all probands, including cytoskeleton-related (PLS3 and LIMA1), cell adhesion (PCDHA12, JAM2, and CD24), and signal transduction (ZYX, S1PR1, and PDE1A) genes (Supplementary Fig. 8d, Supplementary Data 7). Expression of specific phenotype-related genes were also reversed by activating distinct 16p12.1 genes. For example, the epilepsy risk gene NNAT⁵⁰ was upregulated in P2C_079 and P1C_007, who both have seizures, and was reversed with CRISPR activation of POLR3E in both probands. CRISPRa also reversed altered isoform usage; activating MOSMO restored isoforms of MECP2 (ENST00000627864) and EPS8L2 (ENST00000318562), genes that have been associated with autism and deafness, respectively, in P1C_077, who manifested these clinical features (Supplementary Data 1 and 7)^51,52.

Next, we investigated whether CRISPRa of 16p12.1 genes rescued cellular phenotypes. In P1C_007, overexpression of NKX2.1 was reversed with CRISPRa of POLR3E and MOSMO, while proliferation and apoptosis phenotypes were rescued by CRISPRa of UQCRC2 and MOSMO (Fig. 4c–d, Supplementary Fig. 9a). Additionally, overexpression of UQCRC2 led to enhanced premature differentiation of NPCs to neurons in this proband (Supplementary Fig. 9b). In P2C_079, CRISPRa of POLR3E rescued hyperproliferation phenotypes (Fig. 4e, Supplementary Fig. 9c). These results underscore the direct role of 16p12.1 genes towards the neural defects observed in deletion carriers, with some functional effects observed with multiple genes and others being gene specific.

We further investigated the pathways altered by CRISPRa of 16p12.1 genes and found enrichment for multiple signaling pathways such as TGF-β, PI3K-AKT, and Wnt (Supplementary Fig. 9d, Supplementary Data 7). Dysregulation of these pathways is known to disrupt cell-fate determination and cell-cycle progression^53–55, as observed in the deletion lines. Therefore, we performed GSEA on the genes with altered expression from CRISPRa lines to assess the direction of change in associated signaling pathways compared to EV controls (Fig. 4f, Supplementary Data 7). We found that CRISPRa of POLR3E and MOSMO led to the restoration of TGF-β signaling, corresponding to the rescue of the NKX2-1 expression level in P1C_007. Furthermore, activation of POLR3E also resulted in the upregulation of immune-related pathways, including interferon and cytokine signaling, which were downregulated in multiple deletion lines. Specific genes with altered expression following CRISPRa, such as STAT3 and TGFB3 within immune pathways, and FZD4, BMP7 and NOG within Wnt and TGF-β pathways, were also found to be dysregulated in these probands, potentially contributing to the reversal of proliferation and apoptosis phenotypes in NPCs⁵⁶ (Fig. 4g, Supplementary Data 7). Together, these results suggest that each 16p12.1 gene modulates distinct signaling cascades in a genetic background-dependent manner, thereby contributing to variable neurodevelopmental trajectories across individuals.

Connectivity of genes with secondary variants in PPI network

To further investigate the role of secondary variants in phenotypic differences across probands, we analyzed how genes with secondary variants whose expression was altered following CRISPRa were connected within a protein-protein interaction network⁵⁷ (Fig. 5a, Supplementary Fig. 10a, Supplementary Data 8). In the POLR3E and MOSMO CRISPRa lines, we observed a higher network connectivity for P1C_007 and P2C_079 compared to P1C_077, correlating with the presence of neural defects in NPCs of P1C_007 and P2C_079 but not P1C_077 (Fig. 5a). These findings were further validated by analysis of PPI networks specifically within signaling pathways, which revealed higher network connectivity in P1C_007 and P2C_079 (Fig. 5b, Supplementary Data 8). Interestingly, in the UQCRC2 CRISPRa lines, P1C_077 and P2C_079 showed higher connectivity compared to P1C_007 (Supplementary Fig. 10a–b, Supplementary Data 8). The highly connected nodes within this network included proteins such as ADCY1, PDE4A, and PRKAR1B, which are involved in cAMP signaling and G protein signaling pathways that modulate synaptic functions in mature neurons^58,59. This result may explain the absence of cellular phenotypes in P1C_077 at the NPC stage. These findings indicate that the extent to which modifiers of individual 16p12.1 genes are connected within signaling pathways is a determinant of phenotypic variation across probands.

Fig. 5. Connectivity of secondary variants across probands in response to CRISPR activation of 16p12.1 genes.

a Density plots show the connectivity (node degree, i.e., the number of connections a node has with other nodes in the network) of POLR3E and MOSMO CRISPRa DEGs (log₂|FC | ≥ 0.5, padj<0.05) that overlap with secondary variants, as measured using the STRING database. Anderson-Darling k-sample test was used to calculate p– values. b Visualization of the PPI network of DEGs (log₂|FC | ≥ 0.5, padj<0.05) that overlap secondary variants following CRISPRa of POLR3E in P1C_077 (top left), P1C_007 (top middle), and P2C_079 (top right), as well as by CRISPRa of MOSMO in P1C_077 (bottom left) and P1C_007 (bottom right). DEGs shown were selected based on their involvement in signaling pathways annotated in the Reactome database. Protein nodes involved in specific pathways are color-coded based on annotations in the STRING database. Average node degrees and PPI enrichment p– values were also calculated using the STRING database. n = 3 independent experiments for each line (see Supplementary Data 10).

Molecular convergence through modulation of FOXG1 and JUN

We sought to identify functional convergence across samples through which the deletion confers disease susceptibility, modulation of which by secondary variants lead to diverse outcomes. As networks of genes within signaling pathways typically converge on TFs⁶⁰, we sought to identify key TFs that regulate downstream effects in the deletion lines. Sequences within Diffpeak regions across the deletion NPCs were enriched for binding motifs of TFs involved in developmental processes, such as FOS, HOXD13, and NR4A1^61–63 (Supplementary Data 9). Furthermore, co-regulatory networks built from TFs with binding motifs enriched in Diffpeak regions across deletion lines identified FOXG1 and JUN as key regulatory hubs, as ranked by ChEA3 (see Methods, Fig. 6a, Supplementary Data 9). Both TFs have been previously implicated in brain development^64,65. Furthermore, we observed altered expression of FOXG1 and JUN, along with distinct changes in gene expression and chromatin accessibility within their regulatory networks across probands (Fig. 6a, Supplementary Fig. 11A–B, Supplementary Data 9). These included genes such as HES1, KLF6, DLX2, and FZD8, which are known to play roles in neural processes such as inhibitory neuron production and NPC proliferation^66–69.

Fig. 6. Modulation of gene regulatory hubs corresponding with variable phenotypes.

a TF co-regulatory network built by ChEA3 using TFs with binding motifs enriched at ATAC-seq Diffpeak regions from comparison of deletion lines with HD lines at the NPC stage. A list of motifs is provided in Supplementary Data 9A-D. The panel below depicts IGV snapshots of ATAC-seq peaks showing increased (DLX2) or decreased (KLF6) chromatin accessibility for select genes in the regulatory network of FOXG1 and JUN, respectively. Quantification and representative images of (b) FOXG1-positive NPCs and (c) JUN-positive NPCs in CRISPR activated lines from P1C_077, P1C_007, and P2C_079. Scale bars, 20 μm. Data presented in bar plots are mean±s.e.m and one-way ANOVA followed by Dunnett’s post hoc test. For FOXG1, P1C_077: EV/M, ****p < 0.0001; EV/P, ****p = 4.12 × 10⁻¹²; EV/U, ns = 0.993, P1C_007: EV/M, ns = 0.979; EV/P, ***p = 0.00016; EV/U, ***p = 0.00043. For JUN, P1C_077: EV/M, ns = 0.628; EV/P, ns = 0.772; EV/U, ***p = 0.00046, P1C_007: EV/M, ****p = 2.01 × 10⁻¹²; EV/P, ****p = 2.17 × 10⁻⁰⁶; EV/U, ns= 0.36, P2C_079: EV/P, ****p = 3.33 × 10⁻¹⁶; EV/U, ns= 0.38. (see Supplementary Data 10 for all statistical test results). 6 images per experiment were quantified. d Overview of CRISPRa results at the NPC stage. Columns represent CRISPRa lines from the three probands; rows indicate the direction of change based on phenotype quantification. Empty bubbles denote no detectable change. Rescue of cellular phenotype refers to the reversal of a cellular phenotype observed in the proband lines (compared to HD controls). The phenotypes of proliferation, apoptosis, inhibitory neuron, and premature maturation were, respectively, measured with EdU, TUNEL assay, VGAT and NKX2.1, and TUBB3-positive cells. e Representative heatmap of normalized expression of genes within the regulatory networks (identified using ChEA3) of FOXG1 (left) and JUN (right) across the CRISPRa lines. Color scale represents the z-scores calculated from TPM within each group. In all panels, n = 3 independent experiments for each line (see Supplementary Data 10). EV, empty sgRNA-MS2 vector; M, CRISPRa of MOSMO; P, CRISPRa of POLR3E; U, CRISPRa of UQCRC2. Source data are provided as a Source Data file.

Based on these findings, we hypothesized that the 16p12.1 deletion affects regulatory hubs within signaling pathway interaction networks, and that modulating the expression of individual 16p12.1 genes would, in turn, alter the expression of these hub genes, including FOXG1 and JUN, as well as their connected genes, leading to changes in cellular phenotypes (Fig. 7). To test this, we further investigated the roles of FOXG1 and JUN in the observed cellular dysregulations in the deletion lines. Since altered activity of the TFs reflects perturbations across multiple associated signaling pathways, we did not expect a direct one-to-one correspondence between specific TF activity and cellular phenotypes. Expression levels of FOXG1 and JUN varied across probands, and these levels were differentially modulated with CRISPR activation of different 16p12.1 genes (Fig. 6b–c). Changes in FOXG1 and JUN expression in the P1C_007 CRISPRa lines and changes in JUN expression in the P2C_079 lines were observed alongside the reversal of cellular phenotypes upon CRISPR activation (Fig. 4, Fig. 6d). We also found that the expression of genes within their regulatory networks, including DLX2 and KLF6, mirrored changes in FOXG1 and JUN expression across CRISPRa lines (Fig. 6e). These findings suggest that individual 16p12.1 gene-associated signaling pathways converge on to key TFs and their regulatory networks that mediate the phenotypic trajectories of the deletion (Fig. 7).

Fig. 7. Conceptual model of the interplay between the 16p12.1 deletion and secondary variants contributing to variable expressivity.

Interactive effects of the deletion and secondary variants are mediated through regulatory hubs, such as TFs (e.g., FOXG1, JUN), within signaling pathways. Bubbles represent genes with gray bubbles indicating 16p12.1 deletion genes. Other genes are colored for the TF regulatory hub, and two-toned colors indicate genes shared among multiple pathways. Patient-specific secondary variants or combinations of variants lead to distinct phenotypic trajectories, shown by the red or blue connecting lines traversing distinct genetic pathways.

Discussion

Our study provides a framework to understand how variable expressivity in neurodevelopmental disorders arises from the complex interplay between disease-associated variants and secondary variants in the genetic background. Using the 16p12.1 deletion as a model, we found that primary and secondary variants jointly modulate key signaling pathways and shape distinct phenotypic trajectories through transcription factors within gene regulatory networks (Fig. 7). Several themes have emerged from our work contributing to the emerging picture that genetic interactions underlie variable expressivity in complex genetic disorders.

First, previous studies using patient iPSC models for disease-associated CNVs reported changes in chromatin accessibility and gene expression^70,71. We find that nonadditive effects of secondary variants can contribute to these changes in the context of the 16p12.1 deletion, both in family-derived lines and in an isogenic line generated from an apparently healthy control. Our results suggest that it may not be possible to fully isolate the functional effects of disease-associated variants from the influence of genetic background. While the roles of genetic background and individual modifier genes have been extensively studied^72,73, we find that secondary variants collectively exert network-level effects to alter cellular phenotypes. Thus, our study highlights the need to rigorously account for genetic background effects to functionally characterize disease-associated variants. This can be achieved by introducing the same variant into diverse genetic backgrounds^74,75, and by evaluating carriers and noncarriers from multiple families.

Second, we did not observe the same cellular phenotypes and the same set of DEGs shared across all deletion lines. However, patterns of similarity emerged among deletion carriers within a family. Within specific families, the 16p12.1 deletion led to changes in inhibitory neuron production, abnormal cell proliferation and apoptosis, and premature neuronal differentiation. Some cellular phenotypes were shared between patient-derived deletion lines and the CRISPR deletion line, such as increased expression of NKX2.1, a ventral telencephalic marker. These findings indicate disruptions in excitatory/inhibitory balance and neurogenesis, leading to altered cell type proportions during brain development, a hallmark of neurodevelopmental disorders^19,38,76. Further studies are needed to dissect the cellular basis of clinical variability across family generations and disease ascertainment in deletion carriers.

Third, to better identify the contribution of specific genes within the 16p12.1 deletion to these varied cellular phenotypes in patient lines, we performed CRISPR activation of MOSMO, UQCRC2, and POLR3E in proband lines. These experiments supported our previous findings from individual and pairwise knockdown of Drosophila homologs⁴²; each gene within this region functions independently, as rescue of different 16p12.1 genes reversed distinct sets of dysregulated genes and cellular phenotypes without affecting the expression of other 16p12.1 genes. We further observed alterations across multiple signaling pathways across differentiation stages and families, including Wnt, TGF-β, PI3K-AKT, and immune-related pathways, which have been implicated in neurodevelopmental disorders^54,77–80. These findings reflect the extensive crosstalk and network-level interconnectivity among these pathways during neural differentiation⁷⁷.

Fourth, we found that the severity of phenotypic outcomes correlated with the PPI connectivity of secondary variants in signaling pathways. Consistent with prior studies demonstrating convergent mechanisms underlying the heterogeneous etiology of autism⁸¹, we found differential modulation of transcription factors such as FOXG1 and JUN, which function as regulatory hubs across deletion lines with distinct genetic backgrounds. Notably, FOXG1 and JUN are known to influence neural development^38,82, providing a mechanistic link between the altered network connectivity and the observed neural phenotypes. As central regulators of cellular function, TFs are sensitive nodes within molecular networks, and their dysregulation has been linked to developmental disorders^83,84. For instance, targeting distinct network hubs associated with 16p11.2 duplication have been shown to rescue specific cellular and behavioral phenotypes^85,86. Together, our results underscore the importance of examining both inter-individual variability and the convergent mechanisms that emerge across diverse genetic contexts.

While our study provides key insights into the functional impact of secondary variants, several aspects could be further refined or expanded in future work. First, we observed strong associations between secondary variants and changes in gene expression and chromatin accessibility in both isogenic and familial settings that fit the proposed model of nonadditive interaction. However, the small sample size and use of a single clone per sample limits our ability to infer causality without further experimental validation. A larger sample size would also allow us to investigate potential interactive effects of the deletion with common variants, including PRS, as well as specific rare variant classes such as noncoding mutations on gene expression and cellular phenotypes. For instance, nonadditive functional effects have been reported for multiple schizophrenia-associated common variants in iPSC models^87,88. Although the CRISPRa assays established a link between deletion genes and cellular phenotypes, the specific contributions of a secondary variant or combinations of variants to altered signaling pathways, regulatory networks, and cellular defects require further investigation. High-throughput approaches such as Perturb-seq and CROP-seq could help confirm these interactions and elucidate how they modulate signaling pathways^89,90. In addition, CRISPR-mediated correction of secondary variants of interest could help validate their contributions to cellular defects⁹¹. Second, we used 2D iPSC differentiation combined with bulk RNA-seq, which may overlook heterogeneous cellular effects, particularly in neuronal subtypes and non-neuronal brain cells relevant to neurodevelopmental disorders⁹². For example, NKX2.1 is known to regulate both inhibitory neuron differentiation and gliogenesis⁹³, thus, model systems with greater cellular diversity could reveal whether these processes are also altered in 16p12.1 deletion lines exhibiting NKX2.1 overexpression. Future work could leverage advanced models such as neural organoids, assembloids, or chimeroids, which provide greater cellular diversity and spatial resolution⁹⁴. Third, we found alterations of TF hubs across deletion lines that correlated with the cellular defects. The causal links between specific TFs and cellular phenotypes or the deletion genes can be established with further experimental validation using genetic manipulation or small molecules targeting the hubs^84–86.

In summary, we provide insights into understanding genotype–phenotype relationships that would help devise personalized therapeutic strategies for these disorders.

Methods

Patient recruitment and clinical phenotype analysis

Patient recruitment and clinical phenotyping were performed as previously described¹⁴. Informed consent authorizing the use and disclosure of protected health information for research purposes was obtained from families recruited directly, and de-identified samples and clinical data were obtained from families recruited through clinics, according to protocols STUDY00000278 and STUDY00017269, respectively, approved by the Pennsylvania State University Institutional Review Board. Detailed medical histories, including clinician-reported, guardian-reported (for children), or self-reported (for adults) information were collected, and standardized questionnaires were administered to assess developmental phenotypes in children and psychiatric features in adults. Questionnaires for children assessed neuropsychiatric and developmental features, anthropometric measurements, congenital anomalies across multiple organ systems, and family history of medical or psychiatric conditions. Assessment of developmental milestones followed the CDC guidelines⁹⁵. The study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Detailed phenotypic data are provided in Supplementary Data 1.

Induced pluripotent stem cell reprogramming and maintenance

Induced pluripotent stem cells (iPSCs) from 12 individuals (seven male and five female participants) were reprogrammed from peripheral blood mononuclear cells using the Cytotune-iPS 2.0 Sendai Programing kit, which includes the four Yamanaka reprogramming factors POU5F1 (OCT4), SOX2, KLF4, and MYC (ThermoFisher Scientific), as described previously⁹⁶. iPSCs were validated by flow cytometry (EPCAM, TRA-1-81, SSEA4 and CD9) and immunofluorescence (NANOG, OCT4, SOX2 and SSEA4). HD_01 and HD_02 were derived from CD34+ cord blood from healthy male and female donors, respectively (BID00274 and BID00271, NINDS). These lines were cultured in mTESR1 (catalog #85850, STEMCELL Technologies) or mTESR Plus medium (catalog #100-0276, STEMCELL Technologies), supplemented with 1% penicillin/streptomycin (catalog #P4333 Sigma-Aldrich) on Geltrex-coated dishes (catalog #A1413302, Gibco. Cells were passaged using 0.5 mM EDTA (#AM9260G Invitrogen) or ReLeSR (catalog #100-0483, STEMCELL Technologies), with ROCK inhibitor (Y-27632) added to improve cell survival (catalog #72304, STEMCELL Technologies). Cells were maintained at 37 °C and 5% CO₂. Mycoplasma testing was performed on iPSCs and NPCs using a mycoplasma PCR detection kit (catalog #MP0035, Sigma-Aldrich).

Generation of CRISPR/Cas9-mediated 16p12.1 deletion iPSC line

The sequences of the two sgRNAs are: TCGGTGCTTAGGATCAGCCT and GCCACTAGCTGACATGGTTG. Two separate vectors were used to deliver sgRNAs: pSpCas9(BB)-2A-Puro (PX459) V2.0 (a gift from Feng Zhang; Addgene plasmid #62988) and pGH020_sgRNA_G418-GFP (a gift from Michael Bassik; Addgene plasmid #85405). The two sgRNA-containing vectors were transfected into the HD_01 line using Lipofectamine Stem Transfection Reagent (#STEM00001, Invitrogen). A CRISPR control line was generated by transfecting empty vectors. Puromycin selection (0.5 ng/ml) was initiated 24 h after transfection and continued for an additional 24 h, after which the selection medium was replaced with mTESR1 medium. The cells were allowed to recover for several days before being diluted and manually distributed into 96-well plates to isolate single-cell clones. Individual clones were cultured for two weeks and genotyped using the following primer sets: WT-F: GACTTCCTCCACATCTTCCTCTA, WT-R: TCAAATAGAGGGGCAGGAGC (546 bp product); Deletion-F: TCCTCAGACTCAATAATTGCCA, Deletion-R: TGACCTTTACTCTGTGACATTGC (942 bp product). PCR products of the expected sizes were purified, and Sanger sequencing was performed to confirm the target sequences. For further conformation, genomic DNA from selected clones was extracted using the GenElute Mammalian Genomic DNA Miniprep kits (#G1N350, Sigma-Aldrich) and PureLink Genomic DNA Mini kit (#K182001, Invitrogen), and the deletion was confirmed using SNP-based microarray.

RNA isolation and RT-qPCR

Total RNA was isolated using the TRIzol reagent (#15596026, Invitrogen) and the PureLink RNA mini kit (#12183018 A, Invitrogen), following a modified version of the manufacturer’s protocol. Cells were collected directly in TRIzol and stored at −80 °C until RNA extraction. Chloroform was added to TRIzol at a 1:5 ratio, samples were shaken vigorously for 30 s, incubated for 3 min at room temperature, and centrifuged at 12,500 × g for 15 min at 4 °C. The aqueous (upper) phase was transferred to a fresh tube, mixed with an equal volume of ethanol, and vortexed. The mixture was loaded onto to a spin column, and washing and elution steps were performed according to the kit protocol. Isolated RNA was treated with TURBO DNase (catalog #AM1907, Thermo Fisher Scientific) to remove residual genomic DNA. RT-qPCR was performed in three independent biological experiments, each with three replicates per sample. For reverse transcription, 1000 ng of RNA was converted to cDNA using the qScript cDNA Synthesis Kit (catalog#95047-100, Quantabio). The resulting cDNA was diluted 1:100 and 2 µl was used in a 10 µl RT-qPCR reaction containing PowerTrack SYBR Green Master Mix (catalog#A46109, Applied Biosystems). Reactions were run on a QuantStudio 3 Real-Time PCR system (Applied Biosystems) using the following cycling conditions: initial denaturation 95 °C for 2 min, followed by 40 cycles of denaturation at 95 °C for 15 s and annealing or elongation at 60 °C for 1 min. Melt curve analysis was conducted to confirm specificity of amplification. The results were analyzed using QuantStudio Design & Analysis software. The following primers were used: GAPDH-F: ATGGGGAAGGTGAAGGTCGG, GAPDH-R: TGACGGTGCCATGGAATTTG; POLR3E-F: GGAGCAGATTGCGCTGAA, POLR3E-R: TTACTGGTGGTCTGGGAAGA; UQCRC2-F: TTCAGCAATTTAGGAACCACCC; UQCRC2-R: GGTCACACTTAATTTGCCACCAA; MOSMO-F: CTGTCACATGTGGTTTGCTGG; MOSMO-R: GGGCAGCCATACAGAAAAGGA. Gene expression levels were normalized to GAPDH as an internal housekeeping control to account for variation in RNA input and cDNA synthesis efficiency. Relative mRNA expression of each 16p12.1 gene, normalized to GAPDH, in proband lines transfected with CRISPRa gRNAs was calculated using the comparative Ct (ΔΔCt) method, with empty vector (EV)–transfected proband lines serving as controls. Data were presented as mean ± s.e.m. across biological replicates.

Neural conversion of iPSCs to NPCs

The neural conversion of iPSCs was performed as previously described with modifications¹⁸. Each iPSC line was differentiated in three replicates and grown separately. On day 0, iPSCs were treated with Collagenase Type IV (catalog #07909, STEMCELL Technologies). Cells were then scraped and transferred to an uncoated 60 mm dish containing mTESR1 media supplemented with 10 µM Y-27632 for embryoid body (EB) formation. After 48 h in suspension, mTESR1 media was replaced with neural differentiation media (N2 media) containing DMEM/F-12 (catalog #10565018, Gibco), 1% N-2 Supplement (catalog #17502001, Gibco), 1% MEM nonessential amino acids (catalog #11140050, Gibco), 2 µg/ml Heparin (catalog #07980, STEMCELL Technologies), 1% penicillin/streptomycin (catalog #P4333, Sigma), 5 µM SB431542 (catalog #72232, STEMCELL Technologies), and 0.25 µM LDN193189 (catalog #72147, STEMCELL technologies). EBs were cultured in neural differentiation media for 4 days, with media changes every 2 days. On day 6, EBs were seeded on Geltrex-coated plates and cultured for 8 days to form rosettes. From days 6-8, rosettes were maintained in N2 media without dual-SMAD inhibitors, and from days 8–14, they were cultured in N2B27 media containing a 1:1 mixture of DMEM/F-12 and Neurobasal Medium (catalog #21103049, Gibco), with GlutaMax supplement (catalog #35050061, Gibco), 1% N-2 Supplement, 2% B-27 supplement minus vitamin A (catalog #12587010, Gibco), 1% MEM nonessential amino acids, 2 µg/ml Heparin, and 1% penicillin/streptomycin. On day 14, rosettes were treated with Collagenase Type IV and transferred to uncoated 60 mm dish containing N2B27 media for neurosphere formation in suspension culture for 12 days. On day 26, neurospheres were dissociated by Accutase (catalog #A1110501, Gibco) plated onto Geltrex-coated plates. Neural progenitor cells were maintained in Stemdiff Neural Progenitor Medium (catalog #05833, STEMCELL Technologies) and passaged at a 1:2-1:3 ration using Accutase. This protocol is expected to create a population of NPCs expressing SOX2 and NESTIN, with the majority of cells expressing PAX6.

Neuronal differentiation of NPCs to neurons

Plates were prepared for neuronal differentiation by coating with Poly-D-Lysine (PDL, catalog #A3890401, Gibco) and Laminin (catalog #L2020, Sigma-Aldrich). PDL was diluted 1:1 with phosphate buffered saline (PBS) to a final concentration of 50 µg/ml and added to the wells for 1 h at room temperature. The PDL solution was then removed, and wells were washed 4–5 times with sterile deionized water and allowed to dry for 15 min. Laminin was diluted to 20 µg/ml in chilled DPBS and added to the wells. After characterization, NPCs were plated 800,000 cells/per well in 6-well plates. The following day, 0.2 µM Compound E (catalog#73952, StemCell Technologies) was added to the media. During the first two weeks, cells were maintained in neuronal differentiation medium containing neurobasal medium, 1% GlutaMAX, 1% penicillin/streptomycin, 1% N-2 supplement, 2% B-27 supplement (catalog #17504044, Gibco), 0.2% MycoZap plus-PR (catalog #195263, LONZA), along with 20 ng/ml BDNF (catalog #450-02-1 mg, PeproTech), 20 ng/ml GDNF (catalog #450-10-1 mg, PeproTech), 200 µM Ascorbic acid (catalog #72132, STEMCELL Technologies), 1uM cAMP (catalog #A9501), and 1 µg/ml Laminin. During the final two weeks, the neuronal differentiation medium was switched to BrainPhys neuronal medium (catalog #05790, STEMCELL Technologies), supplemented with 40 µM 5-Fluoro-2′-deoxyuridine (FUDR) (catalog #F0503, Sigma-Aldrich) to eliminate mitotic cells. Half medium changes were performed every 3 days.

Cell proliferation and apoptosis

Proliferation and apoptosis assays were adapted from a previous study¹⁸. On day 0, passage 6 NPCs were seeded at 1 × 10⁵ cell per coverslip onto 18 mm diameter coverslips coated with PDL (1:1 in PBS) and Geltrex (1:100 in DMEM/F12) in 12-well cell culture plates. Cells were maintained overnight in neuronal differentiation medium containing Neurobasal media, 1% N-2 supplement, 2% B-27 supplement minus vitamin A, 1% GlutaMax, and 1% penicillin/streptomycin. On the morning of day 1, neural progenitor cells were treated with 10 µM 5-ethynyl-2’-deoxyuridine (EdU) (catalog#C10337, Invitrogen) for 4 h, after which the medium was replaced with fresh medium lacking Edu. Coverslips were collected and fixed at three timepoints: no chase (0 h), 24-h chase, and 48-hour chase. The Edu incorporation was detected using Click-iT chemistry according to the manufacturers’ instructions. At the 24-h chase timepoint, coverslips were also immunostained with Ki-67 markers (1:200 catalog #12202, Cell Signaling Technology). Apoptosis was assessed using a TUNEL assay performed using Click-iT Plus TUNEL Assay kit (catalog #C10617, Invitrogen), on cells cultured for five days.

Immunostaining of iPSCs and NPCs

NPCs at passage 6 or iPSC clones were seeded on coverslips coated in PDL (1:1 in PBS) and Geltrex (1:100 in DMEM/F12) in 6-well plates. Three replicates in separate wells were used for each line. Media was aspirated, and cells were washed three times with PBS. Cells were fixed in 4% paraformaldehyde (PFA) (catalog #158127, Sigma-Aldrich) for 15 min, washed three times with PBS, and then permeabilized in 0.2% Triton X-100 (catalog #X100, Sigma-Aldrich) in PBS for 15 minutes. Blocking was performed using a 2% BSA (catalog #A7906, Sigma-Aldrich), 0.2% Triton X-100, and 5% goat serum (catalog #G9023, Sigma-Aldrich) in PBS for 1 hour at room temperature or overnight at 4 °C. Coverslips were incubated with primary antibodies for 2 h at room temperature or overnight in the 4 °C. Coverslips were then washed three times with PBS (5 min each), and incubated with the secondary antibody for 2 h at room temperature. After three additional washes, coverslips were mounted using DAPI antifade mounting reagent (catalog #P36931, Thermo Fisher). All antibodies were diluted in blocking buffer. Primary antibodies used were: SOX2 (1:200, catalog #3579S Cell Signaling Technology); NESTIN (1:800, catalog #33475S Cell Signaling Technology); NKX2.1 (1:200, catalog #MAB5460, Millipore Sigma); PAX6 (1:800, catalog #60433S, Cell Signaling Technology); TUBB3 (1:400, catalog #5568 T, Cell Signaling Technology); FOXG1 (1:400, calalog #29642S, Cell Signaling Technology); c-JUN (1:200, catalog #9165 T, Cell Signaling Technology); Ki-67 (1:200, catalog #9129 T, Cell signaling Technology); SSEA4 (1:200, catalog #4755 T, Cell Signaling Technology); NANOG (1:500, catalog #4893 T, Cell signaling Technology); OCT4 (1:200, catalog #2750S, Cell Signaling Technology). Secondary antibodies used were: Anti-mouse IgG (H + L), F(ab’)2 Fragment (1:1000, catalog #4408S, Cell Signaling Technology); Anti-rabbit IgG (H + L), F(ab’)2 Fragment (1:1000, catalog #8889S, Cell Signaling Technology).

Immunostaining of neurons

Neurons were differentiated for 28 days then fixed by adding 20% PFA directly to the culture media to achieve a final concentration of 4%, followed by incubation at room temperature for 15 min. PFA was removed, and cells were washed once with PBS for 5 min. Cells were permeabilized with 0.2% Triton X-100 in PBS for 15 min at room temperature. After permeabilization, cells were blocked for 1 h at room temperature in a solution containing 2% BSA, 0.2% Triton X-100, and 5% goat serum in PBS. Coverslips were incubated with primary antibodies for 2 h at room temperature or overnight at 4 °C. Coverslips were washed with PBS twice for 5 min each and incubated with secondary antibodies for 2 h at room temperature. After two additional 5-min washes with PBS, coverslips were mounted using DAPI antifade mounting reagent. All antibodies were diluted in blocking buffer. Primary antibodies used were MAP2 (1:500, catalog #4542S, Cell Signaling Technology); VGAT (1:250, Synaptic Systems); VGLUT1 (1:500, Synaptic Systems); and NeuN (1:400, catalog #94403S, Cell Signaling Technology). Secondary antibodies used were Anti-mouse IgG (H + L), F(ab’)2 Fragment (1:1000, catalog #4408S, Cell signaling Technology); Anti-rabbit IgG (H + L), F(ab’)2 Fragment (1:1000, catalog #8889S, Cell signaling Technology).

Imaging and quantification

Revolve- R4 (Discover Echo) was used to detect Click-iT signals. An LSM 800 upright and LSM 800 inverted microscope (ZEISS) were used to detect fluorescence signals for immunofluorescence assays. Images were analyzed using Fiji 2.17.0 (ImageJ). For VGAT analysis, fluorescence intensities of the VGAT and MAP2 channels were measured in ImageJ using the mean gray value. The VGAT signal was normalized to MAP2 by dividing the VGAT intensity by the MAP2 intensity. Key VGAT findings were validated by puncta counting. Briefly, five neurites per image were selected, puncta along each neurite were counted, neurite length was measured, and the puncta count was divided by neurite to obtain normalized values. The five normalized puncta values were then averaged for each image.

Whole genome sequencing (WGS) sample processing and analysis

Genomic DNA from five iPSC samples (HD_01, HD_02, CRISPR deletion, P2C_079, and FNC_079) were isolated using GenElute Mammalian Genomic DNA Miniprep Kit (#G1N79-1KT, Sigma-Aldrich). DNA was fragmented, end-polished, A-tailed, and ligated with full-length adapters for Illumina sequencing, followed by size selection. PCR amplification was performed unless libraries were specified as PCR-free. Purification was carried out using the AMPure XP system. Libraries were assessed on the Agilent Fragment Analyzer System and quantified to 1.5 nM through Qubit and qPCR. WGS was performed on the Illumina Novaseq X platform by Novogene (Durham, NC, USA). For an additional 22 participants, DNA was isolated from blood samples, and Illumina TruSeq DNA PCR-free libraries were prepared for 150 bp paired-end WGS using Illumina HiSeq X by Macrogen Labs (Rockville, MD, USA), as previously described¹⁴. Samples were sequenced at an average 37.9X coverage, with an average of 810 million reads per sample and 98.0% of reads mapping to the human genome.

We followed the GATK Best Practices pipeline⁹⁷ to identify single nucleotide variants (SNVs) and small indels using GATK v.4.5.0. Adapter sequences were first marked using GATK’s MarkIlluminaAdapters and reads were aligned to the GRCh38 reference genome using BWA v.0.7.173⁹⁸. Duplicate reads were removed using the Picard’s MarkDuplicates and base quality score recalibration was performed using the BaseRecalibrator and ApplyBQSR. Variants were called for each sample using HaplotypeCaller, merged into a single GVCF, and joint genotyping was performed with GenotypeGVCFs. Variants were then annotated using the Python API for Hail and Ensembl Variant Effect Predictor (VEP) v.109⁹⁹. After splitting multi-allelic sites, variants were filtered for those with (i) ≥90% call rate, (ii) Hardy-Weinberg equilibrium p ≥ 10^-15, (iii) read depth ≥8 across all samples, and (iv) allele balance ≥0.2 in at least one sample. Rare SNVs and indels were further filtered for gnomAD v.2.17¹⁰⁰ frequency <0.1%. Variant consequences were annotated based on VEP transcript consequences, and filtered to include coding variants such as LOF (“transcript_ablation”, “stop_gained”, “frameshift_variant”, “stop_lost”, “start_lost”), missense (“missense_variant”), and splice LOF (“splice_acceptor_variant”, “splice_donor_variant”), splice (“splice_donor_5th_base_variant”, “splice_region_variant”, “splice_donor_region_variant”, “splice_polypyrimidine_tract_variant”), as well as noncoding variants such as upstream (“upstream_gene_variant”), downstream (“downstream_gene_variant”), 5’ UTR (“5_prime_UTR_variant”), 3’ UTR (“3_prime_UTR_variant”), or intronic (“intron_variant”). We used dbNSFP8–10 v.4 annotations to retain missense variants predicted to be deleterious by at least five of nine selected tools (SIFT¹⁰¹, LRT¹⁰², FATHMM¹⁰³, PROVEAN¹⁰⁴, MetaSVM¹⁰⁵, MetaLR¹⁰⁵, PrimateAI¹⁰⁶, DEOGEN2¹⁰⁷, and MutationAssessor¹⁰⁸). Variants were finally restricted to those present in a single individual or those private to related individuals to account for any technical differences between our data and gnomAD.

To identify common SNPs associated with gene expression changes, isoform usage, or chromatin accessibility, we gathered expression quantitative trait loci (eQTLs) data from GTEx v10 (brain cortex) and PsychENCODE (DER-08a_hg38_eQTL.significant)^30,31. We also gathered isoform quantitative trait loci (isoQTLs) based on cortex brain expression from PsychEncode (DER-10a_hg38_isoQTL.significant), and chromatin accessibility QTLs (cQTLs) from PsychEncode (DER-09_hg38_cQTL.significant). Loci meeting the same variant quality criteria described above were extracted. The regression slopes derived from these databases were used to assess whether the direction of gene or transcript expression changes and chromatin accessibility changes were concordant with our data.

STR loci were initially identified using GangSTR v2.5.0¹⁰⁹, followed by quality control and filtering with DumpSTR v6.0.1. STRs with low-quality calls and those with read depths outside the range of 20 to 1000 were excluded using the --gangstr-min-call-DP and --gangstr-max-call-DP parameters. Additional quality control steps included retaining only spanning and bounding reads (--gangstr-filter-spanbound-only) and filtering regions with poorly estimated confidence intervals (--gangstr-filter-badCI). Individual VCF files were then merged into a multi-sample VCF using MergeSTR v.6.0.1. After merging calls, loci were filtered again to retain only those with (i) Hardy-Weinberg equilibrium p ≤ 10⁻⁵, (ii) locus call rate ≥80%, and (iii) loci not overlapping segmental duplication regions as annotated in the UCSC Genome Browser^110–112. Loci were additionally restricted to include only those present in EnsemblTR (https://github.com/gymrek-lab/EnsembleTR) and with at least one sample carrying a non-reference allele. GangSTR was then rerun jointly across all samples for this filtered set of loci.

To identify rare STRs, summary statistics from EnsemblTR were calculated using StatSTR v.6.0.1¹¹³. Alleles at STR loci present in more than five individuals in our cohort were removed. Rare alleles were further filtered for those with frequency <1% in EnsemblTR or representing expansions ≥2 standard deviations above the EnsemblTR mean. Genic consequences of STRs were annotated using Func.wgEncodeGencodeBasicV38 from ANNOVAR¹¹⁴ and variants were classified as exonic (“exonic”), splice (“splicing”), intronic (“intronic”), 5’ or 3’ UTR (“UTR5” and “UTR3”, respectively), upstream (“upstream”), and downstream (“downstream”) STR variants.

CNVs were called using CNVpytor v.1.3.1¹¹⁵ with bin size of 500. Rare CNVs were defined as those with a population frequency of <1% based on gnomAD SV v.4.1 using 50% reciprocal overlap. CNVs overlapping low-confidence of the genome ( > 50% reciprocal overlap) identified from the UCSC Genome Browser were removed¹¹⁰. These regions included (i) segmental duplication regions^111,112, (ii) centromeres¹¹⁶, (iii) and other problematic regions for processing short read sequencing data, including the UCSC Unusual Regions, ENCODE Blacklist, and GRC Exclusions^98,117–120. We additionally removed CNVs on the X and Y chromosomes, as well as low-quality CNVs, defined as those with (i) q0 > 0.5, (ii) pN>0.5, (iii) e_va11 > 10^-5, or (iv) <1500 bp ( < 3 bins). Adjacent CNVs within each sample were merged if they were separated by <5 kbp and the gap was <20% of the combined CNV length. We retained CNVs that overlapped exons, introns, or 5’ or 3’ UTRs or 1 kbp up- or downstream of genes, as defined by GENCODE (GRCh38). Finally, CNVs present in ≤ five individuals in our cohort (based on 50% reciprocal overlap) were removed. For the isogenic CRISPR deletion and HD_01 samples, only rare variants (SNVs, STRs, and CNVs) and QTLs (eQTLs, sQTLs and caQTLs) present in both lines were used for downstream analyses. DEGs, altered isoform usage, and Diffpeaks overlapping with QTLs or unique variants in the isogenic setting were excluded from downstream analyses. For kinship calculations, we used somalier¹²¹ to calculate genetic relatedness between all pairs of samples.

RNA-sequencing sample processing and analysis

Each iPSC line was grown in three replicates, which were differentiated separately. RNA was extracted from iPSCs, neural progenitor cells (NPCs; passage 5), and neurons differentiated from NPCs at day 10 (immature neurons, iMNs) and day 28 (mature neurons, MNs). Only samples with RNA integrity number (RIN) ≥ 6.0, as measured by the Agilent TapeStation 4200 (Agilent Technologies), and purity (A260/280 ratio >1.8) were used for RNA sequencing. RNA sequencing libraries were prepared using the NEBNext Ultra II RNA library Prep kit (#E7770S, NEB) for Illumina according to the manufacturer’s instructions. Paired-end 150 base pair (bp) sequencing was performed using the Illumina NovaSeq 6000 platform by Genewiz (Azenta Life Sciences, South Plainfield, NJ) targeting 30 million reads per sample. Adapter trimming and removal of low-quality reads were performed using Trimmomatic v0.39¹²² with parameters: leading:3, trailing:3, slidingwindow:4:15, and minlen:36. Transcript abundances were quantified using Kallisto v.050.0¹²³ with n = 100 bootstrap samples, and the Kallisto index was built from hg38 cdna. Differential gene expression analysis was conducted using DEseq2 v1.44 package in R v.4.3.1¹²⁴, and catch effects were corrected using Combat-seq from sva v.3.35.2 package in R v.4.3.1¹²⁵. To ensure the correct identity of each sample, deletion carrier status was confirmed by expression of 16p12.1 genes, including UQCRC2, POLR3E, MOSMO, CDR2, EEF2K, PDZD9, and sex was confirmed using sex-specific genes such as SRY and RPS4Y.

Detection of alternative isoform usage

Alternative isoform usage was analyzed using the IsoformSwitchAnalyzeR v.2.4.0¹²⁶ package in R. Abundance files from Kallisto were imported using the importIsoformExpression function. Low abundance genes or isoforms were removed using the prefilter function with default settings. Isoform switches were identified by isoformSwitchTestDEXSeq function. To identify alternative isoform usage independent of overall gene expression changes, isoforms belonging to genes that were also identified as DEGs were excluded from the analysis.

Enrichment analysis of genes associated with secondary variants

We performed over-representation analysis (ORA) and gene set enrichment analysis (GSEA) using clusterProfilter v.4.12.6 and goseq v1.56.0 packages in R (v. 4.3.1), following the described guidelines¹²⁷. For ORA shown in Supplementary Fig. 7b and Supplementary Fig. 9d, we used enrichGO, entichKEGG, and enrichPathway functions in clusterProfilter package with proper control for population genes (top 10,000 most variable genes for Supplementary Fig. 7b and cell-type specific genes for Supplementary Fig. 9d). For ORA in Fig. 2b using DEGs and altered isoform usages overlapping secondary variants, we used the nullp function in goseq package to calculate the probability weight function (PWF), which accounts for gene length bias. We then performed enrichment analysis by goseq function using GO, KEGG, and Reactome databases retrieved by AnnotationDbi v1.66.0, KEGGREST v1.44.1, and ReactomePA v1.48.0 packages. GSEA was performed with the curated gene sets obtained using the msigdbr v.10.0.1 and clusterProfilter packages in R. Enrichment of DEGs within published gene lists was calculated using the GeneOverlap v.1.36.0¹²⁸ package in R. Enrichment of DEGs in DisGeNET was calculated using enrichR v.3.4¹²⁹ package in R.

Weighted gene co-expression network analysis (WGCNA)

WGCNA was performed using the WGCNA 1.72 package in R¹³⁰. For the analysis of transcriptome data from NPCs, iMNs, and MNs (shown in Supplementary Fig. 7a–c), raw counts were first normalized using variance stabilizing transformation (VST) from DESeq2 1.44 package in R. The top 10,000 most variable genes were selected based on median absolute deviation (MAD). The softthresholding power (β) was determined using the pickSoftThreshold function with an R² cutoff of 0.8. A signed hybrid co-expression network was constructed (TOMType = “signed hybrid”, minModuleSize = 100, mergeCutHeight =0.25) using adjacency defined as:

$$a_{ij}={\left[\mathrm{cor}\left(x_i,x_j\right)\right]}^{\beta },\mathrm{for}\mathrm{cor}\left(xi,xj\right)>0$$

$$a_{ij}=0,\mathrm{for\; cor}\left(x_i,x_j\right)\le 0$$

Module eigengenes, representing the expression profiles of each module, were calculated as the first principal component. Pearson’s correlation coefficient was used to assess correlations between module eigengene scores per sample and differentiation stages of samples.

ATAC-seq sample processing and analysis

Live cell samples were collected and passed through a 30 µm filter to obtain a single cell suspension before cryopreservation. Thawed cells were washed and treated with DNAse I (catalog #EN0521, Life Technologies) to remove genomic DNA contamination. Cell number and viability were assessed using a Countess Automated Cell Counter (ThermoFisher Scientific). After cell lysis and removal of cytosol, nuclei were treated with Tn5 enzyme (catalog #20034197, Illumina,) for 30 minutes at 37 °C and purified using the MinElute PCR Purification Kit (catalog #28004, Qiagen) to generate tagmented DNA samples. Tagmented DNA was barcoded using Nextera Index Kit v2 (catalog #FC-131-2001, Illumina), and amplified by PCR, followed by SPRI bead cleanup to obtain purified DNA libraries. The sequencing libraries were multiplexed and clustered onto a flow cell on the Illumina NovaSeq X or NovaSeq 6000 platforms according to the manufacturer’s instructions at Azenta Life Sciences (South Plainfield, NJ, USA). The samples were sequenced using a 2 × 150 bp paired-end configuration. Image analysis and base calling were performed using NovaSeq Control Software (NCS). Raw sequence data (.bcl files) were converted to fastq files and demultiplexed using Illumina bcl2fastq 2.20 software, allowing one mismatch per index sequence. Quality control and processing of fastq sequencing files were performed using the ENCODE ATAC-seq Data Standards and Processing Pipeline (https://github.com/ENCODE-DCC/atac-seq-pipeline). DESeq2 was used to identify differential peaks (Diffpeaks) with an adjusted p-value (padj) cutoff of 0.01. Motif enrichment analysis on Diffpeak regions was performed using findMotifsGenome.pl from HOMER v4.11¹³¹.

ATAC-seq peak annotation

We annotated ATAC-seq peaks using a combination of three approaches: ChIPseeker¹³², the Activity-by-Contact (ABC) model¹³³, and nearest gene assignment. ChIPseeker analysis was performed via Galaxy (version 1.28.3) using GENCODE Release 46 (GRCh38.p14) as the reference genome annotation. The ABC model was applied using RNA-seq and ATAC-seq data from HD_01 and HD_02 at the iPSC and NPC stages. Publicly available Hi-C and H3K27ac ChIP-seq datasets were included to improve contact estimation and enhancer activity scoring^134,135. Expressed genes were defined as those with expression >1 TPM and promoter activity in the top 0.4 quantile. For each chromosome, enhancer-gene pairs within 5 Mbp of a transcription start site (TSS) were scored, and pairs with an ABC score >0.022 (default threshold) were retained as predicted regulatory interactions. Peak-to-gene associations were also established by linking each peak to the nearest gene using the closest function from Pybedtools v.0.12.0.

Analysis of combined effects on gene expression

To assess combined effects, DEGs in individuals carrying both the 16p12.1 deletion and a specific secondary variant compared to all other individuals in the same family were extracted. Rare variants within these DEGs were then identified and filtered. Two-way ANOVA was performed using the statsmodels.formula.api 0.14.0 and statsmodels.api 0.14.0 modules in Python v.3.11.7, with TPM (Transcripts Per Million) values of genes annotated from the filtered variants. Genes and variants were retained only when the interaction p value (16p12.1 deletion × secondary variant, “PR( > F)”) was <0.05. The overlap of these variants with Diffpeak regions ( ± 1 kb) was determined using Pybedtools¹³⁶ v.0.12.0. To identify genes with outlier expression across iPSC, NPC, iMN, and MN stages, TPM count matrices were first filtered to retain genes with expression >0.1 TPM in at least six samples. TPM values were then log₂ transformed and z-scores were calculated for each gene across all the samples at each differentiation stage. Genes with expression z-scores ≥2 or ≤–2 were flagged as overexpression or underexpression outliers, respectively. Genes were considered robust outliers in a given sample if they met this threshold in at least two biological replicates.

Lentivirus packaging

HEK293T cells were cultured in DMEM (catalog #D6429, Sigma-Aldrich) supplemented with 10% FBS (catalog #F2442, Sigma-Aldrich) and 1% penicillin/streptomycin. On day 0, cells were seeded 18-24 hours prior to transfection to reach 80-95% confluency on the day of transfection. On day 1, fresh media was added 30 minutes prior to transfection. Transfection was performed using TransIT-Lenti Transfection Reagent (catalog #MIR6604, Mirus Bio), Opti-MEM (catalog #31985070, Gibco), and LV-MAX Lentiviral Packaging Mix (catalog#A43237, Thermo Fisher). On day 2, fresh media containing ViralBoost Reagent (catalog #VB100, ALSTEM Cell Advancements) was added. On day 3, the supernatant was collected and centrifuged at 300 × g for 10 min at room temperature to pellet debris and then passed through a 0.45 µm filter. Lentivirus Precipitation Solution (catalog #VC100, ALSTEM Cell Advancements) was added to the supernatant at a 1:4 ratio, mixed thoroughly, and incubated overnight at 4 °C. The mixture was then centrifuged at 1500 × g for 30 min at 4 °C. and the resulting virus pellet was resuspended in mTESR1 medium and stored at −80 °C.

CRISPR-based transcriptional activation (CRISPRa) experiments

iPSCs derived from probands were first transduced with lenti dCAS-VP64_Blast and lenti MS2-P65-HSF1_Hygro. Lenti dCAS-VP64_Blast (Addgene plasmid #61425; http://n2t.net/addgene:61425; RRID: Addgene_61425) and lenti MS2-P65-HSF1_Hygro (Addgene plasmid # 61426; http://n2t.net/addgene:61426; RRID: Addgene_61426) were both gifts from Feng Zhang. Cells were then infected with lenti sgRNA(MS2)_zeo, whereas the EV line was infected with only the backbone. To insert sgRNA, the backbone was digested with BsmBI-v2 (NEB) and ligated using Quick Ligase (NEB #M2200S) with phosphorylated and annealed sgRNA oligos. The lenti sgRNA(MS2)_zeo backbone was also a gift from Feng Zhang (Addgene plasmid #61427; http://n2t.net/addgene:61427; RRID: Addgene_61427). sgRNA sequences were: POLR3E: CACGGCCTGCATGAATGGCG; MOSMO: GAGCCGGGAGGACGGAGCTG; UQCRC2: ATAAAGAGAGCAGTAGAGCG. All the transductions were performed with 4 µg/ml polybrene to improve efficiency. After 24 hrs, the medium was replaced with fresh culture medium and cells were incubated for additional 48 h. Selection at the iPSC stage was performed using 10 ng/ml Blasticidin, 250 ng/ml Hygromycin, and 250 ng/ml Zeocin until no further cell death was observed. A second round of selection, at half the concentration, was performed from day 10 to day 12 of neural rosette formation during NPC differentiation. Comparisons between CRISPRa lines to their corresponding EV controls allowed us to identify DEGs resulting from the restoration of specific deletion genes in an isogenic background. DEGs from these experiments were subsequently used in downstream functional analyses. Additionally, we defined “reversed genes” as those DEGs from CRISPRa experiments that showed gene expression changes in the opposite direction compared to DEGs identified in the proband line relative to healthy donor lines.

Protein-protein interaction network analysis

Protein–protein interaction (PPI) network analysis was performed using the STRING database with default settings (network type: full STRING network; required confidence score: ≥0.400; FDR stringency: medium, 5%). To create a density plot, active interaction sources included text mining, experiments, databases, co-expression, neighborhood, gene fusion, co-occurrence. The Anderson-Darling k-sample test was performed using the ad.test function in kSamples v1.2-10 R package. To visualize proteins involved in signal transduction based on Reactome data, only experiments, databases, and co-expression data were used as active interaction sources. Average node degree, PPI enrichment p-value, and functional enrichment results, including FDR-corrected p-values and enrichment strength, were obtained directly from the STRING database.

Construction of TF regulatory networks

We used the ChEA3 (https://maayanlab.cloud/chea3/) web application to construct TF-TF co-regulatory networks using the top 15 TFs whose binding motifs were significantly enriched in Diffpeak regions from deletion lines versus HD lines comparisons at the NPC stage. In comparisons with fewer than 15 enriched motifs, only motifs with p < 0.01 were included. FOXG1 and JUN ranked highest among the enriched TFs based on the Integrated Scaled Rank. TF-TF co-regulatory networks were constructed in ChEA3 using the top-ranked TFs, with edges between TFs defined by supporting evidence from integrated libraries. To identify regulatory network genes for each TF, DEGs from each comparison were input into ChEA3, and genes listed under “Overlapping Genes” column were extracted. The top integrated rank across ChEA3 libraries was used for downstream analysis.

Statistics and reproducibility

Investigators were blinded to experimental condition when quantifying fluorescence intensity. Whenever possible, comparisons between deletion carrier and noncarrier lines within the same family were performed to control for the effects of shared secondary variants and to specifically analyze the joint effects of the 16p12.1 deletion and secondary variants. In the family-based analyses, we compared deletion lines from GL_077 and GL_079 to both nondeletion lines within the same family and HD lines. For GL_007, due to the absence of nondeletion family lines, we compared the deletion lines to HD lines. Comparisons with healthy donor lines provided a consistent baseline across families and served as appropriate controls for cellular assays and functional analysis, as many noncarrier lines also exhibited clinical features due to underlying disease liability in the genetic background. The same bioinformatics pipelines were applied across all conditions for each assay. Statistical analyses were conducted using R v4.3.1 and Python v3.11.7. Participant sex information obtained from clinical records was incorporated as a covariate in the RNA-seq and ATAC-seq analyses. Details of the statistical tests and number of replicates are provided in the corresponding figure legends and Supplementary Data files.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.S. and S.G. conceived and designed the study. S.G. supervised the experiments and analyses. J.S., S.N. and M.D. maintained iPSC culture, performed the neural conversion and extracted RNA. J.S., M.D. and A.P. generated the CRISPR-edited lines. J.S. and S.N. performed immunostaining, imaging and image quantification. J.S. and V.H.B. performed assays on cell proliferation and apoptosis. J.S. designed CRISPRa experiments and J.S. and V.H.B. performed CRISPRa experiments. J.S., D.B., C.S., and B.G. performed analyses on ATAC-seq, RNA-seq and WGS data. C.S. summarized the clinical phenotype information. D.A., K.P. and S.G. recruited 16p12.1 deletion families. P.L. reprogrammed iPSCs from PBMCs derived from recruited patients. J.S., S.N. and S.G. wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA-seq and ATAC-seq data generated in this study are available through NCBI dbGaP under accession code phs002403 [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs002403.v1.p1]. Whole-genome sequencing (WGS) data for HD_01, HD_02, the CRISPR deletion line, P2C_079, and FNC_079 are also available through NCBI dbGaP under accession code phs002403 [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs002403.v1.p1]. WGS data for the remaining individuals are available through NCBI dbGaP under accession code phs002450 [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs002450.v2.p1]. Because these datasets contain individual-level human data, access is controlled through the dbGaP Authorized Access system. Requests for access can be submitted through dbGaP using the accession numbers listed above and are generally reviewed by the NIH Data Access Committee within about two weeks. Approved requests allow access for one year, with the option to renew. Additionally, datasets for training ABC model can be assessed at GSE52457 and GSE16256. Source data are provided with this paper. Detailed results from statistical analyses are available in the Supplementary Data and Source Data files. Source data are provided with this paper.

Code availability

All code generated for this project including pipelines for running bioinformatic software and custom analysis scripts, are available at https://github.com/Jiawan1023/iPSC_integrated_framework¹³⁷.

Competing interests

J.S. and S.G. intend to file a provisional patent application related to this work. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72598-z.