An X-linked long non-coding RNA, PTCHD1-AS, and the core features of autism

Abstract

There are around 100 genes or copy-number variations used in genetic testing for autism spectrum disorder (ASD)^1,2. The established genes are protein coding, and the associated phenotypes usually extend beyond sociobehavioural traits seen in autism, including cognitive/medical complexities and attention deficit hyperactivity disorder (ADHD)^3,4. We examined whole-genome sequencing data in cases of ASD (9,349) and controls (8,332) and identify 27 male individuals with ASD with X-chromosome microdeletions that implicate the long non-coding RNA PTCHD1-AS as an ASD-susceptibility gene (odds ratio = 2.56, P = 0.01). Two Ptchd1-as-knockout mouse models, which were created by disrupting/deleting the evolutionarily conserved exon 3, show ASD-like features in male mice, including increased repetitive behaviours and impaired social behaviour and communication without cognitive comorbidities or ADHD-like behaviours. Hippocampus-dependent synaptic function, complex learning and locomotor activity are unaffected in knockout mice. Native nuclear-enriched mouse Ptchd1-as showed sustained expression from postnatal day 7 onwards in the dorsal striatum, a predominantly GABAergic brain region that is implicated in ASD⁵. Multi-omics analysis revealed transcriptomic alterations in striatal oligodendrocytes, astrocytes and neurons impacting myelination and synaptic plasticity. Disrupting Ptchd1-as led to reductions in conventional protein kinase C (cPKC) isoforms, altered SRC and GSK-3α/β phosphorylation and enhanced striatal synaptic plasticity (long-term potentiation and long-term depression). Together, these findings implicate striatal molecular and circuit-level dysregulation through PTCHD1-AS in ASD aetiology.

PTCHD1-AS, which encodes a long non-coding RNA, is associated with the aetiology of autism spectrum disorder in humans through striatal molecular and circuit-level dysregulation.

Main

ASD, which presents in a fourfold excess in male individuals, is a neurodevelopmental disorder (NDD) characterized by social communication deficits, restrictive, repetitive behaviours and sensory processing differences^3,6. Analyses of ASD cohorts have identified genes/copy-number variants (CNVs) bearing rare (<1% frequency) inherited or de novo variants meeting evidenced-based criteria to be considered for use in ASD genetic testing^1,2,4. Considering only gold-standard autism diagnoses alone, NRXN1, SCN2A, CHD8 and SHANK3 (Supplementary Table 1) are prototypical protein-coding genes that are often found to be involved in ASD (note that these four genes are also involved in other NDDs). Collectively, ASD-associated genes function broadly in cellular pathways involved in synaptic signalling, chromatin organization and transcriptional regulation^3,4,7. When pathogenic variants are found (around 7–40% of cases depending on the cohort examined and technology used), these individuals with ASD tend to show an increased number of medical complexities, and have on average lower adaptive behaviour, IQ and socialization scores^8–10. Distinguishing what constitutes a core autism phenotype from the total is therefore complex and, consequently, little is known about the molecular contributors to the social, repetitive and communicative deficits involved.

Our studies of CNVs in ASD have implicated the Xp22.11 locus (Fig. 1), including the PTCHD1-AS long non-coding RNA (lncRNA), which has three isoforms (PTCHD1-AS1, PTCHD1-AS2 and PTCHD1-AS3)^11–13 spanning around 1.1 Mb. This complex region also includes two protein-coding genes: three-exon PTCHD1 and single-exon DDX53 intronic within PTCHD1-AS. We also identified pathogenic sequence-level variants and CNVs in DDX53 in individuals with autism¹⁴, and separately PTCHD1 has been implicated in the broader category of NDDs with a few individuals also having autism.

Fig. 1. Chromosome Xp22.11 PTCHD1-AS deletions in male individuals are associated with ASD.

a, Male individuals with microdeletions that affect mainly PTCHD1-AS exons have ASD or BAP (7-0391-003) in the ASD-WGS cohort. The deletions affect canonical exon 3 (purple box; orthologous to mouse, on the basis of conservation). Human PTCHD1-AS has multiple isoforms, labelled as PTCHD1-AS1, PTCHD1-AS2 and PTCHD1-AS3. Excerpts are modified from the UCSC Genome Browser. The asterisk indicates de novo mutation, and all others are maternally inherited microdeletions. RefSeq accession codes are shown in parentheses. b, The number of deletion variants overlapping PTCHD1-AS, DDX53 and PTCHD1 in the ASD-WGS cohort (left). Right, the frequency of unique deletions in each gene among the ASD and control WGS cohorts. Significant enrichment (ORs and 95% confidence intervals are shown) was observed for PTCHD1-AS and DDX53, but not for PTCHD1. *P < 0.05, **P < 0.01.

To define the role of PTCHD1-AS in ASD, we used whole-genome sequencing (WGS) to determine the precise microdeletion breakpoints within PTCHD1-AS in 27 male individuals with autism. We searched for genotype–phenotype correlation in these individuals, as well as in a broader collection of 118 other individuals with NDDs. Guided by the genomic findings in ASD, we generated two models disrupting the canonical exon 3 of mouse Ptchd1-as and characterized the resulting behavioural, multi-omic and physiological attributes.

Microdeletions in PTCHD1-AS in ASD

We examined WGS data of male individuals from the Autism Speaks-MSSNG cohort (n = 4,693)⁴, the Simons Simplex Collection cohort (SSC) (n = 2,095)¹⁵ and the SPARK cohort (n = 2,561)¹⁶ (collectively termed the ASD-WGS cohort) and WGS data of population controls (n = 8,332). In total, 27 male individuals with ASD carrying a deletion of one or more exons of PTCHD1-AS were identified from 23 unrelated families. The most common region deleted encompassed exons 1–4 (549 kb), inclusive of DDX53 (Fig. 1). Considering only unrelated individuals with an ASD diagnosis or broader autism phenotype (BAP) alone, 12 out of 23 had deletions that spanned PTCHD1-AS only, 9 out of 23 had deletions encompassing PTCHD1-AS and DDX53, 1 out of 23 individuals had deletions impacting PTCHD1-AS and PTCHD1, and 1 out of 23 individuals had deletions impacting all three genes (Fig. 1b).

Microdeletions within PTCHD1-AS exons were associated with an increased risk of ASD (odds ratio (OR) = 2.56, P = 0.01). DDX53 deletion, of which 10 out of 12 (83%) also involved PTCHD1-AS (OR = 10.7, P = 0.004), exhibited even higher risk for ASD (Fig. 1b). By contrast, we did not find PTCHD1 to be significantly linked to ASD (OR = 0.59 P = 0.67), noting that other data implicate it predominantly in broader NDDs, including intellectual disability (ID) (see below).

PTCHD1-AS ASD with lower comorbidities

Male probands and siblings from the ASD-WGS cohort with microdeletions affecting PTCHD1-AS and PTCHD1-AS/DDX53 predominately exhibit ASD/BAP in 82% of cases (27 out of 32), with some overlap for developmental delay (DD) or ID in 42% of cases (10 out of 24) and ADHD/attention deficit disorder (ADD) in 33% of cases (8 out of 24; Supplementary Table 2). Full-scale IQ levels varied among individuals, ranging from 34 to 94 (12 reporting IQs) with an overall mean of 75 and median of 78, placing the group within the low-average IQ range based on 2 s.d. of the mean.

Given the initial sample size of the PTCHD1-AS ASD-WGS genotype–phenotype data and potential for clinical heterogeneity, we incorporated additional cases of ASD or NDD (from published studies and neurodevelopmental cohorts/datasets; hereafter, the NDD cohort) studied with any genomic technology (Extended Data Fig. 1a and Supplementary Table 3). A total of 118 deletions or loss-of-function variants impacting PTCHD1-AS/DDX53/PTCHD1 were observed in male individuals. When analysed by diagnostic criteria, an appreciable proportion of individuals with a PTCHD1-AS deletion were reported with other/unspecified conditions (78%; 7 out of 9 cases; mostly with speech impairment or gross morphological changes), 37% were reported with DD/ID (13 out of 35 cases of DD/ID) or 30% were reported with combined ASD and DD/ID (9 out of 30 cases of ASD + DD/ID). The majority of those diagnosed with ASD/BAP (around 82%, 36 out of 44 cases of ASD/BAP) carried a microdeletion in PTCHD1-AS, whereas those carrying loss-of-function variants in PTCHD1, or variants spanning the promoters of both genes, accounted for ~9% (4 out of 44) and ~9% (4 out of 44), respectively.

Extended Data Fig. 1. PTCHD1-AS in autism, and mouse models.

a) PTCHD1-AS lncRNA is associated with a primary ASD outcome. Male individuals with deletion variants at the PTCHD1- PTCHD1-AS locus in neurodevelopmental disorder and autism specific cohorts (N = 118). Total individual variant counts per gene impacted: PTCHD1-AS, N = 65; PTCHD1-AS + PTCHD1, N = 31; PTCHD1, N = 22 and primary phenotype (case counts) are shown. Abbreviations: Autism spectrum disorder (ASD), Broader autism phenotype (BAP), Developmental delay/ Intellectual deficits (DD/ID). b) Genotype confirmation of male Ptchd1-as KO-1 and KO-2 mice from CRISPR/Cas9 genome editing at canonical exon-3. This experiment was completed 30 times. c-f) Structural alterations in male Ptchd1-as KO-2 mice. We used anatomical MRI across development to localize changes in brain structure between KO-2 (N = 23, 104 scans) and their littermate controls (N = 27, 121 scans). c) The areas with the greatest degree of slope differences between mutants and controls. d) and e) Plot the time courses for the Intermediary nucleus of the endopiriform claustrum (INEC) and the extra pyramidal fibre system (EFS), respectively. INEC shows differences in developmental slopes around puberty resulting in volume increases in young adulthood, whereas EFS shows early overgrowth around weaning followed by undergrowth in young adulthood resulting in a loss of volume starting shortly after puberty. f) Timing of maximal slope difference and maximal volume differences. Grey shaded area indicates the proportions expected due to chance. Slope differences occur around weaning and again around puberty, resulting in maximal volume differences in young adulthood. d-f). The error bars are all 95% confidence intervals.

In general, male individuals in the NDD-cohort with PTCHD1-AS-only deletions (65) and ASD/BAP show moderate-to-low comorbidities (Supplementary Table 3); around 8% (3 out of 36) have motor delay, 11% have seizure reports (4 out of 36), 17% (6 out of 36) show speech impairment and 25% (9 out of 36) have co-occurring ADD/ADHD—60% lower compared with the pooled lifetime rate in general for ASD (40%)¹⁷.

Of those individuals with a PTCHD1-AS microdeletion (65), around 32% (21) had DD/ID, with or without ASD, suggesting a moderate association. However, given that PTCHD1-AS is adjacent to PTCHD1 (an NDD/ID gene), it is also possible that microdeletions impacting intronic DNA elements within the PTCHD1-AS locus may regulate PTCHD1, resulting in ID/NDD. To functionally discern the deletion impact in these two genes, we previously created an isogenic CRISPR deletion of PTCHD1-AS2 exon 3 (based on family 7-0342 with three ASD male individuals with exon-3-only deletions; Fig. 1) in an induced pluripotent stem cell line and then derived cortical neurons¹³. We saw changes in synaptic transmission indicating a role for PTCHD1-AS in neurons, but no concurrent changes in PTCHD1 or DDX53 expression.

Mouse model targeting Ptchd1-as exon 3

In mice, the Ptchd1-as (Gm15155) orthologue spans around 700 kb on chromosome X qF3-qF4 with five isoforms (Fig. 2a). Ptchd1-as canonical exons 1 and 3 share 87% and 62% base pair sequence identity, respectively, between mice and humans. Notably, no functional orthologue to DDX53 exists in mice¹⁴. To investigate PTCHD1-AS function, we prioritized exon 3 on the basis of human initial microdeletion data, cellular evidence (family 7-0342) and conserved DNA regions between mice and humans.

Fig. 2. Transcriptional consequences in disruption models of exon-3.

a, Mouse Ptchd1-Ptchd1-as (Gm15155) and the predicted/annotated isoforms are evolutionarily conserved with human exon 3 (purple box). b, Schematic of the generation of the Ptchd1-as ^−Ex3 model (KO-1) repaired by non-homologous end joining (NHEJ) mechanisms and the Ptchd1-as^−Ex3-is model (KO-2), replacing exon-3 with an intronic spacer (is) single-stranded oligodeoxynucleotide (ssODN) integrated through homology-directed repair (HDR). c, ddPCR analysis of Ptchd1-as KOs relative to WT mice in the striatum confirmed loss of exon 3 in two transcripts: Gm15155-201 (left) exons (ex) 2–3 (KO-1; P = 0.0032; KO-2, P < 0.0001) and exons 3–4 (KO-1; P = 0.0132; KO-2, P = 0.0093) without consistent alterations to downstream exons (exons 5–6: KO-1; P = 0.0406; KO-2, P = 0.8981); and XR_878287.1 (right) loss of exons 4–5 (KO-1; P = 0.082; KO-2, P = 0.0002) with some upregulation of earlier exons in KO-2 (exons 1–2: KO-1, P = 0.6448; KO-2, P = 0.0350; exons 2–3: KO-1, P = 0.8005; KO-2, P = 0.0066; exons 3–4: KO-1, P = 0.7963, KO-2, P < 0.0001). Two-way repeated-measures analysis of variance (ANOVA; two-sided). n = 5 (WT-1), n = 5 (KO-1), n = 6 (WT-2) and n = 6 (KO-2). d, Assessment of Ptchd1 expression in Ptchd1-as KO lines adult brain relative to WT. Left, confirmation of Ptchd1-as exon 2–3 knockdown in KO-1 (P < 0.0001) and KO-2 (P < 0.0001) mice. Right, Ptchd1 is unaffected by Ptchd1-as exon-3 deletion in the KO-1 (exons 1–2, P = 0.2298; exons 2–3, P = 0.0737) and KO-2 (exons 1–2, P = 0.1163; exons 2–3, P = 0.3221) models. Two-way repeated-measures ANOVA (two-sided). n = 5 (WT-1), n = 5 (KO-1), n = 5 (WT-2) and n = 5 (KO-2). e, Gapmer-mediated knockdown targeting PTCHD1-AS2 exon 1 has no effect on PTCHD1 expression, but does alter DDX53 in human neural stem cells. n = 7/7 independent cell culture batches. One-way repeated-measures ANOVA (two-sided); from left to right versus gapmer scrambled control, P = 0.9139, P = 0.2300, P < 0.0001, P < 0.0001, P < 0.0001, P < 0.0001; ***P < 0.001, ****P < 0.0001. Data are mean ± s.e.m.

We developed two different knockout (KO) models in mice (Fig. 2b): KO-1 (Ptchd1-as^−Ex3), with an exon-3 deletion, and KO-2 (Ptchd1-as^−Ex3-is), in which exon-3 is replaced with a 90 bp intronic sequence (is). Genotype analysis using WGS confirmed successful gene editing (Extended Data Fig. 1b) with no exonic off-target effects, and this was further confirmed using droplet-digital PCR (ddPCR) for Ptchd1-as isoforms (Fig. 2c; Gm15155-201 and RefSeq sequence XR_878287.1).

Structural brain alterations in KO mice

We used 7T magnetic resonance imaging (MRI) to determine whether Ptchd1-as KO-2 is associated with any developmental issues in mesoscopic neuroanatomy. There were developmentally regulated differences in the mean volume and slope of change in restricted structures associated with autism, as well as white-matter/extrapyramidal fibre tracts (Extended Data Fig. 1c–f). Relative to the controls, KO-2 mice exhibited developmental pattern differences in five brain areas (false-discovery rate (FDR) < 10%), including the intermediate nucleus of the endopiriform claustrum, the dorsal intermediate entorhinal cortex, field CA2 of the hippocampus, anterior cingulate cortex (area 24a) and the fibre tract ventral tegmental decussation (Extended Data Fig. 1c–f). Alterations of the last region coincides with developmental windows characterized by active myelination remodelling.

Spatiotemporal Ptchd1-as expression

We used RNA sequencing (RNA-seq) analysis to study known and novel exon expression, as well as ddPCR to ‘walk’ across exon junction pairs of the canonical transcript, to determine the spatiotemporal pattern of native Ptchd1-as in key brain regions relevant to autism (Extended Data Fig. 2a,b,d–g, respectively). We also confirmed exon-3 KO using unbiased RNA-seq analysis (Extended Data Fig. 2b). Orthologous exon 3 of Ptchd1-as (Extended Data Fig. 2c–g) exhibited unique spatiotemporal patterns depending on the regions assayed (hippocampus, cerebellum, striatum and cortex) relative to whole-brain tissues. Expression levels typically peaked at postnatal day 7 (P7) except for in the cerebellum, in which the levels declined precipitously after birth. By contrast, the dorsal striatum, an ASD-relevant brain region linked to restricted repetitive behaviours⁵, and the cortex frequently associated with ASD, showed consistent exon 2–3 expression into young adulthood, with the dorsal striatum surpassing cortex expression beyond P35 (Extended D ata Fig. 2h).

Extended Data Fig. 2. Ptchd1-as exon-exon structure and spatiotemporal expression in key ASD brain regions.

Deep RNA sequencing validates all predicted exons from 5 alternate isoforms in various brain regions in C57bl/6 J adult male mouse. a) Sashimi plot showing representative exon-exon junctions (curved lines) and relative expression of native Ptchd1-as exons in various parts of the brain. b) RNA sequencing confirmed canonical exon-3 knockdown in both Ptchd1-as KOs. Bar chart showing similar Total RNA-seq normalized read counts per exon in Ptchd1-as KOs and WT littermate striatal samples (WT-/KO-1, WT-/KO-2, N = 6/6, 4/4; data is displayed as mean +/− SEM.) with expected deletion in Exon position 7 (red box). Red dotted lines, canonical Gm15155-201 exons. Purple dotted lines, XR_878287.1 exons. c-h) DdPCR walking across exon-exon pairs of canonical Ptchd1-as shows exon 2-3 to be highly expressed relative to other exon-exon pairs at all ages in c) Brain (N = 6, E10.5-P70) d) Cortex (N = 6, E18-P70), e) Hippocampus (N = 6, E18-P70), f) Striatum (N = 4, E18, P7; N = 6, P35, P70) and g) Cerebellum (N = 6, E18-P70). In the hippocampus Ptchd1-as is ~5-fold lower expressed than in whole brain tissues. Ptchd1-as is declining in cerebellar tissues not long after birth but stays elevated into young adulthood in the striatum. h) Comparison summary of Exon 2-3 expression across all tissues at various ages. (N values as described in c-g). c-h) Data is displayed as mean +/− SEM.

PTCHD1 is independent of PTCHD1-AS

Given the genomic proximity of PTCHD1 and PTCHD1-AS, it is plausible that disruption of PTCHD1-AS could influence PTCHD1 expression. In our previous study, we examined induced-pluripotent-stem-cell-derived cortical neurons from male participants with ASD carrying PTCHD1-AS deletions and observed clear electrophysiological phenotypes without changes in PTCHD1 transcript levels¹³. To validate these findings in another biological system, we next assessed Ptchd1 expression in adult male mouse brain tissue from Ptchd1-as KO models. We found that neither KO-line altered Ptchd1 expression relative to wild-type (WT) controls (Fig. 2d). Thus, Ptchd1-as deletion does not affect Ptchd1 mRNA levels in vivo. Moreover, previously described Ptchd1-mutant mice had impairments in attention and cognition but did not exhibit ASD-associated behaviours^18,19.

To further test whether the PTCHD1-AS2 RNA product contributes to local gene regulation, we performed gapmer-mediated knockdown targeting PTCHD1-AS2 exon 1 in human neural stem cells (Fig. 2e). This reduced downstream PTCHD1-AS2 exons by around 80–90% without affecting upstream PTCHD1-AS3 exon 1 or PTCHD1 expression. Notably, DDX53 expression was significantly reduced exhibiting a dependence on PTCHD1-AS expression.

Social behaviour and repetitive grooming

To investigate whether Ptchd1-as KO males exhibit atypical social behaviour, we used the three-chamber social interaction test, and recorded time spent with a conspecific female mouse (M1) compared with an inanimate object (O). While WT mice exhibited a preference for M1 over O, KO-1 and KO-2 mutants were equally interested in O versus M1 (Fig. 3a,b), indicating reduced sociability. Next, the inanimate object (O) was replaced with a novel female (M2). As expected, WT mice interacted more with M2, the socially novel mouse, than with M1. By contrast, KOs showed no preference for M1 versus M2 (Fig. 3c,d and Supplementary Table 4) and no novel social discrimination suggestive of deficits in social learning and memory.

Fig. 3. Ptchd1-as is critical for normal social interaction and communication patterns, and disruption results in ASD-like behaviours in mice.

a,b, Sociability was reduced in KO-1 (a) and KO-2 (b) mice in the three-chamber social interaction. Representative heat maps during sociability (left), the time spent in each chamber (middle) and the direct interaction time with a conspecific female mouse (M1) and an inanimate object (O) (right) are shown. C, centre. c,d, Preference for social novelty was reduced in KO-1 mutant mice (c) and KO-2 mutant mice (d). For a–d, n = 9 (WT-1), n = 11 (KO-1), n = 9 (WT-2) and n = 10 (KO-2). Two-way repeated-measures ANOVA (two-sided). Representative heat maps during social novelty recognition (left), the time spent in each chamber (middle) and the direct interaction time with M1 and a novel female mouse (M2) (right) are shown. In heat maps, red and dark blue indicate maximum time (red) and no time (dark blue) spent within the box locations. e, Reactivity to social odour cues was reduced in KO-1 (left) and KO-2 (right) mice. n = 8 (WT-1), n = 7 (KO-1), n = 8 (WT-2) and n = 7 (KO-2). Two-way repeated-measures ANOVA (two-sided). B, blank odour; F, female urine odour; M, male urine odour. f, Repetitive self-grooming was increased in KO-1 (left) and KO-2 (right) mice. n = 12 (WT-1), n = 8 (KO-1), n = 10 (WT-2) and n = 10 (KO-2). Two-sided unpaired t-test. g, Representative USV sonogram of the entire recording during the sociability test (top) and a magnified 200 ms time window (bottom). h–j, Analysis of individual USV syllables, including the number of USV syllables (h), the mean frequency (i) and a syllable intensity analysis (j) of KO-1 (top) and KO-2 (bottom) mice. n = 9 (WT-1), n = 11 (KO-1), n = 9 (WT-2) and n = 10 (KO-2). Two-sided unpaired t-test. NS, not significant. Statistical details are provided in Supplementary Table 4. Data are mean ± s.e.m.

Social odours, which contain information on social status, colony membership and sex, are crucial in rodent communication²⁰. Mice naturally sniff novel social odours, but quickly habituate to them²¹. In the test of reactivity to a social odour cue, WT mice showed high initial interest in female mouse urine, which they quickly habituated to (Fig. 3e) and dishabituated when male mouse urine (M) was presented. By contrast, both KO-1 and KO-2 mutant mice exhibited reduced interest in novel social odours. Moreover, they exhibited negligible habituation and dishabituation to social odours, suggesting that the KO mice have limited social odour responsiveness and discrimination.

Repetitive behaviour is another key feature of ASD²². Both KO-1 mice and KO-2 mice spent significantly more time self-grooming than their WT littermates (Fig. 3f), indicating enhanced repetitive behaviour.

The revised DSM-5 no longer divides ASD subtypes on the basis of language impairments, or those seemingly without, to enable grouping of broader social communication differences²³. In fact, 75–80% of individuals with ASD with no language delay exhibit functional language impairments²³ that we aimed to analyse in KO mice. Thus, during the sociability test, we recorded mouse ultrasonic vocalization (USV) to assess social communication. We used a human speech-processing-inspired analysis²⁴ (Fig. 3g) to characterize individual USV syllables and full communication repertoires in a courtship-style scenario in which males typically vocalize more towards female conspecifics²⁵. Mice of both KOs emitted fewer USV syllables compared with their WT littermates (Fig. 3h). We also found that USV syllable frequency (Fig. 3i) and utterance intensity (Fig. 3j) is reduced in the KO mice compared with their WT littermates (Fig. 3i,j), indicating reduced social communication in KO mice^26,27.

We also examined the repertoire of USV syllables in KO mice to determine how many unique syllables they use. Using a machine-learning approach²⁴, we generated four integrated syllable repertoires for each genotype (Extended Data Fig. 3a–d), evaluating the degree of similarity among them (Extended Data Fig. 3e). WT littermate groups showed a high degree of similarity, while the Ptchd1-as mutants displayed a broader range, but not significantly different syllabic repertoire (Extended Data Fig. 3f). Together, these results suggest that KOs have normal communication ability but display low utterance levels consistent with social communication deficits. We noted similar reductions in KO mice using a pre-pulse inhibition test for sensorimotor gating function (Extended Data Fig. 4a)—a test of sensory processing differences—but found no changes to locomotor activity levels or anxiety in the open field test (Extended Data Fig. 4b). Similarly, there were also no changes in motor function of Ptchd1-as-KO mice (Extended Data Fig. 4c–e).

Extended Data Fig. 3. Generation and comparison of integrated USV repertoires of Ptchd1-as KO male mice.

a-d, Integrated ultra sonic vocalization (USV) repertoires for the groups of a) WT-1 littermates (N = 9), b) KO-1 mutants (N = 11), c) WT-2 littermates (N = 9), and d) KO-2 mutants (N = 10). Each syllable type displays a syllable identification number (left top, black) and the number of occurrences (left bottom, blue). e) Pearson’s correlation coefficient matrix showing the relationship between WT littermates and Ptchd1-as KOs. f) Violin plots describe similarity scores computed from the Pearson correlation coefficient. (Data is shown as median interquartile range).

Extended Data Fig. 4. Behavioural phenotypes are confined to autistic traits, sensorimotor gating deficits but no general motor function alterations in Ptchd1-as KO male mice.

a) Decreased pre-pulse inhibition in KO-1 (left) and KO-2 (right) mice (two-way repeated measures ANOVA (two-sided); WT-/KO-1, WT-/KO-2; N = 8/6, 8/8; p = 0.0058, p = 0.0019). b) Normal locomotor activity and anxiety level in the open field test in KO-1 (top) and KO-2 (bottom) mice (unpaired t-test for total distance analysis; two-way repeated measures ANOVA (two-sided) for zone time analysis; WT-/KO-1, WT-/KO-2; N = 9/9, 8/9). c) Normal gait length and width in KO-1 (top) and KO-2 (bottom) mice (two-sided, unpaired t-test; WT-/KO-1, WT-/KO-2; N = 9/7, 7/8). d) Normal motor coordination in ladder walking task in KO-1 (top) and KO-2 (bottom) mice (two-sided, unpaired t-test; WT-/KO-1, WT-/KO-2; N = 8/7, 9/11). e) Normal motor learning and coordination in rotarod in KO-1 (top) and KO-2 (bottom) mice (two-way repeated measures ANOVA (two-sided); WT-/KO-1, WT-/KO-2; N = 6/9, 8/7; p = 0.9864, p = 0.1184). See Supplementary Information Table 4 for statistical information). Data are mean ± s.e.m. *p < 0.05, ns, not significant.

Normal cognition and hippocampal function

Given the potential overlap of DD/ID associated with human PTCHD1-AS microdeletions, we next tested Ptchd1-as KO-2 male mice for alterations in learning and memory (Extended Data Fig. 5a–c): contextual fear conditioning²⁸ and two tests of complex learning and executive function (the puzzle box²⁹ and touchscreen pairwise discrimination task³⁰). In contrast to the Ptchd1-KO model¹⁸, which had significant learning and memory impairments on the contextual fear conditioning task, Ptchd1-as KO-2 mice were consistently normal on all of these learning and memory tasks, further supporting Ptchd1-as being more involved in sociobehavioural attributes of autism, and less in intellectual deficits.

Extended Data Fig. 5. Learning and memory tasks and hippocampal synaptic plasticity are unaffected in Ptchd1-as KO-2 male mice.

Learning and memory was preserved in both the a) contextual fear memory paradigm (two-sided, unpaired t-test; WT-/KO-2, N = 7/13) and the b) Bussey-Saksida touchscreen pairwise discrimination (PD) assay. PD acquisition (left) and both the PD and PD-reversal learning tasks (right) were within normal range (two-way repeated measures ANOVA two-sided,; uncorrected Fisher’s LSD test; WT-/KO-2, N = 14/10). c) KO-2 mice also performed similarly to WT in the puzzle box test with no alterations to short- or long-term memory (two-way repeated measures ANOVA two-sided,; post-hoc Sidak; WT-/KO-2, N = 12/13). d, e) Hippocampal synaptic plasticity was unaffected in KO-2 mice. Long-term potentiation (LTP) was induced by a theta-burst stimulation (TBS) protocol in the CA1 region of dorsal (dCA1; WT-/KO-2 N = 12/10) and ventral (vCA1; WT-/KO-2 N = 11/10) hippocampal slices prepared from adult mice, or by a high-frequency stimulation (HFS) protocol in the dentate gyrus region of dorsal (dDG; WT-/KO-2 N = 8/11) and ventral (vDG; WT-/KO-2 N = 7/10) hippocampal slices prepared from young adult mice d). Long-term depression (LTD) was induced by a low-frequency stimulation (LFS) protocol in the CA1 region of dorsal (dCA1; WT-/KO-2 N = 6/9) and ventral (vCA1; WT-/KO-2 N = 6/8) hippocampal slices prepared from juvenile (P14) mice e). The proportions of LTP or LTD (bar graphs) were quantified from the last 10 min of the recording period (unpaired t-tests). Sample traces show superimposed fEPSPs recorded during baseline (1, dashed line) and after induction of plasticity (2, solid line) from representative WT-2 and KO-2 slices (scale bars: 5 ms, 0.5 mV). The data is displayed as mean +/− SEM. ns, not significant. t-tests were all two-sided.

A phenotype that is commonly seen in mouse models of ASD is a reduction in NMDA receptor (NMDAR)-mediated synaptic plasticity in the hippocampus³¹. We therefore studied the dorsal and ventral hippocampus separately, given their roles in general cognition underlying contextual and social learning, respectively, and investigated both area CA1 and the dentate gyrus (DG), where NMDAR-dependent plasticity is most extensively characterized. There was no effect of genotype on long-term potentiation (LTP) or long-term depression (LTD) (Extended Data Fig. 5d,e), or on basal synaptic properties (Extended Data Fig. 6a–d). Instead, Ptchd1-as KO-2 mice exhibited normal levels of both LTP and LTD in the hippocampus, corroborating the specific tissue impact of Ptchd1-as and the normal ability of KO-2 mice for complex learning tasks.

Extended Data Fig. 6. Basal synaptic function and intrinsic neuronal properties are unaffected in Ptchd1-as KO-2 male mice hippocampus, mPFC and striatum.

a-d) Extracellular field potential recordings from hippocampal slices comparing axon connectivity and basal synaptic function in KO-2 and WT-2 littermate mice. a) The graphs plot input/output (I/O) curves of the fibre volley (FV) amplitude as a function of stimulus intensity (left graphs) and fEPSP slope as a function of FV amplitude (right graphs) for dorsal CA1 (dCA1; upper graphs) and ventral CA1 (vCA1; lower graphs) regions of adult hippocampal slices (WT-/KO-2, N = 12/10 and 11/10, respectively). b) Equivalent plots for the dentate gyrus region of dorsal (dDG) and ventral (vDG) slices prepared from young adult mice (WT-/KO-2, N = 8/11 and 10/11). c) Equivalent plots for the dCA1 and vCA1of slices prepared from juvenile (P14) mice (WT-/KO-2, N = 6/9 and N = 6/9). There were no differences in the linear I/O relationships between KO-2 and WT-2 slices (unpaired t-tests). Sample traces show superimposed responses evoked over a range of stimulus intensities (scale bars: 5 ms, 1 mV). d) Paired-pulse facilitation (PPF) ratio was assessed across a range of inter-pulse intervals in the dCA1 and vCA1 of slices prepared from adult (WT-/KO-2, N = 2/10 and 11/10) and P14 (WT-/KO-2, N = 6/9) mice. Similar levels of PPF were observed in KO-2 and WT-2 slices (two-way repeated measures-ANOVA). Sample traces show response pairs elicited with a 50 ms inter-pulse interval (scale bars: 10 ms, 1 mV). e–g) Whole-cell patch clamp analysis of the intrinsic and synaptic properties of layer 5/6 mPFC pyramidal neurons in coronal brain slices prepared from adult KO-2 and WT-2 mice. Similar passive e) and active f) membrane properties were observed for KO-2 and WT-2 neurons in current clamp recordings. e) The resting membrane potential (RMP) was measured in the absence of injected current (I = 0), while input resistance and membrane time constant were measured from the membrane voltage response to a 500 ms hyperpolarising (−50 pA) current injection (two-sided, unpaired t-tests; WT-/KO-2, N = 12/15; sample traces show representative membrane voltage responses in WT-2 and KO-2 neurons, scale bars: 100 ms, 2 mV). f) Action potential (AP) firing frequency was assessed in response to a series of 500 ms depolarizing current steps (+50 to +500 pA, in 50 pA increments) from a starting potential of −70 mV (two-way repeated measures-ANOVA), and AP waveform properties (threshold, amplitude, half-width, and rise tau) were analysed using the first AP elicited in response to the +500 pA current step (two-sided, unpaired t-tests; WT-/KO-2, N = 10/13; sample AP traces from representative WT-2 and KO-2 neurons, scale bars: 100 ms, 20 mV). g) Evoked synaptic currents were assessed in voltage clamp recordings in response to stimulation of layer 2/3. For each cell, AMPAR-EPSCs were first recorded at −70 mV and NMDAR-EPSCs were subsequently recorded at +40 mV with AMPARs blocked by 10 µM NBQX (sample traces show superimposed AMPAR- (negative-going) and NMDAR- (positive-going) EPSCs from representative WT-2 and KO-2 neurons; scale bars: 50 ms, 100 pA). KO-2 and WT-2 neurons exhibited similar NMDAR/AMPAR ratios and EPSC rise and decay kinetics (two-sided, unpaired t-tests; WT-/KO-2, N = 26/30). h-j) Synaptic properties of striatal MSNs measured in whole-cell voltage clamp recordings from adult KO-2 and WT-2 mouse coronal brain slices. h) Evoked AMPAR- and NMDAR-mediated EPSCs were recorded from each cell at −70 mV and +40 mV, respectively, in response to stimulation of the corpus callosum. AMPARs were blocked by NBQX (10 µM) when recording NMDAR-EPSCs (sample traces show superimposed AMPAR- (negative-going) and NMDAR- (positive-going) EPSCs from representative WT-2 and KO-2 neurons; scale bars: 50 ms, 100 pA). KO-2 and WT-2 neurons exhibited similar NMDAR/AMPAR ratios (two-sided, unpaired t-test; WT-/KO-2, N = 11/18). Analysis of AMPAR- i) and NMDAR- j) mediated sEPSCs revealed no differences in the frequency or amplitude, nor in the rise and decay kinetics, of spontaneous synaptic events (unpaired t-tests; WT-/KO-2, N = 11/18). AMPAR-sEPSCs were recorded at −70 mV. NMDAR-sEPSCs were recorded at +40 mV in the presence of 10 µM NBQX. Sample traces show segments of the gap-free recordings (upper; scale bars: 2 s, 30 pA) and the aligned average waveforms of all events (lower; scale bars: 5 ms, 5 pA (AMPAR) and 100 ms, 5 pA (NMDAR) from representative WT-2 and KO-2 neurons. ns, not significant. The data is displayed as mean +/− SEM.

The effects on social novelty learning and memory in mice could be due to altered intrinsic properties (for example, action potential firing) or synaptic properties (especially AMPA receptor (AMPAR)-mediated and NMDAR-mediated synaptic transmission) in brain regions that are involved in these behaviours, such as the medial prefrontal cortex (mPFC)³². We therefore compared the neuronal properties in layer 5/6 neurons of the dorsal mPFC of WT and KO-2 mice. We analysed eight intrinsic neuronal properties and observed no differences between the genotypes (Extended Data Fig. 6e-g). We next investigated cortical–cortical synaptic transmission in this neuronal population by stimulation in layer 2/3. There was no difference in the NMDAR/AMPAR ratio between genotypes or in the kinetics of AMPAR- and NMDAR-mediated excitatory postsynaptic currents (EPSCs) (Extended Data Fig. 6h–j). These findings indicate that intrinsic and synaptic properties are not altered in the mPFC of Ptchd1-as-KO mice.

Differentially expressed striatal genes

We focused on the striatum, a region that is highly relevant to autism^5,33 and where Ptchd1-as is consistently expressed throughout development. As molecular-, cellular- and circuit-level deficits in our Ptchd1-as model could be highly specific, we used an omics approach (Fig. 4a), paralleling human ASD transcriptomics studies, to gain a broader perspective on Ptchd1-as function.

Fig. 4. Multi-omic analysis of the adult dorsal striatum in KO mice indicates glial and synaptic changes.

a, Total RNA-seq, snRNA-seq and proteomic workflows. The diagram was created using BioRender; Bradley, L. https://BioRender.com/dm7byac (2026). b, Total RNA-seq analysis of Ptchd1-as KO. Inset: DEG counts. n = 10 (WT) and n = 10 (KO). P values are provided in Supplementary Table 5. c–e, KO snRNA-seq DEG counts by cell type (c), the top 10 affected DEGs by cell type (snRNA-seq) (d) and the top 20 common cell type DEGs (e) (ranked FDR level). n = 2 (WT) and n = 2 (KO) mice. Nucleus counts per cell type are provided in Supplementary Table 6 and P values are provided in Supplementary Table 10. f, Total RNA-seq versus pseudo-bulk snRNA-seq DEG analysis shows a significant positive Pearson r correlation at P < 0.05 in the two datasets; the top 20 most correlated DEGs (crosses) confirm dysregulated myelination and synaptic plasticity genes. n = 682 DEGs. Two-sided unpaired t-test. g, Comparison of KO-1 and KO-2 pseudo-bulk snRNA-seq DEGs (P < 0.05) are highly congruent in log₂[FC] direction of change (n = 682 DEGs), consistent with KO-1 and KO-2 affecting the same DEGs in a similar manner. P values are provided in Supplementary Table 9. h, Differentially expressed proteins, including downregulated HISTONE 1 family, cPKCs, annexins and upregulated GFAP. n = 4 (WT-2) and n = 4 (KO-2). DEP P values are provided in Supplementary Table 16. Ecc, eccentric; D1- and D2-, D1-MSNs and D2-MSNs; micro, microglia; oligo, oligodendrocytes; astro, astrocytes; poly, polydendrocytes. Differential expression was determined using two-sided DESeq2 (transcriptomics; Wald test) and edgeR (proteomics; likelihood ratio test). Data are the mean log₂[FC]. Multiple-testing correction was performed using the Benjamini–Hochberg FDR procedure. DEG counts: FC > ±10% and *P < 0.05 or FDR < 0.1. For the volcano plots, the horizontal dotted line represents the cutoff P value of 0.05, and vertical dotted lines represent fold change (FC) ± 10%. In b,h, grey dots represent DEGs with FC < 10%, yellow dots represent DEGs with P < 0.05, and purple dots represent DEGs with FDR and <10% FC significant. In d, RNAseq grey dots indicate P < 0.05 and FC < 10%, and multicolored DEGs indicate FDR, <10% FC significant.

We assessed alterations in transcription using both total/bulk tissue RNA-seq (total RNA-seq) and single-nucleus RNA sequencing (snRNA-seq) in the dorsal striatum of young-adult male mice (Fig. 4a–g). KO-1 and KO-2 mice exhibited relatively few total RNA-seq differentially expressed genes (DEGs): 1,202 DEGs in KO-1 and 379 DEGs in KO-2 (P < 0.05) (Supplementary Table 5). These numbers were further reduced under more stringent criteria (FDR < 0.1, fold change (FC) > 10%), yielding 244 DEGS for KO-1 (n = 6) and 2 DEGs for KO-2 (n = 4).

We next used snRNA-seq to enable greater detection of nuclear-enriched transcripts influenced by Ptchd1-as in each cell type (Fig. 4c,d) using a cell-type balancing strategy (Extended Data Fig. 7 and Supplementary Table 6). At the single-cell level, both Ptchd1-as and Ptchd1 are expressed mainly in D1 and D2 medium spiny neurons (MSNs), consistent with more significant DEGs in D1- and D2-MSNs and interneurons (INs) than in astrocytes. While neither Ptchd1 nor Ptchd1-as was identified as a DEG in individual-cell-type snRNA-seq analysis, under pseudo-bulk conditions, nuclear-expressed Ptchd1-as showed modest increased expression in both KO lines (Supplementary Tables 7 and 8), whereas Ptchd1 remained unchanged. While top KO-line DEGs varied (Extended Data Fig. 7g,h), comparison of snRNA-seq from KO-1 and KO-2 mouse pseudo-bulk DEGs revealed a 77.8% overlap in congruently regulated DEGs (3,970 out of 5,103; Supplementary Table 9).

Extended Data Fig. 7. Primary snRNA-seq processing for Ptchd1-as KO line DEG analysis.

a) Uniform Manifold Approximation and Projection (UMAP) plot of weighted nearest neighbour integrated map with nuclei plotted by genotype. b) UMAP heat map showing RNA transcript counts per nuclei across all cell types and all samples. c) Mean RNA transcript counts per cell type from each genotype. d) Dot plots show normalized expression level counts for cell type major markers and genes of interest (GOI) after clustering of nuclei in each genotype. e) Sample nuclei with cell type proportions obtained from initial nuclei isolation (left panel) in Ptchd1-as KO-1 (N = 2,394) WT-1 littermate (N = 7,495), KO-2 (N = 1,856), and WT-2 littermate (N = 2,073) and after down sampling balanced nuclei counts were used (see Supplementary Table 6). f) Circo plot cell type cluster matching prior to DEG analysis. WT-/KO-1; WT-/KO-2 biological N = 1/1, 1/1. g, h) Striatal neurons and astrocytes are impacted in Ptchd1-as KO-1 and KO-2 male mice. snRNA-seq DEG counts by cell type in KO-1 g) and KO-2 h) (FC > ± 10% and p < 0.05 or FDR < 0.1). Volcano plots showing top 10 significant DEGs by cell type (coloured circles representing FDR < 0.1, FC > ± 10%) in KO-1 g) and KO-2 h) WT-/KO-1; WT-/KO-2 biological. N = 1/1, 1/1 mice and see Supplementary Table 6 for summary of N values (nuclei counts per cell type). Differential expression was performed using two-sided statistical tests DESeq2 (transcriptomics; Wald test) and edgeR (proteomics; likelihood ratio test). Data are shown as mean log2 fold changes. Multiple testing correction was performed using the Benjamini-Hochberg FDR procedure.

Given the high congruency of DEGs in KO-1 and KO-2 mice, we next tested for common transcriptional impact by combining the KO lines (collectively termed KO) and WT littermate controls (Fig. 4b–d). KO total RNA-seq analysis yielded twice as many significant DEGs at our cut-off criteria (Fig. 4b; FDR < 0.1, FC > ±10%, 436 DEGs; Supplementary Table 5) relative to KO-1 or KO-2 alone, with most genes falling within a modest range between ±0.5 log₂[FC], similar to human studies^34,35. Total RNA-seq analysis of KO transcriptomes (Supplementary Table 5) also found no significant changes in aggregate Ptchd1-as or Ptchd1 gene expression.

Myelination and synaptic function

The majority of the top significant DEGs from total RNA-seq (Fig. 4b) show deficits associated with striatal oligodendrocyte (Apod, Klk6, Mog) and myelination genes (Mbp, Plp1, Pmp22)³⁶. We observed a significant downregulation of the ASD-susceptibility genes Gfap, S100b and Ttyh1, which are highly expressed in mature astrocytes. Other top impacted DEGs encode molecules involved in dopaminergic transmission, including reduced expression of Qdpr (which produces the essential cofactor BH4 required for dopamine synthesis) and Drd1 (which encodes dopamine receptor 1). Upregulation of excitatory transmission (Gria2, Grik3) and inhibitory transmission (Gabra5, Gabrb2, Gabrb3), as well as altered axon conduction (Scn3a, Kcnq5) (Supplementary Table 5) also suggests synaptic deficits.

KO snRNA-seq analysis revealed that neurons, particularly INs, followed by D2-MSNs and D1-MSNs, exhibited the highest number of FDR significant DEGs (Fig. 4c), compared with glia, with roughly equal proportions of significant upregulated DEGs and downregulated DEGs (Supplementary Table 10), consistent with the predominant neuronal-type expression patterns of Ptchd1-as. Within glial cell types, the same downregulated DEGs affecting myelination (Mbp, Mog) and mature oligodendrocytes (Klk6) that were found to be impacted in total RNA-seq analysis (Fig. 4b) were also detected in snRNA-seq (Extended Data Fig. 10a). Relative to MSNs, INs had more top FDR significant DEGs with larger log₂[FC] values (Fig. 4d), including two downregulated potassium channel genes (Kcnc, Kcnmb2), downregulated synaptic kainate receptor subunit (Grik1) and downregulated glutamate binding protein (Grip1). TCF4 is a high-confidence ASD-susceptibility gene⁴ and a significant DEG in striatal INs, along with another E-box-binding transcription factor, Zeb2, both of which are linked to rare NDDs³⁷. The top 20 common DEGs ranked by FDR significance (Fig. 4e) showed high congruency in log₂[FC] direction and degree of change for the D1-MSN and D2-MSN subtypes, which also impacted glia. Common cell type DEGs included the lncRNA Malat1, the maternally imprinted lncRNAs Meg3 and Mirg, the ubiquitin gene Ubb and the proneuronal cell-fate driver gene Ddx5, as well as the autism-relevant genes Gria2, Camk4, Rbfox1 and Csmd1. We also observed upregulation of Mir124a-1hg, which has been shown to directly interact with Gria2 transcripts sequestering them and influencing synaptic plasticity³⁸.

Extended Data Fig. 10. Additional western blotting.

a) Heatmap of Log2FC from KO snRNA-seq shows overall pseudo-bulk downregulation of oligodendrocyte markers and myelin genes are consistent with top significant DEGs in Total RNA-seq (Glia, balanced aggregate of all glia nuclei expression). Astrocytes are overexpressing Mbp and oligodendrocytes are upregulating Klk6 and Mog suggestive of neuroinflammation responses and myelination impairments. GFAP is upregulated in the striatum as assessed by b) immunohistochemistry and c) western blots (WT-/KO-2, N = 9/7; p = 0.041). d) FMRP protein levels are unaltered in the striatum (WT-/KO-2, N = 9/7) and cortex (WT-/KO-2, N = 7/9). e) PKC protein levels are unaltered in cortex (WT-/KO-2, N = 7/9). f) Levels and phosphorylation of Src, GluN2A, GSK3α and GSK3β are unaltered in the cortex (WT-/KO-2: N = 9/7). Western blot data normalized to in-gel total protein stain (see Supplementary Fig. 1 for full blot images and Supplementary Table 18 for full statistical information). Data are mean ± s.e.m., two-sided unpaired t-test, * p < 0.05.

We identified common dysregulated pathways and clusters using gene set enrichment analysis (GSEA) in total RNA-seq DEGs from Ptchd1-as KO, KO-1 and KO-2 (Extended Data Fig. 8a and Supplementary Table 11) in the striatum. Impacted pathways affected mainly synaptic transmission, synaptic plasticity, voltage-gated channel activity, microRNA activation and translation and were further supported by similar results in snRNA-seq (Supplementary Tables 12–14). We further examined the top ranked 150 pathways for each striatal cell type (Extended Data Fig. 8b), requiring that each pathway be significant in at least one cell type (FDR < 0.05). As expected, given the higher expression of native Ptchd1-as in D1- and D2-MSNs, these were the most affected cell types with highest pathway normalized enrichment scores (NES) in common pathways: upregulated pathway themes included myelin sheath, translation, transcription, immune and inflammatory signalling, and energy production pathways such as glycolysis and oxidative phosphorylation, whereas downregulated pathways involved voltage-gated channels and ion regulation. Neuronal pathways related to synaptic function showed MSN-subtype-specific NES directionality and, within glia, oligodendrocytes exhibited an upregulated myelin sheath pathway like MSNs, consistent with neuronal expression of Mbp. Together, Total RNA-seq and top snRNA-seq pathways from our GSEA analysis converged on highly similar pathways that involved alterations in energy metabolism, ion-channel regulation and neuronal and synaptic function.

Extended Data Fig. 8. Gene set enrichment analysis of Total- and snRNA-seq transcriptomes from adult Ptchd1-as KO striatum.

a) Individual colonies (KO-1, KO-2) and combined (KO) demonstrate common synaptic transmission/plasticity related pathways. (WT-/KO-1, WT-/KO-2, WT/KO, N = 6/6, 4/4, 10/10). Significant pathways (FDR < ± 0.1; NES, normalized enrichment score). b) Heat map with NESvalues from a cell type GSEA for KO vs WT mice are shown for the top-ranked 150 pathways across eight striatal cell types. Dominant alterations occurred in D1- and D2-MSNs with upregulated pathway themes primarily in translation, mixed effects on synaptic/synaptic transmission/plasticity pathways and downregulation in voltage gated channels/ion regulation gene sets. snRNA-seq is WT/KO, biological N = 2/2 mice (see Supplementary Table 6 for summary of N values (nuclei counts per cell type) Warmer colours indicate positive NES and cooler colours indicate negative NES. Pathway ranking by FDR × |NES| where at least one cell type has FDR < 0.05.

Comparison of total versus pseudo-bulk DEGs showed that 682 were common significant DEGs with a positive Pearson correlation between P values (r = 0.33; Fig. 4f and Supplementary Table 15) and highly consistent in log₂[FC] congruency (82%; Fig. 4g). Eight out of the top twenty correlated DEGs (P < 0.05) are related to myelin production (Apod, Mbp, Mog, Plp1, Mobp, Trf, Mal) and glial cell differentiation³⁹. Moreover, the ASD-susceptibility genes Gria2 and Grm5 are prominently correlated DEGs involved in synaptic plasticity regulation. Collectively, our data suggest that disruption of Ptchd1-as in the striatum alters myelination and synaptic function.

Striatal astrocytic gliosis

The striatal proteome of KO-2 mice was used to analyse the impact of transcriptional changes and to detect additional translation-level alterations (Fig. 4h and Extended Data Figs. 9 and 10). Out of around 2,455 proteins, captured by tandem mass tag mass spectrometry (TMT-MS), we obtained 107 differentially expressed proteins (DEPs, P < 0.05), of which 30 DEPs were significant at FDR < 0.1 and FC > ±10% (Fig. 4h). By contrast, no FDR significant DEPs were observed in the cortex (prefrontal) and only 38 DEGs reached P < 0.05, (Supplementary Table 16), demonstrating high tissue specificity.

Extended Data Fig. 9. Ptchd1-as KO multi-omics comparisons and striatal western blotting.

a) Multi-omics cross-analysis of all significant (p < 0.05; see specific p-values in Supplementary Tables 5, 9, 16; Wilcoxon rank-sum test and p-adjusted value determined post-hoc by Benjamini-Hochberg FDR correction) DEPs and DEGs (Total and Pseudo-bulk snRNA-seq) show 20 significant genes found in all three data sets including, myelin basic protein (MBP) and Ppp3r1, the calcium sensing subunit of the serine/threonine protein phosphatase complex calcineurin (also known as PP2B) involved in synaptic plasticity and all three cPKCs. General translation co-DEG and DEPs include downregulated ribosomal subunits Rps28, 20S proteasome subunit and Psmd2 suggesting impaired translation and proteasome instability. Proteins involved in transcription processes were also detected DEPs (Sub1, Hnrnpdl, Hnrnpa0, Hnrnpd and Ncl; see Supplementary Table 16). Dagger symbol represents ASD risk genes. Transcriptomics WT/KO, biological N = 2/2 mice (snRNA-seq, see Supplementary Table 10 for summary of N values (nuclei counts per cell type). Proteomics (WT-/KO-2, N = 4/4). Differential expression was performed using two-sided statistical tests DESeq2 (transcriptomics; Wald test) and edgeR (proteomics; likelihood ratio test). Data are shown as mean log2 fold changes. Multiple testing correction was performed using the Benjamini-Hochberg FDR procedure b-d) Western blots of striatal tissue validating the reduction in b) PKCα p = 0.0383 (WT-/KO-2 N = 7/5). c) Increases in phosphorylation state (pSrc Y416; p = 0.0363) and (pGluN2A; p = 0.0383), but no change in total Src or GluN2A or d) Increases in phosphorylation state (pGSK-3α S21, p = 0.0328; pGSK-3 β S9, p = 0.0439);WT-/KO-2: N = 9/7; but no change in total GSK3-α and GSK-3β. Data normalized to in-gel total protein stain see Supplementary Fig. 1 for full blot images). Data are mean ± s.e.m., two-sided unpaired t-test, * p < 0.05.

The top transcriptional glial DEGs Gfap and Mbp were also represented as significantly upregulated proteomic DEPs (GFAP and MBP; Fig. 4h and Extended Data Fig. 9a, respectively) in the striatum but not the cortex (Supplementary Table 16). MBP/Mbp was identified as differentially expressed in all multi-omic datasets (Extended Data Fig. 9a) and across multiple cell types but only upregulated in astrocytes (Extended Data Fig. 10a). MBP/Mbp expression is normally restricted to oligodendrocytes but is expressed in reactive astrocytes for immune responses in the brain⁴⁰. Astrocytic GFAP upregulation (a hallmark of neuroinflammation processes⁴¹ in the striatum) was further validated by immunohistochemistry and western blotting (Extended Data Fig. 10b,c). Together, increased astrocytic MBP and GFAP, without overt hypertrophy, suggests that Ptchd1-as disruption causes a mild striatum-specific reactive astrocytic gliosis, consistent with a mild neuroinflammatory response.

Altered striatal protein expression

Top significant striatal DEPs by FDR (Fig. 4h) were downregulated and clustered into families comprising linker histone-1 family (H1.3 and H1.2, except H1.4, which was significant at P < 0.05), cPKC family (PKCα (Prkca), PKCβ (Prkcb) and PKCγ (Prkcg)) and the PKC scaffolding family annexins (Anaxa5, Anaxa6, Anaxa7). Cross-comparison of each dataset yielded 20 high-confidence DEPs, which were also significant transcriptome DEGs, including ASD-susceptibility proteins such as the post-synaptic density protein DLG2, sodium channel SCN2A, and two cPKCs, PRCα and PRCβ (Extended Data Fig. 9a). This cross-analysis also identified other proteins of interest, such as PKCγ and the regulatory subunit of the phosphatase calcineurin (PPP3R1).

Given the links to plasticity in our multi-omics analysis, we validated PKCα alongside key synaptic molecules, including the NMDA receptor subunit, GluN2A and regulatory kinases (SRC and GSK3α/β)^42,43 (Extended Data Fig. 9b–d) and ASD, ID-relevant fragile-X messenger ribonucleoprotein (FMRP) (Extended Data Fig. 10d), which was unchanged in our western blot experiments. Consistent with the proteomics, we observed reductions in PKCα levels in the striatum (Extended Data Fig. 9b) but not in the cortex (Extended Data Fig. 10e). Notably, we found increased SRC phosphorylation (indicative of its activation) along with elevated phosphorylation of GluN2A (Extended Data Fig. 9c) and both GSK-3α and GSK-3β (Extended Data Fig. 9d), within the striatum but not in the cortex (Extended Data Fig. 10f). These phosphorylation events, combined with downregulated cPKCs, suggest that Ptchd1-as is a key regulator of striatal synaptic function. While cell-type-specific effects cannot be ruled out, the lack of changes observed in prefrontal cortical tissues using proteomics and western blotting suggests a larger role for striatum in the ASD phenotype.

Altered corticostriatal synaptic plasticity

Human brain imaging and mouse model studies implicate a role for the striatum in ASD^44,45. Given the comparatively higher expression of Ptchd1-as in the striatum, and the specific multi-omic alterations found implicating dysregulated proteins that are involved in striatal plasticity, we investigated synaptic function in this region^46,47. Recordings were made from the medial dorsal striatum and stimulation was applied in the lateral dorsal striatum, close to the corpus callosum, to activate the corticostriatal pathway. The NMDAR/AMPAR ratio (Extended Data Fig. 6h) and properties of AMPAR- and NMDAR-mediated spontaneous EPSCs (sEPSCs) (Extended Data Fig. 6i,j) were similar in KO-2 and WT-2 mice. Moreover, there was no alteration in the relationship between stimulus intensity, fibre volley amplitude and synaptic response, indicating that synaptic transmission is unaltered in Ptchd1-as KO mice (Fig. 5a). However, we found that, in Ptchd1-as mice, there is an enhancement of both NMDAR-mediated LTP (Fig. 5b) and mGluR-mediated LTD (Fig. 5c). In contrast to in the striatum, mGluR LTD was unaffected in the hippocampus (Fig. 5d). Notably, the enhancement of LTD in the corticostriatal pathway was mimicked in WT-2 mice by inhibition of cPKC isoforms using a highly selective antagonist: Gö6983 (Fig. 5e). Furthermore, Gö6983 had no additional effect in Ptchd1-as KO-2 slices (Fig. 5e), showing that cPKC inhibition both mimicked and was occluded by the effects of Ptchd1-as disruption. This observation provides a functional correlate of the reduced cPKC activity in the striatum of KO mice identified in the transcriptomic and proteomic studies.

Fig. 5. Ptchd1-as disruption impacts synaptic plasticity in striatal slices from adult mice.

a, Input–output curves plotting fibre volley (FV) amplitude as a function of stimulus intensity, fEPSP amplitude as a function of FV amplitude and paired-pulse ratio for WT-2 and KO-2 mice. n = 18/9 (slices/mice), for each group. b, A subthreshold theta-burst stimulation (TBS; 30 bursts of 4 pulses at 100 Hz, delivered at 5 Hz) can induce LTP in KO-2 mouse striatal slices. WT-2, n = 17/13; KO-2, n = 16/13. Two-sided unpaired t-test; P = 0.0041. c, LTD, induced by 50 µM (S)-DHPG delivered for 10 min (black bar), was decreased in WT-2 mice (8.6 ± 2.9%) compared with KO-2 mice (19.7 ± 2.6%). n = 16/13 in both groups. Two-sided unpaired t-test; P = 0.0073. d, DHPG LTD was similar in WT-2 (15.9 ± 3.8) and KO-2 (14.4 ± 2.5) mouse interleaved hippocampal slices. n = 8/4 in both groups. Unpaired t-test; P = 0.7456. e, Gö6983, a specific PKC inhibitor, enhanced DHPG LTD in WT-2 (22.4 ± 4.9%) mice but had no additional effect in KO-2 mice (19.8 ± 4.5%). n = 8/5 in both groups. Unpaired t-test; P = 0.7032. For b–e, the time courses and bar plots show mean ± s.e.m. The traces are averages of synaptic responses for baseline (region i) from WT-2 (grey) or KO-2 (blue) mice and either after DHPG or after TBS (both red; region ii). *P < 0.05, **P < 0.01. f, A summary of the main findings of this study. PTCHD1-AS/Ptchd1-as (exon 3) is an autism-susceptibility gene, whereas the protein coding gene PTCHD1/Ptchd1 is predominantly involved in ID/NDD. Ptchd1-as KO altered many genes, particularly those involved in euchromatin organization, myelination and synaptic plasticity. Deletion of exon 3 results in reduced cPKC and enhanced DHPG LTD, which may contribute to some of the autistic-like behaviours that are exhibited by the KO mice, such as repetitive behaviour. The diagram in f was partially created using BioRender; Bradley, L. https://BioRender.com/63pmpwc (2026).

Discussion

Here we demonstrate that disruption of a segment of DNA (exon 3), part of the complex PTCHD1-AS lncRNA locus, in male human individuals and mice, is strongly associated with the social, communicative and repetitive behavioural traits consistent with ASD core features. At the molecular level, our previous¹³ and current data also indicate that genetic or transcript knockdown disruptions in PTCHD1-AS/Ptchd1-as human/mouse lncRNA do not alter the expression of the PTCHD1/Ptchd1 protein-coding gene.

Given that our findings in this study strongly implicate PTCHD1-AS in autism, that parallel work also implicates the protein coding gene DDX53 in ASD¹⁴ and that these two genes interact (Fig. 2e), PTCHD1, PTCHD1-AS and DDX53 were subjected to a formal evaluation of autism gene link (EAGLE)² assessment. EAGLE determines the strength of evidence linking specific genes to individuals with a formal diagnosis of ASD only (excluding broader NDD or ID alone). Considering all available data currently in the literature, including using the novel exons from this study, EAGLE scores for PTCHD1-AS, DDX53 and PTCHD1 were 33.6, 10.2 and 5.2, respectively (note that the cut-off score for definitive ASD-association is >12; Supplementary Table 17). In addition to the rare pathogenic variants in PTCHD1-AS described here and also in DDX53 in autism¹⁴, along with the potential role of PTCHD1 in the neurodevelopmental phenotype, we have also recently found common genetic variants by XWAS analysis⁴⁸ directly adjacent to PTCHD1-AS exon 4/DDX53.

The unique phenotypic profile of Ptchd1-as mutants compared with other ASD-associated-gene mouse models is summarized in Supplementary Table 1. There are some commonalities with models such as Fragile X/FMRP, MeCP2 and ADNP, in that dysregulation at the transcription and translation level are seen for chromatin and synaptic proteins⁴⁹. As Ptchd1-as-mutant mice do not exhibit impairments in learning and memory, it may now be possible to start to delineate the underlying basis for the social and intellectual alterations. For example, whereas Fragile-X mice show altered synaptic plasticity in the hippocampus, a region that is highly associated with learning and memory function, Ptchd1-as-mutant mice do not. In the corticostriatal pathway, the threshold for inducing NMDAR LTP was lowered and the magnitude of mGluR LTD was enhanced in the KO. This latter observation supports the idea that altered mGluR function is relevant to the autism phenotype⁵⁰.

A working model linking human genetic findings found to be associated with core features in autism to altered synaptic function in mice is shown in Fig. 5f. The absence of PTCHD1-AS/Ptchd1-as lncRNA in male individuals/male mice affects the expression and activation state of numerous proteins, including those involved in myelination, histone regulation and synaptic function. One such effect of the KO of Ptchd1-as is a reduction in the levels of conventional PKC isoforms and alterations in synaptic plasticity in the striatum—effects that may be relevant to repetitive behaviours observed in autism. Further definition of PTCHD1-AS lncRNA-influenced molecular, cellular and circuit-level pathways may offer common modulation targets influencing the core features of autism.

Methods

Phenotype–genotype analysis

Ethics statement

This research involving human participants complies with all relevant ethical regulations, including obtaining informed consent from all participants through the recruitment sites. Ethical review and approval were obtained from The Hospital for Sick Children Research Ethics Board (REB 1000080561).

ASD-associated genetic variants at the PTCHD1-PTCHD1-AS locus in ASD cohorts

We analysed MSSNG, Simons Simplex Cohort and SPARK ASD WGS databases for genomic variants at the PTCHD1-PTCHD1-AS locus, with ethics approval and informed consent for the MSSNG database as previously described^4,51. MSSNG cohorts contain family data with at least one child having a diagnosis of ASD. Rare microdeletion variants are defined as those less than 1 Mb in size and occur at a frequency of below 1% in control population cohorts. Variants meeting these criteria are then evaluated for ASD risk, and corresponding phenotype data are collected.

Frequency impact of rare deletions across the PTCHD1-PTCHD1-AS locus

We conducted a comparative analysis of the frequency of rare copy-number deletions (populational frequency less than 1%) overlapping the target exons on PTCHD1-AS, as well as PTCHD1 and DDX53. We assessed deletions with a length of 1 kb or more, identified as previously described⁵² across three independent ASD cohorts (MSSNG, SSC, and SPARK)^4,15,53, contrasting with six independent control cohorts (1000 Genomes Project, 1,234 male individuals⁵⁴; MGRB, 1,756 male individuals^55,56; HostSeq, 4,235 male individuals⁵⁷; CHILD, 203 male individuals⁵⁸; INOVA, 100 male individuals, www.inova.org); cardiomyopathy⁵⁹ and congenital heart disease cohorts, 804 male individuals⁶⁰). Our analysis exclusively considered high-quality deletions, meeting the following criteria: (1) length of at least 1 kb; (2) identification by both ERDS⁶¹ and CNVnator⁶², with at least 50% reciprocal overlap between the two methods; (3) less than 70% overlap with repetitive or low complexity genomic regions (such as telomeres, centromeres and segmental duplications); (4) exclusion of CNVs in the pseudoautosomal regions or X chromosomal calls in males. We then used Fisher’s exact tests to compare the occurrence of overlapping deletions in male individuals within each target region between probands and control cohorts, resulting in final P values and ORs. The final comparison includes all PTCHD1-AS from exons 2 to 5, including the novel exons. We also analysed deletions overlapping DDX53 and PTCHD1 for comparison purposes.

Analysis of genotype–phenotype relationship in NDD cohorts at the PTCHD1-PTCHD1-AS locus

In ASD-specific cohorts with WGS data or any technology and primary diagnosis of ASD, individual data from the publicly available database DECIPHER v.11.27 and Lineagen (a private genetic diagnostic company) were accessed in September 2024 and February 2017, respectively. The individual selection and cohort descriptions are summarized in Supplementary Table 18. Here the primary diagnosis is the main NDD. Secondary diagnoses may include other disorders (such as ADHD, ADD, obsessive–compulsive disorder), other neurological conditions (epilepsy, presence of seizures) or additional mental health conditions (anxiety). Records of each database were obtained first by searching for PTCHD1-AS or PTCHD1 loss of function genetic variants in all databases or published cohorts.

Mouse models

We used Benchling (http://benchling.com/) to design guide RNAs and excised mouse Ptchd1-as (Gm15155-201) exon 3 (237 bp) and 150 bp on either side, generating two mutant mice both containing a patient-like deletion of exon 3: (1) Ptchd1-as^−Ex3 (KO-1); and (2) with an insertion of 168 bp intronic tandem poly(A) sequence⁶³ within the lncRNA transcript, termed Ptchd1-as^−Ex3-is (KO-2). Guide RNAs (5′-AGGTTAGCATTATACCACTG-3′ and 5′-GTCTCCACATTTACATACTC-3′) on each side of Ptchd1-as exon 3 were used to create double-stranded breaks and KO-1 mice. A single-stranded intronic oligonucleotide (TCATGATGTTTTGTCCAGGAATAGAAACCCTGACTAAGATACTAGGTTAGCATTATACCAAAATAAAATACGAAATGTGACAGAAAATAAAATACGAAATGTGACAGACTCAGGTTTGTCTACTTTTCTTCATGTTTTAGGAATACGAGACTTATGGACACCAATAAT), including the tandem copies of the neuropilin-1 poly-adenylation sequence (bold), was included with guide RNAs and Cas9 in the injection mix to generate KO-2 mice.

CRISPR–Cas9 editing was performed at the Centre for Phenogenomics as previously described⁶⁴. Founder mice were backcrossed 3–5 generations onto strain C57BL/6J and to refresh the line every third generation. KO-1 and KO-2 mice were validated by PCR break-point analysis. Off-target analysis was performed by WGS and comparison to the C57BL6/J reference sequence using Mutect2 (GATK3.7). No exonic mutations were detected in the initial backcrossed line for either strain.

The primers for genotyping (forward, 5′-GAACAGTGGTTTGGAGGTGTAA-3′; reverse, 5′-TGTTCTGTGAGTTGGGCATATC-3′) detect a single band at 364 bp for KO-2 and at 314 bp for KO-1 mice, representing hemizygous males. WT male mice have a 553 bp fragment.

Mouse management conditions

KO-1 and KO-2 (male hemizygous; C57Bl/6J) mice and their WT littermate mice were used for all behaviour experiments. The mice were bred at the Hospital for Sick Children or The Centre for Phenogenomics, Neurobehavior Core and group-housed in cages with 3–5 mice per cage. The housing conditions maintained a constant temperature of 22 °C and a 12 h–12 h light–dark cycle, with food and water available ad libitum. All of the procedures were approved by the Hospital for Sick Children Animal Care and Use Committee (AUP 49151) and The Centre for Phenogenomics Animal Care and Use Committee (AUP 25-0307; 20-0307H) and conducted in accordance with Canadian Council on Animal Care guidelines. Electrophysiological recordings were carried out at the Lunenfeld Tanenbaum Research Institute in Mount Sinai Hospital, Toronto (AUP 24-0292H) and at the University of Toronto’s, Tanz Centre for Research in Neurodegenerative Diseases, University Hospital Network (AUP 6668.8). Mice were bred and group-housed under similar conditions as described above. Tissues samples for electrophysiology, ddPCR, RNA-seq and proteomics were obtained mainly from mice housed at the TCP unless otherwise stated. Age-matched mice between 8 and 12 weeks of age were used for testing in 2 to 3 different behaviour tests. All of the experiments took place during the light phase and were conducted and analysed in a blinded manner, with the experimenter unaware of the experimental conditions. The sample sizes were determined on the basis of established standards in the field and previous experience with phenotype comparisons. No statistical methods were used to predetermine the sample size.

MRI brain imaging

In total, 50 mice were used for the brain imaging experiments (23 hemizygous male mutants and 27 WT littermate male controls). Each mouse was scanned repeatedly from the early post-natal period into young adulthood according to methods and procedures described previously⁶⁵. The precise time of scans was jittered for each mouse, and not all mice were continued into adulthood.

Image acquisition

Twenty-four hours before each scan, a 0.4 mmol kg⁻¹ dose of 30 mM manganese chloride (MnCl₂) was administered as a contrast agent. For mice that were 10 days old or younger, the dam was intraperitoneally injected with MnCl₂, and pups received MnCl₂ through maternal milk. Mice that were over 10 days of age received intraperitoneal injections of MnCl₂. Up to four mice were scanned simultaneously. Custom-built 3D-printed holders, which allowed anaesthetic delivery and scavenging, and heating were used. During the scan, mice were anaesthetized with 1–2% isoflurane, and the respiratory rate was monitored using a self-gated signal from a modified 3D gradient echo sequence⁶⁶.

A multi-channel 7.0 T MRI scanner with a 30 cm diameter bore (Bruker), equipped with four individual cryogenically cooled coils was used to acquire images of the mouse brains. Parameters of the scan are as follows: T1-weighted FLASH 3D gradient echo sequence, TR = 26 ms, TE = 8.250 ms, flip angle = 23°, field of view = 25 × 22 × 22 mm, with a matrix size of 334 × 294 × 294, yielding an isotropic imaging resolution of 75 μm. The imaging time was 58 min. After imaging, mice were transferred to a heated cage for 5–10 min to recover from the anaesthesia and then returned to their home cage.

Image processing

Images were grouped into age bins centred around post-natal days 3, 5, 7, 10, 17, 23, 29, 36 and 65. The images were then iteratively aligned within their bins using a mix of linear and nonlinear registration within the PydPiper framework⁶⁷. Averages of each timepoint were then aligned towards each other in a chain, registering the P3 average to the P7 average, P7 to P10, and so on. The images were also segmented with the DSURQE atlas⁶⁸ using the MAGeT multi-atlas framework⁶⁹.

All analyses were conducted at the automatically segmented ROI level. First, we aimed to detect whether genotype influenced the development of any brain structures using equation (1):

$${\text{Volume}}_{i,t}={\beta }_0+{\beta }_1\times {\text{bv}}_{i,t}+{\beta }_2\times {\text{Genotype}}_i+f_1\left({\text{age}}_{i,t}\right)+f_2\left({\text{age}}_{i,t},{\text{Genotype}}_i\right)+f_3\left({\text{subject}}_i\right)+{\mathit{ϵ}}_{i,t}$$ 1

where β₀ is the intercept; β₁ is the fixed-effect coefficient to remove the effect of overall brain volume, assumed constant for both genotypes; β₂ is the fixed-effect term for genotype representing any global offsets present across the developmental period; f₁(age_i,t) is a cubic regression spline smooth function, constrained to have a maximum basis dimension of k = 5, estimated for the WT mice (as they are the reference level factor for genotype); f₂(age_i,t,Genotype_i) is a cubic regression spline smooth function, constrained to have a maximum basis dimension of k = 5, estimating the difference in slopes between hemizygous mutants and WT mice; f₃(subject_i) represents the random intercept to account for the longitudinal nature of the data; ${\mathit{ϵ}}_{i,t}~N\left(0,{\sigma }^2\right)$ residual error at each timepoint.

This equation was implemented as a general additive model using the mgcv⁷⁰ packages in R. P values for the smooth term f₂(age_i,t,Genotype)i were then then computed using summary.gam and corrected for multiple comparisons using the FDR.

Next, with a slightly more relaxed model that estimates separate slopes for hemizygous mutants and WT controls, we estimated when and where differences in mean volumes and differences in developmental slopes emerged using Equation 2:

$$\begin{array}{l}{\text{Volume}}_{i,t}={\beta }_0+{\beta }_1\times {\text{bv}}_{i,t}+{\beta }_2\cdot {\text{Genotype}}_i\\ +f_1\left({\text{age}}_{i,t},{\text{Genotype}}_i\right)+f_2\left({\text{subject}}_i\right)+{\mathit{ϵ}}_{i,t}\end{array}$$ 2

To understand the developmental patterns of brain growth, we next tested whether genotypes were different in volume and/or in slope at every day of age between P5 and P90 using estimated marginal means.

Mouse behaviour

Three-chambered social interaction

Mouse sociability was assayed as adapted from the original study⁷¹. Each mouse was placed into a plexiglass three-chambered apparatus and habituated in the centre chamber I for a period of 5 min. For the sociability assay, while the experimental mouse remained in the chamber C, a female mouse (M1) was introduced into a cylindrical wire cage on one of the side-chambers, and an identical empty wire cage (O) was placed on the other side-chamber. The guillotine doors were lifted to allow the experimental mouse to freely explore the three chambers and interact with M1 and O for 10 min. For the social novelty recognition assay, after the sociability assay, the experimental mouse was returned to chamber C by closing the guillotine doors for 5 min. A novel non-cagemate female mouse (M2) was then introduced into the formerly empty wire cage. The guillotine doors were lifted, and the experimental mouse was allowed to interact with M1 and M2 for 10 min. All behaviours within the chambers were recorded using a video camera and subsequently analysed using the ANY-maze tracking system (ANY-maze). The amount of time spent in each chamber during the sociability and social preference assays was automatically calculated for statistical analysis.

USV recording and analysis

During the sociability assay, USVs from mice were recorded by a microphone hung on the three-chambered arena. The USV calls occurring at frequencies ranging between 25 Hz and 125 Hz were automatically filtered by the UltraSoundGate system (Avisoft Bioacoustics). The audio files were then analysed using MUPET²⁴ in MATLAB R2022a (MathWorks). For data processing and individual syllable analysis, the default configuration parameters were used, except for noise reduction set to 10, to minimize background noise and extract individual syllables. The number of syllables, syllable frequency and energy (decibel) were analysed using MUPET. For syllable repertoire analysis, on the basis of the extracted individual syllables, a syllable size of 40 was chosen to build syllable repertoires for the groups of KO-1, KO-2 and their WT littermates. Then, two of the group repertoires were compared, whereby the repertoire elements of one group were sorted according to the best match to the elements of the other group. A similarity matrix and score were generated for statistical comparison. All data processing in the MUPET analysis was adapted from previous studies^24,72.

Olfactory cue-reactivity

Anosmia was assessed as adapted from the olfactory habituation–dishabituation test²¹. Each mouse was placed into an empty cage and presented with a series of conspecific urine odours over seven trials, each lasting for 3 min with a 2-min intertrial interval. Urine was collected from a total of 8–10 mice from 2–3 cages of female or male (8–12 weeks; C57BL/6J) WT mice. During the initial two blank trials (trials 1 and 2), a small, empty plastic mesh chamber was placed at one end of the cage. From trials 3 to 6, the mouse was exposed to urine pooled from a cohort of female mice (the same odour was used for all trials). In total, 50 μl of urine was placed on filter paper inside the chamber and positioned at one end of the cage at the beginning of each trial. On the final trial, the mouse was exposed to urine from a different cohort of non-cagemate male mice. The trials were recorded and digitized using Limelight software (Actimetrics). The amount of time spent sniffing the chamber was recorded for statistical analysis.

Grooming

Each mouse was placed into a plexiglass circular chamber and behaviour was video recorded for a period of 10 min. The amount of time spent on repetitive self-grooming (such as stroking the face or body and licking the forepaws or body) was quantified for statistical analysis.

Pre-pulse inhibition

Each mouse was placed into a plexiglass cylinder of the SR-LAB startle control box (San Diego Instruments). Acoustic startle and pre-pulse stimuli were delivered using a high-frequency speaker placed 20 cm away from the testing cylinder. The startle amplitude was determined as the maximum response within 100 ms after presenting the startle stimulus. Background noise levels were maintained at 65 dB. After a 2-min acclimation period, mice were presented with a series of trials. The startle stimulus (40 ms duration, 120 dB) or prepulse/startle stimuli were presented. A total of five different prepulse intensities (69 dB, 73 dB, 77 dB, 81 dB, 85 dB) was tested. For each prepulse intensity, there were 12 prepulse stimulus-only trials and 12 prepulse/startle stimulus trials, plus 24 startle-only trials. The trials were spaced 15 s apart and intermixed. Analysis was adapted from previous studies⁷³.

Contextual fear conditioning

Each mouse was placed into the chamber for 5 min. After an initial 2 min of free exploration, the mouse received three foot-shocks (0.5 mA, 2 s duration, 1 min apart). The mouse was then returned to its home cage 1 min after the last shock. Retrieval testing was performed the next day, in which the conditioned mouse was placed back into the same context. The time spent immobile (that is, freezing) was recorded for 5 min using an automated FreezeFrame scoring system (Actimetrics) for statistical analysis.

Open field

Each mouse was placed into an opaque white open-field arena. Ambulatory behaviour was video recorded for 30 min and automatically analysed using LimeLight software (Actimetrics). For the analysis of locomotor activity, the total distance travelled in the open field arena was calculated. To assess anxiety-like behaviour, the time spent in each zone (that is, inner zone, middle zone and outer zone) was measured. An increasing time spent in the peripheral outer zone was considered an indicator of anxiogenic behaviour.

Gait analysis

Non-toxic paint was applied to the forepaws (blue) and hindpaws (orange) of each mouse, and then the mouse was allowed to walk through an alley (10 cm in width × 50 cm in length), the floor of which was lined with white paper. The distance between each step (that is, stride length) was measured and averaged, as was the distance between left and right paws (that is, stride width) for statistical analysis.

Ladder walking

A horizontal ladder was set up 20 cm above the floor to assess motor coordination. The metal rungs of the ladder were arranged in an irregular pattern with pseudorandom spacing. During the trial, each mouse was placed onto one side of the ladder and allowed to traverse to the other side. The trial was recorded and subsequently reviewed in slow motion to count the number of times the paws slipped through the rungs for statistical analysis.

Rotarod

Each mouse was placed onto an accelerating Rota-rod system (Med Associates) to assess motor learning. The mouse underwent a 5-day training period, consisting of four trials per day on the Rota-rod. Each trial had a maximum duration of 5 min, and a 10 min intertrial interval was provided for recovery and rest. During each trial, the speed of rotation on the Rota-rod accelerated linearly from 4 rpm to 40 rpm. The latency of mouse retention on the rod before falling was automatically recorded. The latency values of the four trials conducted each day were averaged and analysed for statistical analysis.

Touchscreen PD task

We used the Bussey–Saksida Touchscreen System (Lafayette Instrument) as a conditioning/cognitive testing paradigm as previously reported with minor modifications³⁰. In brief, after mild food restriction was introduced in which mice received one to two food pellets (Bioserve dustless precision 1 g pellets) per day and maintained 80% initial body weight throughout the experiment, 10-week-old WT and KO-2 mice were habituated to the chambers and to food rewards (strawberry milkshake (Neilson), stored at 4 °C for no more than 3 days after opening) for at least two daily sessions in their home cage. We then introduced the pairing of a visual stimulus and delivery of reward in a series of consecutive behaviour shaping steps. The pairwise discrimination (PD) task was then introduced, in which mice must learn that one of two images displayed simultaneously on the screen in the chambers is correct. Mice were rewarded while touching the correct conditioned stimulus (CS⁺), while touching the incorrect conditioned stimulus (CS⁻) resulted in house lights on and a correction trial, then transition to a new trial. Mice first learned to discriminate between marble versus fan (dissimilar spectral characteristics), then the primary PD was introduced containing images with equal spectral characteristics (left versus right) sloping lines. Once this PD task had been learned and the mice reached mastery criteria (2 consecutive days achieving 80% correct trials), stimuli were reversed so that the CS⁺ stimuli became the CS⁻ stimuli (PD-reversal). Activity in the chamber was assessed automatically. All mice were singly housed to ensure that each received the standard feed of 1–2 g every 24 h undergoing food restriction. Mice received 120 µl of reward per successful trial. Trials lasted for 1–2 min depending on the success and speed at which the mouse had learned the task. Sessions ended after 60 min or completion of 30 trials. Sessions were run for 5–6 days per week.

Puzzle box test

A puzzle box test was used to evaluate mouse executive function, as well as learning and memory, as previously reported⁷⁴. The apparatus is a plexiglass box, divided by a removable barrier in an enclosed dark goal box (14 × 28 × 27.5 cm³) and a larger open white start box (58 × 28 × 27.5 cm³) with an underpass that allows mice to move freely between the two compartments. The mice were positioned in the open box facing the wall at the start of the test and latency time for mice to enter the goal box was recorded. All of the mice underwent a total of nine trials (T1–T9) over 3 consecutive days with increasingly difficult puzzles to solve that involved moving from the open light to the closed dark goal box. On day 1, mice could use an open doorway (T1, baseline) to access the goal box. In T2, the doorway was blocked, and the mice must learn to use an underpass to reach the goal box. This T2 challenge was repeated in T3 to assess recall of the solution (short-term memory) after putting them in a new cage for 2 min. On day 2, the previous day’s puzzle was repeated to measure long-term memory (T4) recall of the puzzle solution. Then, a new puzzle was introduced in which the underpass (T5) was obstructed with corncob bedding requiring mice to burrow through the bedding to gain access to the goal box, with subsequent memory trials performed afterwards. On day 3, a third puzzle was introduced in which the underpass was blocked with a cardboard plug (T8), and the mice had to learn to remove the plug to enter the goal box. A 2 min interval was maintained between T8 and T9 for measuring short-term memory again. A maximum time of 5 min was assigned to each mouse to reach the goal box.

Electrophysiology

Acute brain slices were prepared from KO-2 mice and their WT littermates. Mice were euthanized by decapitation under isoflurane anaesthesia. Brains were rapidly extracted and sectioned with a VT1200S vibratome (Leica) in ice-cold cutting solution composed of 205 mM sucrose, 26 mM NaHCO₃, 10 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 0.5 mM CaCl₂ and 5 mM MgSO₄ (saturated with 95% O₂ and 5% CO₂; pH 7.4). Dorsal (DH) and ventral (VH) hippocampal slices (400 µm) for field potential recordings were prepared by first hemisecting the brain and then slicing the hemispheres along their sagittal or horizontal planes, respectively. The CA3 region of slices was removed immediately after slicing for recordings from area CA1. For mPFC or striatum, coronal slices (300 or 350 µm, respectively) were prepared for whole-cell patch clamp recordings and striatal field potential recordings.

Slices were allowed to recover for a minimum of 1 h at room temperature (for hippocampal slices) or at around 35 °C for 30 min followed by room temperature recovery (for mPFC and striatal slices) in standard artificial cerebrospinal fluid (ACSF) composed of 124 mM NaCl, 26 mM NaHCO₃, 10 mM glucose, 3 mM KCl, 1.4 mM NaH₂PO₄, 2 mM CaCl₂ and 1 mM MgSO₄ (saturated with 95% O₂ and 5% CO₂; pH 7.4). After recovery, slices were transferred to a submerged-type chamber where they were continuously perfused with ACSF (2.5 ml min⁻¹) and maintained at 30 °C for electrophysiological recordings. Evoked responses were elicited by constant current stimulus pulses (100 µs) delivered through a platinum/iridium bipolar stimulating electrode (FHC). The signals were amplified using the Multiclamp 700B (Molecular Devices) system, filtered at 2–6 kHz, digitized at 20–40 kHz, and recorded to a personal computer for offline analysis using WinLTP software⁷⁵.

Extracellular field potentials were recorded using glass microelectrodes filled with ACSF (~1.5 MΩ). In hippocampal slices, FVs and field EPSPs (fEPSPs) were recorded from either CA1 stratum radiatum in response to stimulation of the Schaffer collateral/commissural pathway, or from the DG molecular layer in response to stimulation of the medial perforant path. Striatal field potentials were recorded from the dorsal striatum in response to stimulation adjacent to the overlying corpus callosum to activate corticostriatal fibres. The GABA_A receptor blocker bicuculline (10 µM) was added to the ACSF for recordings from the DG and striatum. In each slice, the stimulus intensity was determined as a function of the threshold required to elicit a visually detectable response. Input–output curves were generated by delivering stimuli at multiples of this threshold value (1–7×). The baseline stimulus intensity for paired-pulse facilitation (PPF) and plasticity experiments was set to 2–3× the threshold value. PPF in area CA1 was assessed across a range of interpulse intervals (50, 100, 150, 200 and 250 ms). Test pulses were delivered every 20–30 s, and four–six consecutive responses were averaged for analysis. For hippocampal recordings, FVs were quantified by their peak amplitudes and fEPSPs were quantified by their initial slopes. Striatal field potentials were quantified by their peak amplitudes or initial slopes. Conditioning stimuli for plasticity experiments were delivered after 20–30 min of stable baseline recording. In the CA1, LTP was induced by a TBS protocol (5 bursts of 5 pulses at 100 Hz repeated at 5 Hz) and LTD was induced by a low-frequency stimulation protocol (900 pulses at 1 Hz). In the DG, LTP was induced by a high-frequency stimulation protocol (4 trains of 100 pulses at 100 Hz repeated every 20 s) with the stimulus pulse width doubled to 200 µs. In the striatum, LTP was induced by a TBS protocol comprising 4 pulses at 100 Hz repeated 30 times at 5 Hz, and mGluR-LTD was induced by a 10 min bath application of the group I mGluR agonist (S)-3,5-DHPG (50 µM, Hello Bio). Gö6983 (1 µM, Hello Bio) was dissolved in 0–0.001% DMSO with ACSF and compared with vehicle alone. Levels of LTP or LTD were measured as the percentage change from the baseline during the last 10 min of the recordings.

Whole-cell patch clamp recordings were made from visually identified layer 5/6 pyramidal neurons of the mPFC, or from MSNs in the dorsal striatum. Cells were visualized using the Olympus BX-51WI upright microscope with infrared DIC optics and a ×40 water-immersion lens. Patch pipettes (2.5–5 MΩ) were fabricated from standard-wall (1.5 mm outer diameter, 0.86 mm inner diameter) capillary glass and filled with potassium- or caesium-based intracellular solutions (see below). For all recordings, giga-Ohm seals were obtained under voltage clamp and cells were allowed to equilibrate at a holding potential of −70 mV for at least 5 min after break-in. No corrections were made for liquid junction potentials.

Current clamp recordings from mPFC neurons were made using an intracellular solution composed of 130 mM K-gluconate, 12 mM KCl, 8 mM NaCl, 10 mM HEPES, 0.2 mM EGTA, 4 mM Mg-ATP and 0.3 mM Na-GTP (285–290 mOsm; pH 7.2–7.3). Series resistance and pipette capacitance were compensated using the bridge balance and capacitance neutralization functions of the amplifier, respectively. The resting membrane potential was measured in the absence of current injection (I = 0), and passive membrane properties were measured in response to hyperpolarizing current steps from rest. Active properties were measured in response to depolarizing current steps delivered from a starting potential of −70 mV, which was maintained by a manually adjusted background current.

For voltage-clamp recordings from the mPFC and striatum, patch pipettes were filled with an intracellular solution composed of 130 mM CsMeSO₃, 8 mM NaCl, 10 mM HEPES, 0.5 mM EGTA, 4 mM Mg-ATP, 0.3 mM Na-GTP and 5 mM QX-314 chloride (285–290 mOsm; pH 7.2–7.25). Pipette capacitance was compensated after formation of a giga-Ohm seal and series resistance was monitored throughout recordings. Cells were excluded if the series resistance was >20 MΩ or varied by more than 20%. Series resistance was not compensated. Bicuculline (10 µM) was included in the ACSF for all voltage-clamp recordings to block inhibitory currents, and external Mg²⁺ was increased to 2 mM for mPFC recordings to dampen excitability. Evoked EPSCs were recorded in response to 0.1 Hz stimulation of layer 2/3 for mPFC recordings, or 0.033 Hz stimulation of the corpus callosum for recordings from striatal MSNs. The stimulus intensity was set to evoke EPSCs with amplitudes of 100–200 pA when holding the cell at −70 mV. NMDAR/AMPAR ratios were obtained by first recording AMPAR-mediated EPSCs at −70 mV for a minimum of 10 sweeps. AMPAR-mediated transmission was then fully blocked by addition of NBQX (10 µM) to the perfusate (5–10 min), and cells were held at +40 mV to record the NMDAR-mediated EPSC for a minimum of 10 sweeps. Consecutive sweeps were averaged for measurement of the EPSC amplitudes, 10–90% rise times and 80–20% decay times.

Spontaneous EPSCs (sEPSCs) were recorded from striatal MSNs in continuous acquisition (gap-free) mode for a minimum of 5 min. sEPSC frequency, amplitude, 10–90% rise time and decay time constants were analysed using Mini Analysis software (Synaptosoft). Averaged waveforms of all detected events from each cell were used for rise and decay time analyses. Decay time constants were calculated from single-exponential (AMPAR) and double-exponential (NMDAR) curves fitted to the decaying phase of sEPSCs.

RNA extraction and ddPCR

Male C57BL6/J mouse brain samples or KO strains, with WT litter-matched controls, were dissected at various developmental timepoints in ice-cold HBSS (−Mg, −Ca; Wisent, 311-512-CL) cut into 5 × 5 mm tissue portions and preserved immediately in RNALater (Qiagen, 76106) according to the manufacturers protocol (4 °C, 16–48 h then transferred to −20 °C). Tissue samples were lysed in 600 μl Buffer RLT per 30 mg of tissue and homogenized by handheld automatic pestle cordless motor (VWR, 47747-370) then manual triturization with 21 G and 30 G needles or with the Fisherbrand bead mill homogenizer (RT, speed = 3.1, cycles = 02, time = 0:15 s, delay = 0.05) and 1.0 mm diameter zirconia/silica beds (Biospec, 11079110).

RNA extraction was performed using the RNeasy mini kit (Qiagen, 74104) according to the manufacturer’s protocol, with on column DNase treatment. RNA samples were checked on the Agilent Bioanalyzer 2100 RNA Nano chip for RNA integrity (Tissue average RIN > 9). Reverse transcription was performed using qScript cDNA Supermix (Quantabio, 95048) according to the manufacturer’s protocol with both random hexamers and oligodT primers. Primers to Ptchd1-as transcripts amplifying the exon–exon boundaries were custom designed by TCAG using Quantasoft (v.1.7.4). These and other probes for ddPCR are listed in Supplementary Table 19.

DdPCR was performed as previously described⁷⁶ on a QX-200 instrument (Bio-Rad). The housekeeping gene Tfrc was run in duplex for an internal loading control. After thermal cycling, the plates were transferred to the Bio-Rad QX200 Droplet Reader and probe signal amplification was analysed in absolute quantification mode. QuantaSoft software was used to analyse resulting data. The normalized copy-number ratio was calculated as the absolute copy number for the gene or exon pair of interest/housekeeping gene Poisson ratio.

Antisense LNA gapmer-mediated knockdown

Human fetal neural stem cells (HF6562) were obtained from P. Dirks (The Hospital for Sick Children). This cell line is derived from human fetal brain tissue. This cell line corresponds to HF6562, as annotated in Cellosaurus (RRID: CVCL_C8ZT). F6562 cells were not authenticated in our laboratory. Cell identity was based on the source laboratory. HF6562 cells were not tested for mycoplasma contamination in our laboratory. HF6562 cells were transfected with an antisense LNA gapmer targeting exon 1 of the PTCHD1-AS AS2 isoform (sequence: 5′-GCATAAGTGAAAGGTA-3′; Qiagen, 339512, LG00818734-DFA) or negative control A (5′-AACACGTCTATACGC-3′; 339515, LG00000002-DDA) at 50 nM in a six-well plate using Lipofectamine RNAiMAX (13778075, Thermo Fisher Scientific), according to the manufacturer’s instructions. Cells were collected 48 h after transfection and analysed using ddPCR for expression of PTCHD1-AS exons, DDX53 and PTCHD1. Copy-number counts were normalized to TFRC, and gene expression was calculated relative to the negative-control gapmer.

Total bulk RNA-seq analysis

Exon–exon splicing of native Ptchd1-as

Deep sequencing of C57BL6/J male mice brain (left hemisphere) and cortex (right hemisphere) tissue at P7 was used to assess Ptchd1-as exon–exon splicing. Libraries were prepared with TruSeq Total RNA Ribo Zero Gold with rRNA depletion and sequenced on the Illumina HiSeq2500 system, with paired-end reads 2 × 150 bp, at 100–70 million read depth.

Total bulk RNA-seq sample processing

Total RNA from KO adult male mice and WT littermates (aged 10–12 weeks) left hemisphere striatum was extracted as described above. Complementary libraries were prepared using total RNA NEB Ultra II Directional polyA mRNA library prep kit, and sequenced on the NovaSeq S4 flowcell, with paired-end reads 2 × 150 bp, at 50 million read depth.

Total bulk RNA-seq DEG analysis

Sequence read quality was assessed using FastQC (v.0.11.5). Adapter trimming and removal of lower-quality ends was performed using Trim Galore (v.0.5.0). The quality of trimmed reads was reassessed using FastQC, then screened for presence of rRNA and mtRNA sequences using FastQ-Screen⁷⁷ (v.0.10.0). The RseQC package⁷⁸ (v.2.6.2) was used to assess read distribution, positional read duplication, gene body coverage and junction saturation and to confirm strandedness of alignments. STAR aligner⁷⁹ (v.2.6.0c) was used to align trimmed reads to GRCm39 genome using modified Gendode M29 with custom annotation for the Gm15155 gene. The custom annotation contained the longest versions of all exons from Gencode transcripts 201 and 202 and RefSeq transcripts XR_004940450.1, XR_387096.4 and XR_878287.1. The alignments were processed to extract raw read counts for genes using htseq-count⁸⁰ (v.0.6.1p2). Two-condition differential expression was performed using the DESeq2 package⁸¹ (v.1.26.0) using R v.3.6.1 (R Core Team, 2019). Litter and identified surrogate variables using the sva R package⁸² (v.3.34.0) were used as covariates in the differential expression analysis.

DESeq2 uses the median of ratios method for normalization, and shrinkage estimation for dispersions and fold changes. DESeq2 fits negative binomial generalized linear models for each gene and uses the Wald test for significance testing. Count outlier genes are automatically detected using Cook’s distance and removed from the analysis. DESeq2 also automatically removes genes of which the mean of normalized counts is below a threshold determined by an optimization procedure. Removing genes with low counts improves the detection power by making multiple testing adjustment of P values less severe. The adjusted P value or false-discovery rate is calculated using the Benjamini–Hochberg procedure.

snRNA-seq

Nucleus extraction

Mouse striatal samples from both hemispheres were dissected at P70, cut into smaller pieces before being flash-frozen and stored at −80 °C. Nucleus extraction was performed according to the Nuclei Isolation from Complex Tissues for Single Cell Multiome ATAC + Gene Expression Sequencing (10x Genomics, Demonstrated Protocol, CG000375, Rev B), with minor modifications in homogenization conditions (see below) to ensure that intact nuclei were obtained. All buffers were supplemented with ribonuclease inhibitors (Sigma-Aldrich), and all samples, reagents and steps were kept and performed on ice. One sample was processed at a time.

Striatal samples (~30 mg) were homogenized 25 times in 300 µl of NP40 lysis buffer (10 mM Tris-HCl, 10 mM NaCl, 3 mM MgCl₂, 0.1% NP40, 1 mM DTT, pH 7.4) using a pellet pestle. After homogenization, the samples were diluted in the same volume of NP40 lysis buffer, incubated on ice for 15 min before being passed through a 70 µm cell strainer. After washing, the cell pellet was resuspended with 1 ml PBS containing 1% BSA, filtered through a FACS tube using a 35 µm cell strainer (Falcon, 352235), and 7-AAD was then added at a final concentration of 1%.

Cell sorting was performed by the SickKids-UHN Flow Cytometry Facility. 7AAD-positive nuclei were sorted into a tube containing 1 ml 5% BSA using the Sony MA900 BRYV cell sorter⁸³. The sorted nuclei were pelleted and permeabilized before being resuspended in diluted nucleus buffer (10x Genomics). The nucleus concentration was determined using a Countess II Automated Cell Counter. Samples containing around 10,000 nuclei were immediately processed at TCAG, using the Chromium Next GEM Single-Cell Multiome ATAC-seq + Gene Expression protocol (10x Genomics, CG000338, RevC) according to the manufacturer’s instructions with the Next GEM 3′ polyA library kit, and were sequenced on the NovaSeq 6000 system, with 150 bp paired-end reads, read depth of 50 million for RNA-seq and assay for transposase-accessible chromatin using sequencing (ATAC–seq). Initial quality control was performed with Cell Ranger (10x Genomics). We used DoubletFinder to remove heterotypic nuclei and SoupX to filter out ambient RNA, SC Transform to normalize features, and weighted nearest neighbour and Seurat (v.4.0) for integrative cluster analysis. Cluster annotation was done using DropViz annotations and partially supervised with published major markers for cell types.

Single-nucleus data processing

We used Seurat (v.4.3.0)⁸⁴ and Signac (v.1.8.0)⁸⁵ to process the sequenced single-nucleus multi-omic library. We reidentified peaks with MACS2 (ref. ⁸⁶) and reconstructed the ATAC–seq matrix to prevent distinct peaks from being merged by Cell Ranger⁸⁵ (10x Genomics). Moreover, peaks that aligned to non-standard chromosomes or overlapped known blacklist regions in mm39 (ref. ⁸⁷) were removed. Then, a multi-omic data object was created with matched expression (RNA-seq) and chromatin accessibility (ATAC–seq) measurements. We calculated quality-control metrics for each nucleus, including RNA read counts, mitochondrial reads percentage, ATAC read counts, transcription start site (TSS) signal and nucleosome (NS) signal. We applied threshold-based quality control to remove low-quality nuclei with a mitochondrial read percentage < 5%, NS signal < 2, TSS signal > 2 and library-dependent values for RNA and ATAC read counts. Nuclei that did not pass either RNA-based or ATAC-based metric thresholds were removed during quality control.

Data normalization

RNA assays were normalized following the Seurat workflow. We applied variance stabilization transformation with SCTransform (v.0.3.5)⁸⁸ to normalize the gene expression level across nuclei. We then applied the principal component analysis to derive a lower-dimensional representation of the normalized data. For the ATAC assay, we applied latent semantic indexing (LSI) normalization by running term-frequency inverse-document-frequency transformation followed by singular value decomposition. The first LSI component has been shown to be highly correlated with sequencing depth and was therefore excluded from the analysis⁸⁵. As a result, the top 50 principal components and the 2nd to 40th LSI components were used for downstream processing.

Clustering and annotation

After normalization, we constructed a weighted nearest neighbour graph using FindMultiModalNeighbors in Seurat with the top 50 principal components and the 2nd to 40th LSI components. Next, we clustered the data with FindClusters in Seurat using the smart local moving algorithm (algorithm = 3)⁸⁹ and resolution = 0.6. On the basis of the initial cluster assignment, we ran DoubletFinder to simulate artificial doublets and estimate the probability of a nucleus being a doublet⁹⁰. This helped to reduce the number of potential doublets in the library. We ran the uniform manifold approximation and projection (UMAP)⁹¹ to find a 2D visualization of data and further cleaned the library by removing nuclei that did not cluster well with others, on the basis of the UMAP visualization.

We identified cell types using DEGs in each cluster with FindAllMarkers(assay = “SCT”, only.pos = T) for each library. We annotated the cell type with marker genes derived from DropViz⁹² and Allen Institute mouse brain atlases⁹³. We identified 10 cell types under 5 categories: MSNs: D1-, D2- and eccentric-MSNs; glia: astrocytes, microglia, oligodendrocytes and the oligodendrocyte precurser cells polydendrocytes; glutamatergic neurons, INs and adult neurogenesis-derived nuclei. We focused on MSNs, glia and striatal interneurons in our downstream analysis.

snRNA-seq DEG analysis

After sorting for high-quality nuclei and ascribing cell type annotations, we next combined unnormalized assays with the function Merge in Seurat to construct a new genotype comparison assay for differential gene analysis. We created a KO-1 assay (WT-1 versus KO-1); a KO-2 assay (WT-2 versus KO-2); and a full assay with all four libraries (KO versus WT) to compare DEGs. Normalized nuclei counts: the KO-1 assay contained 9,892 nuclei (7,495 WT-1 and 2,397 KO-1 nuclei); the KO-2 assay contained 3,929 nuclei (2,073 WT-2 and 1,856 KO-2 nuclei); the KO assay contained 13,821 nuclei (9,568 WT and 4,253 KO nuclei). To further validate the uniformity of cell type annotation, we used ClusterMap⁹⁴ to match the DEGs of each cell type and created circle plot visualizations to assess the cell type identity matching across genotype libraries.

DEG analyses were conducted using a randomized DEG approach to balance the nucleus counts (n values) between WT-1 and KO-1 or WT-2 and KO-2 groups for each DEG assay using the lowest nucleus n value from the compared genotypes.

We assessed for DEGs as follows: we assume without loss of generality that the two populations, p₁ and p₂, that we will be testing have n₁ and n₂ nuclei where $n_1\ge n_2$, and q = ⌈n₁/n₂⌉. First, we randomly subsample p₁ without replacement such that the subsampled population $\widetilde{p_1}$ has $|\widetilde{p_1}|=n_2$ nuclei. Then, we run conventional DEG analysis with FindMarkers(assay = “SCT”) in Seurat to identify genes that are differentially expressed between $\widetilde{p_1}$ and $p_2$. Next, we repeat the above two steps for q − 1 times (q repetitions in total) and tracked the DEGs in each repetition. In probability, all nuclei will be sampled and tested over repetitions. Finally, we summarize DEGs identified in each repetition by keeping genes of which the frequency is higher than t = 0.8 over repetitions and rejecting the remaining. Statistical testing was performed using the Wilcoxon rank-sum test and the P value was adjusted post hoc using Benjamini–Hochberg FDR correction. The summarized statistics (P, FDR-adjusted P, log₂[FC]) are reported as the mean of repetitions and s.d. Cell type n values for each assay and sampling repetitions are reported in Supplementary Table 6.

The combined WT versus KO assay contains all four libraries, and we adjusted the sampling strategy accordingly by balancing the nucleus count at the individual-library level. Specifically, we subsampled the three larger populations so that their nucleus counts matched the counts of the smallest population. When calculating the times of repetition q, we took the ceiling of the ratio between the maximum and the minimum nucleus counts across all four libraries, which guaranteed that all nuclei will be tested in probability. This adjustment also equalized the nucleus counts when testing between WT and KO libraries. We then ran the randomized DEG approach as above.

DEG correlation analysis

All DEGs from Ptchd1-as-KO total RNA-seq and pseudo-bulk snRNA-seq analysis were compared to find DEGs in common. Common DEGs were considered to be significant at P < 0.05; the DEGs were then assessed for the correlation degree of significance using Pearson correlation coefficient, and significance levels were determined using a one-sample t-test. Statistical calculations were performed, and data were visualized using ChatGPT 4.0 (Data analyst) assistance.

Gene set enrichment and pathway analysis

Pre-ranked gene-set enrichment analysis^95,96 (GSEA) was done using results from the differential gene expression analyses (total RNA-seq and snRNA-seq as inputs, with the ranking score for each gene equal to the product of −log₁₀[P] multiplied by the sign of the log₂[FC], and a subset of the gene-set collection available from the GSEA website (https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp, downloaded in December 2023), which included C2:CP, C3:MIRs and C5:GO. We used 2,000 permutations and excluded gene sets with fewer than 2 or more than 500 annotated genes; all other parameters for GSEA were left as default.

GSEA results for total RNA-seq were imported into the same Cytoscape session⁹⁷ for visualization using the Enrichment map plug-in⁹⁸, using a FDR threshold of 0.01 and a similarity coefficient threshold of 0.375; each node section was mapped to the product of the FDR and sign of NESs from the GSEA results for each genotype. The autoannotate plugin was then used to cluster similar gene sets using the community cluster algorithm (GLay); the resulting clusters were then manually annotated to highlight the most relevant biological terms in each. The manually annotated clusters from the cytoscape session were used to tag the GSEA results from each comparison.

For GSEA comparison across cell types, pathways were retained if they reached significance (FDR < 0.05) in at least one cell type, and the top 150 pathways were selected on the basis of the composite ranking metric FDR × |NES|. NES values from these 150 pathways were organized into a cell-type × pathway matrix, grouped by manually curated pathway themes and plotted as a heat map. Pathways were ordered within each theme by mean absolute NES across cell types.

Proteomics and western blot analysis

Sample preparation

Tissue samples from adult (aged 8–12 weeks) mouse dorsal striatum and frontal cortex were collected from the same animals, as described above in electrophysiological experiments for whole-cell recordings, and dissected for each region before storage at −80 °C for protein expression studies.

Before western blotting and proteomics analysis, the samples were defrosted and homogenized in RIPA buffer with a protease/phosphatase inhibitor cocktail (Cell Signalling, 5872), centrifuged (20 min at 25,000g) and supernatant protein concentration was determined using the Pierce BCA protein assay (Thermo Fisher Scientific, 23225).

Proteomics analysis using TMT-MS

Adult mouse dorsal striatum and frontal cortex samples were processed by the Network Biology Collaborative Centre (NBCC) at the Lunenfeld-Tanenbaum Research Institute. Samples of 100 µg protein in RIPA lysis buffer (20-188, MilliporeSigma) were resuspended in 5% SDS and 50 mM triethylammonium bicarbonate. Then, 25 µg of protein was reduced (20 mM DTT) and alkylated (40 mM iodoacetamide in dark). The samples were processed using the standard S-trap micro spin column (C02-micro, Protifi) digestion protocol. Next, 10 µg per sample was resuspended in 10 µl of 100 mM HEPES pH 8 and labelled with 80 µg of a unique TMT 16-plex channel (A44520, Thermo Fisher Scientific) in 4 µl of acetonitrile for 1 h at room temperature. The reaction was quenched with 1% hydroxylamine. One-twentieth of each sample was pooled and acquired with data-dependent acquisition using nano high-performance liquid-chromatography tandem MS. A Vanquish Neo UHPLC system was coupled to an Orbitrap Fusion Lumos Tribrid mass spectrometer (FETD2-10002, Thermo Fisher Scientific) and peptides were eluted off a nano-spray emitter generated from fused silica capillary tubing (75 µm inner diameter, 365 µm outer diameter and 5–8 µm tip opening, and packed to 15 cm with C18 reversed-phase material (Reprosil-Pur 120 C18-AQ, 3 µm)) with a 120 min gradient. The flow rate was maintained at 400 nl min⁻¹ with a consistent 0.1% formic acid background. There were three acetonitrile ramps: (1) 3.2% to 16.8% over 72 min; (2) 16.8% to 24.8% over 28 min; and (3) 24.8% to 35.2% over 20 min. The first MS scan ran with an accumulated time of 50 ms, a mass range of 400–16,000 m/z, orbitrap resolution of 120,000, 30% radio frequency lens, 200% automatic gain control and 1,800 V. The subsequent tandem MS scan had cycle times of 2 s, 35 ms accumulation time, 33% higher-energy collisional dissociation collision energy and a first mass to charge ratio of 120–1,800 m/z. All candidate ions had a charge state from 2 to 7, and automatic gain control target of 400% isolated using an orbitrap resolution of 50,000.

Differential protein expression analysis

The resulting raw files underwent the proteomics analysis workflow (PAW) pipeline^99,100 (https://github.com/pwilmart/PAW_pipeline). The Bioconductor package edgeR¹⁰¹ was used for TMT16 data normalization using the exact test for negative binomial data as for statistical differences in expressed proteins, and with Benjamini–Hochberg P-value correction to control the FDR. The TMT16 workflow with the mouse UniProt ID UP000000589 database was used with decoys and contaminants appended.

Western blot

Samples (20 µg total, as 1 µg µl⁻¹ protein in 1× Laemmli sample buffer (Bio-Rad, 1610747) and 2.7% β-mercaptoethanol (Sigma-Aldrich, M3148)) were loaded onto 7.5% TGX Stain-Free FastCast gels (Bio-Rad, 1610181) with an internal batch control lane across gels consisting of a lysate combined from all samples. Gels underwent electrophoresis at 200 V and were light activated for total protein signal on the ChemiDoc MP imaging system (Bio-Rad). Proteins were transferred onto low-fluorescence PVDF membranes (Bio-Rad, 1620264) using the Trans-Blot Turbo system (25 V, 2.5 A, 3 min; Bio-Rad). The membranes were briefly washed in Tris-buffered saline and Tween-20 (TBST) and total protein images were captured. Membranes were blocked for 30 min (Bio-Rad, 120-100-20) and incubated in primary antibodies (Supplementary Table 20) overnight rocking at 4 °C. The membranes were washed with TBST (6 × 5 min) and incubated with fluorescent secondary antibodies for 1 h.

Target protein band signals were normalized to full lane total protein staining (band volume) (Bio-Rad Image Lab software 6.0) and further adjusted according to the internal control lane of their respective membranes. Normalized values relative to the WT were calculated as a ratio of the sample/group WT mean average.

Sample usage across experiments

ddPCR expression knockdown analysis was performed using the same samples when probing for each isoform of Ptchd1-as. Total RNA-seq contained a few samples that were also used in ddPCR and all KO and WT littermates were used in plots of normalized exon counts from total RNA-seq experiments. Brain and cortex sashimi plot samples were obtained from C57BL/6J mice used in spatiotemporal analysis of native Ptchd1-as expression.

Spatiotemporal analysis across brain regions was performed using samples from the same mouse: brain (one hemisphere was used whole) and the other hemisphere was subsectioned for the hippocampus, cortex and cerebellum at P7, P35 and P70. Embryonic day 18 samples were processed from both hemispheres and tissues were collected from different animals. Behaviour experiments used the same mice across 2–3 experiments. All samples for striatal and cortical proteomics were also used for validation in western blotting experiments. Electrophysiology studies were performed using independent samples.

Statistical analysis

Unless otherwise stated, statistical analyses were carried out in R (v.3.6.1) or using statistical packages Prism built-in analysis (v.10 or earlier; GraphPad). The number (n) of biologically independent samples is described in the figure legends and Methods above, and the individual datapoints are shown in the bar plots. Tests used to assess statistical significance between genotypes are described in the respective figure legends and additional behavioural statistics are provided in Supplementary Table 4.

Reporting summary

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

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10515-6.

Author contributions

S.W.S. and C.A.B. conceived and designed the genetics, genotype–phenotype and transcriptomic experiments. Experiments for behaviour were conceived and designed by P.W.F., S.Y.K., with contributions from C.A.B., G.L.C. and S.W.S. Proteomics and western blotting were conceived by L.T.R., C.A.B., J.G. and G.L.C. Experiments for electrophysiology were conceived by G.L.C., M.T., L.T.R., J.L., P.T. and J.G. with contributions from C.A.B. and M.W.S.; C.A.B., S.Y.K., M.T., L.T.R., L.D., J.L., T.L., J.W., P.T., M.M., X.F., J.L.H., G.C., T.P., N.B.S., O.R., A.E.A., J.d.R., A.K., F.J., J.R.M., P.J.R., N.H., E.S., T.S., A.R., M.S., C.K.R., J.C., D.I.F., B.H.Y.C., J.P., A.I., K.M.W., C.W.N., M.L.H. and A.G. collected the data. C.A.B., S.Y.K., M.T., L.T.R., J.L., T.L., J.W., P.T., M.M., J.L.H., G.P., R.A., G.C., T.P., L.E.W.-G., W.E., B. Thiruvahindrapuram, B. Trost, M.S.R., J.J.D.-M., E.D., P.J.R., J. Ellis, C.S., C.R.M., M.W.S., E.A., P.W.F., G.L.C. and S.W.S. provided methodology, formal analysis or interpretation of the data. C.A.B., S.Y.K., M.T., L.T.R, L.D., J.L., T.L., J.W., P.T., X.F., J.d.R., A.K., F.J., E.D., P.J.R., C.S., A.E.A., A.R., N.R.-A. and J. Ellegood performed specific experimental investigations. C.A.B., J.L.H., J.G., P.W.F., G.L.C. and S.W.S. provided support for project administration. C.A.B., M.S., H.M.K., J.C., D.I.F., B.H.Y.C., J.P., D.G.A., C.P.S., M.E., R. Landa, S.M., R. Lesurf., A.J., E.A., G.L.C. and S.W.S. provided resources for the study. C.A.B., M.D.W., J.P.L., L.J.L., B.J.F., M.W.S., J.A.S.V., P.W.F., P.T., J.G., G.L.C. and S.W.S. supervised the project. C.A.B., S.Y.K., M.M., T.L., L.T.R., P.T., J.d.R., G.P., J.L.H., C.S., E.A., G.L.C. and S.W.S visualized the data. C.A.B., S.Y.K., G.L.C. and S.W.S. wrote the paper with input from C.S. and J.G.; C.A.B., J. Ellis, P.W.F., M.W.S., G.L.C. and S.W.S. acquired the funding.

Peer review

Peer review information

Nature thanks John Mattick, Zilong Qiu, Christian Schaaf and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

WGS data from MSSNG and SFARI datasets can be obtained at https://research.mss.ng and https://www.sfari.org/resource/sfari-base/, as was done for this study. Public databases used in the study can be accessed through the following sites: 1000 Genomes Project (https://www.internationalgenome.org/); and Medical Genome Reference Bank (MGRB; 10.1038/s41467-019-14079-0). Additional control datasets can be requested at CHILD (https://ega-archive.org/dacs/EGAC00001002953); HostSeq (https://www.cgen.ca/hostseq-databank-access-request) and Inova (www.inova.org). Data from the cardiology datasets are available online (https://ega-archive.org/studies/EGAS50000000586 and https://ega-archive.org/studies/EGAS00001004929). RNA-seq data can be found at NCBI BioProject (RNA accession: PRJNA1195946) and the proteomic data are available at MassIVE under proteomics accession number MSV000096677. Source data supporting the findings of this study are provided with this Article and can be requested from the corresponding authors.

Competing interests

At the time of this study and its publication, S.W.S. served on the scientific advisory committee of Population Bio, Deep Genomics and Diploid Genomics. Intellectual property from aspects of his research held at The Hospital for Sick Children are licensed to Athena Diagnostics and Population Bio. J.J.D.-M. is an employee and equity holder with Phenomic AI, a biotech firm developing therapeutics targeting cancer. J.A.S.V. served as a consultant for NoBias Therapeutics and has received speaker fees for Henry Stewart Talks. S.M. is on the advisory board of Bristol Myers Squibb, Rocket Pharmaceuticals, Tenaya Therapeutics and Edgewise Therapeutics. E.A. has received consultation fees from Acadia, Roche, Impel, Cell EI and Ono, research funding from Anavex, Roche and Maplight and has patents (14/755/084, USA; 2,895,954, Canada). These relationships did not influence the data interpretation or presentation during this study but are disclosed for potential future considerations. The other authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41586-026-10515-6.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-026-10515-6.