De Novo Mutation in Genes Regulating Neural Stem Cell Fate in Human Congenital Hydrocephalus

SUMMARY

Congenital hydrocephalus (CH), featuring markedly enlarged brain ventricles, is thought to arise from failed cerebrospinal fluid (CSF) homeostasis and is treated with lifelong surgical CSF shunting with substantial morbidity. CH pathogenesis is poorly understood. Exome sequencing of 125 CH trios and 52 additional probands identified three genes with significant burden of rare damaging de novo or transmitted mutations: TRIM71 (p = 2.15 × 10⁻⁷), SMARCC1 (p = 8.15 × 10⁻¹⁰), and PTCH1 (p = 1.06 × 10⁻⁶). Additionally, two de novo duplications were identified at the SHH locus, encoding the PTCH1 ligand (p = 1.2 × 10⁻⁴). Together, these probands account for ~10% of studied cases. Strikingly, all four genes are required for neural tube development and regulate ventricular zone neural stem cell fate. These results implicate impaired neurogenesis (rather than active CSF accumulation) in the pathogenesis of a subset of CH patients, with potential diagnostic, prognostic, and therapeutic ramifications.

In Brief

Congenital hydrocephalus (CH) is a major cause of childhood morbidity and mortality, affecting 1 in 1,000 live births and representing up to 3% of all pediatric hospital charges. Using data from the largest CH exome sequencing study to date, Furey et al. identify four genes (TRIM71, SMARCC1, PTCH1, and SHH) not previously implicated in CH. Remarkably, all four genes regulate ventricular zone neural stem cell fate and, together, explain ~10% of CH cases. These findings implicate impaired neurogenesis in pathogenesis of a significant number of CH patients, with potential diagnostic, prognostic, and therapeutic ramifications.

INTRODUCTION

Hydrocephalus has been defined as the active, progressive distension of the cerebral ventricular system resulting from inadequate passage of cerebrospinal fluid (CSF) from its point of production to its point of absorption (Rekate, 2008). This mechanism is evident in secondary hydrocephalus associated with brain tumors, infection, or hemorrhage, in which intracranial pressure is often elevated (Kahle et al., 2016). However, when infantile hydrocephalus occurs without a known antecedent, it is classified as primary or congenital hydrocephalus (CH) (Tully and Dobyns, 2014). CH can occur in the absence of obstruction to CSF flow (communicating hydrocephalus) or with complete/partial intraventricular obstruction (non-communicating hydrocephalus), most often due to aqueductal stenosis. Cases of communicating CH with normal or low intracranial pressure (Bret and Chazal, 1995; Tully and Dobyns, 2014) raise the question of whether the development of ventriculomegaly is a primary active process or a secondary passive process.

CH is a major cause of childhood morbidity and mortality, affecting 1 in 1,000 live births (Munch et al., 2012; Simon et al., 2008) and costing the U.S. health care system alone $2 billion annually (Shannon et al., 2011; Simon et al., 2008). Current therapy consists of catheter-based CSF shunting and endoscopic approaches, surgeries with high rates of failure and morbidity (Kahle et al., 2016). Significant gaps in our understanding of the molecular pathophysiology of CH impede the development of preventive, diagnostic, and therapeutic measures (McAllister et al., 2015).

Few bona fide human CH-causing genes have been identified (Kousi and Katsanis, 2016). These include L1CAM mutation in X-linked hydrocephalus and aqueductal stenosis (OMIM# 307000; Rosenthal et al., 1992), which constitutes up to 3% of CH, and rare recessive mutations in MPDZ (OMIM# 603785; Al-Dosari et al., 2013), CCDC88C (OMIM# 236600; Ekici et al., 2010), EML1, and WDR81 (Shaheen et al., 2017). These genes affect multiple cellular processes, confounding efforts to formulate a uniform paradigm of CH pathophysiology (Kousi and Katsanis, 2016). While ~40% of all CH cases are predicted to have a genetic etiology (Zhang et al., 2006), mutations in currently identified genes account for less than 5% of primary CH cases (Adle-Biassette et al., 2013; Haverkamp et al., 1999).

The sporadic nature of most CH cases limits the utility of traditional genetic approaches. This has motivated whole-exome sequencing (WES) of large patient cohorts, searching for genes mutated in affected subjects more often than expected by chance. This approach has proven powerful in other neurodevelopmental disorders (Deciphering Developmental Disorders Study, 2015, 2017), including brain malformations (Barak et al., 2011; Bilgüvar et al., 2010), epilepsy (Allen et al., 2013), craniosynostosis (Timberlake et al., 2016, 2017), autism (Awadalla et al., 2010; O’Roak et al., 2011; Neale et al., 2012; Sanders et al., 2012; Krumm et al., 2015), and others. We hypothesized that the apparent sporadic occurrence of CH may reflect, in some cases, damaging de novo mutations and/or transmitted mutations with incomplete penetrance.

RESULTS

Cohort Characteristics and Whole-Exome Sequencing

We recruited 177 probands with non-L1CAM-mutated CH in which hydrocephalus was the predominant phenotypic feature (see STAR Methods). All probands had undergone surgery for CSF diversion. The cohort included 125 parent-offspring sporadic CH trios, 47 singleton cases, and 5 multiplex kindreds (see STAR Methods; Table S1). 88 probands had communicating hydrocephalus and 89 had aqueductal stenosis. All study protocols were approved by the Yale Human Research Protection Program (HRPP) Institutional Review Board.

DNA was isolated, and WES was performed (Timberlake et al., 2016). 95.9% of targeted bases had 8 or more independent reads, and 92.7% had 15 or more (Table S2; Figure S1). 1,789 control trios that comprised unaffected siblings from the Simons simplex autism cohort were sequenced on the same platform and analyzed in parallel (Krumm et al., 2015). Variants were called using the Genome Analysis Toolkit (GATK) Haplotype Caller (McKenna et al., 2010; Van der Auwera et al., 2013), and allele frequencies were annotated in the Exome Aggregation Consortium (ExAC) and gnomAD databases (Lek et al., 2016; see STAR Methods). TrioDeNovo was used to identify de novo mutations, and MetaSVM was used to infer the impact of missense mutations (Dong et al., 2015). Direct Sanger sequencing of PCR amplicons containing the mutation verified variants in genes of interest.

We identified mutations in known CH genes in four probands, including one proband with compound heterozygous mutations in MPDZ, the cause of autosomal recessive (AR) non-syndromic hydrocephalus type 2 (OMIM# 615219; Al-Dosari et al., 2013); two male probands with transmitted D-mis mutations in FLNA, the cause of X-linked dominant (XLD) periventricular heterotopia (OMIM# 300049; Kamuro and Tenokuchi, 1993; Sheen et al., 2004); and one proband with compound heterozygous mutation in CRB2 (Hydro122), the cause of a AR ventriculomegaly with cystic kidney disease (OMIM# 219730; Slavotinek et al., 2015).

Global Analysis of Burden of De Novo Mutation

We determined a de novo mutation rate of 1.4 × 10⁻⁸ per base pair, with 1.09 de novo coding region mutations per proband (Table 1), consistent with expectation and previous work (Homsy et al., 2015; Timberlake et al., 2017; Ware et al., 2015). The burden of de novo mutations in the control cohort was similar (Table S3; Figure S2).

Table 1.

Observed and Expected De Novo Mutation Rates in 125 Probands with Congenital Hydrocephalus and CH Cases with Aqueductal Stenosis for Variant Classes by Phenotype and High Brain Expression

	Observed		Expected
	n	Rate	n	Rate	Enrichment	P
Hydro (125)
All	137	1.10	139.7	1.12	0.98	0.60
Synonymous	32	0.26	39.7	0.32	0.81	0.91
Missense	89	0.71	87.8	0.70	1.01	0.46
D-Mis	16	0.13	16.5	0.13	0.97	0.58
LOF	16	0.13	12.3	0.10	1.30	0.18
Protein-Altering	105	0.84	100.0	0.80	1.05	0.32
Damaging	32	0.26	28.7	0.23	1.11	0.30
Hydro (125) - High Brain Expressed
All	48	0.38	36.9	0.30	1.30	0.04a
Synonymous	12	0.10	10.3	0.08	1.16	0.34
Missense	30	0.24	23.2	0.19	1.29	0.10
D-Mis	5	0.04	4.4	0.04	1.14	0.45
LOF	6	0.05	3.4	0.03	1.78	0.13
Protein-Altering	36	0.29	26.5	0.21	1.36	0.04a
Damaging	11	0.09	7.8	0.06	1.42	0.16
Hydro + AS (66)
All	78	1.18	73.8	1.12	1.06	0.33
Synonymous	22	0.33	20.9	0.32	1.05	0.44
Missense	47	0.71	46.3	0.70	1.01	0.48
D-Mis	12	0.18	8.7	0.13	1.38	0.17
LOF	9	0.14	6.5	0.10	1.39	0.21
Protein-Altering	56	0.85	52.8	0.80	1.06	0.35
Damaging	21	0.32	15.2	0.23	1.38	0.09
Hydro + AS (66) - High Brain Expressed
All	30	0.45	19.5	0.30	1.54	0.01a
Synonymous	9	0.14	5.5	0.08	1.65	0.10
Missense	17	0.26	12.2	0.19	1.39	0.11
D-Mis	4	0.06	2.3	0.04	1.72	0.21
LOF	4	0.06	1.8	0.03	2.25	0.11
Protein-Altering	21	0.32	14.0	0.21	1.50	0.04a
Damaging	8	0.12	4.1	0.06	1.95	0.05a

n, the number of de novo mutations; Rate, the number of de novo mutations per individual; Enrichment, ratio of observed to expected number of mutations; Hydro, congenital hydrocephalus; High Brain Expression, top quartile of expression; AS, aqueductal stenosis; Missense, tolerated missense mutations as predicted by MetaSVM; D-Mis, damaging missense mutations as predicted by MetaSVM; LOF, loss of function denotes premature termination, frameshift, or splice site mutations; Damaging, D-miss and LOF mutations.

^ap value ≤ 0.05

We compared the observed and expected number of de novo mutations in all genes and among genes in the top quartile of brain expression at embryonic day 9.5 (E9.5; see STAR Methods). Protein-altering de novo mutations were significantly enriched over expectation in the latter group, contributing to an estimated 8% of cases (Table 1). Controls showed no enrichment of de novo mutations in any gene class (Table S3).

Five genes contained two or more protein-altering de novo mutations: TRIM71, SMARCC1, PTCH1, PLOD2, and SGSM3 (Table S4). The first three genes have a probability of loss-of-function (LOF) intolerance (pLI) score ≥ 0.99; the latter two genes have pLIs of zero (Lek et al., 2016). The probability of finding five genes with two or more protein-altering de novo mutations in a cohort this size by chance is very low (p = 1.1 × 10⁻⁴). Four of these genes were in the upper quartile of expression in the developing brain. The probability of finding two or more protein-altering de novo mutations in four such genes in our cohort is very low (p = 1.1 × 10⁻⁴; Table S5).

Recurrent De Novo and Transmitted Mutations in TRIM71

LIN41/TRIM71, encoding lineage variant 41 (LIN41)/tripartite motif 71 (“TRIM71”) (Reinhart et al., 2000; Slack et al., 2000), harbored three novel de novo mutations, including the identical heterozygous de novo p.Arg608His mutation in two unrelated probands with prenatally diagnosed communicating hydrocephalus (Hydro101–1 and Hydro102–1) (Figure 1; Table S6). During the course of this study, Hydro102–1 became pregnant and transmitted the p.Arg608His mutation to her son, who was prenatally diagnosed with communicating hydrocephalus (Figure 1; Figure S5). A third de novo TRIM71 mutation (p.Arg796His) was identified in Hydro100–1, who had prenatally diagnosed communicating CH (Hydro100–1). All CH patients harboring TRIM71 mutations underwent surgical CSF shunting at birth because of extreme ventriculomegaly. The probability of finding three or more protein-altering de novo mutations in TRIM71 by chance in a cohort of this size is 2.15 × 10⁻⁷, surpassing genome-wide significance (Table 2) (Ware et al., 2015). TRIM71 is highly intolerant to LOF mutation (pLI = 0.99) and missense variation (Z score = 5.69) (Lek et al., 2016). With 105 protein-altering de novo missense mutations in our cohort, the probability of seeing any instances of the identical mutation at any position in the coding region by chance is very remote (p = 5.24 × 10⁻⁴).

Figure 1. Recurrent, Identical De Novo Mutations in LIN-41/TRIM71 Encoding the let-7 miRNA Target TRIM71.

(A) Representative sagittal (left) and axial (right) brain magnetic resonance images of CH probands Hydro100–1 and Hydro102–5 show communicating hydrocephalus.

(B) Pedigrees with Sanger-verified mutated bases (red) and the corresponding wild-type bases marked on the chromatograms.

(C) Structural modeling of TRIM71 mutation impact. The positively charged guanidinium side chain of p.Arg796 interacts with the negatively charged sugar-phosphate backbone of target RNAs to aid in maintaining the spatial position of the nucleic acid. Mutation of this p.Arg796 residue to the imidazole ring of histidine (ΔΔG = 2.0 kcal/mol) is predicted to disrupt this interaction (right). The side chain of p.Arg608 makes hydrogen bonds with uracil in target RNAs. The p.Arg608His mutation (ΔΔG = 1.6 kcal/mol) is predicted to disrupt these hydrogen bonds (left).

(D) Mapping of TRIM71 mutations. p.Arg608His and p.Arg796His affect conserved residues in the 16^th position of the respective 1^st and 5^th blades of TRIM71’s NHL domain, which mediates the binding of target RNA.

(E) In situ hybridization of wild-type E12.5 and adult mouse brains for Trim71 showing signals in the ciliated neuroepithelium and ventricular zone (V, ventricle; arrow, neuroepithelium at E12.5 and ependymal layer in adulthood; 2.5× and 10× magnification).

Table 2.

Meta-analysis of Protein-Altering De Novo Mutations and Loss-of-Function Heterozygous Mutations in Probands for Genes with Multiple De Novo Mutations

Gene	De Novo Mutations				Transmitted Loss-of-Function (LOF)				Meta p value	pLI	High Brain Expression Rank
	# observed	# expected	enrichment	p value	# observed	# expected	enrichment	Binomial p value
SMARCC1	2	2.3 × 10−3	861.7	2.7 × 10−6a	3	4.3 × 10−2	70.5	1.2 × 10−5	8.2 × 10−10a	1.0	97.4
PTCH1	2	1.5 × 10−3	1374.3	1.1 × 10−6a	1	3.8 × 10−2	26.1	3.8 × 10−2	7.2 × 10−7a	1.0	93.2
TRIM71	3	1.1 × 10−2	274.8	2.2 × 10−7a	0	2.5 × 10−2	0	1	3.5 × 10−6	1.0	74.6
PLOD2	2	3.5 × 10−3	580.2	5.9 × 10−6	0	7.8 × 10−2	0	1	7.7 × 10−5	0	82.8
SGSM3	2	8.6 × 10−3	232.2	3.7 × 10−5	0	6.9 × 10−2	0	1	4.1 × 10−4	0	67.4

Meta-analysis performed by combining the p values from protein-altering de novo mutations and LOF heterozygous mutations using the Fisher’s method with 4 degrees of freedom.

^ap values surpassing the Bonferroni multiple testing correction (2.6 × 10⁻⁶, 0.05/18,989) for p values tabulated by either de novo, heterozygous, or meta-analysis. Of note, the denovolyzeR p value of damaging SMARCC1 mutations was calculated using 126 case-parent trios given the sporadic de novo mutation occurs in the proband’s father (Hydro106–3).

TRIM71 is the homolog of C. elegans Lin-41, a target of the let-7 (lethal 7) microRNA (miRNA) in the heterochronic pathway that regulates stem cell fate (Ecsedi and Grosshans, 2013). TRIM71, like Lin-41, mediates post-transcriptional silencing of mRNAs via direct interactions of its NHL domain and the 5′ or 3′ UTRs of target genes (Aeschimann et al., 2017; Ecsedi and Grosshans, 2013; Slack and Ruvkun, 1998; Vella et al., 2004). TRIM71 also contains ubiquitin ligase activity conferred by its RING domain (Nguyen et al., 2017). The NHL domain in TRIM71 consists of six repeats, each 40–50 residues long, that jointly comprise a barrel-like six-bladed β-propeller (Loedige et al., 2015). The de novo p.Arg608His and p.Arg796His mutations are at homologous positions in different blades of NHL domains. Arginines at these positions are conserved among orthologs from H. sapiens to C. elegans and reside in RPQGV motifs (Figure 1). The NHL domain of TRIM71 is the closest homolog of the NHL domain of D. melanogaster Brat (Loedige et al., 2013, 2014), which has been co-crystallized bound to a target RNA (Loedige et al., 2015). In this structure, the amino groups of side chains of the mutated arginines form hydrogen bonds with either the phosphate backbone (p.Arg796) or a uracil base (p.Arg608) of target RNA (Figure 1; Figure S3). These interactions are predicted to be altered by histidine substitution. While these mutations were predicted as tolerated by MetaSVM, they are predicted as damaging by 43 of 45 other predicting algorithms, including the very conservative MPC-D algorithm (see STAR Methods; Samocha et al., 2017). The recurrence of the identical de novo mutation and the fact that the third de novo mutation occurs at a homologous position in a different NHL domain suggest that TRIM71 mutations may not be simple LOF mutations.

In situ hybridization in E12.5 mouse brain (Figure 1; Figure S4) revealed abundant Trim71 expression in the neuroepithelium and ventricular zone (Cuevas et al., 2015; Maller Schulman et al., 2008). Analogous to its heterochronic expression in C. elegans (Kanamoto et al., 2006; Slack et al., 2000), Trim71 expression significantly decreases during development (Figure 1) (Schulman et al., 2005; Yu et al., 2010). Trim71 is highly expressed in neural progenitor cells (NPCs) of early mouse embryos (until E11.5) but declines secondary to increased expression of let-7 and mir-125 miRNAs as neural differentiation proceeds (Chen et al., 2012; Schulman et al., 2005). Trim71 deletion in mice results in exencephaly and embryonic lethality (failure of closure of the cephalic end of the neural tube) by E10 (Maller Schulman et al., 2008). Trim71 maintains NPC pluripotency NPCs by regulating the balance between self-renewal and differentiation via the post-transcriptional silencing of its target mRNAs (Chang et al., 2012; Ecsedi and Grosshans, 2013; Mitschka et al., 2015; Worringer et al., 2014). The neural tube closure defect in Trim71 knockout mice results from decreased proliferation and precocious differentiation of NPCs (Chen et al., 2012).

Multiple De Novo and Transmitted Mutations in SMARCC1

Two novel, damaging de novo mutations were identified in SMARCC1, encoding BAF155, a 155-kD subunit of the BRG1/BRM-associated factor (BAF; S. cerevisiae SWI/SNF) chromatin remodeling complex (Wang et al., 1996). These mutations included a p.Lys891fs*4 mutation (in Hydro106–3) and a predicted damaging missense p.His526Pro mutation (in Hydro105–1) located in a conserved position in the SWIRM domain, which mediates BAF complex subunit interactions (Da et al., 2006) (Figure 2). p.His526Pro occurred in a proband with obstructive CH with aqueductal stenosis and no other affected family members. p.Lys891fs*4 was de novo in an unaffected father (Hydro106–3) of three children with prenatally diagnosed CH with aqueductal stenosis. Two of these offsprings died in utero from obstructive hydrocephalus, while the surviving child inherited p.Lys891fs*4. In this cohort, the probability of seeing two damaging de novo mutations in SMARCC1 was 2.69 × 10⁻⁶ (Table 2).

Figure 2. De Novo and Transmitted Mutations in SMARCC1 Encoding the SWI/SNF Chromatin Modifier BAF155.

(A) Representative sagittal (left) and coronal (right) brain magnetic resonance images of CH probands Hydro106–1 and Hydro108–1 with obstructive hydrocephalus..

(B) Pedigrees with Sanger-verified mutated bases and the corresponding normal alleles marked on the chromatograms..

(C) In situ hybridization of wild-type E12.5 and adult mouse brains for Smarcc1 showing signals in the ciliated neuroepithelium and ventricular zone (V, ventricle; arrow, neuroepithelium at E12.5, 2.5×, and ependymal layer in adulthood; 10× magnification).

Three additional, previously unidentified, transmitted LOF mutations in SMARCC1 were found in three other CH probands with severe obstructive CH with aqueductal stenosis. In each of these kindreds, the proband was the sole affected member with mutation transmission from an unaffected parent (Figure 2; Figure S6; Table S7). The probability of two de novo damaging mutations and three rare transmitted LOFs occurring in SMARCC1 in this cohort was 8.15 × 10⁻¹⁰; this is a conservative estimate since SMARCC1 is highly intolerant to LOF mutation (pLI of 1), with far fewer LOF mutations than expected (3 LOF SMARCC1 mutations among 60,706 individuals in the ExAC database, and 2 among 3,578 autism controls).

In situ hybridization demonstrated that Smarcc1, like Trim71, is highly expressed in the neuroepithelium and ventricular zone but reduced in later brain development (Ho et al., 2009; Tuoc et al., 2013a; Yan et al., 2008) (Figure 2; Figure S4). Smarcc1 knockout causes embryonic lethality in mice (Kim et al., 2001). Roughly 20% of Smarcc1^+/− mice (Kim et al., 2001) and ~80% of mice homozygous for a missense allele (Baf155^msp/msp) (Harmacek et al., 2014) exhibit exencephaly similar to that of Trim71 knockout mice (Harmacek et al., 2014; Narayanan et al., 2015; Nguyen et al., 2016). The exencephaly phenotype has been attributed to inappropriate proliferation and increased apoptosis of NPCs in the neural tube (Harmacek et al., 2014; Kim et al., 2001). Neuronal specific knockout of Smarca4, encoding a Smarcc1 binding partner in the BAF complex, causes hydrocephalus and aqueductal stenosis (Cao and Wu, 2015). NPC-specific BAF complexes regulate the proliferation, differentiation, and survival of mouse NPCs via transcriptional regulation of genes critical for telencephalon development (Narayanan et al., 2015).

De Novo and Transmitted Mutations in PTCH1

Two previously unidentified de novo LOF mutations were identified in PTCH1, encoding the Sonic Hedgehog (SHH) receptor (Eggenschwiler and Anderson, 2007) Patched-1 (Figure 3). Heterozygous LOF PTCH1 mutations have been previously implicated in Gorlin syndrome (OMIM# 109400), featuring basal cell carcinomas and mandibular tumors, along with variable expressivity of other skeletal and non-skeletal phenotypes (Hahn et al., 1996). De novo PTCH1 mutations included a start-loss mutation (p.Met152fs*1) and a splice donor site mutation (c.1503+3T>C) in unrelated CH probands with aqueductal stenosis (Figure 3; Figure S7; Table S8). Both mutations were heterozygous and occurred in probands of uniplex CH kindreds. A transmitted frameshift mutation in PTCH1 (p.Leu664fs*12) was also found in two brothers with CH (Figure 3; Figure S7; Table S8). This mutation was transmitted from a mother who did not have hydrocephalus but had been diagnosed with Gorlin syndrome on the basis of numerous basal cell carcinomas. The probability of seeing at least two damaging de novo mutations by chance in a cohort of this size is 1.06 × 10⁻⁶, surpassing genome-wide significance (Table 2) (Ware et al., 2015).

Figure 3. De Novo and Transmitted Mutations in PTCH1 Encoding the Sonic Hedgehog Receptor Patched-1.

(A) Representative sagittal (left) and axial (right) brain magnetic resonance images of CH probands Hydro103–1 and Hydro104–1 with obstructive hydrocephalus..

(B) Pedigrees with Sanger-verified mutated bases and the corresponding normal alleles marked on the chromatograms..

(C) In situ hybridization of wild-type E12.5 and adult mouse brains for Ptch1 demonstrates expression in the hindbrain neuroepithelium (V, ventricle; arrow, neuroepithelium at E12.5, 2.5×, and ependymal layer in adulthood; 10× magnification).

Consistent with previous results (Eggenschwiler et al., 2001; Goodrich et al., 1996, 1997; Takahashi and Osumi, 2002), in situ hybridization showed Ptch1 expression in hindbrain neuroepithelium (Figure 3; Figure S4). Ptch1 knockout in mice results in exencephaly, neural tube overgrowth, and lethality between E9.0 and E10.5 (Ellis et al., 2003; Goodrich et al., 1997; Milenkovic et al., 1999). A significant fraction of Ptch1^+/− mice develop hydrocephalus in two different genetic backgrounds (Svärd et al., 2009; Wetmore et al., 2000), similar to the variable expressivity seen in one of the multiplex kindreds. Penetrance of hydrocephalus in these mice increases to 100% for the homozygous quaking viable mutation (Ptch1^+/−; qk^v/v) (Gavino and Richard, 2011). Primary cilia in neuroepithelial cells sense gradients of SHH via PTCH1 and transduce these signals to regulate NPC proliferation, differentiation, and fate specification (Palma et al., 2005; Palma and Ruiz i Altaba, 2004).

De Novo Mutations in PLOD2

Two heterozygous de novo D-mis mutations were identified in the highly brain expressed (HBE) gene PLOD2 (Table S9). PLOD2 did not surpass thresholds for genome-wide significance. PLOD2 encodes a lysyl hydroxylase previously implicated in AR Bruck syndrome type 2 (OMIM# 609220), a variant of osteogenesis imperfecta (McPherson and Clemens, 1997). Further work will be required to assess the significance of these mutations in CH.

De Novo Mutations in Neural Tube Closure and Formation

Gene ontology (GO) pathway analysis identified our significant CH gene set to be highly enriched for genes involved in “neural tube closure and formation.” We searched for additional de novo mutations in genes in this GO pathway (Table S10). We found a single de novo missense mutation in CELSR2 (p.Arg2812Trp), encoding cadherin EGF LAG seven-pass G-type receptor 2, and a single de novo missense mutation in LRP6 (p.Val1415Phe), encoding a co-receptor with frizzled proteins in the WNT signaling pathway. Celsr2 and Lrp6 are highly expressed in ciliated neuroepithelia and regulate neurogenesis, and knockout of each gene results in lethal hydrocephalus in mice (Allache et al., 2014; Cuevas et al., 2015; Gavino and Richard, 2011; Harmacek et al., 2014; Maller Schulman et al., 2008; Tissir et al., 2010).

De Novo Duplications at the SHH Locus

Using the XHMM algorithm (Fromer and Purcell, 2001), we identified seven putative de novo copy number variants (CNVs) (see STAR Methods). Six of these, including five duplications and one deletion, were validated by independent tests (distortion of allelic ratios of heterozygous SNPs in implicated intervals or qPCR; Table S11). Two of these were duplications at the SHH locus, encoding SHH, the canonical ligand for PTCH1 that regulates neurogenesis by conferring positional information to ventral NPCs in the neural tube (Jessell and Sanes, 2000; Lupo et al., 2006). The probability of finding two duplications at the same locus among five duplication events with these gene compositions is very low (p = 1.22 × 10⁻⁴) (see STAR Methods).

DISCUSSION

Using exome sequencing, we implicate four new genes in human CH: TRIM71, SMARCC1, PTCH1, and SHH (Figure 4). High enrichment of mutations in these genes suggests their large effects on phenotypic risk. The genes showing heterozygous protein-altering de novo mutations are all highly intolerant to LOF mutation (pLI ≥ 0.99), and murine knockouts of each gene produce neural tube defects, including exencephaly (Feng et al., 2013; Gavino and Richard, 2011; Goodrich et al., 1997; Hahn et al., 1996; Harmacek et al., 2014; Maller Schulman et al., 2008). SHH encodes the PTCH1 ligand. Ptch1-deficient mice develop fatal hydrocephalus (Celen et al., 2017; Gavino and Richard, 2011). All four genes are expressed in the ciliated neuroepithelium and regulate NPC fate. Mutations in these four new genes account for 8.5% of cases; another 2.3% of cases are explained by previously identified genes.

Figure 4. Meta-analysis of Rare De Novo and Transmitted Mutations in Mutation-Intolerant Genes.

Quantile-quantile plot of observed versus expected p values from meta-analysis of protein-altering de novo and LOF transmitted heterozygous variants comparing the burden of rare variants in genes intolerant to LOF mutation (pLI ≥ 0.90, MAF ≤ 2 × 10⁻⁵).

From the inferred contribution of de novo point mutations to 8% of CH probands in our cohort and the number of genes with more than one de novo mutation, we estimate that additional genes contribute to CH by de novo point mutation (maximum likelihood estimate ~8 genes, with wide confidence interval) (Figures S8 and S9). This estimate is substantially smaller than autism and congenital heart disease (De Rubeis et al., 2014; Homsy et al., 2015; Iossifov et al., 2014; Sanders et al., 2012; Zaidi et al., 2013), consistent with the identification of multiple disease-causing genes from this relatively small cohort. Variants in non-coding elements of these genes might add to their contribution to CH. Sequencing of additional trios and isolated probands has high potential to detect additional rare mutations with large effect on disease risk.

All four cases of CH with TRIM71 mutation had communicating hydrocephalus, whereas all five cases with SMARCC1 mutation and all three cases with PTCH1 mutation had aqueductal stenosis. From the prevalence of these two sub-phenotypes in our cohort (50.2% communicating versus 49.8% aqueductal, respectively) we calculate that this correlation was highly unlikely to occur by chance (p = 0.002). These observations support the pathogenicity of the mutations and suggest phenotypic subsets of CH are influenced by genetic determinants.

Our results suggest incomplete penetrance and variable expressivity of mutations in two CH genes. Four first-degree relatives of probands with the same mutation in SMARCC1 and one relative of a PTCH1 mutation proband had no evidence of hydrocephalus. Moreover, heterozygous LOF mutations in PTCH1 are a well-described cause of Gorlin syndrome (Gorlin, 2004; Gorlin and Goltz, 1960), which can be associated with obstructive hydrocephalus due to medulloblastoma (Amlashi et al., 2003). Similar to murine hydrocephalus with PTCH1 mutation (Wang et al., 2010a; Wetmore et al., 2015), our patients had hydrocephalus without CSF obstruction from medulloblastoma. In addition, a mutation-bearing parent of one of our probands had been diagnosed with Gorlin syndrome owing to numerous basal cell carcinomas but did not have hydrocephalus. This suggests that LOF mutations in our probands are functionally equivalent to those causing Gorlin syndrome and that these probands may be at risk of developing basal cell carcinomas. The determinants of incomplete penetrance and variable expressivity remain obscure but may be explained by common genetic (Timberlake et al., 2016) or environmental (Stuart et al., 2015) modifiers or by stochastic elements.

Similar to other Mendelian forms of CH (e.g., L1CAM) (Rosenthal et al., 1992), several CH patients harboring TRIM71 and SMARCC1 mutations exhibited other neurological findings and structural brain abnormalities in addition to hydrocephalus. All CH patients with TRIM71 mutations and 3 of 5 CH patients with SMARCC1 mutations exhibited neurodevelopmental delay and epilepsy. A mother and son with mutant TRIM71 and ventriculomegaly had similar open schizencephalic clefts (Figure 1). Two patients with SMARCC1 mutations exhibited schizencephalic clefts and possibly other structural brain abnormalities (Figure 2). These findings are consistent with TRIM71 and SMARCC1 regulating key aspects of brain development (Cuevas et al., 2015; Harmacek et al., 2014; Kim et al., 2001; Maller Schulman et al., 2008; Narayanan et al., 2015; Nguyen et al., 2016).

SMARCC1 and PTCH1 in CH are LOF mutations. TRIM71 exhibits a highly specific mutation spectrum involving either of two homologous arginines in the NHL domain that directly binds mRNAs (Chang et al., 2012; Loedige et al., 2013, 2014, 2015). While these mutations likely impair binding to normal target mRNAs, other gene functions, such as its putative ubiquitin ligase activity, may be preserved. Alternatively, these mutations could impart neomorphic effects that cause binding to new mRNAs. Similarly, the observed duplications of SHH suggest that these mutations are gain of function. Dosage loss of PTCH1, which inhibits downstream signaling via effect on the SMO receptor, and dosage gain of SHH would be similarly expected to increase downstream Hedgehog signaling (Briscoe and Thérond, 2013).

Ciliated neuroepithelial stem cells in the neural tube ventricular zone undergo proliferation accompanied by minimal differentiation during early embryogenesis. These cells become differentiated neurons (Ever and Gaiano, 2005; Merkle et al., 2004) in mice between E8 and E14 (Chen et al., 2006; Noctor et al., 2004), a transformation driven by significant alterations in gene expression (Yao et al., 2016). TRIM71, SMARCC1, and PTCH1 are all highly expressed in the ciliated neuroepithelium (Cuevas et al., 2015; Gavino and Richard, 2011; Tuoc et al., 2013b). SHH engages its receptor PTCH1 on neuroepithelial cilia. All five genes regulate the proliferation, fate specification, and/or differentiation of NPCs the development of the neural tube (Kim et al., 2001; Nguyen et al., 2017; Shikata et al., 2011).

The gene mutations reported here suggest the importance of de novo or transmitted mutations that alter the balance between NPC proliferation and differentiation in the pathogenic mechanism of CH. In support of this human genetic data, Dusp16 deficiency in mice causes brain overgrowth and associated obstructive hydrocephalus due to hyperproliferation and expansion of ventricular zone NPCs at the level of the aqueduct (Zega et al., 2017). In contrast, the communicating hydrocephalus phenotype in a mouse model of the human ciliopathy Bardet-Biedl syndrome results from decreased proliferation and survival of NPCs (Carter et al., 2012). Interestingly, intraventricular hemorrhage in neonates, the most common cause of infantile secondary hydrocephalus (Mazzola et al., 2014; Robinson, 2012), is also associated with decreased NPC viability and associated cortical thinning (McAllister et al., 2017; Yung et al., 2011). A “neural stem cell” paradigm of hydrocephalus could provide a unifying mechanism to explain multiple forms of infantile hydrocephalus.

These findings have potential implications for CH treatment. Multiple non-neural mechanisms have been proposed to account for communicating hydrocephalus, including elevated venous pressure, arachnoid granulation immaturity, and lymphatic dysplasia (Govaert et al., 1991). However, given the new genetic data, a primary driver of some communicating forms of CH may not be attributable to active CSF accumulation and associated ventricular distension but rather impaired neurogenesis. In neonatal CH cases with normal or low intracranial pressure (Tully et al., 2016), it is unclear whether surgical CSF shunting addresses a pathologic feature of CH. Instead, it may expose CH patients to surgical morbidity without improving neurodevelopmental outcomes. The implications for families with CH children and their neurosurgeons will stimulate discussion and catalyze further investigation into CH pathogenesis and treatment.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Kristopher T. Kahle (kristopher.kahle@yale.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Subjects and Samples

All study procedures and protocols comply with Yale University’s Human Investigation Committee and Human Research Protection Program. Written informed consent for genetic studies was obtained from all participants. Inclusion criteria included patients with primary congenital hydrocephalus without known genetic causes such as L1CAM or any other large chromosomal deletions or rearrangements. Hydrocephalus cases with secondarily acquired etiologies such as intraventricular hemorrhage (IVH), meningitis, obstruction due to tumors or cysts, and stroke were excluded. Children with hydranencephaly, large cysts and cephaloceles, posterior fossa crowding, myelomeningocele (Chiari II syndrome), or benign extra-axial CSF accumulation (i.e., benign external hydrocephalus) were also excluded. Sequenced trios were composed of two unaffected parents and one affected child with primary congenital hydrocephalus. All probands had undergone surgery for therapeutic CSF diversion (shunt placement and/or endoscopic third ventriculostomy). Patients and participating family members provided buccal swab samples (Isohelix SK-2S DNA buccal swab kits), medical records, radiological imaging studies, operative reports, and congenital hydrocephalus phenotype data.

The control cohort was composed of 1,789 previously whole-exome sequenced and analyzed families from the Simons Foundation Autism Research Initiative Simplex Collection (Fischbach and Lord, 2010; Iossifov et al., 2014; Krumm et al., 2015; O’Roak et al., 2011; Sanders et al., 2012). Sequenced families were comprised of two unaffected parents, one affected child with autism, and one unaffected sibling. Only the unaffected sibling and parents, as designated by the Simons Simplex Collection, were analyzed and served as controls for this study (Krumm et al., 2015).

Whole-Exome Sequencing and Variant Calling

DNA was isolated from buccal swab samples in accordance with manufacturer protocol. Whole-exome sequencing was performed using the IDT xGen capture reagent followed by 99 base paired-end sequencing on the Illumina HiSeq 2000 instrument at the Yale Center for Genome Analysis as previously described (Timberlake et al., 2016). Exome sequencing quality metrics are shown in Table S2 and Figures S1 and S2.

Sequence reads were mapped and aligned to the GRCh37/hg19 human reference genome using Burrows-Wheeler Aligner-MEM (Li and Durbin, 2009). In accordance with GATK Best Practices recommendations, the data were further processed using Genome Analysis Toolkit (GATK) base quality score recalibration (McKenna et al., 2010), indel realignment, duplication marking and removal, and base quality score recalibration (DePristo et al., 2011; Van der Auwera et al., 2013). Single nucleotide variants and small insertions and deletion were called using GATK Haplotype Caller and annotated using ANNOVAR (Wang et al., 2010b), NHLBI exome variant server (Fu et al., 2013), 1000 Genomes (Auton et al., 2015), DbSNP (Sherry et al., 2001), and gnomAD and ExAC databases (Lek et al., 2016).

The average depth of coverage of the whole-exome sequencing data was 54.3x, with greater than 8x coverage in 95.9% of the target region for exome capture (Table S2).

The sporadic or autosomal recessive mode of inheritance exhibited in our cohort pedigrees led us to prioritize de novo, compound heterozygous, and homozygous variants. Variants were filtered for predicted deleteriousness and conservation using a series of in silico prediction algorithms, including Meta-analytic support vector machine (MetaSVM) and Meta-analytic logistic regression (MetaLR) (Dong et al., 2015), Polymorphism Phenotyping (PolyPhen) (Adzhubei et al., 2010), Combined Annotation-Dependent Depletion (CADD) (Kircher et al., 2014), Sorting Intolerant From Tolerant (SIFT) (Kumar et al., 2009; Ng and Henikoff, 2001, 2002, 2003, 2006), conservation across 46 orthologs (cons46diff) (Cromer et al., 2012; Stuart et al., 2015), Likelihood Ratio Test (LRT) (Chun and Fay, 2009), MutationTaster (Schwarz et al., 2010), Mutation Assessor (Reva et al., 2011), Functional Analysis Through Hidden Markov Models (FATHMM) (Shihab et al., 2013), FATHMM–Multiple Kernel Learning (FATHMM-MKL) (Shihab et al., 2015), FATHMM-Coding (Shihab et al., 2014), Protein Variant Effect Analyzer (PROVEAN) (Choi et al., 2012), Variant Effect Scoring Tool (VEST3) (Carter et al., 2013), Mendelian Clinically Applicable Pathogenicity (M-CAP) (Jagadeesh et al., 2016), deleterious annotation of genetic variants using neural networks (DANN) (Quang et al., 2015), Eigen-PC (Ionita-Laza et al., 2016), Genomic Evolutionary Rate Profiling (GERP++ and GERPP++ GT2) (Davydov et al., 2010), phylogenetic p values (phyloP100way and phyloP20way) (Pollard et al., 2010), phastCons100way and phastCons20way (Siepel et al., 2005), Site-specific Phylogenetic analysis (SiPhy) (Garber et al., 2009; Lindblad-Toh et al., 2011), REVEL (Ioannidis et al., 2016), and MPC (Samocha et al., 2017).

METHOD DETAILS

Kinship Analysis

Pedigree information and participant relationships were confirmed utilizing pairwise PLINK identity-by-descent (IBD) calculation (Purcell et al., 2007). The IBD sharing between the probands and parents in all trios is between 45% and 55%. Pairwise individual relatedness was corroborated using KING (Manichaikul et al., 2010).

Haplotype Phasing and Analysis of Inbreeding

Haplotype phasing, inbreeding coefficient, and the longest homozygosity-by-descent (HBD) fragment were estimated using Beagle v3.3.2 (Browning and Browning, 2007) as described previously (Jin et al., 2017). The criteria of consanguinity are defined as runs of homozygosity in segments of 2cM or greater length that collectively comprise at least 0.35% of the genome.

Principal Component Analysis

In order to determine the ethnicity of each participant, we utilized the EIGENSTRAT software (Price et al., 2006) to analyze SNPs in cases, controls, and HapMap subjects as previously described (Jin et al., 2017; Timberlake et al., 2016).

De Novo and Inherited Variant Analysis and Filtering

De novo mutations were called in parent-offspring trios, each consisting of an affected child with primary congenital hydrocephalus and his or her unaffected biological parents, using the Bayesian framework TrioDeNovo (Dong et al., 2015). Candidate de novo variants were filtered based on the following criteria: (1) minor allele frequency (MAF) ≤ 5 × 10⁻³ in ExAC, 1000 Genomes, and EVS, (2) GATK variant quality score recalibration (VQSR) of ‘pass’, (3) minimum sequencing depth of 8 reads in the proband and each parent, (4) genotype quality (GQ) score ≥ 20 and alternate allele ratio ≥ 40%, (5) TrioDeNovo data quality (DQ) score ≥ 7, and (6) exonic or splice-site variant.

Transmitted dominant variants were filtered by similar criteria of rareness and quality: (1) MAF ≤ 2 × 10⁻⁵ in ExAC, (2) GQ ≥ 20 and alternate allele ratio ≥ 40%, (3) GATK VQSR of ‘pass’, and (4) minimum sequencing depth of 8 reads in each participant. Recessive variants were also filtered for rare (MAF ≤ 10⁻³ in ExAC) bi-allelic events (homozygous and compound heterozygous mutations) that met read quality criteria as above (GQ ≥ 20, alternate allele ratio ≥ 40%, ‘pass’ GATK VQSR, and minimum sequencing depth ≥ 8). Hemizygous recessive variants were filtered for rare events (MAF ≤ 2 × 10⁻⁵) using the same quality criteria described above.

The impact of nonsynonymous single nucleotide variants on protein function was predicted using the MetaSVM algorithm (Dong et al., 2015), identifying mutations with rank scores greater than 0.83357 as deleterious (‘D-mis’). D-Mis and loss-of function mutations (nonsense, frameshift insertions and deletions, and splice-site) were considered potentially damaging to protein function.

All de novo and transmitted calls were verified by in silico visualization of aligned reads using the BLAT search (Kent, 2002) and Integrative Genomics Viewer (IGV) (Robinson et al., 2011). Salient de novo and compound heterozygous calls were then verified in all participants that provided DNA by direct Sanger sequencing of PCR amplicons containing the mutation.

Burden of De Novo Mutations

The burden of de novo mutations in congenital hydrocephalus cases and unaffected autism controls was determined using the denovolyzeR package (Ware et al., 2015) as previously described The probability of observing a de novo mutation in each gene was calculated as illustrated previously (Jin et al., 2017), with the exception that the coverage adjustment factor was based on the full set of 125 case trios (or 126 case trios in the SMARCC1 analysis given the inclusion of a de novo in parent Hydro106–3) and 1,789 control trios (separate probability tables for each set). The expected number of de novo mutations across variant classes in case and control cohorts was calculated and compared to the observed number of de novo mutations in each cohort using the Poisson test (Samocha et al., 2014). Gene-set enrichment analyses and statistical tests considered only mutations observed or expected in genes within the specified set (i.e., high brain-expressed, aqueductal stenosis).

Contribution of De Novo Mutation to Congenital Hydrocephalus

The number of de novo mutations in risk genes that contribute to congenital hydrocephalus was calculated based on the observed count of protein-altering de novo mutations compared to expectation ( = Nx((M₁–M₂)/M₁), where N is the total number of trios and M₁ and M₂ are the observed and expected count of protein-altering de novo mutations per trio, respectively)

Enrichment Analysis for the Dominant and Recessive Variants

To quantify the enrichment of LOF heterozygous variants, we calculated the expectation for a gene using the following formula:

$$\text{Expected }{\text{LOF}}_j=L\times \frac{{\text{ mutability}}_j}{{\sum }_{\text{Genes}}{\text{mutability}}_j}$$

where ‘j’ denotes the ‘jth’ gene and ‘L’ denotes the total number of LOF heterozygous mutations. A one-tailed binomial test was conducted to compare the observed number of heterozygous variants to expectation.

For damaging recessive genotypes (RGs) in a specific gene in cases, we conducted a one-tailed binomial test to evaluate enrichment as described previously (Jin et al., 2017). RG can also be modeled separately as compound heterozygotes or homozygotes. The expected number of compound heterozygotes for each gene is derived from distributing the observed number of RGs, N, across all genes according to the ratio of the squared de novo probabilities:

$$\text{Expected Compound }{\text{RG}}_i=N\times \frac{{\mathit{\text{probability}}}_{de\mathit{\text{novo }}}^2}{{\sum }_{\mathit{\text{Genes}}}\left({\mathit{\text{probability}}}_{de\mathit{\text{novo }}}^2\right)}$$

The expected number of homozygotes is derived similarly, but using the linear ratio of de novo probabilities:

$$\text{Expected Homozygous }{\text{RG}}_i=N\times \frac{{\mathit{\text{probability}}}_{\mathit{\text{de novo }}}}{{\sum }_{\mathit{\text{Genes}}}\left({\mathit{\text{probability}}}_{\mathit{\text{de novo }}}\right)}$$

The total number of expected RG for each gene is the sum of the derived expected compound heterozygous and homozygous values.

Gene Ontology Enrichment Analysis

Three genome-wide significant genes PTCH1, SMARCC1, and TRIM71 were input into GOrilla (Eden et al., 2009) to identify enriched GO terms compared to the background set of genes (M = 18,715). For gene-set enrichment analyses, each statistical test considered observed or expected mutations in genes within the specified gene set.

In Silico Splice-Site Prediction

In order to assess the impact of a missense mutation at the splice donor site of intron 10 of PTCH1 that changed the donor site sequenced from GTA to GTC, we utilized Human Splicing Finder (Desmet et al., 2009) and MaxEntScan (Yeo and Burge, 2004). Both programs predicted that this mutation is likely to affect splicing. MaxEntScan assigned a MaxEnt score of 7.64 to the wild-type canonical splice donor site; however, after mutation of the wild-type sequence to GTC, the alternate splice donor bases upstream was assigned a score of −0.18. The difference in MaxEnt scores between the mutant and the reference sequence is −7.82, which provides strong support for this mutation being a potential 5′ donor splice site.

Copy Number Variation Analysis

XHMM was run to call CNVs from WES as previously described (Ruderfer et al., 2016). GATK DepthOfCoverage was used to calculate mean read depth per targets from the alignment files. The data were normalized by removing the highest variance principal components (variance > 70%) and z scores were calculated from the mean read depths. CNVs were called using the Viterbi Hidden Markov model (HMM) and the quality scores were calculated using the forward-backward HMM. After filtering out common CNVs present at allele frequencies greater than 0.1% in 1000 Genomes (Sudmant et al., 2015) and 10% in the cohort, high quality CNVs (SQ > 60 where SQ indicates the phred-scaled quality score for the presence of a CNV event within the interval) were subjected to visual inspection. De novo CNVs were assessed using PLINK/Seq command-line tools (Fromer et al., 2012).

QUANTIFICATION AND STATISTICAL ANALYSIS

De Novo CNV Permutation Test

The probability of finding one exon covered by multiple de novo CNVs was calculated by comparing the observed distribution to an empirical distribution derived from 1 million permutations. For each permutation, five de novo duplications were randomly distributed across the genome; each duplication contained the same number of exons as predicted from XHMM with adjustment if the CNV is partially validated by experiment. In each permutation, the experiment was considered a success if at least one locus contained exactly one exon was covered by multiple duplications. The number of successes was tallied and the p value was obtained by dividing the number of successes by the number of iterations.

Estimating the Number of Risk Genes

A maximum likelihood approach was used to estimate the number of genes contributing to congenital hydrocephalus via de novo events as described previously (Jin et al., 2017). We defined K to be the number of observed protein-altering de novo mutations in high brain-expressed (HBE) genes among cases. R1 indicates the number of HBE genes mutated exactly twice in cases and R2 indicates the number of high HBE mutated three times or more. We set the proportion (E) of protein-altering mutations in risk genes based on point estimate of enrichment in cases compared to expectation (E = (M1–M2)/M1, where M1 and M2 are the observed and expected count of protein-altering de novo mutations per trio, respectively). We then simulated the likelihood function as follows: First, we randomly selected G risk genes from the HBE gene set. Next, we simulated the number of contributing protein-altering mutations in risk genes, i.e., C, by sampling once from Binomial(K,E) distribution. Then, we simulated C contributing protein-altering mutations in G risk genes and K-C non-contributing protein-altering mutations in the complete HBE gene set, using each gene’s protein-altering mutability score as probability weights. We performed 20,000 simulations for G from 2 to 100, and calculated the likelihood function L(G) as the proportion of simulations in which the number of genes with two protein-altering de novo mutations equals to R1 and the number of genes with three or more protein-altering de novo mutations equals to R2. We then estimated the number of risk genes using the maximum likelihood estimate (MLE). Based on the likelihood function, we calculated the Fisher information and constructed the confidence interval based on the MLE and estimated Fisher information using the following equation

$$\text{MLE}\pm 1.96\times \left(\frac{1}{\sqrt{\text{ Fisher Information }}}\right)$$

In Situ Hybridization

Mouse brains and embryos were fixed in 4% paraformaldehyde by overnight immersion and sectioned (10–15 μm, cryostat sections for digoxigenin [DIG] probes). Antisense RNA probes corresponding to murine Ptch1, Smarcc1 and Trim71 (approx. 200bp for DIG-labeled probes respectively) were synthesized to detect Ptch1, Smarcc1 and Trim71 transcripts in murine tissue, using methods previously described (Duncan et al., 1994; Petersen et al., 2002; Schaeren-Wiemers and Gerfin-Moser, 1993).

DATA AND SOFTWARE AVAILABILITY

The sequencing data for congenital hydrocephalus case-parent trios reported in this manuscript have been deposited in the NCBI database of Genotypes and Phenotypes (dbGaP). The accession number for the data reported in this paper is dbGaP: phs000744.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited Data
Whole-exome sequencing data from CH trios (n = 125)	This paper	dbGaP: phs000744; https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000744
Whole-exome sequencing data from SSC control trios	Iossifov et al., 2014	https://ndar.nih.gov/study.html?id=352
Software and Algorithms
Genome Analysis Tool Kit (GATK)	DePristo et al., 2011; McKenna et al., 2010; Van der Auwera et al., 2013	https://software.broadinstitute.org/gatk/
BWA-mem	Li and Durbin, 2009	http://bio-bwa.sourceforge.net/
Annovar	Wang et al., 2010b	http://annovar.openbioinformatics.org/en/latest/
PLINK/SEQ	Fromer et al., 2014	https://atgu.mgh.harvard.edu/plinkseq/
Other
1000 Genomes GRCh37 h19 genome build	1000 Genomes Project	http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/human_g1k_v37.fasta.gz
RefSeq hg19 gene annotation	UCSC Genome Browser	http://genome.ucsc.edu/cgi-bin/hgTables?command=start
Intervals file for IDT xGen v.1.0	Integrated DNA Technologies	http://www.idtdna.com/pages/products/next-generation-sequencing/hybridization-capture/lockdown-panels/xgen-exome-research-panel

Highlights.

• Exome sequencing identifies novel genetic drivers of congenital hydrocephalus (CH)

• De novo and inherited rare variants in four genes explain ~10% of CH cases

• All four CH genes (TRIM71, SMARCC1, PTCH1, and SHH) regulate neural stem cell fate

• These data implicate aberrant neurogenesis in the pathogenesis of a subset of CH