Multi-omic analysis elucidates the genetic basis of hydrocephalus.

We conducted PrediXcan analysis of hydrocephalus risk in ten neurological tissues and whole blood. Decreased expression of MAEL in the brain was significantly associated (Bonferroni-adjusted p < 0.05) with hydrocephalus. PrediXcan analysis of brain imaging and genomics data in the independent UK Biobank (N = 8,428) revealed that MAEL expression in the frontal cortex is associated with white matter and total brain volumes. Among the top differentially expressed genes in brain, we observed a significant enrichment for gene-level associations with these structural phenotypes, suggesting an effect on disease risk through regulation of brain structure and integrity. We found additional support for these genes through analysis of the choroid plexus transcriptome of a murine model of hydrocephalus. Finally, differential protein expression analysis in patient cerebrospinal fluid recapitulated disease-associated expression changes in neurological tissues, but not in whole blood. Our findings provide convergent evidence highlighting the importance of tissue-specific pathways and mechanisms in the pathophysiology of hydrocephalus.

INTRODUCTION

Hydrocephalus is a heterogeneous disease resulting from abnormal accumulation of cerebrospinal fluid (CSF) and subsequent elevations in intracranial pressure resulting in impaired neurodevelopment and morbidity (Kahle et al., 2016; Tomycz et al., 2017). Hydrocephalus affects nearly 1 in 1,000 babies born in the United States (Simon et al., 2008), yet the genetic basis of the disease is largely unknown (Kousi and Katsanis, 2016). While clinical trials have attempted pharmacological strategies to treat hydrocephalus (Whitelaw et al., 2001), no pharmacological approaches have been successful. The current treatments for hydrocephalus are surgical interventions such as insertion of a ventriculoperitoneal (VP) shunt as well as endoscopic third ventriculostomy (ETV) with or without choroid plexus cauterization (CPC) (Kahle et al., 2016; Kulkarni et al., 2017). While many studies have evaluated the efficacy and cost of these procedures (Lim et al., 2018), long-term morbidity remains high.

Hydrocephalus can be a secondary consequence of intraventricular hemorrhage (IVH), spina bifida, infection, brain tumor, or congenital form. In addition, anatomic obstruction (i.e.,aqueductal stenosis) impairing the flow of CSF can be caused by a rare X-linked mutation in L1 cell adhesion molecule (L1CAM) (Rosenthal et al., 1992). Proposed pathophysiological mechanisms of hydrocephalus include impaired development of the neural stem cell niche (Furey et al., 2018; Lehtinen et al., 2011, 2013; Lehtinen and Walsh, 2011; Carter et al., 2012), abnormal ciliated ependymal cells (Takagishi et al., 2017; Wilson et al., 2010; Wodarczyk et al., 2009), disruption of the ventricular zone (Castaneyra-Ruiz et al., 2018; McAllister et al., 2017), and dysfunction of CSF absorption and/or secretion (Karimy et al., 2017; Lun et al., 2015b). However, very little is known about the underlying germline genetic contributions to hydrocephalus (Kousi and Katsanis, 2016; Zhang et al., 2006).

While numerous studies have sought to identify causative genetic mechanisms leading to hydrocephalus, largely based on isolated human case studies and murine models (Kousi and Katsanis, 2016), critical limitations include cost, patient/family recruitment, number of patients (small by standards of population genetics), individual variant validation (typically de novo mutations), and very important species differences between model organisms and human disease. As hydrocephalus is a component of a wide array of genetic syndromes and Mendelian disorders, as well as a secondary consequence of many pathologies, we hypothesized that hydrocephalus is a polygenic and complex disease, but may converge on a limited number of pathways. Thus, elucidation of the genetic basis of hydrocephalus may lead to new insights into pathophysiological mechanisms and identification of targets for pharmacological intervention.

Genome-wide association studies (GWASs) have become ubiquitous as a tool for identifying genetic predispositions to complex traits, but these analyses require very large sample sizes. Importantly, the underlying mechanisms for the identified loci are, for the most part, unclear. Elucidating the mechanistic basis of a complex, polygenic disorder (Manolio et al., 2009) requires understanding the molecular events that give rise to the disease process. We hypothesized that the summative risk for hydrocephalus results from small variation in the expression of many genes, leading to alterations of a limited number of pathways or biological processes that ultimately predispose individuals to disease. Because of the unavoidably smaller sample size for hydrocephalus (relative to other complex disorders), we aimed to test genes whose expression can be reliably determined using genetic variation (rather than millions of genetic variants, as in conventional GWASs, with unknown gene targets) in order to substantially improve statistical power.

To this end, we used whole-genome genetic data (from blood collected for routine clinical care) linked to the deidentified electronic medical record (BioVU; Roden et al., 2008) to perform the largest genetic analysis of hydrocephalus to date. We hypothesized that genetically determined gene expression contributes to the development of hydrocephalus. Thus, we applied PrediXcan (Gamazon et al., 2015), a gene-based method that utilizes the genetic component of gene expression for disease gene identification, leveraging imputation models derived from a reference human transcriptome panel of neurological tissues (10 brain regions from 889 individuals) and whole blood (from 338 patients) (Battle et al., 2017). Our approach identified genetically determined gene expression traits and pathways associated with hydrocephalus. Although correlation does not imply causation, a disease association with genetically determined expression substantially improves on a SNP-based association from conventional GWASs (in which the relevant gene is generally unknown; Nicolae et al., 2010). We considered whether the gene-level associations were driven by linkage disequilibrium (LD) contamination. We observed a notable degree of tissue specificity in the association of genetically determined gene expression with hydrocephalus risk. We replicated our experiment-wide significant finding in an independent GWAS dataset (UK Biobank) (Bycroft et al., 2018). Furthermore, we sought functional support for our findings using PrediXcan analysis of imaging-based phenotypes in the UK Biobank (Elliott et al., 2018) and transcriptome analysis of a murine model of hydrocephalus. Finally, we compared protein expression analysis from CSF isolated from infants with hydrocephalus to gene expression changes identified by PrediXcan, mirroring our findings in neurological tissues, but not in whole blood. In sum, we illuminate the complex genetic architecture of hydrocephalus and offer crucial insights into the pathophysiological basis of the disease.

RESULTS

A diagram illustrating our study design and approach can be found in Figure 1. In this study, we performed a systematic genetic study of hydrocephalus using PrediXcan (Gamazon et al., 2015, 2018), a methodology that estimates the genetic component of gene expression, to identify gene-level associations with disease. We performed systematic validation of our genetic findings using independent replication in the UK Biobank, analysis of brain structural imaging phenotypes linked to genetic information in the UK Biobank, transcriptome analysis of choroid plexus isolated from a murine model of hydrocephalus, and comparison of genetically determined expression changes to directly measured differential proteomic expression analysis of CSF isolated from infants with hydrocephalus.

Figure 1.

Overview of our approach

We leverage a large DNA biobank, BioVU (Roden et al., 2008), linked to deidentified electronic health record (EHR) data. We applied PrediXcan (Gamazon et al., 2015, 2018), which estimates the tissue-specific, genetically determined component of gene expression (i.e., the “germline genetic profile” of the gene expression trait) based on common variants (minor allele frequency >1%) and imputation against a reference transcriptome panel. For this study, we used GTEx transcriptome data in 10 neurological tissues and whole blood (Battle et al., 2017). The genetic component of expression was then tested for association with phenotype to identify gene-level associations. We performed systematic validation using independent replication of genetic results in the UK Biobank (Bycroft et al., 2018), analysis of structural brain magnetic resonance imaging (MRI) phenotypes in the UK Biobank (Elliott et al., 2018), and analysis of choroid plexus isolated from a mouse model of hydrocephalus (Robledo et al., 2008) and compared genetically determined gene expression changes to proteomic analysis of cerebrospinal fluid (CSF) isolated from infants with hydrocephalus compared to non-affected controls. A summary of each phenotype and the corresponding sample size can be found in Table S4.

Generation of imputed transcriptome (PrediXcan) to identify hydrocephalus-associated genes

Estimation of the genetically determined transcriptome in 10 neurological tissues and whole blood was performed using PrediXcan (Figure 1; see STAR Methods). Differential expression analysis in the frontal cortex highlighted maelstrom spermatogenic transposon silencer (MAEL) and lysine demethylase 1A (KDM1A) as outliers (Figure 2A). After multiple testing correction (Benjamini-Hochberg adjusted p < 0.05), MAEL, but not KDM1A, in the frontal cortex was significantly associated with hydrocephalus (Figure 2B). Notably, MAEL was experiment-wide significant, satisfying a highly stringent Bonferroni threshold for statistical significance (pBonferroni < 0.05) based on the total number of gene-tissue pairs tested (see STAR Methods). We note that the p value threshold for significance for PrediXcan is not the same as that for traditional GWAS, with PrediXcan enjoying a substantially reduced multiple testing burden (Gamazon et al., 2015; see STAR Methods). Genetically determined MAEL expression did not differ between varying etiologies of hydrocephalus (i.e., idiopathic congenital, neural tube defects, and others), and the MAEL association was not specific to any one specific etiology. Finally, as a Manhattan plot illustrates, MAEL was uniquely associated with hydrocephalus within the locus and across all genes tested in the frontal cortex (Figure 2C), suggesting that the MAEL association was not merely the result of LD with a different causal gene within the locus. The MAEL association was also observed in the hypothalamus (Figures 2D–2F), based on the total number of genes tested in this tissue. In both frontal cortex and hypothalamus, decreased expression of MAEL conferred increased predisposition to hydrocephalus. A complete list of all SNPs (and their relative weights) in the MAEL cis region (i.e., within 1 Mb) included in the PrediXcan models for frontal cortex and hypothalamus are included in Table S1.

Figure 2.

Genome-wide scan identifies tissue-specific gene-level associations with hydrocephalus

(A) Volcano plot showing odds ratio (OR, x axis) versus −log (p value, y axis) for gene expression differences between cases and controls in the frontal cortex.

(B) Q-Q plot demonstrating a significant association between MAEL and hydrocephalus after correction using Benjamini-Hochberg FDR in the frontal cortex. MAEL is study-wide significant (adjusted p < 0.05) after Bonferroni adjustment for the number of gene-tissue pairs tested in the study.

(C) Manhattan plot showing the gene-level association p values and chromosomal location of the signal from MAEL in the frontal cortex. The experiment-wide significant gene MAEL is also the unique gene in the cis region with a nominal gene-level association (p < 0.05) with hydrocephalus.

(D–I) Analogous analyses were performed in hypothalamus tissue (D–F) and whole blood (G–I).

See also Data S1.

Interestingly, gene expression differences between cases and controls in whole blood, the tissue most often acquired for human genetics studies due to easy accessibility and low cost, showed no significant departure from null expectation (Figures 2G–2I) despite its much larger sample size and, hence, statistical power, demonstrating the importance of considering tissue-specific expression differences in relevant tissues when performing genetic screens of hydrocephalus. However, whole-blood analysis revealed nominally significant associations with previously reported candidate genes—CCDC28A (Cardenas-Rodriguez et al., 2013), MARCKS (Chen et al., 1996; Lang et al., 2006), and TNFRSF10D (Habiyaremye et al., 2017; Jiménez et al., 2014)—that have been independently implicated in human and mouse studies of hydrocephalus.

The list of genes nominally associated (p < 0.05) with hydrocephalus disease status and the distribution of association p values varied across neurological tissues (Figure 3A; Data S1). Gene expression differences between cases and controls in the frontal cortex were the most significant (Figure3A). The distribution of gene effect sizes on hydrocephalus differed across neurological tissues (Figure 3B), highlighting the importance of examining tissue-specific effects on disease risk in neurological tissues. We performed hierarchical clustering of nominally associated genes across neurological tissues (Figure 3C). Using multiscale bootstrap resampling (with “average” as the agglomerative method, “correlation” as distance method, and 1,000 bootstrap replicates; Shimodaira, 2004), we found that, in contrast to the genes nominally associated with hydrocephalus (p < 0.05), randomly sampled genes did not lead to stable clusters, i.e., the hypothesis “the cluster does not exist” cannot be rejected at the significance level of 0.05 (with an “approximately unbiased” value of zero), and do not recapitulate known relationships between Genotype-Tissue Expression (GTEx) tissues (between the cerebellar hemisphere and cerebellum and between the frontal cortex and cortex) (Battle et al., 2017). The gene-level associations (nominal p < 0.05) are highly tissue-specific, with the majority detected in one tissue (Figure 3D). We provide a list of all nominally significant (p < 0.05) genes in neurological tissues for reference (Data S1).

Figure 3.

Differentially expressed genes in neurological tissues and whole blood

(A) Gene-level associations (PrediXcan) with hydrocephalus status, determined by logistic regression (with sex and age and the genotype-based principal components as covariates), in each neurological tissue and whole blood, including genes that depart from null expectation. MAEL expression in frontal cortex was experiment-wide significant (Bonferroni-adjusted p < 0.05) across all tissue-gene pairs tested.

(B) Significance and effect size of gene-level associations in neurological tissues identifying outliers. Gene associations within each neurological tissue are color coded.

(C) Hierarchical gene clustering of the nominally significant gene-level associations (p < 0.05), with whole blood an outlier relative to the neurological tissues.

(D) Number of tissues in which gene-level nominal associations (p < 0.05) with hydrocephalus are detected.

See also Data S1.

Contrary to MAEL’s most commonly studied role in spermatogenesis, MAEL is actually expressed in both males and females, albeit absent in female sex organs, and is highest in the frontal cortex (the neurological tissue deviating the most from null expectation, Figure 1), outside of the testis (Figure 4A). Interestingly, MAEL-deficient mice have been generated, but they do not develop hydrocephalus (Soper et al., 2008). However, since MAEL expression is one of the most highly tissue-specific genes (in expression profile) in the genome (Figure 4B; see STAR Methods) and because MAEL function differs substantially across species (Chen et al., 2015), further mechanistic validation of MAEL in hydrocephalus needs to be performed specifically on human neurological tissue samples, building on our results from human GWASs.

Figure 4.

MAEL expression profile and model of MAEL-mediated trait effect

(A) Population-based expression of MAEL in males (blue) and females (red) across 44 tissues included in GTEx. MAEL is most highly expressed in frontal cortex among brain regions and displays low expression in whole blood. Box edges show interquartile range, whiskers 1.5 × the interquartile range, and center lines the median.

(B) Tissue specificity (x axis) as quantified by τ (see STAR methods) versus frequency of genes (y axis) across the genome. MAEL is one of the most tissue-specific genes in the genome (τ = 0.997).

(C) Canonical model of MAEL-mediated alteration of transposon movement and depression of H3K9me3, leading to recruitment of RNA polymerase II (RNA Pol II) and transcription of previously repressed genes.

Since the experiment-wide significant genetic signal to emerge from our analyses was decreased expression of MAEL in the frontal cortex, we present the canonical model for MAEL function (Figure 4C). MAEL is a critical regulator of PIWI-interacting RNA (piRNA)-mediated repression of transposable elements and methylation (Soper et al., 2008). Thus, decreased MAEL leads to increased transposon mobilization and differential methylation patterns, leading to broad changes in gene expression. However, the specific genes in the brain targeted by MAEL-mediated transposon movement and histone modification H3K9me3, which plays a role in targeting DNA methylation (Lehnertz et al., 2003), are unknown. Thus, our study implicates transposon and histone modification (trimethylation) in the pathophysiology of hydrocephalus; however, additional molecular studies are needed to confirm this finding.

Replication in the UK Biobank

Applying PrediXcan to brain magnetic resonance imaging (MRI) white matter and total brain volume data in the UK Biobank (Elliott et al., 2018; Miller et al., 2016) (N = 8,428; see STAR Methods), we found that MAEL expression in the frontal cortex was significantly associated with white matter volume (p = 0.011) and total brain volume (p = 0.015) at Bonferroni-adjusted p < 0.05. Various structural brain alterations have been implicated in the pathophysiology of hydrocephalus (Del Bigio, 2010). Remarkably, we found a significant enrichment for gene-level associations with white matter (Figure 5A) and total brain volume (Figure 5B) among the top differentially expressed genes (p < 0.05) in frontal cortex.

Figure 5.

Genomic analysis of brain MRI data in the independent UK Biobank validates gene-level associations with hydrocephalus in the same (significant) discovery tissue

Using PrediXcan analysis of brain imaging and genomic data, we validated the study-wide significant association of MAEL in frontal cortex (Bonferroni-adjusted replication p < 0.05). We then considered the associations with the imaging-based phenotypes of the top differentially expressed genes in frontal cortex.

(A) For the hydrocephalus-associated genes (p < 0.05; in red) in the frontal cortex from the BioVU analysis, the Q-Q plots show the PrediXcan p values for their association with white matter volume (A) and total brain volume (B) in the UK Biobank. The departure from the diagonal line indicates enrichment for gene-level associations with the imaging-based phenotypes among the hydrocephalus-associated genes. For comparison, a Q-Q plot for a random set of genes (of equal count; in blue) is included.

For additional support, we analyzed SNPs within the MAEL cis region (i.e., the MAEL locus that is used in PrediXcan analysis) and their association with hydrocephalus in the UK Biobank. We conducted extensive quality control (QC) on the SNPs within the locus and excluded all low-confidence variants (see STAR Methods). We identified nine variants, in LD (r2 > 0.70), nominally associated with hydrocephalus (p < 0.05, Table S2). The most significant SNP (rs75008967) overlaps an enhancer element in fetal brain (male and female) and in several brain regions (Figure S1, H3K27ac) (Creyghton et al., 2010), including hippocampus, substantia nigra, anterior caudate, cingulate gyrus, and dorsolateral prefrontal cortex, based on Roadmap chromatin immunoprecipitation sequencing (ChIP-seq) data (Ward and Kellis, 2012). In addition, a variant in the cis region of MAEL (rs72687818), independent of the lead SNP above (r2 < 0.20), was the most significant association with a common cause of hydrocephalus, namely “subarachnoid hemorrhage from intracranial artery” (Germanwala et al., 2010; Graff-Radford et al., 1989) in the UK Biobank (p = 1.1 × 10−20). We evaluated the gene-level significance from the SNP-level association results in the cis region with hydrocephalus (p = 0.03) and with subarachnoid hemorrhage (p = 4.77 × 10−12), strongly confirming the discovery signal. Collectively, these results provide robust support to the importance of the MAEL locus for hydrocephalus predisposition.

MAEL-mediated associations with hydrocephalus-related neurological traits

To explore how genetically determined expression of MAEL may exert its phenotypic effect on hydrocephalus predisposition in these specific neurological tissues, we identified a cohort of patients in BioVU with related neurological traits that may facilitate further insights into hydrocephalus pathophysiology (see STAR Methods). Consistent with the hydrocephalus associations, decreased MAEL expression in frontal cortex and hypothalamus was found to be associated with “cerebral edema and compression of brain” (p = 0.0021 in frontal cortex and p = 0.0018 in hypothalamus), “other cerebral degenerations” (p = 0.0009 in frontal cortex and p = 0.001 in hypothalamus), and intracerebral hemorrhage (p = 0.016 in frontal cortex and p = 0.021 in hypothalamus). These data are consistent with the association of MAEL expression with regulation of white matter and total brain volumes obtained through analysis of brain MRI phenotypes linked to genetic information (Figure 5) in the independent UK Biobank.

In addition, STRING analysis (version 10.5; Szklarczyk et al., 2017) revealed a number of direct interacting partners that may play a role in MAEL-mediated hydrocephalus pathophysiology, offering insights into potential targets for intervention (Figure S2A). Notably, the other genes in the proposed biological network from the STRING analysis showed nominally significant associations with neurological phenotypes (Figure S2B) in BioVU (see STAR Methods), including “type 2 diabetes with neurological manifestations” and “polyneuropathy in diabetes for decreased TDRKH expression” (p = 7.9 × 10−4 and p = 2.2 × 10−3, respectively) in whole blood and multiple sclerosis for increased DDX4 expression (p = 0.017) in frontal cortex.

To assess the significance of the observed MAEL-mediated neurological trait associations, we performed permutation analysis (n = 1,000) that preserves the pairwise (gene-gene) correlation (within each tissue) of the estimated genetic component of expression (as well as the gene count) among the tested genes. We observed a significant enrichment (empirical p < 0.001) for associations with neurological phenotypes among MAEL’s direct interacting partners. Thus, there may be some shared underlying risk for other neurological disorders in patients with alteration of MAEL-dependent biological networks, which lends additional support for (as noted above) further validation studies specifically in humans to elucidate MAEL’s role in hydrocephalus risk.

Transcriptome analysis and validation in a murine model of hydrocephalus

Many laboratories focus on the use of model organisms in studying hydrocephalus for pre-clinical translational studies. Thus, we conducted differential expression analysis (see STAR Methods) of ChIP-based gene expression data from the choroid plexus of α-adducin knockout mice, which develop hydrocephalus secondary to IVH (Robledo et al., 2008), one of the most common causes of hydrocephalus. Interestingly, there was a marked degree of concordance in both the genes involved and the direction of effect between the transcriptome of α-adducin knockout mouse choroid plexus and tissue-specific gene expression differences observed in our human cohort (Table S2). Although MAEL was validated in human GWASs but not in this murine model, this is consistent with the substantial evidence describing the species and tissue-specificity of MAEL gene expression and regulation (Chen et al., 2015; Genzor and Bortvin, 2015). Finally, nominally significant associations for TMEM50B and KCTD21 (in frontal cortex and hypothalamus) were replicated in mice with a concordant direction of effect (Table S3).

Genetically determined pathway associations with hydrocephalus

To probe relevant biological processes and molecular functions, we utilized the full set of nominally significant associations (p < 0.05) of genetically determined expression with hydrocephalus in each neurological tissue, across all tissues, and in whole blood (Data S1). We hypothesized that pathway analysis of nominally significant genetically determined expression changes (see STAR methods) would confirm a number of pathways and biological processes previously associated with hydrocephalus, and identify additional pathways associated with the disease. This approach is in contrast to conventional approaches for differential expression analysis (which utilize the total, directly measured gene expression), enabling us to identify “genetically determined” and potentially “causative,” rather than consequential disease-relevant networks (see Figure 1).

We performed gene set enrichment analysis (GSEA) (Subramanian et al., 2005) using the Molecular Signatures Database (MSigDB) on nominally significant genetically determined gene expression changes in the frontal cortex, hypothalamus, whole blood (Data S2 and S3), and all other neurological tissues individually as well as cross-tissue analysis (Data S2 and S3, which present significantly enriched gene sets at a false discovery rate [FDR] of <0.05 in each tissue), as has been previously applied to GWAS data (Wang et al., 2007, 2010). Experiment-wide significant gene sets were identified using Bonferroni correction (p < 2.9 × 10−7) (see STAR Methods). The level of significance for a gene set is not correlated with the number of genes in the set (e.g., Spearman correlation p = 0.68 in frontal cortex). Many of our top gene-level associations were identified in two or more neurological tissues (Figure S3; Figure 2D; Data S2 and S3). GSEA/MSigDB (with default options) was performed on the 225 genes that were nominally significant in two or more tissues as the input set (Data S2 and S3), recapitulating the involvement of many pathways identified by single-tissue analysis.

Exome scan and functional genomics suggest potentially pathogenic variants associated with hydrocephalus

We analyzed exome data for additional support for the role of the top differentially expressed genes in hydrocephalus risk. To this end, we tested the nominally significant hydrocephalus-associated genes detected in at least five tissue types (Figure S3; see STAR Methods) for coding variation effects on disease susceptibility, using a cohort of 29,713 patients (Figure S4A; Data S4). TMEM50B, which was also supported in our transcriptome analysis of a murine model of hydrocephalus (Table S3), contains a rare missense variant (rs34327244 or A139T, minor allele frequency [MAF] = 0.4% in Europeans [non-Finnish]) with nominally significant association with hydrocephalus (p = 7.4 × 10−3) (Figure S4B). After Benjamini-Hochberg correction, this association corresponded to a FDR of 11%, warranting additional functional studies given the convergent evidence for the gene. Interestingly, we found that rs34327244 was associated with CSF volume (normalized for overall head size) in the independent UK Biobank (p = 0.016).

In addition, using ChIP-seq data derived from human embryonic stem cells (HUES64) as part of Roadmap Epigenomics (Leung et al., 2015), we found that A139T disrupts an active or primed enhancer element marked by monomethylation of histone H3 at lysine 4 (H3K4me1), suggesting a role for the variant in regulation of transcription. Furthermore, TMEM50B is co-expressed with aquaporin 1 (AQP1) in the frontal cortex (Figure S4C), and A139T alters a regulatory motif that leads to differential allelic affinity of the NKx-2 homeodomain containing transcription factor thyroid transcription factor 1 (TTF1) (Ward and Kellis, 2012). Since decreased expression of TTF1 leads to decreased expression of AQP1 in the apical membrane of the choroid plexus and alteration in CSF formation (Kim et al., 2007), it is possible that A139T leads to decreased availability of TTF1 to promote AQP1 expression (Figure S4D), although additional experimental validation is needed. These studies provide preliminary evidence for aquaporin dysregulation in genetically determined risk for hydrocephalus.

CSF proteomic data from patients with hydrocephalus mirror PrediXcan results

Next, we analyzed proteomic data from CSF isolated from patients with hydrocephalus secondary to IVH, from a previously published study (Morales et al., 2012) for mass spectroscopic analysis (see STAR Methods), and compared the differentially expressed proteins to the differentially expressed genes identified by PrediXcan. This analysis revealed a significantly greater overlap between differentially expressed genes in the frontal cortex and the proteomic signature identified by liquid chromatography-mass spectrometry (LC-MS) than expected by chance (enrichment p = 0.016), identifying three proteins that deviate from null expectation for differential expression in frontal cortex between cases and controls (Figure 6A). The most significant of the three proteins, catalase (CAT), showed a 500-fold decreased expression (p = 1.56 × 10−48) between patients with hydrocephalus versus controls. CSF protein levels of ADAM20 and SCUBE1 also showed nominal association with hydrocephalus (PrediXcan p < 0.05).

Figure 6.

CSF proteomic signature of patients with hydrocephalus recapitulates gene expression associations identified by PrediXcan

(A) Q-Q plot showing significance for three proteins implicated by LC-MS analysis of CSF isolated from patients with hydrocephalus in the association of their genetically determined expression in the frontal cortex (−log10 p value from the PrediXcan analysis shown on the y axis).

(B) PrediXcan p values in frontal cortex of the proteomic signature from CSF (true) versus the remaining proteins (false), demonstrating greater statistical significance (i.e., lower p value from PrediXcan) for the proteins in the CSF signature in frontal cortex (left, p = 0.04) but not in whole blood (right, p = 0.83). Significance was assessed using a Mann-Whitney U test. Box edges show interquartile range, whiskers 1.5 × the interquartile range, and center lines the median.

The proteomic signature identified in the CSF tended to have significantly lower (i.e., more statistically significant) p values from the PrediXcan analysis of frontal cortex (Mann-Whitney U test, p = 0.04, Figure 6B, left) than the remaining proteins. In contrast, we observed no significant expression difference between cases and controls in whole blood (Mann-Whitney U test, p = 0.83, Figure 6B, right) for the proteins identified by LC-MS, further showing the limitation of whole blood as a sub-strate for hydrocephalus genetic analyses. Overall, these results demonstrate that while CSF proteomic analyses may be useful in identifying biomarkers for prognostic purposes, its use, without integration of genetic data, as a discovery platform of genetically determined pathogenic mechanisms is limited.

DISCUSSION

We describe the largest genetic study of hydrocephalus and identify trait-associated genes and signaling pathways, laying the groundwork for molecular studies of hydrocephalus. We report genetically determined contributions to hydrocephalus, the functional consequences of identified genes on brain structure, the transcriptomic signature of differentially expressed genes in choroid plexus of a murine model of hydrocephalus, and hydrocephalus-associated genetically determined expression changes for proteins as measured by an unbiased proteomic screen of CSF isolated from patients with hydrocephalus. The use of genetic information can be used to disentangle changes in expression that influence disease versus secondary expression changes as a result of hydrocephalus, a question that has remained elusive for more than 30 years (Del Bigio, 1989).

PrediXcan imputes the genetic component of gene expression in tissues from which it is nearly impossible to obtain clinical samples, such as neurological tissues. We find a remarkable degree of tissue specificity in gene regulation in hydrocephalus. However, the degree of tissue, cell-type, and single-cell variation in hydrocephalus has yet to be fully appreciated. While common variants and de novo mutations have been shown to play a role in a range of neurodevelopmental disorders, including hydrocephalus (Niemi et al., 2018; Short et al., 2018; Furey et al., 2018), this study implicates common variant-mediated regulation of tissue-specific gene expression as a potential driver of hydrocephalus.

Our analysis revealed a significant association between differential expression of MAEL and hydrocephalus status. MAEL contains two domains, that is, (1) a high mobility group (HMG)-box domain and (2) an RNase H-fold domain that lacks catalytic residues conserved in RNA-H nucleases, but it displays single-stranded RNA (ssRNA)-specific endonuclease activity (Genzor and Bortvin, 2015; Matsumoto et al., 2015). MAEL is essential for piRNA-mediated transcriptional transposon silencing. Interestingly, transposons are DNA sequences capable of changing their position in the genome, with the potential to induce mutations, as well as change a cell’s identity and genomic size. Loss of maelstrom, the MAEL homolog in Drosophila melanogaster, has been shown to perturb RNA polymerase II recruitment, nascent RNA output, and steady-state RNA levels of transposons, leading to increased heterochromatin spreading despite modest changes in H3K9me3 patterns (Sienski et al., 2012), which suggests that MAEL may act independently or downstream of H3K9me3. Consistent with this function, we identified H3K9me3-dependent changes as one of the most significantly enriched curated gene sets for differentially expressed genes (between hydrocephalus cases and controls) across neurological tissues (Data S2 and S3). However, the expression of MAEL is one of the most tissue-specific in the genome (Figure 4B), and a human-specific role for MAEL in neurological disorders has not previously been described. Thus, additional molecular studies of MAEL in hydrocephalus risk and pathogenesis should be performed in human neurological tissue.

Epigenetic modification (H3K9me3) underlying the MAEL association with hydrocephalus is supported by mechanistic data on the role of piRNA biogenesis factor Mili (a target of MAEL) in mice. Mili-deficient mice demonstrate broad changes in CpG hypomethylation across the genome (Nandi et al., 2016). Extending this observation, one of the significantly enriched gene sets (FDR < 0.05) across neurological tissues with hydrocephalus was alteration in CpG methylation (Data S2 and S3), providing convergent functional evidence for the potential role of these pathways in human hydrocephalus. Furthermore, deletion of PIWI (another piRNA biogenesis factor) inhibits axon regeneration that is dependent on the slicer domain of PIWI, indicating that post-transcriptional gene silencing may be involved (Kim et al., 2018). Since loss of MAEL in mice results in selective transposon insertions (Aravin et al., 2008), this is certainly a plausible mechanism. However, the complete repertoire of cell type-specific transposon insertion sites in humans remains largely unknown (Elbarbary et al., 2016). Furthermore, there is evidence that transposon elements are enriched in neural stem cells (Upton et al., 2015), a cell type widely hypothesized to be dysfunctional in hydrocephalus.

Collectively, PrediXcan analysis, rare-variant exome scan, a murine model of hydrocephalus, and genomic analysis of imaging-based brain structural phenotypes justify additional functional follow-up studies on the role of TMEM50B in hydrocephalus. Interestingly, the chromatin state annotation suggests that A136T (rs34327244) may play a role in transcriptional regulation. Indeed, ChIP-seq analysis shows that rs34327244 overlaps an enhancer region that is marked by monomethylation of H3K4me1. Analysis of transcription factor binding profiles revealed that rs34327244 alters a regulatory motif, resulting in differential allelic affinity for NKx-2 homeodomain containing transcription factor TTF1 (Ward and Kellis, 2012). Intriguingly, AQP1, a critical regulator of CSF formation and intracranial water movement (Iliff et al., 2012), is a direct transcriptional target of TTF1, and AQP1 transcript levels are directly correlated with TTF1 expression (Kim et al., 2007). Remarkably, we discovered that rs34327244 was associated with normalized CSF volume in the UK Biobank. These data suggest that in conferring hydrocephalus risk, AQP1 dysregulation may be causative; however, additional detailed molecular studies are required to definitively clarify the role of TMEM50B in tissue-specific AQP1 regulation and hydrocephalus pathophysiology. Nonetheless, our findings suggest that integrated analysis of brain imaging and functional genomics (human and model organism) data can be used to identify a previously inaccessible molecular mechanism relevant to hydrocephalus pathophysiology.

Several functional and structural features on brain MRI have been associated with hydrocephalus (Del Bigio, 2010); however, the genetic basis for these observations is not known. A repository of brain imaging phenotypes from 8,428 individuals tied to GWAS data in the UK Biobank provides a unique resource to probe these questions (Elliott et al., 2018). Since CSF, white matter, and total brain volumes have been independently associated with hydrocephalus (Mandell et al., 2015), we considered the differentially expressed genes in the frontal cortex identified by PrediXcan (Figures 2A–2C; Data S1) and their PrediXcan associations with these imaging-derived traits. We observed a significant enrichment for top associations with these brain structural phenotypes among the hydrocephalus-associated genes. Testing the effect of hydrocephalus-associated genes on brain-imaging traits may be relevant to understanding both normative brain development and neurocognitive outcomes in hydrocephalus (Mandell et al., 2010, 2015; Peterson et al., 2018). Specifically for the aims of our study, understanding the role of these genes in conferring a risk for increased (e.g., macrocephaly) and decreased (e.g., cerebral degeneration) brain volume enhances our mechanistic understanding of hydrocephalus risk, as brain volume influences CSF circulation (Brinker et al., 2014). Alterations in white matter volume have also been observed in congenital hydrocephalus (Lockwood Estrin et al., 2016), suggesting a shared genetic component. Collectively, our data suggest that hydrocephalus-associated genes may exert their effect on disease risk through their role in regulating brain structure and integrity.

Human and murine studies have demonstrated the importance of the CSF proteome in neurodevelopmental disorders as well as in hydrocephalus. We determined that decreased catalase (CAT) gene expression in the frontal cortex had a consistent effect on hydrocephalus as decreased CAT protein in the CSF (Figure 6A). The canonical role of CAT is to cleave hydrogen peroxide into water and oxygen, an important biological process mediating production of reactive oxygen species. CAT has been shown to regulate DNA damage, which causes chromosomal aberrations at regions of the genome centered around transposons (Argueso et al., 2008). These data provide potential links to chromosomal modifications in hydrocephalus pathophysiology.

CSF-derived signals have been shown to play a major role in mediating neural tube closure by signaling to the neural stem cell niche (Chau et al., 2015). In addition, the CSF proteome mediates the localization of Igf1R to the apical membrane of the choroid plexus, in part through regulation by PTEN (Lehtinen et al., 2011). Interestingly, CSF from patients with glioblastoma multiforme (GBM) selectively alters this response (Lehtinen et al., 2011), suggesting that underlying genetic risk may lead to selective development of hydrocephalus. Notably, PTEN signaling is recapitulated here by pathway analysis of genetically determined expression changes, consistent with previous reports on the role of PTEN/phosphatidylinositol 3-kinase (PI3K) signaling in hydrocephalus (Kousi and Katsanis, 2016; Yung et al., 2011; Zheng et al., 2018). Age-dependent alterations in the CSF proteome can also influence adult neural stem cells (Silva-Vargas et al., 2016), which could potentially underlie development of normal pressure hydrocephalus (NPH) in adults. Interestingly, there is a correlation with neurodegeneration-associated proteins in both CSF and cortical biopsies with NPH, suggesting some overlapping pathophysiology (Jeppsson et al., 2016; Leinonen et al., 2012).

It has already been demonstrated that alterations in choroid plexus gene expression (Lun et al., 2015a) and the neural stem cell niche (Carter et al., 2012; Furey et al., 2018) play significant roles in hydrocephalus. The relationship between neuronal and choroid plexus development, CSF dynamics, and genetic regulation underpin a highly complex physiological system with multiple modes of regulation (Lun et al., 2015b). Thus, while we were not able to directly analyze epithelial cells of the choroid plexus and cells from the neural stem cell niche in the ventricular/subventricular zone from humans, we identified a number of pathways in those compartments that had previously been associated with hydrocephalus through targeted molecular and genetic studies. For instance, analysis of genetically determined expression changes in frontal cortex revealed protein biogenesis as being significantly associated with hydrocephalus. Notably, regulation of the genes encoding the protein biosynthetic machine has been shown to be essential in forebrain development and development of macrocephaly (Chau et al., 2018). These authors showed mechanistic evidence detailing mTOR and MYC signaling as critical regulators of forebrain development, recapitulated here through pathway analysis of genetically determined expression changes (Data S2 and S3). Furthermore, mTOR signaling-mediated regulation of cilia has been shown to be required for ventricular morphogenesis (Foerster et al., 2017), further highlighting the potential importance of mTOR in hydrocephalus.

Our PrediXcan analysis is currently limited to common variant-mediated gene expression. However, as our understanding of the contribution of rare variants to human genetic regulation increases (with the requisite increase in study sample sizes and investment in functional genomic studies in diverse ancestries; Zhong et al., 2019), so too will our comprehension of genetically determined expression across the full range of allele frequency spectrum. Despite this limitation, our analysis of rare variants in the most differentially expressed genes in several neurological tissues provides another layer of functional evidence for the contribution of these genes to disease susceptibility.

Electronic health records linked to DNA biobanks, as vast repositories of disease and medication data, will enable rapid discovery and replication of genetic associations, as UK Biobank replication studies in this study have provided support for the role of MAEL in conferring hydrocephalus risk. Finally, validation studies of the implicated genetic components of gene expression in related phenotypes—such as the association between decreased MAEL expression and “cerebral edema and compression of brain” and “other cerebral degenerations” as well as the significant enrichment for neurological trait associations among MAEL’s direct interacting partners—may provide additional mechanistic insights into hydrocephalus pathophysiology.

We present the largest genetic analysis of hydrocephalus and extensive genomic analyses to identify hydrocephalus-associated genes and pathways. Our study highlights the complexity of hydrocephalus pathophysiology and the polygenicity of its genetic architecture. We propose transposon-mediated genetic regulation through MAEL for future mechanistic validation in hydrocephalus. In addition, we identify a potential molecular basis for the role of aquaporin in hydrocephalus risk, a mechanism long hypothesized to be involved in human hydrocephalus. We integrate PrediXcan analysis of imaging-based structural brain phenotypes related to hydrocephalus and demonstrate an enrichment of genes associated with alterations in brain structure and integrity among risk genes. Finally, we observed enrichment for germline-genetic expression changes in neurological tissues among the differentially expressed proteins in CSF between hydrocephalus patients and controls, demonstrating the value of our methodology for the discovery and characterization of the genetic determinants of hydrocephalus.

Limitations of study

We conducted a transcriptome-wide association study (TWAS) of hydrocephalus in 10 brain regions and whole blood. Among the top differentially expressed genes in the brain, we observed an enrichment for gene-level associations with neuroimaging phenotypes, indicating an effect on disease risk through regulation of brain structure and integrity. The molecular mediators that underlie the functional and structural brain variability contributing to disease risk require detailed follow-up studies. Choroid plexus transcriptome analysis of a murine model of hydrocephalus and proteomic analysis of CSF isolated from patients provide additional evidence for the top differentially expressed genes that reach only nominal significance. A larger sample size will improve the resolution to detect disease-associated genes. In the present study, the gene MAEL in frontal cortex attained experiment-wide statistical significance in BioVU with additional support in the UK Biobank. Future studies should be aimed at elucidating the underlying molecular mechanism in human tissue samples, given the high degree of evolutionary divergence for the gene in the tissues (i.e., neurological) of interest.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Eric R. Gamazon (eric.gamazon@vumc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

All results are available in Data S1, S2,S3, and S4.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

BioVU, one of the largest DNA biobanks tied to an electronic health records database containing 2.6 million unique patient records, is a genomics resource at Vanderbilt University Medical Center (Roden et al., 2008). Detailed information on the construction, utilization, ethics and policies of the BioVU resource is described elsewhere (Roden et al., 2008). Per the policies of BioVU, use of these data fall under non-human subject determination and are approved by the Vanderbilt University IRB (#170502). We leveraged this resource to identify patients with a diagnosis of communicating hydrocephalus (Phecode: 331.1, https://phewascatalog.org; Denny et al., 2013; Wei et al., 2017) who have undergone permanent CSF diversion (VP shunt, ETV, or ETV/CPC). In BioVU, we identified patients of European ancestry (287 patients with hydrocephalus and 18,740 controls).

The median age at first CSF diversion operation (VP shunt, ETV/CPC, or ETV) was 4.1 years [0.46–11.13, interquartile range] and 58% of patients were male. Seventy-four patients (26%) had post-hemorrhagic etiologies of hydrocephalus, whereas 95 (33%) of patients were diagnosed with idiopathic (congenital) hydrocephalus. Eighty-five patients (29%) of patients were diagnosed with neural tube defects. The remaining 33 patients (12%) had other etiologies of hydrocephalus (post-infectious, brain tumor, Chiari Malformation Type I and Dandy-Walker syndrome). Control patients had no diagnosis of hydrocephalus or any other neurological or developmental disorder. Genomic ancestry was quantified using principal components analysis (Price et al., 2010). To avoid potential confounding due to population stratification (Derks et al., 2017), we performed our genetic analyses only on patients of European ancestry. We included 3 genotype-based principal components within the European-ancestry dataset as covariates in downstream analyses (Price et al., 2006).

METHOD DETAILS

Estimating the genetically determined expression

We implemented PrediXcan (Gamazon et al., 2015), a gene-based method that estimates the genetic component of expression using imputation models derived from the GTEx reference transcriptome panel in 10 neurological tissues (for a total of 889 brain samples), including frontal cortex and hypothalamus, and whole blood (from 338 individuals, Data S1) (Battle et al., 2017; Gamazon et al., 2018; Battle et al., 2017). PrediXcan utilizes a patient’s germline genetic profile to estimate the genetic component of gene expression in target tissues of interest. The weight (beta) from the imputation model and the number of effect alleles X at the variant j for individual i are used to infer the genetic component of gene expression for the ith patient: To identify genes associated with communicating hydrocephalus, we performed logistic regression with the genetically-determined expression as independent variable and disease status, with sex and age as covariates. Although the reference transcriptome panel (GTEx) consists of healthy controls, the genetically-determined expression was tested for association with disease status in the GWAS (BioVU) samples. PrediXcan seeks to test the effect on disease risk of the genetic component of gene expression (which is to be distinguished from the disease-altered component). One advantage of the gene-based test is statistical, i.e., the reduced multiple testing burden compared to conventional (SNP-based) GWAS (~5–10M statistical tests) (Gamazon et al., 2015). Another advantage is the ability to explicitly test a biologically-meaningful mechanism (gene expression regulation), that is known to be contributory to complex diseases (Gamazon et al., 2018; GTEx Consortium, 2015) versus genetic variants with mostly unknown gene targets. We emphasize that even if measured gene expression (RNA-seq) is available in the GWAS samples for differential expression analysis, this in no way negates the importance of estimating (and then testing for association with disease risk) just the genetic component of gene expression, as PrediXcan aims to do.

Odds ratios were calculated, as in case-control studies, and genes with nominally significant p < 0.05 are reported. Experiment-wide significance for a gene association was evaluated using Bonferroni correction for the total number of gene-tissue pairs tested (n = 9,868) for which expression imputation quality in a tissue (assessed in the reference GTEx panel) satisfied r > 0.01. False discovery rate (Benjamini-Hochberg) was set at 0.05 for each tissue.

Tissue specificity of gene expression was quantified using the tau statistic (Kryuchkova-Mostacci and Robinson-Rechavi, 2017) applied to the GTEx tissues: where Here, x provides the expression values for the gene , and is the number of tissues.

Genetically determined pathways and networks

Since the heritability of traits that have been interrogated through GWAS has been shown to be enriched for regulatory variation (Gamazon et al., 2018), we hypothesized that using the genetically determined component of gene expression, as quantified through PrediXcan, rather than the total expression (with potential environmental and technical confounding components) would enhance statistical power to identify pathophysiologically relevant pathways and networks. For each tissue type, we used as input the genes whose genetic component showed nominal association (p < 0.05) with hydrocephalus. Gene Set Enrichment Analysis (GSEA) (Mootha et al., 2003; Subramanian et al., 2005) was performed by determining enrichment of input genes in the curated gene sets in the Molecular Signatures Database (MSigDB; http://www.gsea-msigdb.org/gsea/msigdb/index.jsp). This list of input genes was compared against the MSigDB gene sets (C2: Curated Gene Sets, C5: Gene Ontology, C6: Oncogenic Signature, and C1: Chromosomal location). We used the default options in GSEA/MSigDB (which internally uses the hypergeometric distribution) to perform the enrichment analysis. Significance of enrichment was assessed using Benjamini-Hochberg adjusted p < 0.05 in each tissue. Experiment-wide significant gene sets were identified using Bonferroni adjustment (adjusted p < 0.05, which corresponded to p < 4.1×10−7) based on the total number of gene sets (4,762 curated sets; 5,917 GO sets; 326 chromosomal locations; and 189 oncogenic signatures; total count = 11,194) and the total number of tissues (n = 11).

BioVU analysis of hydrocephalus-related neurological traits

For our experiment-wide significant finding, we considered the PrediXcan associations in the exact discovery tissue (frontal cortex) with “Other cerebral degenerations” (Phecode: 331; 417 cases and 17,257 controls) and “Cerebral edema and compression of brain” (Phecode: 348.2; 635 cases and 17,257 controls) within BioVU to gain further mechanistic insights into the pathogenesis of hydrocephalus.

UK Biobank replication analysis

Integrated analysis of brain imaging and genomics data can facilitate validation and additional insights into the functional consequences of identified disease-associated genes. We performed PrediXcan analysis in the exact discovery tissue (frontal cortex) on CSF, white matter, and total brain volumes in the UK Biobank (N = 8,428) to validate our experiment-wide significant finding. From the T1 MRI structural image and normalized for overall head size, these brain imaging traits had been quantified (Elliott et al., 2018); we used these phenotypes in the PrediXcan validation analysis. The significance of replication was assessed using Bonferroni-adjusted p < 0.05 based on the total number of gene-tissue-phenotype tuples (1*1*3 = 3) tested. We also tested for enrichment of gene-level associations with these imaging phenotypes among the top differentially expressed genes identified in the frontal cortex (p < 0.05). In addition, because PrediXcan, the rare-variant exome scan, and a murine model of hydrocephalus appear to converge on TMEM50B (albeit only nominally), we interrogated the UK Biobank data for the association of the rare variant rs34327244 (A139T, MAF = 0.4%) within TMEM50B, with Bonferroni-adjusted p < 0.05 (adjusting for the 3 imaging phenotypes) as the replication significance criterion.

For additional support, we tested the SNPs within the MAEL cis region for association with hydrocephalus (Phenotype code = G6_HYDROCEPH) in a much larger UK Biobank dataset (N = 361,194, of which 133 are cases). We utilized public association results from the Benjamin Neale/HAIL team (Bycroft et al., 2018; Ge et al., 2017). Because of the much smaller number of cases than number of controls in this dataset, we followed a widely used variant quality control (QC) set of recommendations. Out of 10,629 SNPs in the region, we excluded all low-confidence variants, defined as follows: (a) MAF < 0.10%; (b) 2*(MAF)*(number of cases) < 25. We used Benjamini-Hochberg adjusted p < 0.05 as the cutoff for significance. We also tested the SNPs within the MAEL cis region (excluding low-confidence variants) for association with subarachnoid hemorrhage (N = 361,070, of which 124 are cases), a common cause of hydrocephalus (Germanwala et al., 2010; Graff-Radford et al., 1989).

Differential expression in mice

We leveraged publicly available choroid plexus expression data in mice (GEO: GSE37098), with and without hydrocephalus, on the C57BL/6J strain to replicate some of our top findings in the two tissues, i.e., frontal cortex and hypothalamus, with significant signals (Benjamini-Hochberg adjusted p < 0.05 within a tissue) in human patients. Only frontal cortex had an experiment-wide significant signal, Bonferroni-adjusted p < 0.05 across tested gene-tissue pairs. Gene expression had been previously quantified using Affymetrix Mouse Gene 1.0 ST Array (Robledo et al., 2008). We conducted differential expression analysis using an empirical Bayes method, as implemented in Linear Models for Microarray Data (limma), to identify gene expression changes associated with hydrocephalus in these mice. We report the genes that were differentially expressed (p < 0.05) in mice among our top genes (p < 0.05) in human frontal cortex and hypothalamus (evaluated separately).

Rare variant exome scan

The goal of this analysis was to identify associations between rare exonic variants in genes detected in at least 5 tissues by PrediXcan and hydrocephalus risk. We developed a cohort of 3,890 children and 25,823 adult patients of European ancestry who had previously undergone genotyping using Illumina Infinium Human Exome Bead Chip platforms. ICD9 codes for hospital billing were used to algorithmically define cases as well as controls, as previously described (Denny et al., 2013, 2010, Karnes et al., 2017, Wei et al., 2017). Fisher’s Exact test and Bonferroni correction were used to detect rare variant associations with hydrocephalus (Denny et al., 2013, 2010, Karnes et al., 2017, Simonti et al., 2016). We tested a total of 15 rare variants (i.e., MAF < 1%, including 1 in TMEM50B).

CSF proteomic data from patients with hydrocephalus

We analyzed a proteomic signature consisting of proteins that were found in a mass spectroscopic analysis of CSF (Morales et al., 2012). We evaluated the genetically determined expression of these genes in frontal cortex for their (PrediXcan) association with hydrocephalus to determine whether the proteins showed significant departure from the null. We then tested whether the signature had greater statistical significance (i.e., lower p value) than the remaining genes (using Mann-Whitney U test). For the latter, we performed the comparisons in frontal cortex (the significant discovery tissue) and whole blood (an easily accessible tissue) to further explore the tissue specificity of these results.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Software and algorithms
BioVU	Denny et al., 2010; Roden et al.,2008	https://victr.vanderbilt.edu/pub/biovu/
PrediXcan	Gamazon et al., 2015, 2018	https://github.com/hakyimlab/PrediXcan
GTEx	GTEx Consortium, 2013, 2015, 2020	https://gtexportal.org/home/
Gene Set Enrichment Analysis	Subramanian et al., 2005	http://www.gsea-msigdb.org/gsea/index.jsphttps://www.gsea-msigdb.org/gsea/index.jsp
Biological samples and deposited data
PrediXcan genetic associations for communicating hydrocephalus	This paper	https://victr.vanderbilt.edu/pub/biovu/
ChIP-seq data from human embryonic stem cells	Leung et al., 2015	Roadmap Epigenomics
Cerebrospinal fluid LC-MS dataset	This paper and Morales et al., 2012	https://pubmed.ncbi.nlm.nih.gov/22186713/
Differential gene expression in mouse choroid plexus	Robledo et al., 2008	Affymetrix Mouse Gene 1.0 ST Array (GEO: GSE37098)