Rare deleterious mutations of HNRNP genes result in shared neurodevelopmental disorders

Abstract

Background: With the increasing number of genomic sequencing studies, hundreds of genes have been implicated in neurodevelopmental disorders (NDDs). The rate of gene discovery far outpaces our understanding of genotype-phenotype correlations, with clinical characterization remaining a bottleneck for understanding NDDs. Most disease-associated Mendelian genes are members of gene families, and we hypothesize that those with related molecular function share clinical presentations.

Methods: We tested our hypothesis by considering gene families that have multiple members with an enrichment of de novo variants among NDDs, as determined by previous meta-analyses. One of these gene families is the heterogeneous nuclear ribonucleoproteins (hnRNPs), which has 33 members, five of which have been recently identified as NDD genes (HNRNPK, HNRNPU, HNRNPH1, HNRNPH2, and HNRNPR) and two of which have significant enrichment in our previous meta-analysis of probands with NDDs (HNRNPU and SYNCRIP). Utilizing protein homology, mutation analyses, gene expression analyses, and phenotypic characterization, we provide evidence for variation in 12 HNRNP genes as candidates for NDDs. Seven are potentially novel while the remaining genes in the family likely do not significantly contribute to NDD risk.

Results: We report 119 new NDD cases (64 de novo variants) through sequencing and international collaborations and combined with published clinical case reports. We consider 235 cases with gene-disruptive single-nucleotide variants or indels and 15 cases with small copy number variants. Three hnRNP-encoding genes reach nominal or exome-wide significance for de novo variant enrichment, while nine are candidates for pathogenic mutations. Comparison of HNRNP gene expression shows a pattern consistent with a role in cerebral cortical development with enriched expression among radial glial progenitors. Clinical assessment of probands (n = 188-221) expands the phenotypes associated with HNRNP rare variants, and phenotypes associated with variation in the HNRNP genes distinguishes them as a subgroup of NDDs.

Conclusions: Overall, our novel approach of exploiting gene families in NDDs identifies new HNRNP-related disorders, expands the phenotypes of known HNRNP-related disorders, strongly implicates disruption of the hnRNPs as a whole in NDDs, and supports that NDD subtypes likely have shared molecular pathogenesis. To date, this is the first study to identify novel genetic disorders based on the presence of disorders in related genes. We also perform the first phenotypic analyses focusing on related genes. Finally, we show that radial glial expression of these genes is likely critical during neurodevelopment. This is important for diagnostics, as well as developing strategies to best study these genes for the development of therapeutics.

Keywords: Cortex development; Gene families; Neurodevelopmental disorders; hnRNPs.

Fig. 1

Study workflow. Candidate NDD HNRNPs were determined from the literature and publicly available information (such as amino acid sequences) and from identification of probands in our novel cohorts. The candidate NDD HNRNPs were finalized by considering only genes in which at least three probands were identified from published and/or novel sources. Functional impacts focused on these finalized NDD HNRNP candidates and included pathogenicity predictions (gnomAD and GEVIR), de novo enrichment analyses (using the Chimpanzee–Human [CH] model and denovolyzeR), missense analyses (using CLUMP and MetaDome), expression analyses of fetal cortex and adult tissues, and phenotypic analyses within HNRNPs, across HNRNPs, and in comparison to other similarly presenting disorders by HPO terms. CNV: copy number variant; pLI: loss-of-function intolerance; LGD: likely gene disrupting; NMD: nonsense mediated decay; HPO: human phenotype ontology

Fig. 2

Protein similarity of hnRNPs. Correlation plot of hnRNPs by canonical amino acid sequence. Pearson correlation values are shown in the bottom half of the plot and are shown visually on the top half of the plot

Fig. 3

Pathogenicity assessment of variation in hnRNPs. pLI and Z-scores were obtained from gnomAD. pLI scores are significantly higher among NDD hnRNPs (n = 12) compared to non-NDD hnRNPs (n = 20), suggesting LGD variants are more likely to be damaging. Z-scores trend towards being significantly higher for NDD hnRNPs, suggesting severe missense variants are likely to be damaging. T-test with Welch’s correction. *p < 0.05. pLI: loss-of-function intolerance

Fig. 4

De novo enrichment and clustering of missense variation analyses of NDD hnRNPs. a De novo variation was assessed for NDD HNRNPs using two statistical models: the CH model and denovolyzeR. Right/above the dotted line indicates the gene achieves exome-wide significance (q < 4.24 × 10^− 7) while right/above the dashed line indicates the gene reaches nominal significance (q < 0.05). HNRNPU reaches exome-wide significance for all protein-impacting variants (Protein) and LGD variants, with severe missense variants reaching significance by only the CH model. SYNCRIP reaches exome-wide significance for LGD variants and all protein-impacting variants by the CH model alone. HNRNPD reaches nominal significance by the CH model. P values are FDR corrected with the number of genes (n = 18,946 for CH model and n = 19,618 for denovolyzeR) with three tests per gene (LGD, missense, and all protein changes) and two tests (CH model and denovolyzeR) per mutation type. Only cohorts with known de novo status were included, as listed in Tables S1 and S7. Statistics can be seen in Table S6. b Analysis of clustered missense variants. Clustering of missense variants was analyzed using CLUMP; scores are shown in Table S6 (paired t-test). Compared to the non-neuropsychiatric subset of gnomAD (n = 114,704, 1958 missense variants), the CLUMP score for NDD hnRNPs (red) among probands is significantly lower than controls in gnomAD (black), indicating more clustering of mutations (shown by arrow). Note that only genes with variants in the current cohort could undergo this analysis. hnRNPH2, hnRNPK, hnRNPR, and hnRNPUL1 each have independent significant clumping compared to gnomAD controls. c CLUMP scores for missense variants in probands with ASD (n = 60). HnRNPH2, hnRNPK, hnRNPR, and hnRNPUL1 reach significance independently. *p < 0.05, **p < 0.01, ***p < 0.001. LGD: likely gene disrupting; Missense: severe missense (CADD ≥ 20); Protein: all protein-affecting variants

Fig. 5

hnRNP proband variants. Protein structure, known binding motif, number of probands by mutation type, location of variants in each protein, and known associated disorders of NDD hnRNPs are shown. Novel cases are above the protein with published cases below. Red indicates LGD variants and blue represents severe missense variants. RRM: RNA recognition motif; qRRM: quasi-RNA recognition motif; KH: K-homology domain; RGG: Arginine-glycine rich (RGG) box; NLS: nuclear localization sequence. Further details of each variant are shown in Table S7. Adapted from Geuens et al. [83]. a Gene reaches exome-wide significance for all protein-impacting variants by CH model. b Gene reaches exome-wide significance for all protein-impacting variants and LGD variants by CH model. c Gene reaches exome-wide significance for all protein-impacting variants and LGD variants by CH model and denovolyzeR. d Gene reaches significance for missense variant clustering

Fig. 6

HNRNP expression in adult and developing fetal cortex tissues. a Heatmap showing transcript-level expression values for NDD hnRNPs for adult brain tissues in GTEx. All tissues are shown in Fig. S4, and p values among individual HNRNPs are shown in Table S3. b Heatmap showing fold change of expression of each NDD HNRNP among 48 different cell types in the developing fetal cortex. Blue indicates increase in fold change of expression and red indicates decreased expression, as determined by Z-scores. NDD HNRNPs have higher fold expression change in MGE progenitors, radial glia, and excitatory neurons, while depleted in inhibitory neurons. All HNRNPs are shown in Fig. S3. Significance indicates enrichment in particular cell types by Wilcoxon ranked sum test with Bonferroni correction based on number of cell types. P values and fold change for scRNA data from developing human cortex can be seen in Tables S4 and S5. c Correlation plot of developing fetal cortex gene expression. Pearson correlation R values are shown in the bottom half of the plot, which are visually in the top half of the plot. P values were corrected by number of genes [23] and number of cell types [48]. HNRNPs in the same homology group tend to have more correlated expression. d Specific brain region enrichment as determined by SEA, showing enrichment of expression of the NDD HNRNPs in the early fetal striatum and early-mid fetal amygdala. *p < 0.05; **p < 0.01; ***p < 0.001, ****p < 0.0001

Fig. 7

Phenotypic information of 189–221 hnRNP-variation probands. a Correlation matrix of phenotypes across hnRNP probands. Genes are in order of protein similarity as determined by Clustal Omega and canonical protein sequences as in Fig. 1. Phenotypes correlate across all HNRNPs, except HNRNPF due to sample size. Size and shade of circle represent correlation coefficients, which are shown on bottom half of matrix. Correlations for LGD and missense variants separately are in Fig. S4. P values, which are corrected by number of genes [23] and phenotypes (88, occurring in at least 20% of any HNRNP group) can be seen in Table S9. b Plot comparing protein and phenotype correlations that are over Pearson’s R = 0.5. Colors are the same as in Fig. 2 protein groups. Those with more similar protein sequences tend to be more phenotypically similar. c Plot of phenotypes of all probands by mutation type. Individual HNRNPs can be seen in Fig. S5. d Heatmap indicating percent of probands with phenotype. Sample sizes can be seen in Table S7 and range from n = 2 (HNRNPF) to n = 83 (HNRNPU). Lines indicate significant differences as determined by pairwise Fisher’s exact tests with Bonferroni correction based on 12 genes, 88 phenotypes, and three mutational categories. Red dashed lines indicate significance with only LGD variants. Raw p values can be seen in Tables S8. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. LGD: likely gene disrupting; MIS: severe missense (CADD ≥ 20)

Fig. 8

Relationship of HNRNP-related disorders to each other and to similarly presenting disorders. a Comparison of average number of HPO terms shared within HNRNPs versus with other similarly presenting disorders, as determined by PhenPath. The HNRNP-related disorders present more similarly to each other than other NDDs. HPO terms are in Table S12. b Heatmap showing fold change of expression of each NDD HNRNP and similarly presenting NDD based on HPO terms among 48 different cell types. Blue indicates increase in fold change of expression and red indicates decreased expression, as determined by Z-scores. c Correlation plot of developing fetal cortex gene expression for HNRNPs and genes implicated in similarly presenting disorders. Pearson correlation R values shown visually, with darker and larger circles indicating higher Pearson R values. HNRNPs are noted with a red line. P values were corrected by number of genes [28] and number of cell types [48]. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. d Comparison of expression Pearson’s R values within HNRNPs and compared to similarly presenting disorders. The expression of the HNRNPs is more similar to each other than to other NDDs