Germline pathogenic variants in HNRNPU are associated with alterations in blood methylome

HNRNPU encodes a multifunctional RNA-binding protein that plays critical roles in regulating pre-mRNA splicing, mRNA stability, and translation. Aberrant expression and dysregulation of HNRNPU have been implicated in various human diseases, including cancers and neurological disorders. We applied a next generation sequencing based assay (EPIC-NGS) to investigate genome-wide methylation profiling for >2 M CpGs for 7 individuals with a neurodevelopmental disorder associated with HNRNPU germline pathogenic loss-of-function variants. Compared to healthy individuals, 227 HNRNPU-associated differentially methylated positions were detected. Both hyper- and hypomethylation alterations were identified but the former predominated. The identification of a methylation episignature for HNRNPU-associated neurodevelopmental disorder (NDD) implicates HNPRNPU-related chromatin alterations in the aetiopathogenesis of this disorder and suggests that episignature profiling should have clinical utility as a predictor for the pathogenicity of HNRNPU variants of uncertain significance. The detection of a methylation episignaure for HNRNPU-associated NDD is consistent with a recent report of a methylation episignature for HNRNPK-associated NDD.


INTRODUCTION
Advances in genomics have resulted in increasingly large numbers of genes being identified as causing neurodevelopmental disorders (NDDs) [1,2]. HNRNPU encodes a component of a multiprotein complex that binds heterogeneous nuclear RNA and scaffold-attached DNA [3,4]. Other members (n = 32) of the large heterogeneous nuclear ribonucleoprotein family that have been implicated in human disease include HNRNPH1, HNRNPH2, HNRNPK, HNRNPR and SYNCRIP [5]. Following suggests that inactivation of HNRPNU might contribute to the neurological and neurodevelopmental phenotype of 1q34q44 microdeletion syndrome [6,7] as de novo mutations in HNRNPU were reported in rare cases of epileptic encephalopathy [8,9]. Subsequently, the phenotype associated with pathogenic variants in HNRNPU was extended to include early-onset seizures, severe intellectual disability, speech impairment, hypotonia, microcephaly and ventriculomegaly [10][11][12][13]. Dysmorphic features (high arched eyebrows, long palpebral fissures, overhanging columella, widely spaced teeth and thin upper lip) have also been described [11,12].
Interpreting the potential pathogenicity of variants of uncertain significance (VUSs) remains a major challenge in many areas clinical genetics, including the diagnosis of NDDs [14][15][16]. A major cause of NDDs are variants in chromatin modifying genes (CMGs) (e.g., histone lysine methyltransferases or histone acetylases etc.) and for many of these disorders, evidence of disordered epigenetic regulation can be detected through alterations of DNA methylation patterns (episignatures) in peripheral blood [2,17]. The identification of CMG-associated NDD specific episignatures can be used to aid variant interpretation and suggest candidate CMGs in unsolved NDDs [15,[17][18][19][20][21][22]. In addition to a role in posttranscriptional RNA processing, HNRNPU (also known as scaffold attachment factor A (SAF-A)) is also reported to have roles in gene transcription, maintenance of higher-order chromatin structure and X-inactivation via Xist [23][24][25]. Recently, a methylation episignature was described for Au-Kline syndrome, a NDD associated with germline mutations in HNRNPK [16]. In the light of this finding, we investigated whether HNRNPU-related NDD was associated with a methylation episignature.

SUBJECTS AND METHODS
We performed genome-wide methylation profiling of >2 M CpGs with a targeted next generation sequencing assay (Illumina TruSeq ® Methyl Capture EPIC NGS) as described previously [17]. Written informed consent was obtained for all participants and the study was approved by South Birmingham Research Ethics Committee.
Genomic DNA with HNRNPU pathogenic mutations (n = 7) were extracted from whole blood by standard methods. Bisulfite conversion, library preparation, target enrichment and sequencing (Illumina NextSeq 2000) were performed at the Cambridge University Department of Medical Genetics Stratified Medicine Core Laboratory (SMCL) as described previously [17]. Raw methylation beta-values were extracted by RnBeads R package (https://rnbeads.org). Data pre-processing and bioinformatics analysis, and detection and visualisation of methylation episignatures were performed according to our standard procedure (see Lee et al. [17,26]). If a significant batch effect (age, gender, batch-based) was detected, the target variables were adjusted by surrogate variable analysis (SVA) using the sva package. The p-value of differentially methylated sites was determined either by a two-sided Welch test or by a linear model employed in the limma package, and the combined p-values (for CpG islands) were determined by Fisher's method. During the process, neighbouring CpGs combined together and assigned as 'DMB (differentially methylated blocks).' DMBs were combined based on their functional similarity. Only DMBs (including CpG Islands) with a p-value lower than 0.01 and a methylation difference between controls and diseases group of more than 20% were considered significant for genome-wide CpG site methylation analysis. A summary of the sequencing coverage and sequencing reads is included in Supplementary Table 1.

RESULTS
Clinical and genetic features of HNRNPU patient cohort The seven individuals studied had been diagnosed with a HNRNPU-NDD after the identification of a germline HNRNPU variant (see Table 1). All HNRNPU variants were assessed as likely pathogenic or pathogenic and were predicted to have a loss-of-function effect (6 were predicted, in the absence of nonsense-mediated mRNA decay, to result in a truncated gene product and one patient was predicted to have PBRM1 haploinsufficiency as a result of a de novo deletion of exons 1-11) ( Table 1). The positions of the truncating variants are plotted on the HNRNPU protein in Fig. 1.
The frequency of the clinical features displayed by the 7 individuals with a HNRNPU variant is summarised in Table 1. The overall frequency of clinical features such as seizures, developmental delay, intellectual disability, and hypotonia (Table 2) in the study cohort was similar to that in a previously reported series [27] of 17 patients with HNRNPU-NDD (1 patient from the current cohort were also represented in this previous series). Table 3 provides detailed overview of clinical characteristics of current (n = 7) cohort with an update on those patients previously published in Taylor et al. and Durkin et al. [12,27].
Inspection of hypermethylation/hypomethylation profiles in individual cases (see Fig. 1) showed some variability in the extent of methylation alterations, but there was no obvious relationship apparent between this variability and the type of variant or, for truncating variants, the position of the predicted effect on the HNRNPU gene product.

DISCUSSION
We found evidence of methylation alterations in blood DNA from patients with HNRNPU-associated NDD and this, to our knowledge, is the first report of a methylation episignature for HNRNPU inactivation by a NGS-based assay. Recently, Rooney et al. [32] published DNA methylation signatures for 9 pathogenic/likely pathogenic and 1 VUS HNRNPU variants using data from Illumina EPIC methylation array assay which interrogates fewer CpGs.
The role of members of the heterogeneous nuclear ribonucleoproteins in NDDs (HNRNPH1, HNRNPH2, HNRNPK, HNRNPR and HNRNPU) and cancer (HNRNPA1, HNRNPA2B1, HNRNPC, HNRNPD, HNRNPF, HNRNPK, HNRNPR, and HNRNPU) has been the subject of recent investigations but the relationships between the individual function of disease-associated HNRNPs and the mechanisms of To eliminate potential biases introduced by normalised data, a PCA clustering analysis was performed based on preprocessed beta-values. The results demonstrated that 227 DMPs were able to effectively differentiate the HNRNPU group from the control group. c Scatter plots were generated by comparing the methylation beta-values of individuals with the mean values of the healthy control group, using a confidence interval of ±3 standard deviations (3 SD). A significant pattern of the number of DMPs with gain of methylation (GOM) compared to loss of methylation (LOM) DMPs was detected (17.6% hypermethylated, 2.08% hypomethylated). relevant disorders is not well defined [5,33]. Our finding of a disordered epigenetic state in peripheral blood DNA from patients with pathogenic variants in HNRNPU is consistent with the report of Choufani et al. [16] who described a methylation episignature for HNRNPK-associated NDD in 9 individuals. Whereas Choufani et al. [16] used a methylation bead array platform targeting 850,000 CpGs for methylation profiling, we used a NGS-based assay targeting >2 M CpGs. The difference in methodology and analytical approaches used by Choufani et al. [16] and us limits the detailed comparison of the methylation episignatures from the two conditions. Whereas Choufani et al. [16] identified 429 statistically significant CpG DMPs in their AKS discovery cohort (n = 6) using a false discovery rate adjusted p-value of 0.05 and a minimum methylation difference of 10%, we employed a p-value of less than 0.01 and a more stringent minimum methylation difference of 20% and identified 227 DMPs in our HNRNPU-NDD cohort (n = 7). However, no overlapping DMBs were found between these 227 DMPs and the 429 CpGs identified from the EPIC array in the HNRNPK-NDD cohort (though in the HNRNPKassociated episignature of the 429 CpGs identified using the EPIC array, only 178 of these CpGs were present within the target regions of the EPIC-NGS analysis). We note that both in our cohort and in the findings from the HNRNPK-AKS cohort studied by Choufani et al. [16], significantly altered DMP events comprised both hypermethylation and hypomethylation alterations (see Fig. 1). However, whereas in our HNRNPU cohort, DMPs showed predominantly hypermethylated DMPs (12/16 DMBs), the episignatures from HNRNPK-AKS from Choufani et. al. indicated more hypomethylated DMPs than hypermethylated DMPs [16].
The differences in methylation profiling platforms between our investigations and those of Rooney et al. [32] limit a direct detailed comparison of the respective results but there are similarities in the overall episignature patterns with both hyper and hypomethylation alterations. For example, whilst we identified 16 differentially methylated blocks (DMBs) (including 10 CpG islands) among 7 HNRNPU patients (12 of which were hypermethylated and 4 were hypomethylated), Rooney et al. [32] identified 18 differentially methylated regions (including 12 CpG islands) with 12 being hypermethylated and 4 hypomethylated. We have previously used EPIC-NGS methodology to identify methylation episignatures in a range of chromatin disorders (e.g., Kabuki syndrome Type 1, KMT2B-DYS28, Luscan-Lumish syndrome (SETD2) and Rabin-Pappas syndrome (SETD2) from healthy controls [17,26], however a much wider range of NDDs have been studied by methylation array profiling and Rooney et al. [32] compared DNAm patterns in their HNRNPU cohort to 56 other NDDs and identified most overlap between the differentially methylated positions within the episignatures for HNPNPU with those for velocardiofacial syndrome and BAFopathy cohorts.
The analysis of methylation episignatures in chromatin disorders can often inform likely pathogenicity of variants of uncertain significance (VUSs). Candidate pathogenic HNRNPU missense variants are rare [27] and in our cohort all of the patients had a pathogenic loss of function HNRNPU variants, so we were not able to formally confirm the utility of DNAm testing for variant interpretation [1]. Nevertheless, the extent of the significant DMPs for our HNRNPU cohort suggest that episignature profiling will have clinical utility as a predictor for clarifying pathogenicity of HNRNPU VUSs (as described previously for HNRNPK variants [16]). In cases of a suspected chromatin disorder in which a VUS is predicted to be non-pathogenic episignature analysis may suggest another diagnosis or suggest the presence of an undetected pathogenic variant [1]. Thus the differential diagnosis of HNRNPK-NDD includes Kabuki syndrome and comparison of the methylation signatures for these two disorders would enable them to be distinguished by methylome analysis [16]. Indeed we note that in their recent paper Rooney et al. [32] reported (a) a HNRNPU in frame deletion (p.(Glu279del)) which did not demonstrate the HNRNPU episignature and (b) an undiagnosed patient who did demonstrate a HNRNPU DNAm profile suggestive of HNRNPU NDD and was then subsequently found to harbour a candidate pathogenic variant in HNRNPU (c.1720_1722delAAG p.(Lys574del)).
The main differential diagnosis of HNRNPU-related NDD can be wide due to a number of causes associated with developmental impairment-epileptic encephalopathy (DEE). However, the common differentials include Rett and Angelman syndromes [12] Although Rett syndrome is caused by mutations in the methyl-CpG-binding protein 2 (MECP2) it is not associated with a DNA methylation signature and though methylation changes occur in a subset of Angelman syndrome patients, these are generally limited to the imprinted SNURF:TSS-DMR at chromosome 15q11q13 [33,34]. Therefore, the presence of the relevant DNAm episignature in a child with a clinical suspicion of HNRNPU-NDD would be consistent with this diagnosis rather than any other conditions causing DEE including Rett or Angelman syndrome.
HNRNPU is abundantly expressed in the developing mouse brain and biallelic loss of HNRNPU function was associated with cortical cell death in a genetically-engineered mouse model [35]. Prominent features of HNRNPU-NDD include developmental delay, epileptiform seizures, speech and language impairment and behavioural alterations (e.g., autistic features or aggressiveness). Abnormal brain imaging is common (but the range of anomalies is variable) and cardiac and renal structural defects also occur. Transcriptomic studies in the brains of homozygous and heterozygous HNRNPU-deficient mouse models demonstrated widespread effects on gene expression, particularly in the homozygote mice affecting multiple signalling pathways including synaptogenesis, neuroinflammation and (cell cycle control [36]. Evidence for disordered RNA splicing (a known role of HNRNPU) was detected in HNRNPU mutant mice brain cortex [31]. Though RNA splicing is critical for brain development, our findings suggest that the pathogenesis of HNRNPU-NDD might also be related to disordered epigenetic regulation of gene expression. Epigenomic and transcriptomic analysis of HNRNPU mutant mice might provide further insights into potential disease mechanisms.
Finally, it has been noted previously that rare, apparently healthy, individuals with HNRNPU truncating variants may be found in the gnomAD data set (https://gnomad.broadinstitute.org) [12]. This might reflect a lack of detailed phenotypic information or variability of phenotypic expression. However, methylation episignature analysis of such individuals might provide novel insights into genotype-epigenotype-phenotype relationships.