HNRNPH2-R114W remodels RNA splicing and neurodevelopmental gene programs in isogenic cerebral organoids

A single letter change in a protein that shepherds RNA through the splicing machinery may be enough to rewire thousands of genetic instructions in the developing human brain. In a study published in Frontiers in Bioengineering and Biotechnology, researchers modeled an ultra-rare neurodevelopmental disorder in cerebral organoids grown from a patient’s own cells, then compared them against genetically matched organoids in which the mutation had been corrected. The comparison revealed a sweeping, yet organized, reorganization of RNA splicing and neurodevelopmental gene programs, offering one of the most detailed molecular portraits to date of how a single RNA-binding protein variant can disturb early cortical development.

The variant in question, HNRNPH2-R114W, affects a gene on the X chromosome that encodes a member of the heterogeneous nuclear ribonucleoprotein family. These proteins bind G-rich RNA motifs and help decide where precursor messenger RNA is cut and rejoined during splicing, the process that generates the diverse protein-coding transcripts a cell needs. Pathogenic variants in HNRNPH2 cause Bain type X-linked syndromic intellectual developmental disorder, a condition marked by severe developmental delay, absence of verbal communication, impaired ambulation, seizures, and prominent autistic features. The individual whose cells seeded this study carries a de novo heterozygous R114W missense variant and exhibits precisely this severe clinical profile.

To study the variant in a controlled setting, the team reprogrammed the patient’s peripheral blood mononuclear cells into induced pluripotent stem cells using non-integrating episomal vectors. One stem cell line, designated Minus, retained the R114W variant. A second line, designated Plus, was generated by correcting the HNRNPH2 locus with CRISPR-Cas9 within the same genetic background, and Sanger sequencing confirmed the correction-associated sequence changes in edited clones. Both lines maintained strong nuclear expression of the pluripotency markers POU5F1 and NANOG, and both were differentiated into cerebral organoids following a modified Lancaster protocol that included embryoid body formation, neural induction, Matrigel embedding, and suspension maturation.

Imaging told a reassuring story about gross development. Whole-mount immunostaining showed PAX6-positive progenitors arranged in radial, ventricular-like zones around central lumens, with TUBB3-positive neurons localizing to peripheral regions in both genotypes. Organoids from the variant and corrected lines were comparable in size and morphology, with no observable structural abnormalities. HNRNPH2 immunofluorescence, however, revealed genotype-dependent differences: corrected organoids showed signal concentrated in neuroepithelial regions, while variant organoids exhibited weaker and more diffuse staining. The authors interpret this as an indirect indicator of altered HNRNPH2 distribution rather than definitive evidence of reduced protein abundance, since the assay relied on immunofluorescence alone without orthogonal protein quantification.

The transcriptomic analysis, though exploratory in design, produced striking results. Bulk RNA sequencing was performed on pooled RNA from approximately twelve organoids per genotype, each pool sequenced in two technical reactions at a single developmental time point. Differential expression analysis with DESeq2 showed near-complete separation of the two genotypes along the first principal component, which accounted for 99.5 percent of the variance. Thousands of genes were differentially expressed at a false discovery rate of 0.05, with roughly equal numbers upregulated and downregulated. Genes downregulated in variant organoids were enriched for neuron differentiation, axon development, and synaptic transmission, while upregulated genes pointed toward WNT and NOTCH signaling, melanogenesis, tyrosine metabolism, and retinoid biosynthesis.

The splicing findings were even more dramatic. Using SGSeq to quantify both annotated and novel splice junctions supported by split reads, the researchers identified 6,310 differentially used junctions out of roughly 98,000 examined, meaning about six to seven percent of all junctions were significantly altered. By contrast, the exon-focused workflow, which used featureCounts and voom-limma diffSplice testing, found only 17 significant exons. This junction-dominant pattern is biologically coherent: HNRNP H-family proteins regulate splice-site selection at G-rich motifs, so a variant that disrupts their RNA engagement would be expected to reorganize local donor-acceptor usage rather than simply delete exons. Corrected organoids showed increased junction usage in genes such as TUBA1A, TUBB3, CNTN2, and CKB, while variant organoids showed higher usage in DCT, PMEL, ALDH1A3, RDH10, CALB1, and XIST.

Computational structural modeling provided a mechanistic hypothesis for how the variant acts. The researchers built a composite model of HNRNPH2’s three quasi-RNA recognition motifs from available crystal structures and docked it relative to the spliceosomal B complex. In this arrangement, residue R114 sits near a predicted RNA-contact interface within the RRM2 domain, positioned in a geometry compatible with direct RNA interactions. The R114W substitution, replacing a positively charged arginine with a bulky tryptophan, would be predicted to alter this interface and change how the protein engages its RNA targets. The authors emphasize that this docking framework is hypothesis-generating and requires extended simulations and direct structural validation, for example through cryo-electron microscopy.

To organize the multi-layered RNA signals, the team developed an integrated prioritization framework that combined differential expression, event-level splicing significance, and isoform-level remodeling into a composite score per gene. This analysis identified regulatory hubs enriched for axon guidance and synaptic signaling, including CNTN2 and NCAM1, which are critical for cortical wiring, along with TRPM3, HES5, HES6, RTN4, and HMGCS1, which participate in neuronal excitability, neurogenesis, and metabolic support. Some genes, such as RUFY3, EFS, SPAG9, CRMP1, HNRNPC, UTRN, and EVL, showed significant splicing changes without detectable expression shifts, indicating post-transcriptional regulation of neuronal projection and cytoskeletal pathways. Others, including CDH6 and PTPRS, exhibited combined transcriptional and exon-usage alterations.

A key question in organoid transcriptomics is whether apparent gene expression changes reflect genuine regulatory effects or merely shifts in cell-type composition. To address this, the researchers applied MuSiC deconvolution against the human neural organoid cell atlas, a single-cell reference of neurodevelopment. Both genotypes contained mixtures of neuroepithelium, dorsal and ventral telencephalic neurons, glioblasts, oligodendrocyte progenitors, and astrocytes. Variant organoids showed modest enrichment of neuroepithelial and glial-associated fractions and a neuronal fraction approximately five percent lower than corrected organoids. Curated marker heatmaps supported this inference, with progenitor markers such as PAX6 and SOX2 and astrocyte markers GFAP, S100B, and AQP4 showing genotype-linked patterns. The authors conclude that these modest compositional shifts are unlikely to account for the extent of transcriptional and splicing alterations observed, suggesting the variant biases developmental timing toward progenitor and glial states rather than causing neuronal loss.

The findings support a model in which HNRNPH2 variants disrupt post-transcriptional regulation through altered protein localization, splicing, and ribonucleoprotein assembly, producing toxic or dominant-negative effects rather than simple loss of function, with limited compensation by the paralog HNRNPH1. Beyond clarifying disease mechanism, the study nominates candidate isoform-level endpoints for therapeutic investigation, including assessment of allele-selective antisense oligonucleotide strategies and HNRNPH2 modulation in human cortical organoids. The authors also propose that their integrative methodology, combining differential expression, multi-layer splicing analysis, and single-cell-based deconvolution, could be applied to other RNA-binding protein-driven neurodevelopmental disorders, opening avenues to examine how HNRNPH2-associated RNA dysregulation evolves across developmental time and recurs across independent biological contexts.

Subject of Research: HNRNPH2-R114W remodels RNA splicing and neurodevelopmental gene programs in isogenic cerebral organoids

Article Title: HNRNPH2-R114W remodels RNA splicing and neurodevelopmental gene programs in isogenic cerebral organoids

Article References: Caiaffa, C. D., Ghousifam, N., Watkins, S. L., Nakaya, H. I., Bowling, R. A., Jr., Finnell, R. H., & Cabrera, R. M. (2026). HNRNPH2-R114W remodels RNA splicing and neurodevelopmental gene programs in isogenic cerebral organoids. Frontiers in Bioengineering and Biotechnology, 14, Article 1922305. https://doi.org/10.3389/fbioe.2026.1922305

Image Credits: AI Generated

DOI: 10.3389/fbioe.2026.1922305

Keywords: HNRNPH2-R114W, remodels, splicing, neurodevelopmental, gene, programs, isogenic, cerebral, organoids, scientific research