Cell-type-specific alternative splicing in the brain and kidney of a Setbp1^S858R Schinzel–Giedion syndrome mouse

ABSTRACT

Schinzel–Giedion syndrome (SGS) is an ultra-rare Mendelian disorder caused by gain-of-function variants in the SETBP1 gene. Although previous studies determined multiple roles for SETBP1 and its associated pathways in disease manifestation, they did not assess whether cell-type-specific alternative splicing (AS) plays a role in SGS. We quantified gene and splice junction expression from single-nuclei RNA-sequencing data from the cerebral cortex and kidney of atypical Setbp1^S858R SGS patient variant and wild-type mice. We identified 33 and 62 genes with statistically significant alterations in splice junction usage in the brain and kidney, respectively. We identified significant splice junction usage in a member of the heterogeneous nuclear ribonucleoprotein family, Hnrnpa2b1. These findings were cell-type specific in the cerebral cortex and cell-type agnostic in the kidney, suggesting tissue specificity of AS in Setbp1^S858R mice. To broaden the impact of our results for the rare disease community, we developed a point-and-click web application that enables users to explore single-cell-resolution changes at the gene and splice junction levels. Overall, our findings implicate AS in a tissue- and cell-type-specific manner in the cerebral cortex and kidney of Setbp1^S858R mice.

Summary: Characterization of alternative splicing in a patient-derived mouse model of Schinzel–Giedion syndrome reveals altered cell-type- and tissue-specific splicing patterns in novel and previously described alternatively spliced genes.

INTRODUCTION

Schinzel–Giedion syndrome (SGS) is an ultra-rare autosomal dominant Mendelian disorder caused by variants in the SET binding protein 1 (SETBP1) gene (Hoischen et al., 2010). SETBP1 codes for a transcription factor expressed ubiquitously in the human body (Whitlock et al., 2024), and patients with SGS manifest clinical phenotypes related to the central nervous system, musculoskeletal system, heart and kidney/urinary tract (Duis and van Bon, 2024). Symptoms of SGS include global neurodevelopmental impairment, progressive neurodegeneration, mild-to-profound intellectual disability, treatment-resistant seizures, distinctive craniofacial structure, muscle hypotonia/spasticity, hydronephrosis and gastrointestinal problems (Schinzel and Giedion, 1978; Liu et al., 2018; Duis and van Bon, 2024). Gain-of-function variants located in (typical SGS) or near (milder atypical SGS) the 12-bp hotspot of the degron region of the SETBP1 protein prevent SETBP1 degradation by proteasomes (Acuna-Hidalgo et al., 2017). When the SETBP1 protein accumulates, it causes aberrant proliferation, deregulation of oncogenes and suppressors, unresolved DNA damage, apoptosis resistance (Banfi et al., 2021) and decreased histone pan-acetylation in neural progenitor cells (Zaghi et al., 2023). Models of SGS are critical for further resolving the molecular etiology and future therapeutics.

Recently, the heterozygous Setbp1^S858R patient variant mouse model (hereby referred to as Setbp1^S858R) was developed based on the S867R variant discovered in two atypical patients with SGS who experienced seizures, developmental delay and genital abnormalities (Acuna-Hidalgo et al., 2017). Setbp1^S858R mice have low female fertility, smaller stature, and reduced brain, liver and kidney organ weight compared to age- and sex-matched wild-type mice. With single-nuclei transcriptomics of the cerebral cortex and kidney in Setbp1^S858R mice compared to those of wild-type mice, we previously reported that although Setbp1 was only differentially expressed in excitatory neurons, many of its targets were differentially expressed and regulated in multiple brain and kidney cell types (Whitlock et al., 2023), suggesting a role for gene regulatory processes in SGS. Additionally, a recent study identified 38 alternatively spliced genes in the peripheral blood of a typical patient with SGS using bulk RNA sequencing (RNA-seq) (Liu et al., 2022), further implicating gene dysregulation in SGS. As an essential gene regulatory mechanism (Chen and Manley, 2009), alternative splicing (AS) increases organisms' functional genomic diversity by creating multiple isoforms/transcripts and proteins from a single gene through the joining of different combinations of exons from a single pre-mRNA. SETBP1 itself is known to undergo AS in humans (Dyer et al., 2025), and two SETBP1 targets, LSM2 and ZMAT2, are splicesome and pre-splicesome components, respectively (Montemayor et al., 2020; Tanis et al., 2018). As cell-type-specific AS is essential for neurodevelopment (Zhang et al., 2016), neurodevelopmental disorders (Patowary et al., 2024) and epilepsy (Huang et al., 2022), we hypothesized that AS was altered in the cerebral cortex and the kidney of Setbp1^S858R mice compared to those of wild-type mice.

Here, we re-processed our previously generated Setbp1^S858R mouse single-nuclei RNA-seq (snRNA-seq) data (Whitlock et al., 2023) to investigate the role of AS in SGS – an ultra-rare disease – and to create a valuable resource for the rare disease community and beyond. To quantify cell-type-specific AS, we used splice junction usage (SJU), which measures AS outputs (i.e. the relative isoform abundance produced from a single gene) from transcriptomic data by counting reads that span splice junctions (SJs). We calculated and compared cell-type-specific SJU in the cerebral cortex and the kidney of Setbp1^S858R and age-matched wild-type mice (Fig. 1A). This comparison identified cell-type-specific genes and their function and disease associations with AS between conditions (Fig. 1B-E). Although we did not identify cell-type-specific AS in Setbp1, we found other genes with significant AS, such as Hnrnpa2b1, a member of the heterogeneous ribonucleoprotein family. Our findings further implicate AS in SGS by identifying cell-type-specific and -shared genes and pathways altered in the cerebral cortex and the kidney of Setbp1^S858R mice. To allow researchers to explore our AS analysis results on the Setbp1^S858R snRNA-seq data, we developed a graphical user interface. This resource, available as a Shiny web application (Chang et al., 2026), allows researchers to easily explore our cerebral cortex and kidney AS results for themselves quickly and without programmatic knowledge.

Fig. 1.

Graphical abstract. (A) Schematic overview of our processing and analysis pipeline. (B) We analyzed pseudobulk gene expression and calculated splice junction usage (SJU) for each cell type and condition. (C) We compared SJU values for each cell type using a permutation test to identify cell-type-specific differences in alternative spicing (AS) between Setbp1^S858R and wild-type mouse tissues. (D) Next, we visualized all annotated transcripts and splice junction (SJ) locations for each significant SJU gene. (E) Finally, we compared the genes with significant SJU between cell types and annotated their functions and disease associations to predict their biological relevance. FC, fold change; OPC, oligodendrocyte precursor cell; snRNA-seq, single-nuclei RNA sequencing; UMAP, uniform manifold approximation and projection. Created in BioRender by Lasseigne, B. N. (2026). https://BioRender.com/j2x78qi. This figure was sublicensed under CC-BY 4.0 terms.

RESULTS

Cerebral cortex and kidney snRNA-seq from Setbp1^S858R mice shows cell-type-specific gene and SJ expression

We processed our previously generated (Whitlock et al., 2023) snRNA-seq dataset from the cerebral cortex and kidney of C57BL/6J-Setbp1^em2Lutzy/J mice heterozygous for Setbp1^S858R and matched wild-type mice, and quantified gene and SJ expression using STARsolo (Kaminow et al., 2021 preprint). We retained 51,318 cerebral cortex and 75,889 kidney nuclei that passed quality control filtering. From there, we annotated cerebral cortex cell types (Fig. 2A,B) using canonical marker gene expression to obtain seven cell types: astrocytes, excitatory neurons, inhibitory neurons, microglia, oligodendrocytes, oligodendrocyte precursor cells (OPCs) and vascular cells. As expected for cerebral cortex tissue, and in agreement with our previous publication (Whitlock et al., 2023), most of the nuclei we annotated were neurons, with the largest group being excitatory neurons (n=33,569; Fig. 2A). Similarly, we performed kidney cell-type annotation using canonical marker gene expression to identify 18 cell types (Fig. S1A,B). Brain and kidney cell types were evenly distributed across Setbp1^S858R and wild-type mice (Fig. 2C,D; Fig. S1C,D), underscoring the high quality of the dataset and sample processing consistency.

Fig. 2.

Cell types are evenly distributed across Setbp1^S858R and wild-type mice. (A) Representative UMAP colored by cell types annotated in our integrated dataset. (B) A dot plot of representative cell type markers for cell types shown in A. Size represents the percentage of cells expressing a given gene, and color saturation represents average expression. (C) Representative UMAP colored by condition. (D) Bar plot of cell-type proportions based on condition. Dark gray denotes Setbp1^S858R mice, and light gray denotes wild-type mice. n=6 mice (three wild type, three Setbp1^S858R).

After annotating cell types, we integrated SJ count information from STARsolo into our analyses using MARVEL, an R package for single-cell splicing analysis (Wen et al., 2023). We identified more SJs in cells with more expressed genes (linear regression, R²=0.8035). For example, excitatory and inhibitory neurons had a higher total number of genes expressed and SJs detected per cell than glial cell types in the cerebral cortex (Table S2, Fig. S2A,B). This is consistent with other research, as the fact that neurons express more genes than glia has previously been reported (Dopp et al., 2024). However, neurons still express more SJs per cell than glial cell types when dividing by the total number of genes expressed (Fig. S2C), suggesting that neuron-specific genes are more transcriptionally complex. This finding corroborates a recent study that identified distinct neuron-specific AS complexity across multiple RNA-binding proteins and in genes involved in synapse, cell projection and ion channel activity between neuronal subpopulations using deep single-cell RNA-seq of the mouse cortex (Feng et al., 2021). Additional work in murine brains has highlighted the differential splicing of neurexin genes in pre- and post-synaptic neurons (Huntley et al., 2020). Finally, previous findings indicate that neuron-specific synaptic genes are longer and have more transcript isoforms (Koopmans et al., 2019). In contrast, the number of SJs per cell, normalized to the number of genes per cell across kidney cell types, is uniformly distributed in the kidney compared to in the cerebral cortex cell types (Fig. S3). The gene we detected with the most SJs in the cerebral cortex was Syne1, which had 156 different SJs in excitatory neurons (Table S2). SYNE1 encodes spectrin repeat-containing nuclear envelope protein 1, or nesprin-1, variants of which can cause multiple Mendelian disorders, including cerebellar ataxia, Emery–Dreifuss muscular dystrophy and arthrogryposis multiplex congenita (Indelicato et al., 2019). Our results corroborate previous findings that Syne1 has over 100 exons, resulting in dozens of transcript variations (Dyer et al., 2025). We saw cell-type-specific differences in the number of detected SJs for Syne1: we detected 156 SJs in excitatory neurons (the most), but only 25 SJs in microglia (the least). In the kidney, either Syne1 or Syne2, a prognostic marker for clear cell renal cell carcinoma (Pontén et al., 2008), had the most SJs in all 18 cell types (Table S3). This highlights the cell-type specificity of SJ expression and, thus, AS in our single-nuclei dataset.

Setbp1^S858R mice have cell-type-specific splicing in all measured cerebral cortex and most kidney cell types, but not in Setbp1

To determine cell-type-specific SJU changes in Setbp1^S858R compared to wild-type mice, we performed a permutation analysis using the MARVEL (Wen et al., 2023) R package for each cell type between conditions. We detected 33 and 62 genes with significant changes in SJU in the cerebral cortex and kidney, respectively (Table S4; permutation test, P<0.05 and delta >1; see ‘Differential SJ and gene expression analysis with MARVEL’ section for details). The cerebral cortex cell type with the most genes with significant SJU was astrocytes (n=6, permutation test, P<0.05), while the cerebral cortex cell type with the least genes was OPCs (n=2) (Fig. 3A). In contrast to in the cerebral cortex, we did not observe any cell-type-specific alternatively spliced genes in three of 18 kidney cell types [collecting duct principal cells (CDPCs), thin ascending limb of the loop of Henle (LOH) cells and dendritic cells; Fig. 3B]. T cells had the most genes with significant SJU (n=6), and six cell types only had one gene with significant SJU (Fig. 3B). Interestingly, almost 82% (27 of 33 genes) of significant SJU genes in the cerebral cortex were cell-type specific, whereas only 53% (33 of 62 genes) of significant SJU genes in the kidney were cell-type specific, in Setbp1^S858R compared to wild-type mice (Fig. 3). Additionally, we found 13 genes with significant SJU that overlapped between the cerebral cortex and the kidney in our Setbp1^S858R mice (Table S4). Of 13 cross-tissue significant SJU genes in our data, seven encode RNA-binding proteins and other splicing regulators, including Pnisr, Hnrnpa2b1, Srrm2, Zcchc7, Son, Mbnl1 and Srek1 (Table S4). Although Hnrnpa2b1, which encodes an RNA-binding protein, was cell-type specific in the cerebral cortex, its AS pattern was broader in the kidney, and Son, which encodes a splicing co-factor with significant changes in SJU for all non-vascular cell types in the cerebral cortex, was cell-type agnostic in both the cerebral cortex and kidney (Table S4), highlighting the cell-type- and tissue-specific diversity of AS patterns in Setbp1^S858R compared to wild-type mice. Consistent with our previous work (Whitlock et al., 2023), Setbp1 was expressed in all cerebral cortex and kidney cell types. However, we did not observe significantly different Setbp1 gene expression, splice junction expression (SJE) or SJU in Setbp1^S858R compared to wild-type mice (Figs S4 and S5). While Setbp1 was not significantly differentially expressed or spliced, our previous work suggests that the S858R variant in Setbp1, which encodes a transcription factor, affects downstream gene regulation, as evidenced by the significant transcriptional changes observed in multiple Setbp1 targets (Whitlock et al., 2023). Therefore, the lack of AS in Setbp1 in our study is unsurprising, as the Setbp1 S858R protein disrupts Setbp1 degradation, allowing it to continue accumulating in the absence of gene expression and regulatory changes (Whitlock et al., 2023). Although we observed cerebral cortex-specific changes in expression and usage (i.e. decrease in inhibitory neurons and vascular cells, and increase in excitatory neurons and oligodendrocytes) of SJ-3, which incorporates the Setbp^1S858R variant located on exon 4, these differences were not statistically significant during permutation testing (Fig. S4C). Additionally, only one Setbp1 isoform has been identified in mice (Setbp1-201; Fig. S4A). Though we did not identify significant changes in AS of Setbp1, our findings indicate altered splicing patterns in Setbp1^S858R mice with varying degrees of cell-type specificity across the cerebral cortex and the kidney.

Fig. 3.

Setbp1^S858R mice have cell-type-specific splicing in all cerebral cortex and most kidney cell types compared to wild-type mice. (A,B) UpSet plot of genes with significant SJU changes between Setbp1^S858R and wild-type mice, split by cell type in cerebral cortex (A) and kidney (B). Brighter colors indicate higher overlap between cell types, with purple showing genes unique to each cell type. Bar graphs on the far right represent the total set size for each cell type, while the top bar graphs denote the intersection size. CDPC, collecting duct principal cells; CDIC, collecting duct intercalated cells; DCT, distal convoluted tubule; LOH, loop of Henle; OPC, oligodendrocyte precursor cell; PCT, proximal convoluted tubule; PST, proximal straight tubule. n=6 mice (three wild type, three Setbp1^S858R). The cerebral cortex and the kidney were from the same mouse.

AS of the heterogeneous nuclear ribonucleoprotein family A2/B1 gene (Hnrnpa2b1) is cell-type and tissue specific in Setbp1^S858R compared to wild-type mice

Because SGS is an ultra-rare Mendelian disorder with fewer than 100 reported cases worldwide (Duis and van Bon, 2024), we sought to compare our genes with significant SJU to previously identified alternatively spliced genes from a bulk RNA-seq study in the peripheral blood of a typical patient with SGS compared to their unaffected parents (Liu et al., 2022). None of our predicted SJU genes from the cerebral cortex or the kidney of this atypical SGS mouse model directly overlapped with the 37 orthologs of the 38 alternatively spliced genes identified in human peripheral blood. However, we predicted significant SJU in the cerebral cortex and the kidney of Hnrnpa2b1, encoding an RNA-binding protein that is closely related to the heterogeneous nuclear ribonucleoprotein A1 (Hnrnpa1) (He and Smith, 2009) identified as an alternatively spliced gene in the peripheral bulk RNA-seq blood study (Liu et al., 2022). Even though the functions of Hnrnpa1 and Hnrnpa2b1 proteins are not identical, the members of the heterogeneous nuclear ribonucleoprotein family act compensatorily in disease (Lu et al., 2022), show structural homology (Mayeda et al., 1994) and play essential roles in neurodevelopment (Tilliole et al., 2024). Previous protein–protein interaction (PPI) data compiled in StringDB indicate a strong, high-confidence interaction between the Hnrnpa1 and Hnrnpa2b1 proteins (StringDB confidence score, 0.982; Fig. S6). Additionally, StringDB data suggest that Setbp1 interacts with both heterogeneous nuclear ribonucleoproteins through its PPI with Set (StringDB confidence score, 0.791; Fig. S6), which is known to be bound by Setbp1 (Wang et al., 2023). SETBP1 protein accumulation causes DNA damage in neural progenitor cells derived from patients with SGS, as accumulated SETBP1 binds and stabilizes SET, thereby blocking P53 signaling and disrupting cell proliferation and DNA repair (Banfi et al., 2021). In our snRNA-seq data, we detected 13 transcripts of Hnrnpa2b1 (Fig. 4A; Fig. S7A) with 17 and 18 SJs in the cerebral cortex and the kidney, respectively (Fig. 4C; Fig. S7C). Significant SJU in Hnrnpa2b1 was specific to inhibitory neurons in the cerebral cortex (Fig. 4C, bottom; Table S4) and shared across multiple cell types in the kidney (Fig. S7C, bottom; Table S4) in Setbp1^S858R compared to wild-type mice. Interestingly, normalized Hnrnpa2b1 gene expression was highest in astrocytes and microglia in our dataset, rather than in inhibitory neurons, in the cerebral cortex (Fig. 4B). In contrast, Hnrnpa2b1 had significant AS in 12 of 18 kidney cell types (i.e. CDPCs, connecting tubule cells, distal convoluted tubule cells, dendritic cells, proximal convoluted tubule cells, podocytes, proximal tubule cells, proximal straight tubule cells, T cells, thick ascending limb of the LOH, thin ascending limb of the LOH, thin descending limb of the LOH; Fig. S7C), indicating a tissue-specific AS pattern in Setbp1^S858R mice. Overall, our findings of SJU in Hnrnpa2b1 suggest that AS of genes encoding heterogeneous nuclear ribonucleoproteins are altered across species (human and mouse) and tissues (cerebral cortex, kidney, peripheral blood) and have a cell-type-specific pattern in the cerebral cortex in SGS.

Fig. 4.

Hnrnpa2b1 has inhibitory neuron-specific AS changes in the cerebral cortex of Setbp1^S858R compared to wild-type mice. (A) Transcript models of all 13 annotated transcripts of Hnrnpa2b1. The color indicates transcript classification: indigo, transcripts flagged for nonsense-mediated decay (NMD); dark teal, protein-coding transcripts; turquoise, protein-coding transcripts, but coding sequence (CDS) is not defined. Arrows indicate the direction of transcription. (B) Split violin plots showing Hnrnpa2b1 normalized gene expression per cell for all cell types, split by condition. Darker shades indicate Setbp1^S858R mice, and lighter shades indicate wild-type mice. (C) Heatmaps of the changes in normalized mean SJ expression (top) and usage (bottom) between Setbp1^S858R and wild-type mice for 17 SJs of Hnrnpa2b1. The top heatmap annotation indicates cell type. A positive delta indicates that expression or usage was higher in Setbp1^S858R mice than in wild-type mice, and a negative delta indicates that expression or usage was higher in wild-type mice than in Setbp1^S858R mice. The asterisk indicates the significant SJU in inhibitory neurons. n=6 mice (three wild type, three Setbp1^S858R).

Development of a web application for visualizing gene expression, SJE and SJU in our SGS model snRNA-seq data

Finally, to expand the accessibility of our analyses to researchers focused on ultra-rare diseases and the broader scientific community, we built an R Shiny (Chang et al., 2026) web application featuring the data presented here. This application is particularly valuable for the ultra-rare disease community, such as those studying SGS, for which data are often scarce, and patient numbers are limited. This application offers several key benefits and uses:

1. Custom visualization: users can create custom gene expression and SJU uniform manifold approximation and projection (UMAPs) (Fig. 5A-C) for any chosen gene in C57BL/6J wild-type and Setbp1^S858R mouse cerebral cortex and kidney. This flexibility enables researchers to visualize specific genes of interest using pre-processed, normalized and scaled gene expression values, as well as pre-calculated SJUs.

2. Exploration of significant genes: researchers can further investigate the 33 cerebral cortex and 62 kidney genes we identified with significant SJU and generate alternatively spliced gene summary plots (as shown in Fig. 4).

3. Comparative analysis: the resource supports exploration within and across conditions (wild type and Setbp1^S858R), cell types, tissues, genes and SJs. Users can quickly search for changes in SJ expression and usage between wild type and Setbp1^S858R.

4. Gene isoform information: the application provides detailed information on gene isoforms, including genomic locations and annotated regions [e.g. nonsense-mediated decay (NMD), protein coding and retained introns]. This feature helps researchers understand the functional implications of different isoforms.

5. Data download: researchers can easily download plots from their custom analyses, making it convenient to use these visualizations for preliminary data analysis, hypothesis generation or as supporting evidence in their own research.

6. Additional resources: the web application includes helpful links to external resources (e.g. genome browsers) and a frequently asked questions (FAQ) section to assist users in navigating the tool and understanding the data.

Fig. 5.

Our web application enables users to access and visualize gene expression and SJU data for the cerebral cortex and the kidney. (A-C) Representative screenshots of our web application. The ‘Gene Expression and SJU UMAPs’ tab allows users to examine and download the UMAPs annotated by cell types (A), gene expression (B) and/or SJU (C) of any gene(s) of interest in our dataset in the cerebral cortex and/or kidney.

Our Shiny web application serves as an excellent resource for the scientific community, facilitating the exploration and analysis of our Setbp1^S858R snRNA-seq dataset. The application is publicly available at https://lasseignelab.shinyapps.io/setbp1_as, providing a user-friendly platform for researchers to explore the data and generate new insights for their own research.

DISCUSSION

In this study, we analyzed cell-type-specific gene and SJ expression and usage from cerebral cortex and kidney single-nuclei profiles of Setbp1^S858R mice compared to age- and sex-matched control mice. We identified SJU differences in all cell types in the cerebral cortex and the kidney. We identified 33 and 62 genes with significant changes in SJU between the Setbp1^S858R and wild-type mice in the cerebral cortex and kidney, respectively. We found that 82% of significant SJUs were cell-type specific in the cerebral cortex, whereas only 53% of significant SJUs in the kidney were specific to a cell type. Additionally, 36 of 84 total significant SJU genes across tissues have already been shown to be associated with various diseases, cancer and autism spectrum disorder (Table S4). Given the emphasis of previous research in the field, specifically on neural progenitor cells and neurons (Banfi et al., 2021; Zaghi et al., 2023; Cardo et al., 2023), this study builds on our prior work (Whitlock et al., 2023), underscoring that the function of SETBP1 as an epigenetic hub (Piazza et al., 2018) leads to cell-type-specific signatures in atypical SGS cell types. For example, although we reported SJU changes in all cerebral cortex cell types we measured, the excitatory and inhibitory neurons had the most significant SJUs, while astrocytes and oligodendrocytes had the most cell-type-specific SJU genes in Setbp1^S858R compared to wild-type mice. Even though Setbp1 was broadly expressed across all cerebral cortex and kidney cell types, we did not identify significant SJU in any cell type across tissues, indicating that the S858R variant does not lead to changes in AS in Setbp1 itself. Although the changes did not reach significance, we observed alterations in SJE and SJU for all cerebral cortex and kidney cell types across multiple SJs of Setbp1, which may be due to the limitations of short-read RNA-seq data for AS analyses and the current knowledge of Setbp1 isoforms, as only one is currently annotated in the mouse. Interestingly, we observed the greatest SJU increase in SJ-3, the junction spanning exon 3 to the variant-impacted exon 4 of Setbp1, in vascular cells – a cell type not previously implicated in SGS. Our results further implicate that disease-causing patient variants in SETBP1 induce disease-associated molecular programs (here, differential SJU) in many cerebral cortex and kidney cell types, which are critical for understanding disease pathogenesis. As the brain has the most complex splicing profile of all tissues (Xu et al., 2002), and splicing is essential for healthy neurodevelopment (Sanders et al., 2020), this finding is not surprising. However, to our knowledge, this is the first report of AS in the cerebral cortex and the kidney of a SETBP1-associated model.

Furthermore, we predicted significant SJU in the cerebral cortex and the kidney of the Hnrnpa2b1 gene, encoding an RNA-binding protein of the heterogeneous nuclear ribonucleoprotein class. This class comprises a large family of RNA-binding proteins that contribute to AS, mRNA stabilization, and transcriptional and translational regulation (Geuens et al., 2016). Additionally, these proteins act compensatorily in disease, exhibiting high amino acid homology (Lu et al., 2022; Mayeda et al., 1994). Although genes encoding heterogeneous nuclear ribonucleoproteins are ubiquitously expressed, genetic variants that alter their sequence lead to various neurodevelopmental and neurodegenerative diseases (Tilliole et al., 2024). A human peripheral blood study from an SGS proband compared to their unaffected parents identified 38 alternatively spliced genes, including Hnrnpa1, the gene encoding a member of the heterogeneous nuclear ribonucleoprotein class. None of our genes with significant SJU in the cerebral cortex or the kidney directly overlapped with the 37 orthologs of the 38 alternatively spliced genes identified in the human peripheral blood SGS study (Liu et al., 2022), which may be due to differences in species (human versus mouse), sequencing technologies (bulk versus snRNA-seq), variant locations (typical versus atypical SGS) and tissue type (peripheral blood versus cerebral cortex and kidney). However, we identified significant SJU in Hnrnpa2b1, another member of the heterogeneous nuclear ribonucleoprotein class, closely related to Hnrnpa1 (He and Smith, 2009). Intriguingly, we identified significant cell-type-specific SJU of Hnrnpa2b1 in the cerebral cortex. Despite inhibitory neurons possessing the second-lowest gene expression of Hnrnpa2b1 in the cerebral cortex cell types, they were the only cell type with significant SJU in this gene. Neurons, being non-dividing cells, require tight regulation of mRNA homeostasis, indicating their vulnerability to dysfunction of RNA-binding proteins (Geuens et al., 2016). Our findings further suggest that AS in members of the heterogeneous nuclear ribonucleoprotein class, such as Hnrnpa2b1, may play a role in neurodegenerative phenotype in SGS. Interestingly, we also identified significant SJU of Hnrnpa2b1 in the kidney of our Setbp1^S858R mice, although we observed AS in the kidney across multiple cell types, indicating tissue-specific AS effects in this protein family. Altogether, this suggests that AS in members of the heterogeneous nuclear ribonucleoprotein class plays a crucial role in SGS, both at the cellular and tissue levels.

To overcome challenges for researchers to quickly explore our data and analyses for themselves, we created a user-friendly point-and-click Shiny web application. Our web application can be used for custom visualization, gene exploration, comparative analysis (across conditions, tissues, cell types, genes and SJs) and detailed isoform annotations. This resource includes the scaled and normalized gene counts for pre-processed Setbp1^S858R and wild-type cerebral cortex and kidney samples for seven and 18 cell types, respectively. It also features pre-calculated SJU across more than 15,000 genes, easy plot downloads and reference links for gene searches. By integrating these tools into rare disease projects, we can make these datasets more accessible to the broader community. This enhanced accessibility allows researchers and clinicians to interact with and understand complex genetic data more easily, fostering collaboration and accelerating discoveries in rare disease research.

Although our study provides additional insight into dysregulated regulatory mechanisms, specifically AS through SJU, there are several limitations to the current study. First, single-cell/nuclei and SJ read count data are sparse by nature. Although we attempted to counteract this with high sequencing depth (∼100,000 reads per nucleus), our SJU values were low compared to those in other publications that used the same computational framework (Wen et al., 2023). Additionally, these prior works used cells, rather than nuclei, which may contribute to the differences we observed. Second, while useful, mouse models are not perfect substitutes for patient profiles because of species differences. For example, Setbp1 has only one annotated transcript in mice, but there are seven annotated protein-coding transcripts in humans (Ensembl, GRCh38.p14) (Dyer et al., 2025) and two known protein isoforms in UniProt (UniProt Consortium, 2025). We did not detect significant SJU changes for Setbp1 in the Setbp1^S858R model, but that does not preclude changes in SETBP1 isoform expression at a different developmental stage or in patients. Finally, we generated sequencing profiles with the Illumina platform using short-read sequencing. Although splicing can be measured in short-read RNA-seq by split reads [i.e. reads spanning SJs (García-Ruiz et al., 2025)], this approach does not typically measure full-length transcripts because the average mammalian transcript length is 2-3 kbp. For example, an increase in SJE or reads between exons 2 and 4 of Son would suggest reduced expression of exon 3 in Setbp1^S858R mouse cerebral cortex. This would help to identify potential isoforms expressed, but cannot be used to detect full-length transcript or isoform expression or usage. Therefore, our analysis was limited to SJ expression and SJU, rather than full-length transcript expression and usage, which may have generated false positives and/or missed splicing events.

To our knowledge, this work is the first to identify cell-type-specific AS patterns resulting from SGS patient variants in Setbp1in the cerebral cortex and kidney. Future studies incorporating additional SGS models and leveraging new technologies, such as long-read single-cell profiling capable of capturing full-length transcripts in an individual cell, are critical for further establishing the potential impact of AS in SGS. Additionally, while there are no documented sex differences in SGS, which is an ultra-rare Mendelian disorder, AS is known to change between sexes (Trabzuni et al., 2013; Blekhman et al., 2010; Karlebach et al., 2020 preprint; Sosnowski et al., 1989; Han et al., 2022), and male and female SGS splicing profiles could be differentially affected by the accumulation of SETBP1. Therefore, changes in additional time points and sexes should be explored. Additionally, although outside the scope of this study, future research is needed to determine how splicing may be impacted in a cell-specific manner, particularly to investigate the effects of differences in Hnrnpa2b1 SJU on disease pathogenesis in SETBP1-associated diseases. In conclusion, our findings further implicate the impact of Setbp1 patient variants on a diverse range of cell types and molecular mechanisms in the cerebral cortex and the kidney, underscoring their importance in contributing to SGS.

MATERIALS AND METHODS

Data acquisition

We obtained raw snRNA-seq data of six male mouse brain cerebral cortex and kidney samples: three 6-week-old C57BL/6J-Setbp1^em2Lutzy/J mice heterozygous for Setbp1^S858R (RRID:IMSR_JAX:033235) and three healthy, C57BL/6J age-matched wild-type mice (RRID:IMSR_JAX:000664) from our recent publication (Whitlock et al., 2023), available at Gene Expression Omnibus (GEO) under accession number GSE237816.

Data processing

We built a conda environment using Anaconda3 version 2023.07-2 to process raw sequencing data and provided the parameters to build this environment in an environment.yml file (available at https://github.com/lasseignelab/230926_EJ_Setbp1_AlternativeSplicing/blob/main/bin/conda/environment.yml). In that conda environment, we used STAR version 2.7.10b (Kaminow et al., 2021 preprint) to build a STAR genome reference from the GENCODE M31 primary assembly. Next, we ran STARsolo version 2.7.10b (Kaminow et al., 2021 preprint) on each sample with the following recommended options to best replicate 10x Genomics' CellRanger's filtering protocol and achieve the most recovered cell barcode similarity with our previous work: --soloType CB_UMI_Simple, --soloFeatures GeneFull_Ex50pAS SJ, --soloCellFilter EmptyDrops_CR, --soloUMIlen 12, --clipAdapterType CellRanger4, --outFilterScoreMin 30, --soloCBmatchWLtype 1MM_multi_Nbase_pseudocounts, --soloUMIfiltering MultiGeneUMI_CR, --soloUMIdedup 1MM_CR.

Data quality control and filtering

We removed ambient RNA from our gene counts using SoupX (Young and Behjati, 2020) using the raw matrix, barcodes and feature files generated by STAR Solo. Because SJ counts already exhibit very high sparsity, we did not repurpose SoupX to apply it to SJ counts, thereby avoiding the exacerbation of their sparsity. We used the filtered barcodes, features and matrix TSV files as input for preprocessing in Seurat. We used the Seurat version 5.0.0 R package for quality control, filtering and clustering analyses (Hao et al., 2023) in the cerebral cortex and the kidney. We imported the gzipped and filtered STARsolo output matrices (barcodes, features and matrix) into R using the Seurat Read10x function (Hao et al., 2023). We created a Seurat object using the CreateSeuratObject function for each sample and condition (Setbp1^S858R and wild type) before merging them into a single Seurat object per condition. We observed that all cells were equally distributed across cell-cycle phases using Seurat's MergeLayers and CellCycleScoring functions and converting default human gene IDs to mouse IDs using bioMart version 2.56.1. We filtered at the cell level (i.e. mitochondrial ratio<0.05, between 1000 and 15,000 genes per cell). We also removed Malat1 from the cerebral cortex because it is frequently over-detected with poly-A capture technologies such as 10x (Clarke and Bader, 2024 preprint). Then, using default parameters, we performed batch correction using Harmony version 1.1.0 (Korsunsky et al., 2019) to preserve biological variation while reducing variation due to technical noise (Luecken et al., 2022). We scaled and normalized expression data using the ScaleData and NormalizeData functions with default settings (i.e. a scale factor equal to 10,000 and natural-log normalization). We then performed principal component analysis (PCA) using the RunPCA function from Seurat (Hao et al., 2023) without approximation (approx=FALSE) in order to improve reproducibility. We plotted UMAPs and cell-type proportions to confirm successful integration across conditions for cerebral cortex and kidney samples.

Clustering and cell-type assignment

We also used the Seurat R package (version 5.0.0) (Hao et al., 2023) for cell-type assignment. We used a clustering resolution of 0.75, identifying 32 clusters in the cerebral cortex and a resolution of 1.25, identifying 50 clusters in the kidney using the Leiden option, which relies upon leidenalg v0.10.1 (Traag et al., 2019). We identified differentially expressed marker genes for each cluster [using the FindAllMarkers from Seurat (Hao et al., 2023)] with a log-fold change threshold >0.2 and a Bonferroni-adjusted P-value <0.05. We assigned cell types using differential expression of cell-type-specific genes identified through PanglaoDB (Franzén et al., 2019), CellMarker 2.0 (Hu et al., 2023) and the Allen Brain Cell Atlas (Table S1). We visually examined the expression of these canonical cell-type markers by making feature and dot plots. Owing to non-specific cell markers and low cell numbers, we combined pericytes, vascular leptomeningeal and endothelial cells into one vascular cell type in the cerebral cortex.

SJE normalization

To normalize SJE for data visualization, we employed a method similar to Seurat's NormalizeData function, which divides feature counts by the total counts for that cell, multiplying by a scale factor and applying a natural-log transformation. Here, we divided SJ counts by the total SJ counts for that cell, multiplied by a scale factor of 1000 (determined by the average number of SJs expressed per cell in our dataset), and finally used the base R function log1p to natural-log transform the resulting values.

SJU calculation

To calculate SJU, we divided the total SJ count for a cell type for each SJ by the total read count for that gene. As SJU is a percentage, we converted SJU values over 100 to a maximum of 100 (which only occurred for low-count genes) and any infinite values (indicating no gene expression) to Not a Number (NaN).

Differential SJ and gene expression analysis with MARVEL

We used MARVEL version 2.0.5 (Wen et al., 2023), an R package with multiple analytical tools for investigating single-cell resolution SJU, to integrate SJ and gene expression information into a single R object per tissue. We filtered the data to include only SJs within a single gene annotated in GENCODE release M31. We imported the filtered gene expression data processed with Seurat and SJE data processed with STARsolo (Kaminow et al., 2021 preprint). Then, we filtered the SJ data for cells that passed the Seurat quality control metrics described above and combined them into a single sparse count matrix. Finally, using the combined sparse matrix, we created cerebral cortex and kidney MARVEL objects using the CreateMarvelObject.10x function.

We also used the MARVEL package for differential SJ and gene expression analyses. For differential SJU, we employed MARVEL's CompareValues.SJ.10x function, which applies a permutation test on the SJU of individual SJs for a given cell group, as transcripts of SJs are rare, creating extreme sparsity, which cannot be overcome with a non-parametric rank-based test (Wen et al., 2023). We compared conditions (Setbp1^S858R and wild-type mice) for each cell type across tissues. We used a P-value cutoff of 0.05 (meaning that, out of 100, the permutation of random condition assignment would randomly have five or fewer absolute values of the delta greater than actual delta values) and an absolute value of delta SJU of greater than one. We used at least 5% of cells expressing a gene or SJ as gene and SJ expression cutoffs.

To identify genes that are differentially expressed and have SJU, we used MARVEL's CompareValues.Genes.10x function. This function employs a Wilcoxon rank sum test on normalized log2-transformed gene expression values.

Significant SJU disease annotation

We observed a total of 84 unique genes with significant SJU in the cerebral cortex and the kidney, mapping to 74 human orthologs. To annotate genes with significant differential SJU for disease associations, we compared COSMIC cancer gene census (version 100), SFARI Gene (2025 Q3 release) and GenCC (accessed 250811) to our genes following mouse to human ortholog mapping (gprofiler2 version 0.2.3). The COSMIC cancer gene census consists of expert-curated cancer-driving genes, tumors associated with cancer-driver genes, and assignment of genes to Tier 1 (strong functional and mutational support) or Tier 2 (strong support either functionally or mutationally, but not both) (Sondka et al., 2018). Of note, none of our genes with significant SJU in either organ were associated with germline tumors in COSMIC, so we only reported somatic tumor associations. SFARI Gene includes autism-associated genes, with gene scoring for the strength of evidence supporting associations (Banerjee-Basu and Packer, 2010; Abrahams et al., 2013). Finally, GenCC is an international gene curation coalition with submitters from diagnostic laboratories and established online resources (e.g. Gene2Phenotype, Online Mendelian Inheritance in Man, Orphanet, ClinGen) (DiStefano et al., 2022).

Transcript structure and SJ visualization

We used the R package ggtranscript version 0.99.9 to visualize SJs and the GENCODE release M31 gtf to annotate known transcripts of genes with significant SJU. We labeled SJs with the letters ‘SJ’ and a number (e.g. ‘-1’) indicating the SJ's genomic location on a transcript. If a gene is transcribed in the forward direction, we assign the SJ numbers in ascending order and vice versa. This way, we labeled SJs consistently from the 5′ to 3′ direction of translation.

PPI network investigation

We accessed the StringDB (Version 12.0) graphical user interface at https://string-db.org/ to identify and visualize known interactions between the transcription factor Setbp1 and the RNA-binding proteins Hnrnpa1 and Hnrnpa2b1. We included all three proteins in our query under the ‘Multiple proteins’ tab and selected Mus musculus as the organism of interest. To investigate the interactions between the proteins we input, we chose to include ten additional nodes in the PPI by selecting the ‘more’ button twice (total nodes, 13). In the ‘Settings’ tab, we opted to have edge thickness represent the strength of the data supporting the interactions. The confidence scores reported in the paper were extracted by selecting the respective edges, which delineate the combined StringDB score and the evidence that supports it.

Web application development

We built our web application using the R Shiny framework version 1.8.1.1. It is hosted with shinyapps.io at https://lasseignelab.shinyapps.io/setbp1_as.

Reproducibility

For all the analyses following data processing, we used R version 4.3.1 through RStudio version 2023.06.2+561 running in docker containers through Singularity version 3.5.2 on the UAB supercomputer, Cheaha. Docker images for each analysis are specified in each script and available at Docker (https://hub.docker.com/r/emmafjones/setbp1_alternative_splicing and https://hub.docker.com/r/tsoelter/setbp1_alternative_splicing) and Zenodo (https://zenodo.org/records/12534825). All code was independently reproduced by two other laboratory members and is available at GitHub (https://github.com/lasseignelab/230926_EJ_Setbp1_AlternativeSplicing) and Zenodo (https://zenodo.org/records/17601458).