Rbm3 deficiency leads to transcriptome-wide splicing alterations

ABSTRACT

Rbm3 (RNA-binding motif protein 3) is a stress responsive gene, which maintains cellular homeostasis and promotes survival upon various harmful cellular stimuli. Rbm3 protein shows conserved structural and molecular similarities to heterogeneous nuclear ribonucleoproteins (hnRNPs), which regulate all steps of the mRNA metabolism. Growing evidence is pointing towards a broader role of Rbm3 in various steps of gene expression. Here, we demonstrate that Rbm3 deficiency is linked to transcriptome-wide pre-mRNA splicing alterations, which can be reversed through Rbm3 co-expression from a cDNA. Using an MS2 tethering assay, we show that Rbm3 regulates splice site selection similar to other hnRNP proteins when recruited between two competing 5$\mathrm{\prime }$ splice sites. Furthermore, we show that the N-terminal part of Rbm3 encompassing the RNA recognition motif (RRM), is sufficient to elicit changes in splice site selection. On the basis of these findings, we propose a novel, undescribed function of Rbm3 in RNA splicing that contributes to the preservation of transcriptome integrity.

Introduction

Rbm3 is a small (17 kDa) RNA-binding protein that was initially shown to increase in expression in response to cold stimulation and hypothermia [1,2] and to enhance the translation of pro-survival target mRNAs [3]. In this context, it was found that Rbm3 mitigates neuronal cell death upon hypothermia treatment [4] and therefore might be considered to be a promising new target for the therapy of neurodegenerative diseases such as Alzheimer’s and Parkinson’s disease [5]. Subsequent studies indicated that Rbm3 is not only induced in response to cold but also upon various other stress stimuli including hypoxia [6], serum deprivation [7], UV and γ-irradiation [8,9] and various toxins and drugs [10], suggesting a more general function of Rbm3 in cellular stress response. Consistently, Rbm3 was described to regulate apoptosis [4], cell cycle control [11], and cancer development [12]. However, the molecular mechanisms underlying the pro-survival activity of Rbm3 are still not fully understood.

Rbm3 is a highly conserved protein in mammals that structurally resembles heterogeneous nuclear ribonucleoproteins (hnRNPs) and hnRNP-like proteins (Figure 1A), a large family of RNA-binding proteins which are important regulators of RNA metabolism. HnRNPs share a common modular gene structure consisting of one or more RNA-binding domains (RBDs) – usually RNA recognition motifs (RRMs) or K-homology (KH) domains – and a glycine-rich domain (GRD) that promotes or represses pre-mRNA splicing [13]. Pre-mRNA splicing is a process in which non-coding sequences (introns) are removed from the nascent RNA transcript, and protein-coding sequences (exons) are joined together. As the vast majority of eukaryotic genes are interrupted by introns, splicing is critical for correct RNA maturation. This process is mediated by a multiprotein/RNA complex called the spliceosome, which consists of five major subunits: the U1, U2, U4, U5 and U6 small nuclear ribonucleoprotein particles (U snRNPs), along with hundreds of dynamically associating proteins [14]. These subunits include RNA binding proteins, such as hnRNPs, which can influence splicing by either promoting or antagonizing spliceosome-binding to conserved sequences at the exon-intron boundaries, which are also referred to as splice sites (SS). Consequently, transcriptome-wide splicing defects caused by hnRNP malfunctions are frequently associated with severe pathological phenotypes ranging from neurodegenerative diseases to cancer [15–17]. Based on the structural similarity of Rbm3 with hnRNP proteins, a predominant localization of Rbm3 within the nucleus [7] and the predicted interaction of Rbm3 with splicing complexes (STRING database [18]), it is tempting to speculate that Rbm3 has a role in pre-mRNA splicing.

Figure 1.

Rbm3 deficiency alters splicing landscape genome-wide. (A) homology between the Glycine-rich domain (GRD) and RNA recognition motif (RRM) of Rbm3 and the two closest hnRnps: hnRNP A0 and Rbmx. (B) graphical representation and quantification of significant as events and affected genes detected by rMATS. SE: skipped exon; A5SS: alternative 5$\mathrm{\prime }$splice site; A3SS: alternative 3$\mathrm{\prime }$splice site; MXE: mutually exclusive exons; RI: retained intron. (C) number of overlapping genes in each as event category. (D) heat map showing the altered splicing landscape of AS genes. Red colour represents increased, while blue colour represents decreased sequencing read inclusion/exclusion ratios (calculated from rMATS Ψ- values). Each column represents a biological replicate. For example, in the SE event box, negative (blue) Ψ- values represent lower exon inclusion frequency in KO cells, indicating that exon skipping is frequent in Rbm3^(def) cells. (E) gene ontology analysis showing biological processes significantly enriched with AS genes.

Here, we demonstrate that Rbm3 deficiency causes transcriptome-wide pre-mRNA splicing alterations and compromised transcriptome integrity, which can be reversed by transient Rbm3 expression from a cDNA. We show that multi-exon genes with a high expression level are particularly sensitive to Rbm3 loss. We demonstrate that artificial tethering of Rbm3 to reporter mRNAs impacts alternative splicing decisions and that the isolated RNA recognition motif of Rbm3 is sufficient for its splicing activity. Taken together, our findings suggest a novel role for the RNA binding protein Rbm3 as a splicing regulator.

Results

Rbm3 is a ubiquitously expressed hnRNP-like gene

As the cell-type-specific expression profile of Rbm3 has only been investigated in a limited number of studies, we aimed to further investigate its expression across various cell populations and tissues to understand its potential function. To achieve this, we explored the Tabula Muris single-cell RNA-seq compendium [19] and found that Rbm3 is expressed in virtually all identified mouse tissues and in most cell lines, even in the absence of any stress stimulus, suggesting that Rbm3 – as other hnRNPs – may have a fundamental cellular role (Supplemental Fig. S1). HnRNP proteins are ubiquitously expressed and well-documented regulators of pre-mRNA splicing and other aspects of RNA metabolism. To better understand Rbm3’s potential physiological steady-state functions, we performed homology searches and identified hnRNP proteins to show the closest sequence similarity to Rbm3 in addition to its well-described homologue Cirbp. Rbm3 has a modular organization with an N-terminal RRM domain and a C-terminal part, which is enriched in glycine (GRD). Both Rbm3 domains displayed significant sequence similarity to individual members of the hnRNP protein family (Figure 1A). While the GRD of Rbm3 was found to have similarity to the GRD of members of the hnRNP A/B protein subfamily (comprising A0, A1, A2/B1 and A3), the RRM domain shares considerable sequence identity with the RRM of Rbmx/hnRNP G. Rbmx is a known regulator of α-synuclein splicing and alternative splicing (AS) patterns upon DNA damage response [20,21], while hnRNP A/B members are known for their role in mRNA metabolism, including splicing [22]. Based on the structural similarities between Rbm3 and hnRNPs and the predicted interaction of Rbm3 with splicing complexes [18], we next analysed whether Rbm3 indeed plays a role in pre-mRNA splicing.

Rbm3 deficiency causes transcriptome-wide splicing alterations

To investigate the effects of Rbm3 deficiency on genome-wide splicing, we used a CRISPR/Cas9-mediated strategy to disrupt the Rbm3 gene (Supplemental Fig. S2A). Mouse L929 cells were transfected with p×330vector expressing a gRNA targeting the 2^nd exon of the Rbm3 gene, which is present in all annotated transcript isoforms. A cell clone characterized by a missense mutation where a T nucleotide was inserted between nucleotides 143 and 144 of the Rbm3 coding sequence was selected and verified by immunofluorescence staining and immunoblot (Supplemental Fig. S2B-C). Consistent with previous data [9,23], Rbm3-deficient L929 cells are viable, but showed slightly slower proliferation rate and moderate delay in recovery upon UV irradiation (Supplemental Fig. S2D). RNA-seq analysis of WT and Rbm3^(def) L929 cells detected 1451 significant differentially expressed genes (DEG) with a false discovery rate (FDR) < 0.05. Of these, 978 were upregulated (log2FC ≥ 1) and 473 were downregulated (log2FC ≤ −1). Gene Ontology (GO) analysis of DEGs revealed enrichment in biological processes related to ‘cell differentiation’, ‘cell membrane organization’ and ‘extracellular matrix organization’ (Supplemental Fig. S2E-F). Next, we performed splicing analysis using rMATS software [24]. Applying FDR < 0.05, rMATS identified 3570 significant alternative splicing (AS) events and 1952 differentially spliced genes across all chromosomes in knockout cells compared to wild-type cells (Figure 1B–C). These results indicated that Rbm3 deficiency causes transcriptome-wide splicing abnormalities in line with a potential role as a splicing regulator. We identified changes in all known types of AS events, including exon skipping (SE; 59,49%), mutually exclusive exons (MXE; 12,15%), intron retention (RI; 12,04%) and alternative splice site selection (A5SS; 8,51% and A3SS; 7,78%) (Figure 1D). GO analyses of the mis-spliced genes revealed biological functions such as ‘chromatin remodelling’, ‘epigenetic regulation of gene expression’, ‘regulation of translation’ and ‘regulation of protein stability’, which are essential for housekeeping and basal cellular functions (Figure 1E). To predict the potential physiological consequences of the aberrant splicing patterns upon Rbm3 deletion, we ran IsoformSwitchAnalyzer [25] to detect novel transcript isoforms, their relative abundance and the functionality of the respective isoform-encoded proteins. IsoformSwitchAnalyzer found 465 genes with significantly altered mRNA isoform expression in Rbm3^(def) cells compared to WT (Figure 2A and Supplemental Table). Among these, 244 genes switch to a shorter mRNA isoform while 209 genes switch to a longer RNA isoform. These RNA isoforms have either fewer or more exon counts, altered frequency of intron retention, differential UTR length or last exon. Moreover, we detected a complete loss of splicing isoforms or the appearance of new, unannotated splicing isoforms. Sixty-nine transcript isoforms are predicted to gain resistance against Nonsense-mediated decay (NMD), and 49 to be sensitive to NMD (i.e. exclusion or inclusion of premature stop codon). IsoformSwitchAnalyzer predicted 117 isoforms with additional protein domains and 106 mRNA with loss of functional domains. The changes in protein structure include inclusion or exclusion of intrinsically disordered regions (IDR), which most likely function in protein-protein interactions as well as domains for signalling and subcellular localization (Figure 2A). The 465 genes with significant alteration of isoform expression are enriched in cell cycle, apoptosis and DNA damage response (DDR) regulators. Splicing changes of several candidate genes (Map3k7, Cdk7, Cdk4, and Clk2) were further validated by RT-PCR (Figure 2B–E). Consistent with our RNA-seq data, the selected genes and their corresponding transcripts showed significantly decreased exon inclusion levels in the absence of Rbm3, resulting in shorter PCR products. Each RT-PCR product was confirmed by sequencing. To ensure that the observed splicing changes were due to Rbm3 deficiency, we co-expressed the major Rbm3 isoform (UniProt ID: O89086) from a cDNA in our Rbm3^(def) cell line and found that exon inclusion could be reversed (Figure 3A, B, and C), with the exception of Clk2 exon 4, which showed an opposite outcome. This might be explained by non-physiological, excessive levels of Rbm3 under overexpression conditions, possibly leading to the displacement of another splicing factor specifically required for Clk2 exon 4 splicing (Figure 3D). Additionally, we cannot fully exclude the possibility that the size of the GFP tag perturbs the interaction of Rbm3 with another splicing factor necessary for Clk2 exon 4 splicing. Nevertheless, the general trend towards the restoration of Cdk7, Map3k7 and Cdk4 splicing suggests that the Rbm3-GFP protein is functional.

Figure 2.

Molecular consequences of Rbm3 deficiency. (A) table containing the detected type and number of transcript isoform switches on mRNA level (green table) and their predicted consequences on their corresponding protein structure (blue table) and subcellular localization (yellow table) in Rbm3^(def) L929 cells. Abbreviations: untranslated regions (UTR), open reading frame (ORF), nonsense-mediated decay (NMD), transcription start site (TSS), intrinsically disordered region (IDR). Semi-quantitative RT-PCR images and quantifications confirm increased exon skipping in candidate genes: Map3k7 (B), Cdk7 (C), Cdk4 (D) and Clk2 (E). Top left panel: graphical representation of transcript isoform structures (thick black bars are exons, thin lines are introns, black arrows show strand directionality and purple arrows show oligo locations for PCR validation). Top right panels: RT-PCR images and quantification of the detected exon-skipping events. Data represent means of the exon inclusion-to-skipping ratio ± standard deviation from three independent biological replicates. Statistical significance was calculated using an unpaired two-tailed Student’s t-test. ** = p < 0.01; **** = p < 0.0001.

Bottom panels are Sashimi plots of the corresponding exon skipping events. Y-scale on the left shows the read coverage on exons. The curves and the numbers within indicate the number of reads with inclusion/exclusion events.

Figure 3.

Rbm3 co-expression restores exon-inclusion rates. (A – F) RT-PCR analysis and quantification of exon skipping events in Rbm3^(wt), Rbm3^(def), and in Rbm3^(def) L929 cells where Rbm3-gfp is ectopically overexpressed. Genes are: Map3k7 (A), Cdk7 (B), Cdk4 (C) and Clk2 (D). Statistical significance was calculated by one-way ANOVA with Tukey’s multiple comparisons test. Error bars represent standard deviation from biological triplicates. * = p < 0.05; ** = p < 0.01; n.S. = not significant. Eif3f (E) and gapdh (F) are negative control genes without detectable as events in Rbm3^(def) L929 cells.

Rbm3 deficiency particularly affects splicing of highly expressed multi-exon genes

To substantiate our findings, we also analysed other publicly available Rbm3 knockout RNA-seq data sets from a different origin: Rbm3 KO murine activated lung innate lymphoid cells (ILCs) [26], Rbm3 KO murine adult hippocampus [27] and a human pancreatic cancer cell line with siRNA silenced RBM3 [28]. We used the same analysis pipeline as before and rMATS identified 4655 significant AS events in Rbm3 KO lung ILCs, from which again SE and RI events were the most frequent: 44.46% (2070) and 22.06% (1027) respectively. Additionally, rMATS detected 11.17% (520) and 17.65% (822) alternative 5$\mathrm{\prime }$ and 3$\mathrm{\prime }$ splice sites, 4.64% (216) MXE and 22.06% (1027) RI events, affecting 2534 expressed genes. GO analysis of the AS genes found in lung ILCs identified ‘mRNA processing’, ‘DNA recombination’ and ‘regulation of DNA repair’ processes, important for regulation and maintenance of proper gene expression, similar to the AS processes in L929 cells. Additionally, protein homeostasis processes were also affected by AS, i.e.: ‘ribosome biogenesis’, ‘protein polyubiquitination’ and ‘protein acylation’ (Figure 4A). In murine adult Rbm3 KO hippocampi, we detected 379 significant AS events, including 40.89% (155) and 20.58% (78) SE and RI events, 13.98% (53) and 15.83% (60) A5SS and A3SS and 8.7% (33) MXE events, affecting 253 expressed genes. We would like to point out that the RNA-seq data from murine hippocampi are a single-end dataset which is less sensitive for AS analysis thus likely explaining the lower AS detection rate. GO analysis of AS genes from mouse hippocampi returned neuronal and synapse function-related processes such as ‘neuron to neuron synapse’, ‘postsynaptic specialization’ and ‘protein localization to synapse’ (Figure 4B). Next, we tested whether RBM3 also influences splicing landscape in human cells, thus we ran rMATS on data from RBM3 WT and siRNA-silenced human pancreatic cancer cells. In RBM3-silenced human pancreatic cancer cells, we detected 2958 significant AS events, from which again SE and RI are the most frequent: 57.43% and 16.6%, respectively. The rest are 9.56% A5SS, 11.73% A3SS and 6.65% MXE events, affecting 2350 expressed genes. GO analysis of AS genes identified regulators of protein transport, cell death, DNA repair and metabolism, again important components of cellular maintenance (Figure 4C). Overall, these data show that Rbm3 deficiency causes transcriptome-wide splicing alterations.

Figure 4.

Characterization of mis-spliced genes upon Rbm3 deficiency. (A – C) gene ontology analysis showing biological processes significantly enriched with as genes in RNA-seq data from Rbm3-deficient activated mouse lung immune cells (A), mouse hippocampus (B) and in human pancreatic cancer cells (C). (D – F) relative frequency plots of as genes (blue line) and all expressed genes (black line) in Rbm3^(def) L929 cells, lung immune cells and mouse hippocampus dataset for the comparison of the following characteristics: (D) gene expression level (log2 scale), (E) gene length (log10 scale) and (F) exon counts (log2 scale). Wilcoxon rank-sum test, **** represent the following p-values from left to right panels: (D) p < 2.2 × 10⁻¹⁶, p < 2.2 × 10⁻¹⁶, p = 5.648 × 10⁻¹², p = 1.617 × 10⁻¹¹; (E) p < 2.2 × 10⁻¹⁶, p < 2.2 × 10⁻¹⁶, p < 2.2 × 10⁻¹⁶, p < 2.2 × 10⁻¹⁶; (F) p < 2.2 × 10⁻¹⁶, p = 1.39 × 10⁻¹³, p = 1.568 × 10⁻⁵ and p = 6.987 × 10⁻⁷.

Intriguingly, the detected AS genes are enriched among relatively highly expressed genes (Figure 4D). To investigate this observation, we employed a Wilcoxon rank-sum test to compare the distribution of expression levels between AS genes and non-AS genes. The Wilcoxon test results were as follows: p-value <2.2 × 10⁻¹⁶ for mouse L929 cells; p-value <2.2 × 10⁻¹⁶ for mouse lung ILCs; p-value = 5.648 × 10⁻¹² for mouse hippocampus; and p-value = 1.617 × 10⁻¹¹. These results indicate that the expression levels of AS genes are consistently higher than those of non-AS genes in these mouse tissues. We also observed that our AS genes are enriched by long genes. Long introns and weak splice sites are linked to increased AS events, especially exon skipping, due to the lack of SS recognition [29]. Therefore, we also measured the gene length, exon counts of AS genes, and the splice site strength of the skipped exons of AS genes using MaxEntScan as described in [30]. Our data show that Rbm3-sensitive genes in general have a higher expression level, gene size and number of exons, where the mis-spliced exons have weaker splice sites (Figure 4E–F and Supplemental Fig. S3B-C), indicating that Rbm3 is particularly required for the maintenance of fidelity of splice site recognition of complex genes with a high transcriptional and splicing isoform burden. Given the considerable homology between the RRM domains of Rbm3 and Rbmx, we computationally tested whether this structural resemblance is also reflected in the splicing landscape of expressed genes upon Rbmx deletion. To do this, we analysed publicly available RNA-seq data from human RBMX knockout and WT pancreatic cancer cells. Similarly to the RBM3^(si) dataset, rMATS detected transcriptome-wide splicing abnormalities in RBMX knockdown (kd) cells, with skipped exons being the most frequent (66.89% SE, 11.41% MXE, 8.99% RI, 7.47% A3SS and 5.23% A5SS). The detected AS genes are enriched in multi-exon genes with an average or higher expression level (Supplemental Fig. S4 A and B). These results indicate that the structural homology between Rbm3 and Rbmx may imply functional similarities as well.

Rbm3 alters splice site selection

As our AS list also included genes linked to splicing and mRNA processing, we wanted to test whether Rbm3 directly regulates splicing. For this, we performed MS2 tethering assays as described previously [13,31] and recruited MS2-Rbm3 and derived protein domain fusions to a tandem MS2 binding site that is placed between two competing test 5$\mathrm{\prime }$ splice sites (5$\mathrm{\prime }$SS). Previously, using this reporter system we demonstrated that related hnRNP proteins promote selection of the upstream 5$\mathrm{\prime }$SS at the expense of the downstream 5$\mathrm{\prime }$SS, which is normally preferentially used [31]. In our current assay, when expressed alone or together with MS2 coat protein only, the HIV-1-derived splicing reporter showed a preferential activation of the downstream 5$\mathrm{\prime }$SS D1_down in mouse fibroblast L929 cells (Figure 5B, lanes 1–2) recapitulating the respective splicing profiles obtained in human cells [31]. While the RS domain of SRSF1 further enhanced activation of the downstream splice site (Figure 5B, lanes 2 and 3), hnRNP-like mouse Fox2α led to a clear shift towards the upstream splice site D1_up (Figure 5B, lanes 2 and 4). Strikingly, the MS2-RBM3 fusion proteins also favoured the upstream 5$\mathrm{\prime }$SS as would be expected for a splicing factor that functionally belongs to the hnRNP protein family (Figure 5B, lanes 2, 4 and 5 and Figure 5C). Unexpectedly, the N-terminus of Rbm3 that encompasses the RRM domain was sufficient to recapitulate a shift towards upstream splice site D1_up (Figure 5B, lanes 2, 4 and 6 and Figure 5C) while the GRD alone had no impact on splice site selection (Figure 5B, lanes 2, 4 and 7). To ensure that the observed splicing effects were not caused by binding of Rbm3 to positions other than the MS2 binding site within our reporter mRNAs, we also co-expressed Rbm3 without the MS2 coat protein domains (Figure 5D-E). Importantly, Rbm3 alone had no impact on 5$\mathrm{\prime }$SS selection. These results indicate that Rbm3 presence regulates splice site choice and this activity is mediated by the RRM domain and its immediate flanking sequences. Overall, these findings highlight Rbm3 as an important regulator of pre-mRNA splicing and transcript integrity.

Figure 5.

Rbm3 alters splice-site selection. (A) graphical representation of the splicing reporter assay which consists of two main components: 1.) Rbm3 (full protein, or RRM or GRD separately) is tagged with the MS2 bacteriophage protein that recognizes and binds to bacterial stem-loop RNA structures. 2.) an artificial mRNA containing stem-loop structures in between two competing (upstream and downstream) splice sites. MS2 tethering of potential splicing regulators could edit various mRNA sizes: splicing repressors cause intron retention, producing a 433 bp fragment with the oligos targeting the exons (grey bars), while splicing regulators with preferential 5$\mathrm{\prime }$splice site recognition might produce either a 238 bp fragment (SD1 down) or a 138 bp fragment (SD1 upstream). (B) RT-PCR results of our assay show increased use of the SD1 upstream 5$\mathrm{\prime }$SS when Rbm3 is bound, indicating that Rbm3 is a splicing regulator. (C) bar diagram showing the quantification of the 5$\mathrm{\prime }$splice site usage. (n = 5 biological replicates, one-way ANOVA followed by post-hoc Tukey’s test, p < 0.01) (D) and (E) PCR assay and quantification of 5$\mathrm{\prime }$splice site usage in the presence of ectopically expressed Rbm3-gfp without MS2 tag (n = 4 biological replicates, one-way ANOVA followed by post-hoc Tukey’s test).

Discussion

Regulation of gene expression by RNA-binding proteins occurs at all stages of the RNA life cycle, from RNA biosynthesis to protein translation. RBPs are multifunctional proteins that regulate multiple aspects of RNA metabolism. Rbm3 was originally described as a cold-induced RBP that acts as a translational enhancer of a specific subset of RNA targets that promote survival. However, our comprehensive gene expression profiling shows Rbm3 as a ubiquitously expressed protein across tissues, pointing towards previously unknown physiological functions of Rbm3 even in the absence of stress. Strikingly, increased Rbm3 protein levels were shown to revert pathologic CD44 mRNA splicing in prostate cancer [32], and Rbm3 is predicted to physically interact with several splicing factors, including other hnRNP proteins as well as the U1 snRNP-specific protein U1-A (STRING database, [18]). Although recent studies could provide interesting insights into the splicing regulation of the Rbm3-encoding RNA itself [33,34], the role of Rbm3 as a regulator of pre-mRNA splicing remains largely undefined.

Here, we demonstrated that Rbm3 deficiency results in transcriptome-wide pre-mRNA splicing defects and compromised transcriptome integrity. While Rbm3 deficiency did not significantly alter mRNA expression levels of splicing factors (Supplemental Fig. S2E-F), we cannot completely exclude indirect effects on splicing such as those caused by changes in the Rbm3-dependent translation of cellular splicing factors. On the other hand, given the functional and structural similarities between the RRM domains of Rbm3 and Rbmx, as well as the similarity between the C-terminal GRD of Rbm3 and hnRNP A/B proteins [13], our new data from Rbm3^(def) cells and MS2 tethering experiments suggest a direct role for Rbm3 in splicing regulation. Contrary to our expectations, the N-terminal part of Rbm3 containing the RRM domain was sufficient to recapitulate the effect of the full-length protein on splicing. Such activity of RRMs of other hnRNP proteins had previously been described to be functionally important for the splicing activity by, e.g. forming contacts with RNA binding proteins or U snRNAs [35–40]. Based on these findings, we hypothesize that Rbm3 is a novel splicing factor that is required for the maintenance of splicing accuracy. We propose a model in which Rbm3 is an integral component of molecular networks composed of various splicing factors (SFs) that determine splicing outcomes across tissues [41]. Analogous to other hnRNP proteins, Rbm3 may cooperate with or compete against other SFs for binding to its target sites [13,42]. The splicing outcome will be dictated by the specific portfolio and the relative concentrations of other SFs expressed within a given cell type or under a particular cellular status. This can also explain the differential sensitivities of tissues to altered Rbm3 expression levels, as the built-in redundancy of hnRNP proteins might also partially compensate for the loss of Rbm3. Although we observed that Rbm3-sensitive splice sites tend to be weaker than the consensus sequences, we failed to extract specific molecular signatures that specify the sensitivity of genes to loss of Rbm3 and the exact molecular mechanisms underlying splicing regulation by Rbm3 need to be addressed by future studies.

In total, we identified 1952 genes which were differentially spliced in our Rbm3^(def) mouse L929 fibroblast. While splicing changes of most of the candidate genes validated by RT-PCR analysis (Cdk7, Map3k7 and Cdk4) were reversed upon ectopic expression of a Rbm3-GFP fusion protein, Clk2 exon 4 showed an unexpected opposite splicing phenotype (Figure 3D). This might be explained by increased or excessive concentrations of Rbm3 causing displacement of other SFs from the Clk2, which are normally required for exon 4 splicing. Alternatively, the size of the GFP tag interferes with the formation of Rbm3-containing, multi-component splicing complexes that are specifically required to promote Clk2 exon 4 inclusion. Accordingly, it is plausible that the GFP tag could block the assembly of additional copies of Rbm3 or other SFs at Clk2 RNAs, which are functionally important for spliceosome recruitment and exon 4 recognition.

Strikingly, the splicing isoforms of the protein kinases Map3k7, Cdk4 and Clk2 identified in Rbm3-deficient cells have already been linked to cancer [43–45] and neurodegenerative diseases [46], consistent with previously described roles of Rbm3 in cell cycle control [47,48] and neuroprotection [34]. These transcripts are predicted to produce protein variants that either partially or completely lack their kinase domain, potentially impairing cell cycle control. Consistent with these findings, we observed a slightly lower proliferation rate in Rbm3-deficient L929 mouse fibroblasts (Supplementary Figure S2C). However, whether different tissues respond variably to Rbm3 deficiency and whether splicing changes mediated by Rbm3 deficiency predispose certain tissues or cell types to cell cycle disorders and diseases are intriguing questions for future studies.

Splicing is temporally and functionally coupled to transcription (for a recent review see [49]). Increased sensitivity of highly transcribed, multi-exon genes might be explained by the ‘window of opportunity model’, which proposes that the speed of the RNA polymerase II (RNAPII) dictates the time frame, in which an upstream splice site can be recognized before transcriptional elongation leads to competition with a newly synthesized downstream splice site [50–52]. It is therefore tempting to speculate that a specific subset of upstream splice sites requires Rbm3 at high transcription rates in order not to be outcompeted by emerging stronger downstream splice sites. Whether slowing down RNAPII can rescue the inclusion of Rbm3-sensitive exons is an intriguing question that remains to be addressed by future experiments. Alternatively, Rbm3 deficiency might directly or indirectly result in a limited availability/compromised allocation of spliceosomes within the cells. Here, highly transcribed, multi-exon genes could be the very first ones to suffer from spliceosome scarcity due to their higher pre-mRNA splicing activity. Such mechanisms could also explain why we identified a different set of Rbm3-sensitive genes and GO terms in each cell type and tissue. Nonetheless, in this study, we demonstrate for the first time that Rbm3 deficiency leads to transcriptome-wide splicing abnormalities and impaired transcriptome integrity. Future experiments need to clarify the molecular principles behind Rbm3-dependent RNA splicing.

Methods

Cell culture and cloning experiments

We used the pX330-vector-based CRISPR/Cas9 system to generate a Rbm3-deficient L929 cell line, with the following gRNA sequences: forward: GATGAGAGCTATGAATGGAG and reverse: CTCCATTCATAGCTCTCATC. Transfection of L929 cells was carried out by using a commercial JetPRIME® transfection kit. Cells expressing the Rbm3 gRNA were enriched by a selective medium containing 2 μg/ml puromycin. Two days later, individual GFP-positive cells were sorted in 96 wells. A cell clone with Rbm3 cDNA containing a T nucleotide insertion between 143 and 144 nucleotides was expanded and used. All L929 cells were cultured in DMEM containing 4.5 g/L D-Glucose (Gibco), supplemented with 2 mm L-Glutamine (Gibco), 10% FCS and 1% PenStrep (Gibco). All cells were cultured at 37°C in a humidified incubator with 5% CO₂. To monitor cell proliferation, cells were cultured in an Incucyte® (Sartorius) automated live-cell analysis system, where images were taken every 2 h for 48 h for wild type and Rbm3^(def) L929 cells, with or without 20J/m² UVC irradiation before monitoring growth. For the rescue experiments, the main isoform of Rbm3 (ENSMUST00000040010.10) was cloned into the mammalian pEGFP-N2 overexpression vector. Rbm3^(def) L929 cell transfections were performed in 6-well plates at ~30% confluence using jetPRIME® as transfection reagent following the manufacturer’s instructions. Forty-eight hours later GFP positive cells were FACS-sorted directly into the cell lysis buffer for RNA extraction (RNeasy kit, Qiagen).

Immunofluorescence

Cells were cultured on glass coverslips. At 70–80% confluence, cells were washed with room temperature (RT) PBS and fixed for 15 min in 4% formaldehyde-containing PBS. Subsequently, cells were washed twice with PBS and permeabilized using 0.1% Triton X-100 in PBS for 15 min at RT. Cells were then washed twice with PBS and blocked with 4% Normal Horse Serum (NHS) for 30 min. Coverslips were incubated overnight with 4% NHS PBS containing Rbm3 primary antibody (Abcam, ab134946). Next day, cells were washed three times with PBS for 5 min and incubated in 4% NHS PBS containing Alexa Fluor conjugated secondary antibodies for one hour. Cells were washed three times with PBS for 5 min and incubated in PBS containing 1:2,000 DAPI (ThermoFisher) for 20 min. After one wash with PBS, samples were mounted using Prolong Diamond (Invitrogen). Images were obtained with a LSM700 ZEISS Axio Imager M2 Microscope and Rbm3 staining intensity was quantified with Fiji (Image J 1.54f).

Immuno-blot assay

Wild-type and Rbm3^(def) L929 cells were grown in 6-well plates. Around 75–85% confluence, the culture medium was removed, and cells were washed twice with PBS. Cells were lysed in RIPA buffer 150 mm NACl; 1% Nonidet® p-40, 0.5% sodium deoxycholate, 0.5% SDS and 50 mm Tris (pH 7.4). Protein samples were loaded on NuPAGE™ 4–12% Bis-Tris gels. Samples were transferred to a 0.2 µm Nitrocellulose membrane (BioRad), washed with PBS and blocked in 4% BSA containing PBS overnight at 4°C. Next day, membranes were incubated in 4% BSA PBS containing anti-Rbm3 (Abcam, ab134946) or anti-αTubulin (Sigma) antibody for 1 h at RT. Membrane was washed with PBS-Tween 0.1% twice for 10-min RT. Finally, the membrane was incubated 4% BSA PBS containing IRDye® (LI-COR) secondary antibodies for 1 h, washed in PBS twice and processed in an Odyssey® imaging system (LI-COR).

Splicing reporter assay

The MS2 splicing reporter SV D1 2MS2 D1 env/GFP was previously described [31] and is based on an HIV-1 glycoprotein/eGFP expression plasmid [53]. The MS2-Rbm3, MS2-RRM and MS2-GDR fusion protein expressing plasmids (SVSD4/SA7 scMS∆FG Rbm3, RRM or GDR) were generated by substitution of the BamHI/XhoI fragment of SVSD4/SA7 scMS∆FG SRSF1(RS) with a BglII/XhoI-digested fragment that had been amplified using appropriate primers (see below) and mouse cDNA as a template. SVSD4/SA7 scMS∆FG (MS2 only), SVSD4/SA7 scMS∆FG SRSF1(RS domain) and SVSD4/SA7 scMS∆FG Fox2α had been described before [31]. L929 cell transfections were performed in 6-well plates at ~30% confluence using jetPRIME® as transfection reagent following the manufacturer’s instructions. Total RNA samples were collected 48 h after transfection using the RNeasy kit (Qiagen). For cDNA synthesis 1 μg of DNase digested-RNA was reverse transcribed by using SuperScript™ reverse transcriptase (ThermoFisher). For semi-quantitative analysis of SV D1 2MS2 D1 env/eGFP-derived reporter mRNAs, cDNA was used as a template for a Platinum™ Taq polymerase-based PCR reaction with forward primer #3210 and reverse primer #3211. PCR products were separated on 2% agarose gels and stained with Ethidium bromide for visualization. For the full primer list, please see the Supplemental Table (Oligos-used tab).

RNA library preparation and sequencing

Total RNA was isolated from WT and Rbm3^(def) L929 cells using the RNeasy kit (Qiagen) including the ‘on-column’ DNase step (RNase free DNase kit, Qiagen). RNA quality was estimated by using Bioanalyzer (Agilent) and only high-quality RNA (R.I.N. value > 8) was used for further analyses. Samples were sequenced on an Illumina NovaSeq 6000. Libraries were prepped with Illumina TruSeq stranded mRNA protocol including ERCC RNA spike-in mix as control.

RNA-seq data analysis

Most computationally intensive processes were done on GALAXY EU servers [54]. For the isoform switch analysis, sequencing reads were quality checked by FastQC (Galaxy Version 0.74+galaxy0) and aligned with the STAR aligner (Galaxy Version 2.7.10b+galaxy4) to mouse reference genome mm39. For de novo transcript assembly StringTie (Galaxy Version 2.2.1+galaxy1) [55] was used with Mus musculus gencode vM32 transcripts as guide [56]. Finally, IsoformSwitchAnalyzeR (2.2.0) [57] Bioconductor (3.18) package was used under R environment (4.3.0) to detect significant changes in transcript abundance. For differentially expressed gene analysis, Kallisto (0.46.0) [58] and RUVSeq [59] were used to estimate gene expression by using mouse gencode vM32 transcripts as template. All genes with detectable expression levels were included in the analysis (>0.001 CPM). Genes with FDR < 0.05 and log2 fold change >1 are considered upregulated, whereas genes with FDR < 0.05 and a log2 fold change < −1 are considered downregulated. For alternative splicing, BAM files were subjected to rMATS turbo (4.2.0) [24]. Sashimi plots were generated by the rmats2sashimiplot app (https://github.com/Xinglab/rmats2sashimiplot). GO analysis was performed by using clusterProfiler (4.10.1) Bioconductor package [60]. Splice site strength of exons was measured by the MaxEntScan web application developed by Burge lab:(http://hollywood.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq_acc.html).

Statistical analyses

Experimental data were obtained from at least three biological replicates. Statistics were performed in GraphPad Prism 7. In general, p-value <0.05 was considered statistically significant. All statistical methods and p-values are described in the figure legends. IsoformSwitchAnalyzeR calls the DEXSeq package [61] for statistical analysis.

Data and code availability

RNA-seq data from L929 cells have been deposited on the NCBI SRA website under the BioProject name PRJNA1090010 and are publicly available as of the date of publication. Human and mouse datasets were downloaded from the European Nucleotide Archive (https://www.ebi.ac.uk/ena/browser/home; human RBMX^(kd): PRJNA1012715, human RBM3^(si): PRJNA717807, mouse Rbm3^(KO) lung ILCs: PRJNA649420 and mouse Rbm3^(KO) hippocampus: PRJNA529585). Microscopy and other data reported in this paper will be shared by the lead contact upon request.

Funding Statement

Á.G. was funded by Project-ID 73111208 – SFB 829, European Joint Project on Rare Diseases RD20-113, acronym TC-NER. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Author contributions

Á.G. conceived and supervised the project. Á.G. designed and performed most experiments. Á.G. prepared the figures. Á.G. and S.E. wrote the manuscript. S.E. designed and interpreted the splicing reporter experiments. Á.G. prepared all sequencing samples. Á.G. and A.P. performed bioinformatics analyses. Á.G. and M.G. designed and performed the knockout experiments. Á.G., J.H.J.H., S.E., A.P., H.S. and M.G. critically read and edited the manuscript.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15476286.2024.2413820