Introduction

Efficient gene expression in eukaryotes requires introns and exons to be correctly recognised by the spliceosome, the macromolecular machine that joins exons together. The spliceosome recognises short sequences that are present at exon-intron junctions within precursor mRNAs called splice sites. In higher organisms there is some flexibility in splice site recognition, as most genes produce multiple mRNAs by alternative splicing. However, aberrant “cryptic” splice sites that are weakly selected or totally ignored by the spliceosome occur frequently in the human genome and could function as decoys to interfere with gene expression (Aldalaqan et al., 2022; Sibley et al., 2016). Many cryptic splice sites are located amongst repetitive sequences within introns, where they are repressed by RNA binding proteins belonging to the hnRNP family (Attig et al., 2018). Cryptic splice sites can also be present within exons, particularly long exons where they can cause formation of exitrons (exon sequences that are removed as if they were introns) (Marquez et al., 2015).

The testis-specific nuclear RNA binding protein RBMXL2 was recently shown to repress cryptic splice site selection during meiosis, including within some ultra-long exons of genes involved in genome stability (I. Ehrmann et al., 2019). RBMXL2 is only expressed within the testis (Aldalaqan et al., 2022; I. Ehrmann et al., 2019), raising the question of how the cryptic splice sites controlled by RBMXL2 are repressed in other parts of the body. RBMXL2 is part of an anciently diverged family of RNA binding proteins. The RBMXL2 gene evolved 65 million years ago following retrotransposition of the RBMX gene from the X chromosome to an autosome (I. Ehrmann et al., 2019). RBMX and RBMXL2 proteins (also known as hnRNP-G and hnRNP-GT) share 73% identity at the protein level and have the same modular structure comprising an N-terminal RNA Recognition Motif (RRM) and a C-terminal disordered region containing RGG repeats (Figure 1A). RBMX and RBMXL2 are also more distantly related to a gene on the long arm of the Y chromosome called RBMY that is deleted in some infertile men (with only ∼37% identity between human RBMXL2 and RBMY) (Elliott et al., 1997; Ma et al., 1993). The role of RBMY in the germline is almost totally unknown, but RBMY protein has been implicated in splicing regulation (Elliott et al., 2000; Venables et al., 2000).

Figure 1.

The X and Y chromosomes are inactivated during meiosis within a heterochromatic structure called the XY body (Turner, 2015; Wang, 2004). Meiosis is quite a long process, and to maintain cell viability during this extended period when the X chromosome is inactive, a number of autosomal retrogenes have evolved from essential X chromosome genes and are actively expressed during meiosis. However, it is unknown whether RBMXL2 and RBMX are functionally similar enough for RBMXL2 to provide a replacement function for RBMX during meiosis, or whether RBMXL2 evolved differently to control meiosis-specific patterns of expression. Suggesting somewhat different activities RBMX was recently shown to activate exon splicing inclusion, via a mechanism involving binding to RNA through its C-terminal disordered domain facilitated by recognition of m6A residues and RNA polymerase II pausing (Liu et al., 2017; Zhou et al., 2019).

Here, we have used iCLIP and RNA-seq to analyse the binding characteristics and RNA processing targets of human RBMX. We identify a novel class of RBMX-dependent ultra-long exons connected to genome stability and transcriptional control, and find that RBMX, RBMXL2 and RBMY paralogs have closely related functional activity in repressing cryptic splice site selection. Our data reveal an ancient mechanism of gene expression control by RBMX family proteins that predates the radiation of mammals and provides a new understanding of how ultra-long exons are properly incorporated into mRNAs.

Results

RBMX primarily operates as a splicing repressor in somatic cells

We first set out to identify the spectrum of splicing events that are strongly controlled by RBMX across different human cell lines. We used RNA-seq from biological triplicate MDA-MB-231 cells treated with siRNA against RBMX (achieving >90% depletion, Figure 1B), followed by bioinformatics analysis using the SUPPA2 (Trincado et al., 2018) and MAJIQ (Vaquero-Garcia et al., 2016) splicing prediction tools. We identified 315 changes in RNA processing patterns in response to RBMX-depletion that were high enough amplitude to be visually confirmed on the IGV genome browser (Robinson et al., 2011) (Figure 1 – Figure supplement 1) (Figure 1 – Source Data 1). Analysis of these splicing events within existing RNA-seq data from HEK293 cells depleted for RBMX (GSE74085) (Liu et al., 2017) revealed 166 high amplitude events that are controlled by RBMX in both HEK293 and MDA-MB-231 cells (Figure 1C). We concentrated our downstream analysis on these splicing events (Figure 1 – Source Data 1). 92% of the splicing events regulated by RBMX in human somatic cells were already annotated on Ensembl, Gencode or Refseq (Figure 1D). Strikingly two thirds of these events are repressed by RBMX, meaning they were increasingly used in RBMX depleted cells compared to control, and include exon inclusion, alternative 5’ and 3’ splice sites, exitrons, and intron retention (Figure 1E). This means that RBMX primarily operates as a splicing repressor in human somatic cells.

Splicing control and sites of RBMX protein-RNA interaction are enriched within long internal exons

The above data indicated that RBMX has a major role in repressing cryptic splicing patterns in human somatic cells. To further correlate splicing regulation to patterns of RBMX protein-RNA interactions, we next mapped the distribution of RBMX-RNA binding sites in human somatic cells. We engineered a stable human HEK293 cell line to express RBMX-FLAG fusion protein in response to tetracycline addition. Western blotting showed that expression of RBMX-FLAG was efficiently induced after tetracycline treatment. Importantly, levels of the induced RBMX-FLAG protein were similar to those of endogenous RBMX (Figure 2A). We next used this inducible cell line to carry out individual nucleotide resolution crosslinking and immunoprecipitation (iCLIP) – a technique that produces a global picture of protein-RNA binding sites (Konig et al., 2011). After crosslinking, RBMX-FLAG protein was immunoprecipitated, then infra-red labelled RNA-protein adducts were isolated (Figure 2B) and subjected to library preparation. Following deep sequencing of biological triplicate experiments, 5 to 10 million unique reads (referred to here as iCLIP tags, representing sites of RBMX protein-RNA cross-linking) were aligned to the human genome. Each individual iCLIP replicate showed at least 70% correlation with each of the others (Figure 2 – Figure Supplement 1A). K-mer motif analysis revealed RBMX preferentially binds to AG-rich sequences (Figure 2C and Figure 2 – Figure Supplement 1B).

Figure 2.

In line with previous work on other RNA binding proteins (Van Nostrand et al., 2020), only 31% of the RNA splicing events that are controlled by RBMX in both HEK293 cells and MDA-MB-231 cells were identified by iCLIP as direct targets for RBMX binding (Figure 2 – Figure supplement 1C). However, RBMX-responsive internal exons that did contain RBMX iCLIP tags were significantly longer than the ones that are not bound by RBMX (Figure 2D). We therefore compared the length of the exons regulated (identified by RNA-seq) and bound by RBMX (identified by iCLIP) within protein-coding genes to all mRNA exons expressed in HEK293 (Liu et al., 2017). We reasoned that larger exons might have a higher chance to be bound by RBMX merely because of their large size. Because of this, we did not take into account the density of RBMX binding and considered all exons that contained at least one iCLIP tag. Strikingly, we found that exons regulated and bound by RBMX were significantly longer than the median size of HEK293 mRNA exons which is ∼130 bp (Figure 2E). This led us to test whether RBMX protein is preferentially associated with long exons. For this we plotted the distribution of exons bound and regulated by RBMX together with all exons expressed from HEK293 mRNA genes (Liu et al., 2017) (Figure 2 – Source Data 1). We found that RBMX controls and binds two different classes of exons: the first have comparable length to the average HEK293 exon, while the second were extremely long, exceeding 1000 bp in length (Figure 2F). We defined this second class as ‘ultra-long exons’, which represented the 20.8% of exons regulated by RBMX and 15% of the ones that contained RBMX iCLIP tags. These proportions were significantly enriched compared to the general abundance of ultra-long exons expressed from HEK293 cells, which was only 3.9% (Figure 2G). Overall, this data revealed a function for RBMX in the regulation of splicing of ultra-long exons.

RBMX is important for proper splicing inclusion of full-length ultra-long exons within genes involved in DNA repair and RNA polymerase II transcription

We next performed Gene Ontology analysis in order to determine whether RBMX regulates splicing patterns for particular classes of genes. Both the genes bound by RBMX (detected using iCLIP, Figure 3 – Figure supplement 1A and Figure 3 – Source Data 1) and regulated by RBMX in both MDA-MB-231 and HEK293 cell lines (detected using RNA-seq, Figure 3 – Figure supplement 1B and Figure 3 – Source Data 1) each showed individual global enrichment in functions connected to genome stability and gene expression. Similarly, Gene Ontology analyses showed that genes that were both bound by and dependent on RBMX for correct splicing were enriched in pathways including transcription and gene expression, DNA replication and repair, and meiosis (Figure 3A and Figure 3 – Source Data 1). In comparison, Gene Ontology analyses of all ultra-long exons in the human genome revealed enrichment of gene expression but not genome stability pathways (Figure 3 – Figure supplement 1C and Figure 3 – Source Data 1), showing that regulation of ultra-long exons within genes involved in DNA metabolism and cell cycle is characteristic of RBMX. These data are consistent with published observations (Adamson et al., 2012; Munschauer et al., 2018; Zheng et al., 2020) that depletion of RBMX reduces genome stability. In addition, comet assays also detected increased levels of genome instability after RBMX depletion (Figure 3 - Figure Supplement 1D, E).

Figure 3.

The above data indicated that RBMX-RNA binding interactions and splicing control by RBMX are particularly associated with long internal exons and enriched within classes of genes involved in genome stability. These exons included the 2.1 Kb exon 5 of the ETAA1 (Ewings Tumour Associated Antigen 1) gene, where RBMX potently represses a cryptic 3’ splice site that reduces the size of this exon from 2.1 Kb to 100 bp (Figure 3B and Figure 3 – Figure supplement 2A). RT-PCR analysis confirmed that RBMX depletion causes a much shorter version of ETAA1 exon 5 to prevail, particularly in MDA-MB-231 and NCI-H520 cells, but less in MCF7 cells (Figure 3C). ETAA1 encodes a replication stress protein that accumulates at sites of DNA damage and is a component of the ATR signalling response (Bass et al., 2016). Selection of RBMX-repressed cryptic 3’ splice sites within ETAA1 exon 5 removes a long portion of the open reading frame (Figure 3 – Figure supplement 2B). Consistent with the penetrance of this high amplitude ETAA1 splicing defect being sufficiently high to affect protein production, no ETAA1 protein was detectable 72 hours after RBMX depletion from MDA-MB-231 cells (Figure 3D).

Another ultra-long exon is found within the REV3L gene that encodes the catalytic subunit of DNA polymerase ζ that functions in translesion DNA synthesis (Martin and Wood, 2019). RBMX similarly represses a cryptic 3’ splice site within the ultra-long exon 13 of the REV3L gene (∼4.2 Kb), that has an extremely high density of RBMX binding (Figure 3E). RT-PCR analysis confirmed a strong splicing switch to a cryptic splice site within REV3L exon 13 after RBMX was depleted from MDA-MB-231, MCF7 and NCI-H520 cells (Figure 3F).

We also detected extremely high density RBMX protein binding within exon 9 of the ATRX gene (3 Kb) that encodes a chromatin remodelling protein involved in mitosis. Depletion of RBMX results in expression of a shortened version of ATRX exon 9, caused by formation of an exitron through selection of cryptic 5’ and 3’ splice sites within exon 9 (Figure 3 – Figure supplement 2C).

RBMX protein-RNA interactions may insulate important splicing signals from the spliceosome

The iCLIP data suggested a model where RBMX protein binding may insulate ultra-long exons so that cryptic splice sites cannot be accessed by the spliceosome. This model predicted that RBMX binding sites would be close to important sequences used for selection of cryptic splice sites. RBMX iCLIP tags mapped just upstream of the cryptic 3ʹ splice sites within ETAA1 exon 5 in HEK293 cells and MDA-MB-231 cells after RBMX depletion (Figure 3B), suggesting that RBMX may bind close to the branchpoints used for these cryptic splicing patterns. However, although usually located close to their associated 3’ splice sites, in some cases branchpoints can be located far upstream (Gooding et al., 2006). We tested the prediction that RBMX may sterically interfere with components of the spliceosome by directly mapping the branchpoints associated with use of these cryptic ETAA1 splice sites. To facilitate mapping of the branchpoint sequences used by the cryptic 3ʹ splice site within ETAA1 exon 5, we made a minigene by cloning the ultra-long ETAA1 exon 5 and flanking intron sequences between constitutively spliced β-globin exons (Figure 3 – Figure supplement 2D). Confirming that this minigene recapitulated cryptic splicing patterns, after transfection into HEK293 cells we could detect splicing inclusion of both the full-length and shorter (cryptic) versions of ETAA1 exon 5 ETAA1 mRNA isoforms using multiplex RT-PCR (Figure 3 – Figure supplement 2E). We then used an RT-PCR assay (Figure 3 – Figure Supplement 2F) to monitor the position of branchpoints just upstream of the cryptic 3ʹ splice sites of ETAA1 exon 5 (Královičová et al., 2021). Sanger sequencing of the amplification product made in this assay confirmed that the branchpoint sequences used by these cryptic 3’ splice sites are adjacent to RBMX binding sites (Figure 3G and Figure 3 – Figure supplement 2G).

RBMXL2 and RBMY can replace the activity of RBMX in somatic cells

The above data showed that although RBMX can activate splicing of some exons, it predominantly operates as a splicing repressor in human somatic cells, and moreover has a key role in repressing cryptic splicing within ultra-long exons. This global pattern of RBMX activity is thus very similar to that previously reported for RBMXL2 in the germline, where RBMXL2 represses cryptic splice sites during meiosis. RBMXL2 is expressed during male meiosis when the X chromosome is silenced. To directly mimic this switch in protein expression we constructed a HEK293 RBMXL2-FLAG tetracycline-inducible cell line, from which we depleted RBMX using siRNA (Figure 4A). Western blots showed that RBMX was successfully depleted after siRNA treatment, and the RBMXL2-FLAG protein was strongly expressed after tetracycline induction, thus simulating their relative expression patterns in meiotic cells (Figure 4B). We then analysed cryptic splicing patterns. Confirming our previous finding, in the absence of tetracycline treatment depletion of RBMX led to increased selection of cryptic splice sites within ETAA1 exon 5 and REV3L exon 13, and to formation of an exitron within ATRX exon 9 (Figure 4C-E, compare lanes 7-9 with lanes 10-12). Strikingly, tetracycline-induction of RBMXL2 was sufficient to repress production of each of these aberrant splice isoforms (Figure 4C-E, compare lanes 1-3 with lanes 4-6). We also globally investigated patterns of splicing in these rescue experiments by performing RNA-seq analysis of each of the experimental groups. For each target gene we analysed we found that RBMXL2 was able to replace the function of RBMX in splicing control (Figure 4 – Figure supplement 1B-D).

Figure 4.

The above experiments showed that RBMXL2 is able to globally replace RBMX activity within somatic cells. RBMX and RBMXL2 are both more distantly related to the Y chromosome-encoded RBMY protein, with RBMX and RBMY diverging when the mammalian Y chromosome evolved (Figure 1A). RBMY has also been implicated in splicing control (Nasim et al., 2003; Venables et al., 2000), but its global functions are very poorly understood. We thus tested whether RBMY might also be performing a similar function to RBMX. Employing a HEK293 cell line containing tetracycline-inducible, FLAG-tagged RBMY protein, we detected successful recovery of normal splicing patterns of the ultra-long exons within the ETAA1, REV3L and ATRX genes within RBMX-depleted cells 24 hours after tetracycline induction of RBMY (Figure 4 – Figure supplement 2). These results indicate that even despite its more extensive divergence, RBMY can also functionally replace RBMX in cryptic splice site control within long exons. Thus, splicing control mechanisms by RBMX family proteins pre-date the evolution of the mammalian X and Y chromosomes

The disordered domain of RBMXL2 is required for efficient splicing control of ultra-long exons

The above data showed that RBMX predominantly operates as a splicing repressor in somatic cells, thus performing a functionally parallel role to RBMXL2 in the germline. Although RBMX contains an RRM domain that is the most highly conserved region compared with RBMXL2 and RBMY, splicing activation by RBMX depends on its C-terminal disordered domain that also binds to RNA (Liu et al., 2017; Moursy et al., 2014). We thus reasoned that if RBMX and RBMXL2 were performing equivalent molecular functions, rescue of splicing by RBMXL2 should be mediated by the disordered region of RBMXL2 alone, independent of the RRM (Liu et al., 2017; Moursy et al., 2014). To test this prediction, we created a new tetracycline-inducible HEK293 cell line expressing the disordered region of RBMXL2 protein and not the RRM domain (RBMXL2ΔRRM, Figure 5B). Tetracycline induction of this RBMXL2ΔRRM protein was able to rescue siRNA mediated depletion of RBMX (Figures 5C-E), directly confirming that the C-terminal disordered domain of RBMXL2 protein is responsible for mediating cryptic splicing repression.

Figure 5.

Discussion

We previously showed that the germ cell-specific RBMXL2 protein represses cryptic splice site selection during meiotic prophase. Here we find that this is part of a bigger picture, where the closely related but more ubiquitously expressed RBMX protein also provides a similar activity within somatic cells. Supporting this conclusion, both RBMX and RBMXL2 proteins most frequently operate as splicing repressors in their respective cell types. Furthermore, RBMXL2 and even the more diverged RBMY protein are able to provide a direct replacement for RBMX function within human somatic cells.

We further find that RBMX binds and is very important for proper splicing inclusion of a group of ultra-long exons, defined as exceeding 1 Kb in length. RBMXL2 similarly represses cryptic splice sites within ultra-long exons of genes involved in genome stability including Brca2 and Meioc (I. Ehrmann et al., 2019). Although many of the splice sites we find to be repressed by RBMX within ultra-long exons are already annotated, they are not usually selected in the human cell lines we investigated and thus represent potential decoy splice sites that would interfere with full-length gene expression. Long human exons provide an enigma in understanding gene expression. Most human exons have evolved to be quite short (∼130 bp) to facilitate a process called exon definition, in which protein-protein interactions between early spliceosome components bound to closely juxtaposed splice sites promote full spliceosome assembly (Black, 1995; Robberson et al., 1990). Exon definition also requires additional RNA binding proteins to recognise exons and flanking intron sequences. These include members of the SR protein family that bind to exonic splicing enhancers (ESEs) and activate exon inclusion, with exons typically having higher ESE content relative to introns. While the mechanisms that ensure proper splicing inclusion of long exons are not well understood, cryptic splice sites would be statistically more likely to occur within long exons compared to short exons, where they could prevent full-length exon inclusion. Cryptic splice sites within long exons could be particularly problematic (compared to an intronic location) since they would be embedded within a high ESE sequence environment. For example some long exons require interactions with the SR protein SRSF3 and hnRNP K and phase separation of transcription factors to be spliced (Kawachi et al., 2021).

Meiotic sex chromosome inactivation represses expression of a panel of essential genes, opening the need for alternative routes to fulfil their function (Turner, 2015). A number of essential X-linked genes have generated autosomal retrogenes that are expressed during meiosis however genetic inactivation of many of these retrogenes causes a phenotype that manifests outside of meiosis. (Wang, 2004). An exception is exemplified by the RPL10 and RPL10L proteins that are 95% identical: RPL10 mutation causes meiotic arrest, and RPL10L has been shown to directly replace its X-linked ortholog RPL10 during meiosis (Jiang et al., 2017; Wang, 2004). RBMXL2 is the only other X-linked retrogene that has been shown to be essential for meiotic prophase (I. Ehrmann et al., 2019). Here we show that ectopic expression of RBMXL2 can compensate for lack of RBMX in somatic cells. This is consistent with a recent model suggesting that RBMXL2 replaces RBMX function during meiosis because of transcriptional inactivation of the X chromosome (Aldalaqan et al., 2022). This general requirement for functionally similar RBMX family proteins across somatic and germ cells further suggest that RBMX-family functions in splicing control have been required for ∼200 million years, since before the divergence of separate RBMX and RBMY genes early in mammalian evolution.

The iCLIP data reported here show a high density of RBMX binding within ultra-long exons, consistent with a model in which RBMX protein binding to RNA masks sequences required for cryptic splice sites selection. Such RBMX binding would block access to spliceosome components or splicing activator proteins (Figure 6). Our data show that the C-terminal disordered domain of RBMXL2 protein is sufficient to control splicing inclusion of ultra-long exons. This is exactly analogous to the mechanism of control of splicing activation by RBMX, which occurs via recognition of m6A modified RNA targets via the C-terminal disordered domain (Liu et al., 2017). Intriguingly, global studies have shown that m6A residues are enriched within some long internal exons (Dominissini et al., 2012), where they might help facilitate RBMX protein-RNA interactions. The C-terminal disordered region of RBMX is also reported to mediate protein-protein interactions therefore shorter exons that show defective splicing in RBMX-depleted cells but are not directly bound by RBMX could rely on different regulatory mechanisms. RBMY, RBMX and RBMXL2 directly interact with the SR protein Tra2β (Elliott et al., 2000; Venables et al., 2000) and have opposing functions during RNA binding and splicing regulation (Nasim et al., 2003; Venables et al., 2000). Hence it is still possible that RBMX family proteins counteract recognition by SR proteins of ESEs near cryptic splice sites via a protein-protein interaction mechanism.

Figure 6.

Extensive literature shows that RBMX is important for genome stability, including being involved in replication fork activity (Munschauer et al., 2018; Zheng et al., 2020), sensitivity to genotoxic drugs (Adamson et al., 2012) and cell proliferation (https://orcs.thebiogrid.org/Gene/27316). Interestingly, many of the ultra-long exons controlled by RBMX are within genes important for genome stability, including REV3L, ATRX and ETAA1. This makes it likely that RBMX contributes to maintaining genome stability through ensuring full-length protein expression of genes important in this process. As an example, we show here that depletion of RBMX protein causes aberrant selection of a high amplitude cryptic splice site within ETAA1 exon 5 which prevents detectable expression of ETAA1 protein, and contributes to genome instability (Bass et al., 2016). Cancer and neurological disorders are amongst the most common human diseases associated with defective DNA damage response (Jackson and Bartek, 2009). The double role of RBMX in genome maintenance via both direct participation in the DNA damage response and splicing regulation of genome stability genes could explain why mutations of RBMX are associated with an intellectual disability syndrome (Cai et al., 2021; Shashi et al., 2015), and why RBMX has been identified as a potential tumour suppressor (Adamson et al., 2012; Elliott et al., 2019).

Materials and methods

Cell culture and cell lines

MDA-MB-231 (ATCC^® HTB-26™), MCF7 (ATCC^® HTB-22™), U2OS (ATCC^® HTB-96™) and NCI-H520 (ATCC^® HTB-182™) cells were maintained in Dulbecco′s Modified Eagle′s Medium (DMEM) high glucose pyruvate medium (Gibco, #10569010), supplemented with 10% fetal bovine serum (FBS, Gibco, #21875034) and 1% Penicillin-Streptomycin (Gibco, #15140130). HEK293 (ATCC^® CRL-1573) were maintained in DMEM plus 10% fetal bovine serum. Cell line validation was carried out using STR profiling was according to the ATCC^® guidelines. All cell lines underwent regular mycoplasma testing.

Generation of tetracycline-inducible cell lines

RMBX, RBMXL2, RBMY and RBMXL2ΔRRM genes were cloned onto a FLAG-pcDNA5 vector and co-transfected with pOG44 plasmid into Flp-In HEK293 cells like previously described (Ehrmann et al., 2016). RBMX-FLAG, RBMXL2-FLAG, RBMY-FLAG and RBMXL2ΔRRM-FLAG expression was induced by the addition of 1µg/ml tetracycline (Sigma-Aldrich) to promote expression via a tetracycline-inducible promoter. The Flp-In HEK293 cells were cultured in high glucose pyruvate medium (Gibco, #10569010), supplemented with 10% FBS (Gibco, #21875034) and 1% Penicillin-Streptomycin (Gibco, #15140130).

siRNA knockdown and tetracycline induction

RBMX transient knockdown was established using two different pre-designed siRNAs targeting RBMX mRNA transcripts (hs.Ri.RBMX.13.1 and hs.Ri.RBMX.13.2, from Integrated DNA Technologies). Negative control cells were transfected with control siRNA (Integrated DNA Technologies, # 51-01-14-04). Cells were seeded onto 6-well plates forward transfected with Lipofectamine™ RNAiMAX transfection reagent (Invitrogen, # 13778150) according to manufacturer′s instructions using 30 pmol of siRNA for 72h at 37°C before harvesting. For tetracycline-inducible cell lines, Flp-In HEK293 cells expressing either RBMXL2-FLAG, or RBMY-FLAG, or RBMXL2ΔRRM-FLAG genes were similarly seeded onto 6-well plates and treated with RBMX and control siRNAs for 72h at 37°C. 24h before harvesting 1µg/ml of tetracycline (Sigma-Aldrich) was added to half of the siRNA-treated samples to promote the expression of RBMXL2-FLAG and RBMY-FLAG.

RNA-seq

RNA was extracted from cells using RNeasy Plus Mini Kit (Qiagen #74134) following manufacturer’s instructions and re-suspended in nuclease-free water. RNA samples were DNase treated (Invitrogen, AM1906). For siRNA treated MDA-MB-231 cells, paired-end sequencing was done initially for two samples, one of negative control and one of RBMX knock-down, using an Illumina NextSeq 500 instrument. Adapters were trimmed using trimmomatic v0.32. Three additional biological repeats of negative control and RBMX siRNA treated MDA-MB-231 cells were then sequenced using an Illumina HiSeq 2000 instrument. The base quality of raw sequencing reads was checked with FastQC (Andrews, 2010). RNA-seq reads were mapped to the human genome assembly GRCh38/hg38 using STAR v.2.4.2 (Dobin et al., 2013) and subsequently quantified with Salmon v. 0.9.1 (Patro et al., 2017) and DESeq2 v.1.16.1 (Love et al., 2014) on R v.3.5.1. All snapshots indicate merged tracks produced using samtools (Li et al., 2009) and visualised with IGV (Robinson et al., 2011). For HEK293 cells treated with either RBMX or control siRNA, either in the absence or in the presence of tetracycline, RNAs were sequenced using an Illumina NextSeq 500 instrument. Quality of the reads was checked with FastQC (Andrews, 2010). Reads were then aligned to the human genome assembly GRCh38/hg38 to produce BAM files using hisat2 v.2.2.1 (Kim et al., 2015) and samtools v.1.14 (Li et al., 2009) and visualised using IGV (Robinson et al., 2011).

Identification of splicing changes

Initial comparison of single individual RNA-seq samples from RBMX-depleted and control cells was carried out using MAJIQ (Vaquero-Garcia et al., 2016), which identified 596 unique local splicing variations (LSV) at a 20% dPSI minimum cut off from 505 different genes potentially regulated by RBMX. These LSVs were then manually inspected using the RNA-seq data from the second RNA sequencing of biological replicates for both RBMX-depleted and control cells, by visual analysis on the UCSC browser (Karolchik et al., 2014) to identify consistent splicing changes that depend on RBMX expression. The triplicate RNA-seq samples were further analysed for splicing variations using SUPPA2 (Trincado et al., 2018), which identified 6702 differential splicing isoforms with p-value < 0.05. Predicted splicing changes were confirmed by visual inspection of RNA-seq reads using the UCSC (Karolchik et al., 2014) and IGV (Robinson et al., 2011) genome browsers. Identification of common splicing changes between RBMX-depleted MDA-MB-231 and HEK293 cells was done comparing data from this study with data from GSE74085 (Liu et al., 2017).

iCLIP

iCLIP experiments were performed on triplicate samples in RBMX-FLAG expressing Flp-In HEK293 cells using the protocol described in (Huppertz et al., 2014). Briefly cells were grown in 10 cm tissue culture dishes and irradiated with 400 mJ cm−2 ultraviolet-C light on ice, lysed and sonicated using Diagenode Bioruptor® Pico sonicator for 10 cycles with alternating 30 secs on/ off at low intensity and 1 mg of protein was digested with 4 U of Turbo DNase (Ambion, AM2238) and 0.28 U/ml (low) or 2.5 U/ml (high) of RNAse I (Thermo Scientific, EN0602). The digested lysates were immunoprecipitated with Protein G Dynabeads™ (Invitrogen, #10003D) and either 5 μg anti-FLAG antibody (Sigma-Aldrich, F1804) or 5 μg IgG (Santa Cruz biotechnology, sc-2025).

Subsequently a pre-adenylated adaptor L3-IR-App (Zarnegar et al., 2016) was ligated to the 3’ of the RNA fragments. The captured Protein-RNA complexes were visualised using Odyssey LI-COR CLx imager scanning in both the 700nm and 800nm channels. The RNA bound to the proteins was purified, reverse transcribed with barcoded RT oligos complementary to the L3 adaptor. The cDNAs were purified using Agencourt AMPure XP beads (Beckman Coulter™, A63880), circularised and linearised by PCR amplification. The libraries were gel purified and sequenced on Illumina NextSeq 500. All iCLIP sequencing read analysis was performed on the iMaps webserver (imaps.goodwright.com) using standardised icount demultiplex and analyse work flow. Briefly, reads were demultiplexed using the experimental barcodes, UMIs (unique molecular identifiers) were used to remove PCR duplicates and reads were mapped to the human genome sequence (version hg38/GRCh37) using STAR (Dobin et al., 2013). Crosslinked sites were identified on the iMAPS platform and the iCount group analysis workflow was used to merge the replicate samples.

K-mer enrichment analysis

K-mer motif enrichment was performed with the z-score approach using the kmer_enrichment.py script from the iCLIPlib suite of tools (https://github.com/sudlab/iCLIPlib). All transcripts for each non-overlapping protein coding gene from the Ensembl v.105 annotation were merged into a single transcript, used for this analysis, using cgat gtf2gtf --method=merge-transcripts (Sims et al., 2014). Each crosslinked base from the merged replicate bam file was extended 15 nucleotides in each direction. For every hexamer, the number of times a crosslink site overlaps a hexamer start position was counted within the gene and then summed across all genes. This occurrence was also calculated across 100 randomizations of the crosslink positions within genes. The z-score was thus calculated for each hexamer as (occurrence – occurrence in randomized sequences) / standard deviation of occurrence in randomized sequences.

Exon size analysis

Annotations of all human exons related to position and size were downloaded from Ensembl Genes v.105 (http://www.ensembl.org/biomart/). Selection of exons expressed in HEK293 was performed using data from control RNA-seq samples of the dataset GSE74085 (Liu et al., 2017), subsequently filtered to focus on mRNA exons using biomaRt v.2.52.0 (Durinck et al., 2005). Size of the mRNA exons containing RBMX-regulated splicing patterns was annotated using IGV (Robinson et al., 2011). iCLIP tags were extended to 80 nt sequences centered at the crosslinked site, and annotated within human exons using ChIPseeker v.1.32.0 (Yu et al., 2015) and Esembl Genes v.105. iCLIP tags present in mRNA exons were filtered using biomaRt v.2.52.0 (Durinck et al., 2005). iCLIP-containing exons were listed once, independently of the number of tags or tag score. Plots were created using ggplot2 v.3.3.6. (Wickham, 2016) on R v.4.2.1. Statistical analyses to compare exon length distributions between samples were performed by Wilcoxon Rank Sum and Kruskal-Wallis tests using base R stats package v.4.2.1 and pseudorank v.1.0.1 (Happ et al., 2020). Significant enrichment of ultra-long exons in RBMX regulated and bound exons was performed using the “N-1” Chi-squared test (Campbell, 2007) on the MedCalc Software version 20.218.

Gene Ontology analyses

Gene Ontology Analyses were performed in R v.4.2.1 using GOstats v.2.62.0 (Falcon and Gentleman, 2007) except for [Figure 3 – Figure supplement 1A] for which clusterProfiler::enrichGO v.4.4.4 (Yu et al., 2012) was used. Entrez annotations were obtained with biomaRt v.2.52.0 (Durinck et al., 2005). Read counts from control treated HEK293 cells (Liu et al., 2017) and the Bioconductor annotation data package org.Hs.eg.db v.3.15.0 were used as background. P-values were adjusted by false discovery rate using the base R stats package v.4.2.1, except for [Figure 3 – Figure supplement 1A] for which the default Benjamini-Hochberg method was used while running enrichGO. Significantly enriched GOBP pathways were filtered with a p-value cut-off of 0.05. Redundant terms identified with GOstats were removed using Revigo (Supek et al., 2011) with SimRel similarity measure against human genes eliminating terms with dispensability score above 0.5. The dot-plots were produced using ggplot2 v.3.3.6 (Wickham, 2016) focussing on representative terms associated to at least 5% of the initial gene list. Full GOBP lists can be found in Figure 3 – Source Data 1.

Comet assay

The comet assay was performed using the Abcam Comet Assay kit (ab238544) according to manufacturer’s instructions. Briefly U2OS cells transfected with RBMX siRNA or control siRNA were harvested after 72 hours, 1x10⁵ cells were mixed with cold PBS. Cells in PBS were mixed with low melting comet agarose (1/10) and layered on the glass slides pre-coated with low melting comet agarose. The slides were lysed in 1x lysis buffer (pH10.0, Abcam Comet Assay kit) for 48 hours at 4°C, immersed in Alkaline solution (300 mM NaOH, pH>13, 1 mM EDTA) for 30 min at 4°C in the dark and then electrophoresed in Alkaline Electrophoresis Solution (300 mM NaOH, pH>13, 1 mM EDTA) at 300mA, 1volt/cm for 20 min. The slide was then washed in pre-chilled DI H2O for 2 min, fixed in 70% ethanol for 5 min and stained with 1x Vista Green DNA Dye (1/10000 in TE Buffer (10 mM Tris, pH 7.5, 1 mM EDTA), Abcam Comet Assay kit) for 15 min and visualized under fluorescence microscopy Zeiss AxioImager (System 3). Comet quantification was performed using OpenComet (Gyori et al., 2014).

RNA extraction and cDNA synthesis for transcript isoform analysis

RNA was extracted using standard TRIzol™ RNA extraction (Invitrogen, #15596026) following manufacturer’s instructions. cDNA was synthesized from 500 ng total RNA using SuperScript™ VILO™ cDNA synthesis kit (Invitrogen #11754050) following manufacturer’s instructions. To analyse the splicing profiles of the alternative events primers were designed using Primer 3 Plus (Untergasser et al., 2012) and the predicted PCR products were confirmed using the UCSC In-Silico PCR tool. ETAA1 transcript isoform containing the long exon 5 was amplified by RT-PCR using primers 5’-GCTGGACATGTGGATTGGTG-3’ and 5’-GTGCTCCAAAAAGCCTCTGG-3’, while ETAA1 transcript isoform containing the short exon 5 was amplified using primers 5’-GCTGGACATGTGGATTGGTG-3’ and 5’-GTGGGAGCTGCATTTACAGATG-3’. RT-PCR with this second primer pair could in principle amplify also a 2313 bp product from the ETAA1 transcript isoform containing the long exon 5, however PCR conditions were chosen to selectively analyse shorter fragments. REV3L 5’-TCACTGTGCAGAAATACCCAC-3’, 5’-AGGCCACGTCTACAAGTTCA-3’, 5’-ACATGGGAAGAAAGGGCACT-3’. ATRX 5’-TGAAACTTCATTTTCAACCAAATGCTC-3’ and 5’-ATCAAGGGGATGGCAGCAG-3’ All PCR reactions were performed using GoTaq^® G2 DNA polymerase kit from Promega following the manufacturer’s instructions. All PCR products were examined using the QIAxcel^® capillary electrophoresis system 100 (Qiagen). Statistical analyses were performed using GraphPad Prism 9.5.0.

Western blot analyses

Harvested cells treated with either control siRNA or siRNA against RBMX were resuspended in 100mM Tris-HCL, 200mM DTT, 4% SDS, 20% Glycerol, 0.2% Bromophenol blue, then sonicated (Sanyo Soniprep 150) and heated to 95°C for 5 minutes. Protein separation was performed by SDS-PAGE. Proteins were then transferred to a nitrocellulose membrane, incubated in blocking buffer (5% Milk in 2.5% TBS-T) and stained with primary antibodies diluted in blocking buffer to the concentrations indicated below, at 4°C over-night. After incubation the membranes were washed three times with TBS-T and incubated with the secondary antibodies for 1 hour at room temperature. Detection was carried out using the Clarity™ Western ECL Substrate (Cytiva, RPN2232) and developed using medical X-ray film blue film in an X-ray film processor developer. The following primary antibodies were used at the concentrations indicated: anti-RBMX (Cell Signalling, D7C2V) diluted 1:1000, anti-ETAA1 (Sigma, HPA035048) diluted 1:1000, anti-Tubulin (Abcam, ab18251) diluted 1:2000, anti-GAPDH (Abcepta, P04406) diluted 1:2000 and anti-FLAG (Sigma, F1804) diluted 1:2000.

Minigene construction and validation

A genomic region containing ETAA1 exon 5 and flanking intronic sequences were PCR amplified from human genomic DNA using the primers 5’-AAAAAAAAACAATTGAGTTAAGACTTTTCAGCTTTTCTGA-3’ and 5’-AAAAAAAAACAATTGAGTGCTGGGAAAGAATTCAA TGT-3’and cloned into pXJ41 (Bourgeois et al., 1999). Splicing patterns were monitored after transfection into HEK293 cells. RNA was extracted with TRIzol™ (Invitrogen, #15596026) and analysed using a One Step RT-PCR kit (Qiagen, #210210) following manufacturer’s instructions. RT–PCR experiments used 100 ng of RNA in a 5-μl reaction using a multiplex RT-PCR using primers: 5ʹ-GCTGGACATGTGGATTGGTG-3ʹ, 5ʹ-GTGGGAGCTGCATTTACAGATG-3ʹ and 5ʹ-GTGCTCCAAAAAGCCTCTGG-3 ʹ. Reactions were analysed and quantified using the QIAxcel^® capillary electrophoresis system 100 (Qiagen).

Branchpoint analysis

Using the RNA from the long minigene transfections with RBMX and RBMXΔRRM, total RNA was extraction using TRIZOL reagent (Life Technologies) following the standard manufacturer’s instructions, RNA concentration was quantified by NanoDrop UV-Vis spectrophotometer and treated with DNase I (Invitrogen, Am1906). 1μg of purified RNA was reverse transcribed with a SuperScript™ III Reverse Transcriptase (Invitrogen, #18080093) using ETAA1 DBR R1 RT-PCR primer 5′-AAGTTCTTCTTCTTGACTTTGTGTT-3′ and treated with RNaseH (New England Biolabs, M0297S). 1μl of the cDNA was used for PCR amplification reactions using using GoTaq® G2 DNA Polymerase (Promega, #M7845) in the standard 25µl reaction following the manufacturer’s instructions. PCR amplification was carried out using 2 different primer sets ETAA1 DBR R2 5′-GCTCTTGAATCACATCTAGCTCT-3′ and ETAA1 DBR F1 5′-AGCCAAACTAACTCAGCAACA-3′, ETAA1 DBR R2 5′-GCTCTTGAATCACATCTAGCTCT-3′ and ETAA1 DBR F2 5′-AGCATTTGAATCCAGGCAGC-3′ with 18 cycles of amplification at an annealing temperature of 56°C. PCR products were sub cloned into the pGEM-T-Easy vector (Promega, A1360) using manufactures instructions. Plasmids were subjected to Sanger sequencing (Source BioScience) and the sequences were checked with BioEdit (Hall, 1999) and aligned to the genome with UCSC (Karolchik et al., 2014).

Accession numbers

Genomic data is deposited at the Gene Expression Omnibus accession GSE233498.