SRSF5 regulates alternative splicing of DMTF1 pre-mRNA through modulating SF1 binding

ABSTRACT

Among the three DMTF1 splicing isoforms, DMTF1α acts as a tumour suppressor through promoting p14ARF expression, while DMTF1β exhibits an oncogenic role likely through antagonizing DMTF1α. However, the molecular mechanism underlying alternative splicing of DMTF1 pre-mRNA has not been delineated. In the current study, we discovered SRSF5 as a regulatory protein binding to a region located between DMTF1β and α acceptor splice sites to promote DMTF1β and γ splicing. We demonstrated that SRSF5 expression positively correlated with DMTF1β/α ratio in breast cancer samples, and ectopically expressed SRSF5 promoted the splicing of DMTF1β and γ, but not DMTF1α, when testing endogenous DMTF1 pre-mRNA and a reporter construct. Upon SRSF5 knockdown, we observed significantly decreased DMTF1β and γ ratios of endogenous transcripts. An RNA sequence just upstream of the α acceptor site contains two adjacent SRSF5 binding elements, one of which overlaps with an SF1 binding site. Our mechanistic studies revealed that SRSF5 binding to both elements in this region could consign SF1 to its distal-binding sites close to the β and γ acceptor sites, favouring splicing of their isoforms. Overall, our study revealed SRSF5 as a key regulator to promote DMTF1β and γ splicing, and consequently reduce DMTF1α splicing.

Abbreviations

ARF: alternative reading frame, that is, p14ARF, or CDKN2A (cyclin-dependent kinase inhibitor 2A); β-gal: β-galactosidase; CLIP-seq: crosslinking and immunoprecipitation-sequencing; DMTF1: the cyclin D binding myb-like transcription factor 1; ESS/ESE: exonic splicing silencer/enhancer; Ex: exon; FBS: fetal bovine serum; Gluc: Gaussia luciferase; hnRNPs: heterogeneous nuclear ribonucleoproteins; In: intron; ISS/ISE: intronic splicing silencer/enhancer; PBS: phosphate-buffered saline; PCR: polymerase chain reaction; PSI: percent-splice-in; qPCR: quantitative real-time PCR; RIP: RNA immunoprecipitation; RNAseq: RNA sequencing; RT: reverse transcription; SF1: splicing factor 1; SR: serine/arginine-rich proteins; SRSF5: serine and arginine-rich splicing factor 5; TCGA: the cancer genome atlas; UCSC: University of California, Santa Cruz. WT: Wild type

1. Introduction

The DMTF1 (cyclin D-binding Myb-like Protein 1, also known as DMP1) gene is located on chromosome 7q21 and can be transcribed into a pre-mRNA with 18 exons [1]. Due to alternative splicing in intron 9 of the pre-mRNA, mature DMTF1 mRNAs consist of three isoforms encoding proteins DMTF1α, β and γ with 760, 272 and 285 amino acids, respectively [1,2]. DMTF1α is a transcription factor recognized as an important tumour suppressor with a haplo-insufficient property through regulating the ARF-MDM2-p53 pathway during oncogenesis [3–5]. RAS-RAF signalling-stimulated DMTF1α can transactivate the transcription of ARF, which blocks MDM2-mediated p53 degradation [6,7]. Additionally, DMTF1α can directly interact with p53 at the protein level to attenuate its ubiquitination and degradation mediated by MDM2 [8]. We first demonstrated that DMTF1β could promote mammary tumour formation in a MMTV promoter-driven DMTF1β transgenic mouse model and its increased expression correlated with poor clinical outcomes of breast cancer patients [9]. Tschan, et al. also showed that DMTF1β could attenuate ARF expression through antagonizing the transcriptional activity of DMTF1α [10]. Recently, we demonstrated that both transcripts and proteins of DMTF1β and γ showed reduced stability in HeLa cells compared to α and DMTF1γ, similar to DMTF1β, exhibited inhibitory activity to DMTF1α-mediated transcription [11].

Dysregulated pre-mRNA splicing can cause or is associated with different human diseases, such as Alzheimer’s disease, autism spectrum disorder, autoimmune diseases and cancers [12–15]. Mis-spliced pre-mRNAs are very common in cancers, leading to aberrant expression of different isoforms, which may exhibit distinct, or even opposite, biological activities. Abnormal pre-mRNA splicing may result from mutations at or close to splice sites, aberrant expression or mutations of splicing factors, and variations of RNA secondary structures [15,16]. According to a previous study, 10% of 80,000 reported mutations in the human genome could impact pre-mRNA splicing [17]. Recent studies revealed that G-quadruplex structure formed by RNA adjacent to splice sites could also alter splicing process [18,19].

Serine/arginine-rich (SR) proteins and heterogeneous nuclear ribonucleoproteins (hnRNPs) regulate pre-mRNA splicing in a sequence-specific manner [20–22]. Early reports suggested that the SR proteins and hnRNPs could promote and repress splicing, respectively [23], but more recent studies indicated that the overall consequences of their regulation could be either activating or repressing, depending on specific binding or interaction contexts [24]. In addition to the SR proteins and hnRNPs, many other proteins with RNA binding affinity can also regulate pre-mRNA splicing, as previously reviewed [24,25]. To properly exert functions, most trans-acting factors need to recognize and bind to specific sequences, including exonic splicing silencer/enhancer (ESS/ESE) and intronic splicing silencer/enhancer (ISS/ISE) sequence [1,21,26]. Mutations in these splicing elements can result in aberrant splicing events. In a hereditary nonpolyposis colorectal cancer, mutated ESE sequences produced many unclassified variants of the mismatch repair genes MLH1 and MSH2 [27]. Moreover, aberrant expression of trans-acting factors in cancer cells also affects splicing processes [15]. In glioma, high levels of SRSF1 could cause alternative splicing of MYO1B pre-mRNA, leading to enhanced expression of MYO1B-fl isoform that promoted glioma genesis [28]. Overexpressed hnRNPF, K and H1 also promoted development of breast cancer through aberrantly regulating alternative splicing of MCL1 pre-mRNA [29].

SRSF5 (serine and arginine-rich splicing factor 5) has been identified as a key factor in regulating the alternative splicing of the pre-mRNAs of several genes, including MCL1, CCAR1 and HNRNPAB [30–32]. Importantly, the crosslinking and immunoprecipitation-sequencing (CLIP) assay has been used to systematically map SRSF5 binding sites in cellular RNAs, and had provided valuable insights in identifying new regulatory targets of SRSF5 [33,34]. As an oncogenic splicing factor, SRSF5 has been reported to promote the development of different cancers [30,35–37]. Previous studies also indicated that posttranslational modifications, including ubiquitination and acetylation, could modulate the stability of SRSF5 [30], but it was unclear whether they could affect its splicing activity. Nevertheless, arginine-methylated SRSF5 could act as a shuttling protein to modulate mRNA export activity in multipotent cells, and subsequently promote the expression of multipotency factors [33].

Splicing factor 1 (SF1) is an essential component of the spliceosome and can specifically recognize the intron branch points in pre-mRNAs [38,39]. It also recruits U2 snRNPs through interacting with U2AF2 [40–42]. The N-terminal half of SF1 contains two structural motifs, an hnRNPK homology domain and a zinc knuckle, involved in RNA binding [43]. Previous studies revealed different binding elements of SF1 in pre-mRNAs [40,44–47], but it is difficult to generalize any consensus motif from them. Very limited research has been reported regarding the functional interplay between SF1 and SR proteins. One report indicated that SRSF1 knockdown could change the pattern of SF1 localization in nuclear speckles of HeLa cells [48].

In this study, we investigate the regulatory elements and splicing factors contributing to the alternative splicing of DMTF1 pre-mRNA. We identify SRSF5 as a key regulator that modulates alternative splicing of DMTF1 pre-mRNA through interfering with SF1 binding to enhance DMTF1β/γ production.

2. Materials and methods

2.1. Plasmids, ASOs and oligonucleotides

Oligonucleotides used in PCR, blocking assays and DNA sequencing were synthesized by Genewiz Inc. To generate the splicing vectors containing a small region (‘minigene’) of DMTF1, the wild type (WT) and mutated sequences encompassing Exon 9 to Exon 11 of DMTF1 were individually subcloned into the pcDNA3 vector using the ClonExpress® II One Step Cloning Kit (cat# C112, Vazyme) with homologous recombination. Mutant reporters were produced by PCR amplification using oligonucleotides with mutated sites, and assembled together using the cloning kit. Three antisense oligonucleotides (ASOs) ASO-1, −2 and −3 targeting human SRSF5 mRNA were synthesized by Tsingke Biological Technology Inc. with their targeting sequences as TGA ATA CCC GAC AGC CAC TCA T, TCT CCC ATT TAT TTC CTT TCC A and ACC CCC CTT CCA AAT CCC ACC C, respectively. An ASO control (ASO-cont) targeting a scrambled sequence, AAT GAC GTA ATA GAG TAG TCC C, was also created. All ASOs had phosphorothioate backbone modifications, which enhanced their stability and target RNA binding affinity.

2.2. Cell culture, transfection, lentivirus packing and infection

All cell lines used in this study were purchased from the ATCC. HeLa, 293 T, MCF-7, MDA-MB-231 and MDA-MB-453 cells were cultured as recommended by the ATCC. HeLa and 293 T cells were cultured in DMEM medium, MDA-MB-231 and MDA-MB-453 cells were cultured in RPMI medium, and MCF-7 cells were cultured in MEM medium, which were all supplied with 10% foetal bovine serum (FBS) and 1% antibiotics (penicillin and streptomycin) (HyClone); MCF-7 cell medium was also supplied with 0.01 mg/ml human recombinant insulin. MCF-10A cells were cultured with DMEM/F12 supplemented with 5% foetal bovine serum, 0.01 mg/ml human recombinant insulin, 20 ng/ml EGF, 0.5 mg/ml hydrocortisone, and 100 ng/ml cholera toxin. Lipofectamine 2000 (Invitrogen) together with Opti-MEM was used in transient transfection following a procedure recommended by the manufacturer. When testing the minigenes, 1 µg of minigene-plasmid was transfected into HeLa cells cultured in a 12-well plate. Transfection of WT minigene reporters into mammary cells cultured in 60-mm dishes used 2.5 µg of the plasmid. Lentiviral production and infection followed a procedure described previously [49].

2.3. Reverse transcription and quantitative PCR

Total RNA was extracted from cultured cells using the Tripure reagent (Roche, USA) according to the manufacturer’s protocol. Reverse transcription (RT) was carried out using the One-Step gDNA Removal and cDNA Synthesis SuperMix (Transgen, China). In this process, 1 µg of total RNA from each sample was treated by DNase I for 30 min and then reversely transcribed by a poly (dT) primer, a customed gene-specific primer, or random hexamers in a volume of 20 µl at 42°C for 30 min. To evaluate gene expression or splicing events in minigenes, HeLa cells were transfected by minigene vectors for 48 h, followed by total RNA extraction and RT transcription using a primer on pcDNA3 (ATT AGG AAA GGA CAG TGG GAG TGG). In the analysis of quantitative PCR (qPCR), cDNA of each sample was amplified in triplicate using the LightCycler® 480 SYBR Green I Master (Roche, cat# 04913914001) in the LightCycler® 480II Real-Time PCR System (Roche, Basel, Switzerland). Percentage of the transcript for each DMTF1 variant (V), produced by either its minigene or endogenous pre-mRNA, against the total transcriptional level (T), was calculated using the comparative threshold method with a formula of (1+ E_V)^−Ct_V/(1+ E_T)^−Ct_T [50]. In the formula, ‘E’ represents amplificatory efficiency of quantitative primers, and was assessed using quantitative amplifications with stepwise diluted cDNAs as templates. For the quantification of the three endogenous DMTF1 RNA isoforms, an upstream primer (GGC TGT AGC TGA TCC ATC CGT) in the 5ʹ-UTR was used together with one of the three isoform-specific downstream primers (DMTF1α-speci-L, DMTF1β-speci-L, and DMTF1γ-speci-L) covering the splicing junction regions of the three isoforms. An upstream primer in pcDNA3 (pri-a: GCA GAG CTC TCT GGC TAA CTA G) could also be used together with these three isoform-specific primers to quantify or compare the splicing events of the three isoforms in exogenous minigene vectors. For the evaluation of total RNA levels of DMTF1 and minigenes, a downstream primer in Exon 9 (pri-b: GAG CTT CTC AAT TTC TTC AGG TG) was used together with a specific upstream primer from the corresponding experiment in PCR amplification.

2.4. Antibodies

The commercial antibodies and their vendors are β-actin (cat#: AF7018, Affinity), GAPDH (cat# 10 R-G109A, Fitzgerald Industries International), SRSF5 (cat# ab67175, Abcam), Flag (M2, Sigma-Aldrich), and HA (cat# 32–6700, Invitrogen). RAD, a pan DMTF1 antibody, recognizing all three isoforms, was reported previously [9].

2.5. Crosslinking immunoprecipitation (CLIP)

After HeLa cells were transfected and cultured for 48 h, the cells were cross-linked by adding 37% formaldehyde with 0.3% as its final concentration, followed by sonication to fragment RNA. The lysates were then IPed using Flag antibody-conjugated agarose (Flag-Ab beads) following the procedure in the Magna Nuclear RIP (Cross-Linked) RNA-Binding Protein Immunoprecipitation Kit (cat# 17–10520, Merck). To release the crosslinking, the IPed samples and their inputs were incubated with 0.1 μg/μl of Proteinase K at 60°C for 30 min. Furthermore, the IPed RNA was extracted by the Tripure reagent and reversely transcribed by random hexamers, followed by qPCR with 10% of the input as a control.

In the CLIP study of Flag-SRSF5 to evaluate its blocked binding by a reversely complementary oligonucleotide, 0.5 nmol of each oligonucleotide with a specific or a scrambled sequence was transfected with 8 µg of Flag-SRSF5 expression plasmid into HeLa cells cultured in 10 cm dishes. Here, three pairs of primers were designed to quantify precipitated RNA, Prs-up (Prs-up-U, Prs-up-L), Prs-βGAA (Prs-βGAA-U, Prs-βGAA-L), and Prs-ex10 (Prs-ex10-U, Prs-ex10-L). The sequences of these primers are presented in Supplementary Table 2.

In the CLIP study of Flag-SF1 to test the effects of cotransfected HA-SRSF5, 6 µg of HA-SRSF5 expression plasmid or an empty vector was cotransfected with 5 µg of Flag-SF1 expression plasmid into HeLa cells in 10 cm dish. Three primer pairs were designed to evaluate precipitated RNA levels, Prs-P1 (Prs-α-SF1-bs-U, Prs-α-SF1-bs-L), Prs-P2 (Prs-β/γ-SF1-bs-U, Prs-β/γ-SF1-bs-L), and Prs-P3 (Prs-in8-SF1-bs-U, Prs-in8-SF1-bs-L). The sequences of these primers are presented in Supplementary Table 2.

In the CLIP study of pcDNA3-minigenes, WT and its mutant minigene vectors (2.5 µg/each) were simultaneously transfected into HeLa cells cultured in 10 cm dishes. After 6 h, the transfected cells were trypsinized and replated into four 10-cm dishes and cultured for 20 h. The four dishes were then individually transfected by 5 µg of Flag-SF1, 5 µg of Flag-SRSF5, 4 µg of HA-SRSF5 + 3 µg of Flag-SF1, and 4 µg of empty vector + 3 µg of Flag-SF1. Individual primer sets were designed to specifically amplify the transcript of each minigene as listed in Supplementary Table 2.

2.6. Protein expression and purification

His×6-tagged SRSF5-ΔRS, a truncated SRSF5 protein with its RS domain deleted (Supplementary Figure 4A), was expressed in E. coli Rosetta cells induced by 0.15 mM IPTG at 16°C overnight. The bacteria were lysed in a lysis buffer (20 mM HEPES, 100 mM KCl, 0.2 mM EDTA, 20% glycerol, 1% Triton, 2 mM PMSF, 1 mM DTT, 1 mg/ml lysozyme, pH 8.0). The bacterial lysate was centrifuged at 12,000 g for 30 min and recombinant SRSF5-ΔRS protein in the supernatant were purified by Ni-NTA agarose (QIAGEN, Hilden, Germany) following a standard protocol.

2.7. SRSF5-blocked in vitro reverse transcription assay

RNA fragments of the DMTF1 pre-mRNA intron 9 and its mutants were synthesized in vitro following a protocol in the T7 RNA Polymerase Kit (cat# P4074, Promega). One μg of purified Hisx6-SRSF5-ΔRS was incubated with 0.1 pg of these in vitro synthesized RNAs, respectively. The samples were then analysed by RT using a downstream primer, Prs-vitro-RT, and assessed through quantitative PCR using a primer pair, Prs-vitro-U and Prs-vitro-L. The sequences of these primers are presented in Supplementary Table 2.

2.8. RNA electrophoretic mobility shift assay (REMSA)

A WT FAM-labelled RNA probe of DMTF1 intron 9 and its three mutants with altered SRSF5 binding sites were synthesized by Genewiz Inc. In the binding reaction, 5 pmol of each labelled RNA probe and 1 µg of purified His×6-SRSF5-ΔRS protein were mixed in a binding buffer (10 mM HEPES, 1 mM EDTA, 0.11 M KCl, 40 mM NaCl, 5% glycerol, 10 µg/ml BSA, 0.01% NP40, 1 mM DTT, pH 8.0), and incubated at room temperature for 10 min. In a super-shift reaction, 0.5 µg of 6× His-tag antibody (cat# 66005-1-lg, Proteintech) was added to the binding reaction. The samples were analysed by 8% native polyacrylamide gel electrophoresis at 200 V and 4°C for 25 min in a 0.5× TBE (45 mM for Tris-Borate and 1 mM for EDTA, pH 8.0). Fluorescent bands were visualized using Typhoon FLA7000 (GE, Boston, USA).

2.9. Bioinformatic analysis

The TCGA breast cancer dataset (ID: TCGA_BRCA_exp_HiSeqV2) and the matched clinical information of the patients, were downloaded from the UCSC Cancer Browser (https://genome-cancer.ucsc.edu). The percent-splice-in (PSI) of DMTF1β/α dataset was extracted from the TCGA Spliceseq browser (http://bioinformatics.mdanderson.org/TCGASpliceSeq). Identification of the splicing elements within the sequence from the 5ʹ-end of Intron 9 and the 3ʹ-end of Intron 10 of DMTF1 pre-mRNA was carried out using a website-based server, Human Splicing Finder 3.1 (http://umd.be/HSF3/). The binding sites of SRSF5 and SF1 in this region were predicted using a website-based sever, SpliceAid 2 (http://193.206.120.249/splicing_tissue). RNA secondary structure of the sequence between the 5ʹ-end of Exon 9 and 3ʹ-end of Exon 10 of DMTF1 pre-mRNA was computed using the RNA Structure 6.2, a suite of software tools, based on an approach of MaxExpect [51].

In CLIP-seq data analyses, reads in the GSE118265 dataset were split according to their tagged nucleotides containing two barcodes adjacent to their 3ʹ-adapters, and PCR duplicates using other sequences of the barcodes were removed. The barcode was constructed with a pattern of NNBBNTTTTTTNN (N = random tag nucleotide, T = tag nucleotide, and B = RY-space tag nucleotide). Adapters were trimmed and libraries were demultiplexed, using Flexbar (3.5.0) [52] with parameters: – barcodes <Sample.barcode.fa> – barcode-unassigned – barcode-trim-end LTAIL – barcode-error-rate 0 – adapter-trim-end RIGHT – adapter-error-rate 0.1 – adapter-min-overlap 1 – min-read-length 15 – umi-tags. Afterwards, reads were mapped to the reference genome hg38 using STAR (2.7.7a) [53] with parameters: – outFilterMismatchNoverReadLmax 0.04 – outFilterMismatchNmax 999 – outFilterMultimapNmax 1 – alignEndsType Extend5pOfReads12 – sjdbGTFfile <gencode.v34.annotation.gtf> – sjdbOverhang maxReadLength-1 – outReadsUnmapped Fastx – outSJfilterReads Unique – outSAMtype BAM SortedByCoordinate. We excluded all reads identified as PCR duplicates by the UMI_tools (1.0.1) [54], and performed peak calling by PureCLIP (1.3.1) [55].

High-throughput data of RNA structure in HeLa cells, extracted from the GSE74353 dataset and subjected to read processing and mapping against the reference genome hg38 by icSHAPE-pipe [56], were introduced here. Sambamba (0.6.6) [57] were employed to remove repeated alignment reads, followed by data exhibition using the Integrative Genomics Viewer (IGV).

2.10. Statistical analysis

Brown–Forsythe and Welch ANOVA tests followed by Dunnett T3 correction for multiple comparison, and two-way ANOVA followed by Tukey correction for multiple comparison, were performed to analyse statistical significance of all data using the GraphPad Prism 8.0 (GraphPad, SanDiego, CA). The criterion for statistical significance was indicated by asterisks (*, ** and ***) for settings of P < 0.05, P < 0.01 and P < 0.001, respectively, in the figures.

3. Results

3.1. Determination of splicing regulatory elements in Intron 9 of DMTF1 pre-mRNA

Splicing factors binding to splice sites regulate the splicing process of a pre-mRNA and determine the relative production of its isoforms. The DMTF1 pre-mRNA has three major splice isoforms. DMTF1α is the splicing product between the 3ʹ donor site of Exon 9 (Ex9) and 5ʹ acceptor site of Ex10, while DMTF1β and γ are the alternative splicing products with 3ʹ-splice sites in Intron 9 (In9) (Fig. 1A). To interrogate which key splicing factor(s) regulates DMTF1 pre-mRNA splicing, we first evaluated the contribution of individual nucleotides found in In9, Ex10 and In10 (i.e. In9-Ex10-In10) using the web-based algorithm Human Splicing Finder 3.1 (Fig. 1B and Supplementary Figure 1A), which predicts potential regulatory splicing elements or motifs in human RNA sequences and assesses the effects of mutated sequences on splicing alterations [58]. Generally, a specific element may act as a silencer to repress, or an enhancer to promote, the splicing process in its vicinity. Based on this analysis, we identified three potential splicing elements in the In9-Ex10-In10 sequence with relatively high scores, suggesting either splicing silencer or enhancer activity of these sequences. The first element, located between nucleotides 697 and 705 (indicated by the blue peak 1 in Fig. 1B, with the first nucleotide of In9 designated as ‘1’), is a potential splicing silencer that we designated ‘γU rich region’ (i.e. γU), due to its location in the γ-specific region and poly U tract. The second element, located between nucleotides 802 and 820 (indicated by the red peak 2), is a potential splicing enhancer that we designated ‘βGAA repeats’ (i.e. βGAA), due to its location in the β/γ-shared region and 6 ‘GAA’ repeats. The third element, located between nucleotides 1167 and 1181 (indicated by the blue peak 3), is another potential splicing silencer that we designated ‘Intron 10 U rich region’ (i.e. In10U), based on its location in the In10 and multiple uridines.

Figure 1.

Evaluation of sequence elements regulating DMTF1 pre-mRNA splicing. (A) Schematic diagrams of the human DMTF1 gene and alternative splicing of its pre-mRNA. The start and stop codon positions are based on the human DMTF1 mRNA sequence of the accession number NM_001142327.2 of the NCBI. ‘Ex’ denotes exon, while the lines connecting exons represent introns, which are not proportionally drawn to their actual lengths. The intron 9 is specially marked by ‘In9’. In the pre-mRNAs, the splicing patterns of DMTF1α, β and γ isoforms are indicated by folded lines and the coding regions of the three isoforms are differentially coloured as indicated at the bottom. (B) Contributions of the sequence elements in DMTF1 pre-mRNA to its splicing. The horizontal axis represents the region containing In9 to In10 of DMTF1 pre-mRNA with the first nucleotide of In9 designated as ‘1’. The vertical axis indicates the scores of different elements’ contributions to the splicing evaluated by the Human Splicing Finder 3.1 [58]. The peaks 1, 2 and 3 correspond to the γU, βGAA and In10U elements, respectively. Blue silencers and red enhancer are indicated with their sequences, which were presented in red letters

To evaluate the functional roles of these sequences, we selectively mutated nucleotides in the γU, βGAA and In10U elements (Fig. 2A; mutated nucleotides are shown in capitalized red letters) based on the evaluation of the contributions of individual nucleotides to the splicing. When analysed by the Human Splicing Finder 3.1, the mutations greatly reduced the scores of the three elements as either splicing silencer or enhancer (Fig. 2A, top-right panel). We also mutated the acceptor splice sites for DMTF1α, β and γ isoforms individually, combinatorially, or in combination with γU or βGAA mutations (Fig. 2A).

Figure 2.

The role of novel splicing elements in DMTF1 isoform abundance. (A) The sequences of wild type (WT) and mutant minigenes in the regions of the γU, βGAA and In10U elements, and three acceptor splice sites. ‘m-’ denotes mutant of the element(s) and the mutated nucleotides are printed in red capitalized letters in the sequences. The scores of the mutants of the three elements predicted by the Human Splicing Finder 3.1 are shown at the right. The acceptor sites of the three isoforms are labelled on the top. (B) Schematic diagram of a representative minigene construct. The Ex9 to Ex11 of the DMTF1 gene sequence is inserted into pcDNA3. The blue and red boxes represent the γU, βGAA and In10U elements. The splicing patterns favouring DMTF1α, β and γ isoforms are labelled on the top, while the isoform-specific primers are labelled below. The upstream primer pair and the primer for reverse transcription (RT) are indicated. (C) Assays to determine splicing ratios of minigene reporters. The minigenes listed in (A) were individually transfected into HeLa cells followed by total RNA extraction and RT using a primer on pcDNA3. Quantitative PCR (qPCR) was performed using an upstream primer in pcDNA3 (labelled as ‘pri-a’) and each isoform specific primer to evaluate the splicing percentages at the DMTF1α, β and γ acceptor sites against its %total level produced by each minigene. The total transcriptional level by each minigene reporter was assessed by qPCR using the pri-a and a downstream primer (pri-b) in Ex9. The colour of each bar matches the colour of its minigene name. The data are shown as mean values ± SEM from a representative experiment with triplicate samples, and the experiments were repeated seven times with similar results

To analyse the splicing regulation in In9, we generated splicing reporters using the region from Ex9 to Ex11 (i.e. Ex9-In9-Ex10-In10-Ex11) of the DMTF1 pre-mRNA driven by the CMV promoter in pcDNA3 (Fig. 2B). These minigene reporters contained the wild type (WT) and mutant (m-) sequences listed in Fig. 2A. The DMTF1 minigene transcripts in reporter assays using HeLa cells were reverse transcribed by a downstream RT primer on the vector (Fig. 2B), while reverse transcription of samples from untransfected cells used the poly (dT) primer. The samples were analysed by PCR using DMTF1-E9U and DMTF1-E10L primers (Supplementary Figure 3A). WT reporter showed all three amplified bands for DMTF1α, β and γ, as the endogenous RNA sample did (Supplementary Figure 3 C). Additional weak bands were also observed with endogenous RNA alone, likely due to non-specific amplification or undetermined DMTF1 transcripts. Low intensity of DMTF1β and γ bands reflected their relatively weak splicing sites, and was also partially due to their longer amplified regions than that of α. Overall, the WT minigene reporter could virtually splice into a similar pattern to that of the endogenous DMTF1 pre-mRNA.

We individually transfected these minigene reporters into HeLa cells. After 48 h, the extracted RNA was subjected to RT using a downstream primer on the vector (Fig. 2B). The splicing of the three DMTF1 isoforms in each reporter was individually quantified by qPCR using a common upstream primer (pri-a) on the vector and a downstream isoform-specific primer (α-speci-L, β-speci-L or γ-speci-L), with the data normalized by qPCR using the primers pri-a and pri-b (Fig. 2B). Importantly, only samples undergoing RT process, but not those without it, showed a band at 138 bps, when amplified by pri-a and pri-b primers (Supplementary Fig. 3B), indicating lack of contaminant minigene plasmids in RT samples. Therefore, the procedure of using the primer pair was reliable as an internal reference in the minigene assay. The three isoform-specific primers were designed based on the sequences of the corresponding splice junction sites and thus could discriminately recognize DMTF1α, β and γ transcripts. Additional information about these primers, including the ones on the pcDNA3 vector, is presented in Supplementary Fig. 2A. To simplify the narration, we numbered each minigene transfection from #1 to #16 (Fig. 2C). Each isoform was assessed as its percentage of all transcripts from a minigene (%total). As shown in Fig. 2C, addition of the percentages of all three isoforms did not reach 100%, which was most likely due to the presence of unspliced DMTF1 transcripts.

To examine unspliced pre-mRNA levels, we used an RT primer on the vector, and primers DMTF1-E9U and DMTF1-I9L in qPCR (Supplementary Figure 3A). Unspliced pre-mRNAs from mutant minigenes showed no significant difference from that of WT minigene (Supplementary Figure 3D), indicating that unspliced pre-mRNA levels would not affect the evaluation of alternative splicing ratios in DMTF1 intron 9.

As predicted, three acceptor site mutants (m-α, m-β and m-γ) specifically diminished corresponding transcripts (#7 to # 9, Fig. 2C, and Supplementary Figure 3 C), while the simultaneous mutations (m-α/β/γ) completely abolished the splicing of all three isoforms, and thus none of them was detected by isoform-specific primers (#10, Fig. 2C, and Supplementary Figure 3 C). γU mutation increased γ but reduced α and β (#2), suggesting its role as a splicing silencer for DMTF1γ. βGAA mutation modestly increased α, but markedly reduced β and γ (#3). However, concurrent βGAA and α site mutations robustly increased both β and γ (#14). Meanwhile, concurrent βGAA and β site mutations increased γ, but not α (#15), and concurrent βGAA and γ site mutations increased α but reduced β (#16). These results suggested that βGAA was a splicing enhancer element for both β and γ, but this activity depended on α and β sites (Supplementary Figure 2B).

In10U mutation increased the levels of all three isoforms (#4, Fig. 2C), while concurrent In10U and γ site mutations exhibited a pattern of isoform levels similar to that of γU mutation (compare #5 with #2). These results suggested that In10U had no functional interplay with γU but was a non-specific silencer element for DMTF1 In9 splicing. Therefore, we hereafter focused on γU and βGAA in the following studies. Additionally, concurrent γU and βGAA mutations showed no synergy (compare #6 with #2 and #3, Fig. 2C), suggesting their lack of interplay; thus, they are likely independent elements regulating DMTF1 pre-mRNA splicing.

Interestingly, β or γ site mutation did not cause marked change of α transcript (#8 and #9, Fig. 2C), and γ site mutation did not alter β levels either (#9); on the contrary, α site mutation (m-α) markedly increased both β and γ (#7), and β site mutation (m-β) increased γ by over twofold (#8). Thus, the α isoform was likely most preferentially spliced, while the γ isoform was the least spliced during DMTF1 pre-mRNA splicing (Supplementary Figure 2B).

3.2. Functional evaluation of splicing factors binding to the γU and βGAA regions in intron 9 of the DMTF1 pre-mRNA

After discovering the γU and βGAA regulatory elements, we asked whether any splicing factor could associate with them to modulate the splicing process of DMTF1 pre-mRNA. We searched the web-based database SpliceAid 2 (http://193.206.120.249/splicing_tissue), which contains human splicing factor expression data and RNA target motifs [59], for the potential splicing factors binding γU and βGAA, and obtained 30 candidate proteins (Supplementary Table 1). We previously demonstrated that increased DMTF1β expression promoted mammary tumour formation in transgenic mice and a high DMTF1β/α ratio correlated with poor survival outcome of breast cancer patients [9]; thus, if any protein regulates DMTF1 splicing, its expression should correlate with the DMTF1β/α ratio. Based on this hypothesis, we analysed expression correlation coefficients, or R values, of each listed splicing factor with the PSI (percent-splice-in) values of DMTF1β/α, (i.e. the β/(α + β) values), using the RNAseq data from 1095 breast cancer patients collected in the TCGA database. In these analyses, TIA1 and DAZAP1 were among the proteins that could potentially bind to γU and had the largest positive and negative correlation coefficients (0.242 and −0.281, respectively), while SRSF5 and TRA2B were among the potential βGAA-binding proteins that exhibited the largest positive and negative correlation coefficients (0.334 and −0.212, respectively). In a coordinate with the correlation coefficients and the -log(10) p values as the two axes, the data points of SRSF5, DAZAP1, TIA1 and TRA2B stood out from those of most other splicing factor candidates (Fig. 3A). Dot graphs with fitted lines also showed the positive or negative correlations between the expression of these genes and their corresponding DMTF1β/α PSI values (Fig. 3B).

Figure 3.

Evaluating the regulation of DMTF1 pre-mRNA splicing by splicing factors. (A) A coordinate with the two axes as correlation coefficients and corresponding -log(10) P values of splicing factor expression versus DMTF1β/α PSI (percent-splice-in) values based on the analyses of 1095 breast cancer patients collected in the TCGA database. Red and green dots with labelled names indicate the selected splicing factors showing positive and negative correlation with DMTF1β/α PSI, respectively. SRSF5 and TRA2B potentially bind βGAA, while TIA1 and DAZAP1 possibly bind γU. (B) The diagrams of scattered dots showing the correlation of splicing factors TIA1, SRSF5, DAZAP1 and TRA2B with DMTF1β/α PSI in the TCGA dataset of 1095 breast cancer patients. The red lines are the linear regression and the R values (or correlation coefficients) are shown. (C) to (F). Effects of ectopically expressed splicing factors on both the splicing of DMTF1 pre-mRNA by WT minigene and the expression endogenous DMTF1 proteins. Increasing amounts (0 to 1.6 μg) of Flag-TIA1, SRSF5, DAZAP1 and TRA2B expression plasmids (all compensated by an empty vector to 1.6 μg, if necessary) were transfected with 2 μg of WT minigene into HeLa cells cultured in 6-well plates. After 2 days, the cells were collected and divided into two parts to extract total RNA and proteins. Upper panel: bar graphs show the %total of α, β and γ by WT minigene, all RNA levels were determined by RT-qPCR using isoform-specific primers. Lower panel: Western blot analyses by Flag, RAD and GAPDH antibodies. The data are shown as mean values ± SEM from a representative experiment with triplicate samples, and the experiments were repeated three times with similar results

To evaluate the effects of the SRSF5, DAZAP1, TIA1 and TRA2B on DMTF1 pre-mRNA splicing, we generated their expression constructs using the pSL5 vector with the β-actin promoter and a Flag tag. We individually cotransfected HeLa cells with a gradually increasing amount of these vectors and the WT reporter construct (Fig. 2A,B), followed by RT-qPCR analyses to quantify each DMTF1 isoform for its %total, and Western blot to determine endogenous DMTF1 protein levels. Based on the RT-qPCR data, only exogenous SRSF5 clearly reduced DMTF1α and increased β and γ levels (Fig. 3C-F), whereas the other three splicing factors showed ambiguous effects. In Western blot analyses, SRSF5 also decreased DMTF1α and increase β expression (Fig. 3D). Among the other three factors, TRA2B favoured DMTF1α, but reduced DMTF1β (Fig. 3F), consistent with correlation analysis in Fig. 3B, but the effects of TIA1 and DAZAP1 could not be clearly determined (Fig. 3C,E). Overall, SRSF5 displayed the most pronounced effect as a potential splicing enhancer on both DMTF1β and γ. This result was consistent with the functional role of βGAA in regulating DMTF1 pre-mRNA splicing shown in Fig. 2C.

3.3. SRSF5 is a splicing factor regulating DMTF1 pre-mRNA in breast cancer

Prior to evaluating SRSF5 regulation in DMTF1 splicing, we examined relative DMTF1α, β and γ levels of pre-mRNAs alternative splicing from both minigene and endogenous gene.

We transfected WT minigene into nontumorigenic mammary MCF-10A cells, and breast cancer MCF-7, MDA-MB-453 and MDA-MB-231 cells, followed by RT-qPCR as designed in Fig. 2B. Based on DMTF1β/α and γ/α values, WT minigene showed preferential DMTF1β and γ splicing versus α in breast cancer cells with MCF-10A as a control (Fig. 4A). To determine splicing ratios of endogenous DMTF1 pre-mRNA, we first assessed its total transcript levels by RT-qPCR, using a poly (dT) primer in RT and an upstream 5ʹ-UTR primer plus a downstream Ex9 primer in qPCR. We then quantified each isoform in the RT products by qPCR using corresponding isoform-specific primer plus the 5ʹ-UTR primer. Based on DMTF1β/α and γ/α values, endogenous DMTF1 pre-mRNA in breast cancer cells showed significantly preferred DMTF1β and γ splicing over α, compared to MCF-10A cells (Fig. 4B), consistent with WT minigene data in Fig. 4A. In addition, the %total of DMTF1α was significantly lower in MDA-MB-453 cells than that in MCF-10A cells, while DMTF1β and γ showed generally higher %total values in breast cancer cells, especially MCF-7 and MDA-MB-453 cells, than that in MCF-10A cells (Fig. 4C, upper panel). In Western blot analyses, DMTF1α and β proteins exhibited the same expression trends as their transcripts (Fig. 4C, lower panel). Among the three breast cancer cell lines, MCF-7 and MDA-MB-231 cells showed markedly and marginally higher SRSF5 levels, respectively, than that of MCF-10A cells. For unknown reasons, we did not observe clear any SRSF5 band in MDA-MB-453 cells.

Figure 4.

Evaluating DMTF1 pre-mRNA splicing mediated by SRSF5 in different mammary cell lines. (A) and (B) PSI assessments of DMTF1β/α ratios in mammary cell lines using minigene assay (A) or qPCR for endogenous transcripts (B). In (A), nontumorigenic mammary MCF-10A cells and three breast cancer cell lines MCF-7, MDA-MB-231 and MDA-MB-453 were individually transfected with WT minigene shown in Fig. 2A and 2B. Total RNAs were extracted from transfected cells to quantify DMTF1α and β transcripts. In both (A) and (B), qPCR was carried out using the isoform-specific primers shown in Materials and Methods. The PSI of DMTF1β/α represents the ratio of β divided by (α + β). (C) Endogenous DMTF1 transcript and protein levels in mammary cells. Upper panel: the %total of endogenous DMTF1 isoforms in four mammary cell lines. The percentage were evaluated following the procedure described in Materials and Methods. Lower panel: Western blot analyses of DMTF1 and SRSF5 proteins in four mammary cell lines. (D) Effects of SRSF5 knockdown on relative expression of DMTF1 isoforms in MCF-7 cells. The cells were individually transfected by 200 pmol of antisense oligonucleotides (ASOs), ASO-cont, ASO-1, −2 and −3 into MCF-7 cells cultured in 6 cm dishes. After 48 h, the collected cells were divided into two portions to extract total RNA and proteins. Upper panel: the %total of endogenous DMTF1 isoform transcripts. Lower panel: Western blot analyses of DMTF1 and SRSF5 proteins. (C) and (D) All Western blot studies used β-actin as loading controls. (A) to (D) All quantitative data are shown as mean values ± SEM from a representative experiment with triplicate samples, and the experiments were repeated three times with similar results. In (D), some error bars are too short to be visible

Due to relatively high DMTF1β/α and γ/α PSI values, and decent SRSF5 expression in MCF-7 cells, we used this cell line to examine the effects of SRSF5 knockdown on DMTF1 splicing. Three antisense oligonucleotides (ASO-1, −2 and −3), targeting different sites of SRSF5 mRNA, and an ASO-cont were individually transfected into MCF-7 cells and total RNAs were extracted after 48 h. Compared to the ASO-cont, all three ASO-1, −2 and −3 could reduce SRSF5 protein levels, but ASO-3 showed the best knockdown (Fig. 4D, lower panel). Also in ASO-3-transfected cells, we observed significantly increased DMTF1α and simultaneously reduced β and γ. ASO-1 and 2 transfections showed similar tendencies of DMTF1 isoform changes to ASO-3, except a decrease of α by ASO-1, likely due to unspecific effects (Fig. 4D, upper panel). In Western blot analyses, DMTF1α and β protein expression showed consistent changes with their transcript quantitation (Fig. 4D). The ASO transfections were also carried out in MDA-MB-231 cells, but we did not observe significant effects of SRSF5 knockdown on DMTF1 isoform splicing (data not shown), likely due to relatively low SRSF5 levels in MDA-MB-231 cells compared to that of MCF-7 cells (Fig. 4C).

3.4. SRSF5 physically binds to Intron 9 of DMTF1 pre-mRNA

To verify the interaction between SRSF5 and the βGAA element of DMTF1 pre-mRNA, we carried out crosslinking immunoprecipitation (CLIP) experiments. We designed an oligonucleotide, rc-βGAA (ATT TCA TCA TTT ATT CTT CAT TCT TCT TCT TCC C), which was reverse complementary to the βGAA region. rc-βGAA and a scrambled oligonucleotide (TTA CTT GTA CAG CTC GTC CAT GCC) were synthesized and individually cotransfected with Flag-SRSF5 expression vector into HeLa cells followed by CLIP as described in the Materials and Methods section. We made three sets of primers covering the upstream region in In9 (Prs-up), the βGAA element region (Prs-βGAA), and the Ex10 (Prs-ex10) (Fig. 5A). In the IP by Flag-Ab beads, we could pull down Flag-SRSF5 (Fig. 5B, left panel). We then carried out RT using random hexamers as primers and qPCR using each of the three primer sets described above. Among these primer sets, only Prs-βGAA showed meaningful amplification, while Prs-up and Prs-ex10 displayed very weak, or neglectable, signal (Fig. 5B, right panel). Importantly, compared to the scrambled oligonucleotide, the presence of rc-βGAA could reduce the FLAG-SRSF5-IPed βGAA region by about sevenfold, suggesting that the rc-βGAA oligonucleotide could competitively block the binding of SRSF5 to the βGAA element, as schematically depicted in Fig. 5A. We used the two-way ANOVA analysis to compare the value of each sample with that of each of the other five samples in Fig. 5B, with the three detecting regions in the CLIP assay and two oligonucleotides (rc-βGAA and the scramble) were set as two individual observational variables. The results revealed significant interaction between these variables (p < 0.0001), indicating that rc-βGAA specifically competed and blocked SRSF5 binding to βGAA element.

Figure 5.

Evaluation of SRSF5 binding to intron 9 of DMTF1 pre-mRNA. (A) Schematic diagram of oligonucleotide binding interfering with crosslinking immunoprecipitation (CLIP). The region from Ex9 to Ex10 of the DMTF1 gene is shown. The rc-βGAA (reverse complementary to the βGAA site) oligonucleotide that matches the βGAA site and a scrambled oligonucleotide are presented in red arrows and labels. The three primer sets (Prs-up, Prs-βGAA and Prs-ex10 located upstream of βGAA, inside βGAA and inside exon 10, respectively) used in qPCR with predicted product sizes were labelled. (B) CLIP studies to evaluate SRSF5 binding to In9 of DMTF1 pre-mRNA. A Flag-SRSF5 expression vector was cotransfected with the rc-βGAA or a scramble oligonucleotide into HeLa cells. After 48 h, CLIP assay was carried out using Flag-Ab beads. Western blot analyses were used to detect Flag-SRSF5 in the input (10%) and IPed samples (upper panel). After RT using random oligonucleotide hexamers, the samples were analysed by qPCR using the primer pairs Prs-up, Prs-βGAA and Prs-ex10 (lower panel). Significant interaction between rc-βGAA and three detected regions was determined by the two-way ANOVA, p < 0.0001. (C) Schematic diagram of the Ex9 to Ex10 region of DMTF1 pre-mRNA with predicted SRSF5 binding sites. The splicing patterns of DMTF1α, β and γ are indicated on the top. The sequences of three SRSF5 binding sites (SRSF5-bs1, -bs2 and -bs3) with their interval lengths are shown. (D) The sequences of the SRSF5-bs1 and -bs2 in the five minigene reporters. (E) qPCR analyses of CLIP studies to examine SRSF5 binding sites. Flag-SRSF5 were cotransfected with all five WT and mutant minigene reporters simultaneously into HeLa cells. IPed RNA covering the SRSF5-bs1 and -bs2 region in the WT, mut1, 2, 3 and 4 shown in (D) was quantified by RT-qPCR using individually designed specific primer pairs (Supplementary Table 2) with random hexamers as a RT primer. The qPCR value of each IPed sample was normalized by the qPCR value using the same primer pair with 10% input as the template. (F) Schematic diagrams of in vitro SRSF5-mediated RT block assay. In vitro synthesized RNA of DMTF1 In9 and its mutants were individually incubated with purified His×6-SRSF5-ΔRS. RT was carried out using a designated downstream primer. Among these minigenes, WT, mut1 and 4 fitted the scenario of the left without qPCR product, while the mut2 and 3 fitted the right. (G) RT block assay to determine SRSF5 binding affinity to its predicted binding sites in DMTF1 In9. The experiments followed the procedure described in Materials and Methods. Purified SRSF5-ΔRS (upper panel) or its heat-inactivated sample (middle panel) were used in the assay. The lower panel is the results of the assay with neither SRSF5-ΔRS nor reverse transcriptase, with the WT in the upper panel as a comparison (the red bar at right). (H) RNA electrophoretic mobility shift assay (REMSA) to evaluate SRSF5 binding sites in DMTF1 In9. FAM-labelled DMTF1 In9 RNA probe WT and its SRSF5-bs1 and/or -bs2 mutants were individually incubated with His×6-SRSF5-ΔRS, and analysed by native polyacrylamide gel. Gel images with both long and short exposure are presented. The His×6-SRSF5-ΔRS/probe complex and super-shifted complex are denoted by red arrow heads and pointed at left. Non-Specific bands and free probes are also pointed at left. (I) Effects of SRSF5 knockdown on DMTF1 pre-mRNA splicing of minigene reporters. SRSF5 ASO-3 and ASO-cont were individually cotransfected with the WT or mut3 minigene into MCF-7 cells. After 2 days, DMTF1 isoform transcripts were quantified using the strategy and primers shown in Figure 2B, with the total transcripts assessed by qPCR using pri-a and pri-b primers. In (B), (E), (G) and (I), all data are shown as mean values ± SEM from a representative experiment with triplicated samples and the experiments were repeated three times with similar results

Using the SpliceAid 2, we identified three potential SRSF5 binding sites in DMTF1 In9, and named them as SRSF5-bs1 (GAAGAAGA), -bs2 (GCUGC) and -bs3 (GCUGC) (Fig. 5C) [60,61]. With WT minigene in Fig. 2B as the backbone, we generated four mutant minigene reporters with SRSF5 binding sites either deleted or mutated (Fig. 5D, lower panel). We individually synthesized the ‘minigene-specific primer sets’ (Fig. 5D, lower panel, and Supplementary Table 2) based on corresponding sequences of the five minigenes to specifically amplify the region covering the SRSF5-bs1 and SRSF5-bs2. Flag-SRSF5 was transfected with all these minigenes simultaneously into HeLa cells, followed by CLIP assay using Flag-Ab beads. With random hexamers as RT primers, the qPCR value of each IPed sample by its ‘minigene-specific primer set’ was normalized by the qPCR value using the same primer set with 10% input as the template. As a result, mut2 and mut3, both lacking SRSF5-bs1, showed similar RNA signal with significantly lower intensity than that of the WT (Fig. 5E), indicating that SRSF5-bs2 alone was insufficient to recruit SRSF5.

The SRSF5 protein consists of four major regions, including two RNA-recognition motifs (RRMs), one glycine-rich region, and one serine and arginine (RS) domain (Supplementary Fig. 4A). A previous study suggested that SRSF5-RNA binding depended on RS domain phosphorylation [62]; however, other groups reported that the RRMs alone were sufficient to promote pre-mRNA splicing [63], and phosphorylation defective mutants of SR proteins, including SRSF5, showed minor splicing alterations [64]. To evaluate how RS domain affected SRSF5-mediated DMTF1 splicing, we generated a RS domain-deleted mutant, SRSF5-ΔRS. After individually transfecting WT SRSF5, -ΔRS and an empty vector into HeLa cells, we observed similar activity of the WT and mutant SRSF5 in promoting DMTF1β, but attenuating α, splicing (Supplementary Fig. 4B). Therefore, our data supported that the RS domain is dispensable in SRSF5-mediated DMTF1 splicing.

To prove SRSF5 binding to βGAA element, we developed an assay to show SRSF5-blocked in vitro RT process. As presented in Fig. 5F, SRSF5 binding to βGAA site of an RNA would impede reverse transcription initiated by a downstream primer. Thus, qPCR amplifying a region upstream of the βGAA site would negatively reflect SRSF5 binding affinity. In this assay, we used SRSF5-ΔRS to eliminate potential interference by RS domain phosphorylation. Thus, we incubated purified recombinant His×6-SRSF5-ΔRS (Supplementary Fig. 4C) with in vitro synthesized RNA fragments of WT and its mutants in DMTF1 In9 (Fig. 5D, lower panel). Among these mutants, mut1, 2 and 3 harboured deletions of SRSF5-bs2, -bs1, or both, respectively, while mut4 held mutated SRSF5-bs2 and was used in later experiments. As shown in Fig. 5G (upper panel), SRSF5-ΔRS could bind WT, mut1 and 4, but not mut 2 and 3, suggesting that SRSF5-bs1 (or βGAA) was an essential element for SRSF5 binding. To assess potential interference in qPCR data, we carried out this assay using heat-inactivated SRSF5-ΔRS, which showed no significant difference in binding these RNA fragments (Fig. 5G, middle panel). Additionally, when the assay was performed without adding SRSF5-ΔRS and reverse transcriptase, we detected a double-digit drop of the signal compared to the standard reactions (Fig. 5G, lower panel). SRSF5 binding to DMTF1 intron 9 was also verified by REMSA. We synthesized FAM-labelled RNA probes with intact SRSF5-bs1 and -bs2 (WT) and their individual or simultaneous deletions (mut1, 2 and 3) based on the sequences in Fig. 5D. Purified His×6-SRSF5-ΔRS (1 μg) was incubated with 5 pmol of each RNA probe, followed by native polyacrylamide gel analysis. The sample using WT probe showed a slowly migrated band in middle region of the gel, which was likely the His×6-SRSF5-ΔRS/probe complex, because it exhibited reduced intensity in other three samples, especially mut3 with both SRSF5-bs1 and -bs2 deleted (Fig. 5H). When a His-tag antibody was added in the binding assay, a super-shifted band was only observed in WT and mut1 and 2, but with monotonically reduced intensity, suggesting that both SRSF5-bs1 and -bs2 were involved in SRSF5 binding, and -bs1 had higher affinity than -bs2 (Fig. 5H). A band below the super-shifted complex was also detected, which was likely non-specific band because of its intensity irresponsive to His-tag antibody. Consistently, with SRSF5 knockdown by its ASO-3, mut3 minigene, lacking both SRSF5-bs1 and -bs2, showed no difference of DMTF1 isoform splicing (Fig. 5I).

Overall, our data indicated that SRSF5 could bind SRSF5-bs1 and -bs2 to regulate DMTF1 pre-mRNA alternative splicing.

3.5. SRSF5 binding changes SF1 binding preference in Intron 9 and promotes DMTF1β and γ isoform splicing

Using the SpliceAid 2, we identified three potential SF1 binding sites in In9 of DMTF1 pre-mRNA, designated as α-SF1-bs, β-SF1-bs and γ-SF1-bs, based on their closeness to the three splice sites (Fig. 6A). In addition, we discovered one potential SF1 binding site in In8, located just upstream of Ex9 and designated as In8-SF1-bs. Interestingly, both α- and β-SF1-bs enclose the GCUGC motif, corresponding to the SRSF5-bs2 and -bs3, respectively (Fig. 6A), suggesting that SF1 binding to these two sites could potentially be blocked by SRSF5. To prove this prediction, we carried out CLIP experiments using Flag-Ab beads in the lysates of HeLa cells transfected by Flag-SF1 and HA-SRSF5 or an empty vector. With random hexamers as primers in RT, we individually amplified the DNA fragments covering the α-SF1-bs, β/γ-SF1-bs and In8-SF1-bs using three primer sets (PriP3, PriP2 and PriP1, respectively) that were specifically designed for these regions (Fig. 6A and Supplementary Table 2). As shown in Fig. 6B, the presence of HA-SRSF5 significantly reduced SF1 binding to α-SF1-bs, but markedly enhanced its binding to β/γ-SF1-bs as compared to the vector control group. The In8-SF1-bs region showed similar binding affinity to Flag-SF1 irrespective of the SRSF5 presence or not, consistent with its lack of SRSF5 binding motifs in its proximity. The bar values of the three regions in Fig. 6B represent the data normalized against 10% of corresponding inputs and thus the heights of all bars could indicate their relative binding affinity in the whole region. Thus, SF1 binding increase to β/γ-SF1-bs caused by HA-SRSF5 was very substantial.

Figure 6.

Effects of SRSF5 on SF1 binding and activity in modulating DMTF1 pre-mRNA splicing. (A) Schematic diagram of the region from In8 to Ex10 region of DMTF1 pre-mRNA with labelled binding sites of SRSF5 and SF1. The splicing patterns of DMTF1α, β and γ are indicated on the top. The three regions of qPCR with positions of their primer pairs (PriP1, PriP2 and PriP3) are labelled as paired arrows beneath the intron-exon diagram. The sequences of the three SRSF5 binding sites (SRSF5-bs1, -bs2 and -bs3) and four SF1 binding sites (α-, β-, γ- and In8-SF1-bs) are shown, and their sequences are marked. (B) and (C) CLIP studies to examine SF1 binding in In9 of DMTF1 pre-mRNA. In (B), Flag-SF1 vector was cotransfected with HA-SRSF5 or an empty vector into HeLa cells. Flag-SF1 and HA-SRSF5 proteins in the input and IPed samples were examined by Western blot analysis (left panel). The IPed endogenous RNA levels of the In8-, β/γ- and α-SF1-bs were assessed by qPCR using the primer pairs Pri-P1, Pri-P2 and Pri-P3, respectively. In (C), Flag-SF and HA-SRSF5/vector were cotransfected with all five WT and mutant minigene reporters simultaneously into HeLa cells. The IPed RNA covering the α-SF1-bs and SRSF5-bs1 region in the WT, mut1, 2, 3 and 4 shown in Figure 5D was quantified by RT-qPCR using random hexamers as primers in RT and individually designed specific primer pairs (Supplementary Table 2) in qPCR. The qPCR value of each IPed sample was normalized by the qPCR value using the same primer pair with 10% input as the template. (D) Effects of SRSF5 on DMTF1 pre-mRNA splicing in minigene reporters. Flag-SRSF5 or an empty vector was cotransfected with the WT or mutant minigene reporters individually into HeLa cells. The quantification of DMTF1 isoform transcripts were determined using the strategy and primers shown in Figure 2B, with the total transcript levels assessed by qPCR using the pri-a and pri-b primers. In (C) and (D), the significant interaction between ectopically expressed SRSF5 and different minigenes was validated by the two-way ANOVA, p < 0.0001. All data are shown as mean values ± SEM from triplicated samples and the experiments were repeated at least three times with similar results

A previous study indicated that the GAAGAAGA motif was a key element for SRSF5 recognition in a pre-mRNA [60], which is presented as two overlapped sequences in SRSF5-bs1 inside the whole βGAA element, as we defined above in Fig. 1B. Thus, SRSF5-bs1 likely played an essential role in SRSF5-mediated regulation of DMTF1 pre-mRNA splicing. Since α-SF1-bs and SRSF5-bs1 were near each other, we examined their potential interplay by the minigenes listed in Fig. 5D. Among these minigene reporters, mut4 had its α-SF1-bs (containing the SRSF5-bs2) replaced by γ-SF1-bs sequence, leading to elimination of SRSF5-bs2. Importantly, the mut4 transcript showed SRSF5-ΔRS binding affinity comparable to that of the WT (Fig. 5E and 5G), suggesting that SRSF5-bs1 alone was sufficient to recruit SRSF5.

To evaluate SF1 binding to the transcripts of the minigenes, we transfected Flag-SF1 with all these minigene reporters simultaneously into HeLa cells, followed by CLIP studies using Flag-Ab beads. With random hexamers as RT primers, the qPCR value of each IPed sample by its specific primer set (Fig. 5D) was normalized by the qPCR value using the same primer set with 10% input as the template. As shown by the black bars of Fig. 6C, mut1 and 3, which lacked α-SF1-bs (overlapping with SRSF5-bs2), exhibited expected CLIP-qPCR reduction of the α-SF1-bs and SRSF5-bs1 region compared to WT and mut2. Importantly, mut4 displayed similar signal to the WT, suggesting that the γ-SF1-bs sequence at the α-SF1-bs site could restore Flag-SF1 binding to this region.

To evaluate the effects of SRSF5 on SF1 binding to its recognized elements in DMTF1 In9, we transfected an HA-SRSF5 expression plasmid into HeLa cells together with Flag-SF1 and all five minigene vectors simultaneously, followed by the CLIP assay using Flag-Ab beads. The sample was then analysed by the same RT-qPCR strategy (i.e. minigene specific primer sets) as shown in Fig. 5D to quantify the RNA IPed by Flag-SF1.

As shown by the grey bars in Fig. 6C, the signal of mut1 and mut3, with deleted α-SF1-bs, represented non-specific binding of Flag-SF1 in this region. With the WT minigene reporter, HA-SRSF5 could very efficiently block Flag-SF1 binding to this region. Interestingly, in mut2 with deleted SRSF5-bs1, this blockage was significantly attenuated, but not completely abolished, suggesting that HA-SRSF5 could bind to SRSF5-bs2 (overlapping with α-SF1-bs) in the absence of its adjacent SRSF5-bs1, but with markedly reduced affinity compared to that of SRSF5-bs1, leading to partially blocked Flag-SF1 binding to α-SF1-bs. This prediction was proved by the result of mut4; with α-SF1-bs being replaced by γ-SF1-bs, which does not contain any SRSF5 binding element, mut4 lacked response to SRSF5 expression (Fig. 6C).

To evaluate how SRSF5 could regulate DMTF1 splicing in In9 in the reporter system, we individually transfected the minigene vectors shown in Fig. 5D with Flag-SRSF5 or an empty vector, followed by RT using the RT primer shown in Fig. 2B and qPCR using the isoform-specific primers together with a primer in pcDNA3. With ectopically expressed Flag-SRSF5, the WT minigene displayed reduced α isoform splicing and increased β/γ splicing (Fig. 6D), suggesting that the binding of SRSF5 to SRSF5-bs1, i.e. the βGAA element, could enhance the β/γ isoform expression and consequently reduce levels of the α isoform (Fig. 6D). Among WT and all minigene mutants, α splicing showed generally high levels. Flag-SRSF5 could cause its fluctuation, but the changes were not statistically significant, except in WT. These data suggested that neither α-SF1-bs nor SRSF5-bs1 was essentially required in maintaining α isoform splicing.

Compared to the WT, both mut1 and mut4 retained about half levels of β and γ, irrespective of Flag-SRSF5 cotransfection, indicating a crucial role of the α-SF1-bs/SRSF5-bs2 region for β/γ splicing (Fig. 6D). With the deletion of SRSF5-bs1, mut2 and mut3 virtually lost β and γ splicing ability, consistent with the essential role of both SRSF5 and SRSF5-bs1 (or the βGAA element) in β and γ isoform expression.

We analysed the CLIP data brought down by Flag-SF1 (Fig. 6C) using a two-way ANOVA analysis to compare the value of each sample with that of each of the other nine samples. In this analysis, the presence or absence of mutations in minigenes (Fig. 5D) and ectopic SRSF5 expression were set as two individual observational variables. The results demonstrated significant interaction between these variables (p < 0.0001). In addition, in the evaluation of each isoform proportion from these minigenes shown in Fig. 6D, the two-way ANOVA analysis with these two variables was also used to compare the value of each sample with that of each of the other nine samples. The results of all three isoforms also indicated significant interaction between these variables (p < 0.0001). Overall, the statistical analyses suggested that SRSF5 could specifically impact SF1 recognition on DMTF1 pre-mRNA transcribed by WT minigene and concurrently interfere with its splicing decision, due to the presence of the α-SF1-bs and SRSF5-bs1 in WT minigene.

Alterations in RNA structure can mediate pre-mRNA splicing [65]. We analysed DMTF1 splicing (Supplementary Fig. 5A) by the RNAStructure 6.2 software, which uses algorithms to predict secondary RNA structure [66]. With the input of the Ex9-In9-Ex10 sequence, the software predicted multiple interactions between the region close to the 3ʹ-donor site of Ex9 and the region near the α-acceptor site (Supplementary Fig. 5B). Meanwhile, the β/γ-acceptor sites were generally covered or protected by other interactions, without direct access to Ex9. However, the binding of SRSF5 to SRSF5-bs1 markedly changed the interaction pattern between the region of the α/β/γ-acceptor sites and the region of the Ex9 3ʹ-donor site (Supplementary Fig. 5C), greatly increasing the splicing chances of the β/γ-isoforms. Thus, the prediction of RNAStructure 6.2 based on the Ex9-In9-Ex10 sequence is consistent with our experimental data indicating that SRSF5 can promote DMTF1β/γ splicing.

Lu, et al. developed a method, PARIS, based on reversible psoralen crosslinking of RNA and used it to globally map RNA duplexes or interactions with a base-pair resolution in living cells [67]. We analysed the GSE74353 dataset of this study and detected two non-gapped, isolated clusters of RNA structures in the vicinity of DMTF1 splicing sites. Neither of them showed interaction with other RNA regions (Fig. 7B), indicating that the two regions (the γ-splicing site and α-SF1-bs/SRSF5-bs1) could form individual or separate RNA structures.

Figure 7.

Analyses of the datasets of RNA duplexes mapping and SRSF5 CLIP-seq. (A) Schematic diagram of the region from exon 9 to exon 10 of DMTF1 pre-mRNA. The splicing patterns of DMTF1α, β and γ isoforms are indicated by folded lines, and the length of the analysed region is shown below the diagram. (B) Analysis of the RNA duplexes or interaction mapping of the DMTF1α, β and γ splice site region based on a dataset [67]. Two clusters of RNA interaction reads are indicated by a red bar (Cluster 1) and a green bar (Cluster 2). (C) Analysis of SRSF5 binding sites based on a SRSF5 CLIP-seq dataset [34]. Both splicing sites and splicing elements are indicated at the bottom

Recently, a CLIP-seq (crosslinking and immunoprecipitation-sequencing) based approach was developed to quickly identify RNA–protein interactions in cells, which was used to analyse RNA binding of various SRSF proteins, including SRSF5 [34]. To corroborate the binding of SRSF5 in In9 of DMTF1 pre-mRNA, we analysed the CLIP-seq data of this study and observed highly enriched SRSF5 binding reads in the region of the α-SF1-bs and SRSF5-bs1 (Fig. 7C), which was consistent with the region of high SRSF5 binding affinity determined in our CLIP studies. Noteworthily, the cluster 2 RNA structure reads and SRSF5 binding reads showed partial overlapping (Fig. 7). This suggested that SRSF5 binding likely interferes with or disrupts the RNA structure in cluster 2, which is in line with our prediction that SRSF5 binding to α-SF1-bs/SRSF5-bs1 may alter the RNA structure to favour DMTF1β and γ splicing.

Based on our experimental data and bioinformatic analyses, we proposed a mechanistic model of SRSF5-promoted DMTF1β and γ splicing (Fig. 8). In normal cells with relatively low levels of SRSF5, SF1 can bind to α-SF1-bs of DMTF1 pre-mRNA and create RNA structures that make β/γ splice sites inaccessible for splicing  (upper panel, Fig. 8). In cancer cells with relatively high SRSF5 levels, it binds to the SRSF5-bs1 (or βGAA) site. This can both prevent SF1 from binding to α-SF1-bs and disrupt the RNA structures in this region, which allows SF1 to bind its consensus element close to the β/γ splice sites and promote the splicing of DMTF1β and γ isoforms (lower panel, Fig. 8).

Figure 8.

Schematic diagram of SRSF5-promoted DMTF1β and γ splicing through modulating RNA structure and SF1 binding. Upper panel: when SRSF5 protein levels are relatively low, SF1 binds to the α-SF1-bs (UGCUGCC) of DMTF1 pre-mRNA and recruits the Spliceosome to the vicinity. This makes the region of β/γ splice sites form RNA structures (represented by two loops) that prevent the splicing of DMTF1β and γ isoform, but did not interfere with DMTF1α splicing (shown by red dash line). Lower panel: in cancer cells with relatively highly SRSF5 levels, it binds to the SRSF5-bs1 (or βGAA) site to prevent SF1 from binding to α-SF1-bs. The SRSF5 binding also disrupts the RNA structures in this region, which allows SF1 to bind its consensus elements close to the β/γ splice sites and recruit the Spliceosome to promote the splicing of DMTF1β and γ isoforms

4. Discussion

The DMTF1 gene was discovered in 1998 by Inoue, et al. as a D-type cyclin-associated transcription factor [68], while its alternative splicing was first described in 2003 by Tschan, et al. [2]. Since then, several groups, including ours, reported the functional roles of DMTF1 isoforms in cancer development [1]. Among the three isoforms, DMTF1α plays a tumour suppressive role in several cancers, including breast, lung and bladder cancers [5,69,70]. We revealed increasing expression of DMTF1β in breast cancer samples compared to normal mammary tissues and its inverse correlation with breast cancer patients’ survival rates. Importantly, we demonstrated an oncogenic role of DMTF1β in breast cancer initiation using a transgenic mouse model [9]. Recently, we reported that DMTF1γ exerted similar activities to DMTF1β in antagonizing DMTF1α, DMTF1β/γ proteins had shorter half-lives than α, and nonsense-mediated decay (NMD) could likely contribute to relatively low expression of DMTF1β/γ [11]. Despite the progress in the functional characterization of DMTF1 isoforms, the molecular mechanisms underlying their alternative splicing remain mysterious. As DMTF1α and β play distinct or opposite roles during mammary oncogenesis, dissecting the process and regulatory factors of DMTF1 pre-mRNA alternative splicing can provide insights into designing novel intervening strategies to reduce β-isoform splicing and attenuate cancer progression.

In this study, we first utilized bioinformatic approaches to identify regulatory sequences in the vicinity of the three DMTF1 acceptor splice sites (Figs. 1 and 2) and discovered several proteins that could potentially bind these elements (Fig. 3). Based on gene expression correlations, reporter assays and evaluation of endogenous DMTF1 isoform levels, we narrowed down these regulatory candidates to SRSF5 (Figs. 3 and 4), which is overexpressed in various cancers and promotes aberrant splicing of multiple cancer-related genes [30,31,36,37]. Our functional studies verified SRSF5 as a key regulator in the alternative splicing of DMTF1 pre-mRNA through binding to its consensus motif βGAA, a splicing enhancer element, near the α acceptor site, which alters SF1 protein-binding preference to promote the splicing of DMTF1β and γ (Figures 5 and 6). Our study is the first report to uncover the regulatory mechanism of DMTF1 alternative splicing.

We identified SRSF5-bs1 as a primary-binding site of SRSF5 in In9 of DMTF1 pre-mRNA. Consistently, SRSF5-bs2 deletion did not alter SRSF5 binding to transcripts of DMTF1 minigene reporter constructs compared to WT (Fig. 5E,G,H). In addition, it is reasonable to predict that RRM2 binding to SRSF5-bs1 may promote RRM1 binding to SRSF5-bs2 inside the α-SF1-bs, which blocks the access of SF1 to this region and causes its relocation to the β/γ-SF1-bs to promote β/γ isoform splicing (Fig. 6). In recent years, the contribution of RNA structure to pre-mRNA alternative splicing has drawn increasing attention from this field [71,72]. A successful splicing requires a donor site and an acceptor site to be sterically adjacent. The preferential splicing of DMTF1α among the three isoforms in normal cells should meet this prerequisite, which must be altered or at least interfered with by the binding of SRSF5, which is increasingly expressed in cancers. Especially, with SRSF5 binding to both SRSF5-bs1 and bs-2, the agitation of RNA structure proximate to the α/β/γ-acceptor sites would be extensive, possibly permitting the 3ʹ-end of Ex9 to reach the β/γ-acceptor sites and consequently increasing β/γ-splicing, as shown in Fig. 8. This projection has been verified by our analyses using RNAStructure 6.2, which predicted that SRSF5 binding to SRSF5-bs1 could create chances of interaction between the Ex9 3ʹ-donor site and β/γ-acceptor sites (Supplementary Fig. 5). Meanwhile, the analyses in the datasets of RNA duplexes mapping and SRSF5 CLIP-seq also supported our proposed model that SRSF5 binding to the SRSF5-bs1/2 region likely altered the RNA structure to favour DMTF1β and γ splicing (Fig. 7).

Pre-mRNA splicing happens co-transcriptionally, and the transcription process can greatly impact alternative splicing decision [73]. A potential splice acceptor site may be preferentially used if it is transcribed prior to other, even stronger, splice sites. However, in the case of DMTF1 pre-mRNA, the γ and β splice sites are located upstream and transcribed before the α site. Thus, co-transcriptional splicing decision is clearly not a mechanism to explain the predominant α isoform splicing by DMTF1 pre-mRNA. We also analysed the three splice sites of DMTF1 pre-mRNA using the maximum entropy modelling algorithm that predicts splicing strength of RNA motifs [74]. To our surprise, α splice site of DMTF1 pre-mRNA gave a score of 4.73, much lower than those of β and γ splice sites (11.78 and 8.13, respectively). The fact that precedingly transcribed β and γ sites with relatively high splice strength are unfavourably spliced compared to the later transcribed α site strongly suggests that additional regulatory factors, most likely a special RNA structure formed by the β/γ acceptor region, prevent β and γ site splicing.

SRSF5 has two RNA recognizing domains, RRM1 and RRM2, recognizing distinct-binding elements [21,44,72]. RRM1 at the N-terminal of SRSF5 recognizes a pyrimidine-containing sequence, while the RRM2 in the middle region binds GGA-rich sequences [30]. Thus, the SRSF5-bs1 (GAAGAAGAAGAAUGAAGAA), containing six GAA repeats in a 19-nt stretch, is likely the primary binding site of SRSF5 using its RRM2 domain, while SRSF5-bs2 (GCUGC) binding to the RRM1 domain possibly has relatively low affinity. How the RS domain and its phosphorylation contribute to SRSF5’s activity of promoting pre-mRNA splicing is controversial [62,63]. In this study, our data supported that the RS domain was dispensable for both SRSF5 binding to the In9 and its regulation of DMTF1 splicing, favouring the model of two previous reports [63,64]. In addition to phosphorylation, SRSF5 was also reportedly modified by acetylation, ubiquitination and methylation [30,33], but it is unclear whether these posttranslational modifications modulate SRSF5 activity during DMTF1 splicing.

Multiple studies demonstrated upregulated SRSF5 expression in cancer cells [30,35–37], consistent with its proliferative character during oncogenesis. When detecting endogenous protein levels in different mammary cell lines, we observed relatively low expression of SRSF5 protein in MDA-MB-231 cells (Fig. 4C), consistent with a previous study showing its lower levels in MDA-MB-231 cells than those in MCF-7 cells [31]. The specific genetic or epigenetic characters leading to SRSF5 downregulation in MDA-MB-231 cells need future investigation.

In the current study, we demonstrated that the βGAA element was crucial in determining the preference of DMTF1 pre-mRNA splicing through recruiting SRSF5 to In9. Similar to our finding, Moschall, et al. reported that the presence and absence of a GAA-repeat element in the foamy viral genome decided the positions of 3ʹ-splice acceptor sites during env splicing through interfering with the branch point recognition by SF1 [75]. In addition, our observation that SRSF5 binding to the βGAA element near the α acceptor splice site in In9 to disfavour DMTF1α isoform production was consistent with previous reports showing that intronic binding of SR proteins typically interfered with splicing processes in their vicinity through preventing the recruitment of ribonucleoproteins [76,77].

When investigating regulatory factors that could bind the enhancer and silencer elements discovered by us in In9 of DMTF1 pre-mRNA, we identified several protein candidates, including SRSF5, TIA1, DAZAP1 and TRA2B. Although we chose SRSF5 for further investigation, the other three proteins also possess potential in modulating DMTF1 splicing. The possible regulatory activities of these proteins in DMTF1 splicing deserve future investigation. In the current study, we identified several regulatory elements, but only extensively investigated the role of βGAA in mediating DMTF1β and γ splicing. The contributions of other elements, especially the γU, to DMTF1 pre-mRNA splicing remain undetermined and need to be characterized in future studies.

5. Conclusions

DMTF1α is an important tumour suppressor, while its alternative splicing isoform DMTF1β plays an oncogenic role in mammary oncogenesis. In the current study, we demonstrated SRSF5 as a key regulator binding to a splicing enhancer element in DMTF1 intron 9 to promote DMTF1β and γ splicing, and consequently reduce DMTF1α expression. SRSF5 exerted this activity through altering RNA structure and modulating SF1 binding to favour DMTF1β and γ expression. Our study for the first time revealed a molecular mechanism underlying DMTF1 pre-mRNA splicing and supported an oncogenic role of SRSF5 in breast cancer development. Importantly, our discovery provided insights into designing novel therapeutic strategies in breast cancer treatment through intervening DMTF1 splicing.

Funding Statement

This work was supported by the Fundamental Research Funds for the Central Universities [2572017AA14] to JL and the National Natural Science Foundation of China [81872293 and 81672795] to GS.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Consent for publication

All authors have agreed on the contents of the manuscript.

Data availability statement

All data generated or analysed during this study are included either in this article or in the supplementary information files.

Author’s contributions

J.L. and G.S. conceived the project, wrote the manuscript and generated the figures. J.L., Y.Q., Y.L., H.W. and K.S. conducted experiments. J.L. and G.L. analysed bioinformatic data. D.L. and J.S. provided technical support to several key experiments. D.S. critically read the manuscript and provided conceptual comments.

Supplementary material

Supplemental data for this article can be accessed here.