SPF45/RBM17-dependent, but not U2AF-dependent, splicing in human short introns


 Human pre-mRNA introns vary in size from under fifty to over a million nucleotides. We searched for essential factors specifically involved in the splicing of human short introns by screening siRNAs against 154 human nuclear proteins for activity on a model short 56-nucleotide intron-containing HNRNPH1 pre-mRNA. We identified a known alternative splicing regulator SPF45 (RBM17) as a general splicing factor that is essential to splice out this 56-nt intron. Whole-transcriptome sequencing of SPF45-deficient cells revealed that SPF45 is specifically required for the efficient splicing of many short introns. Our crosslinking and biochemical analyses demonstrate that SPF45 specifically replaces the U2AF heterodimer on the truncated poly-pyrimidine tracts in these short intron. To initiate splicing, the U2AF-homology motif (UHM) of the replaced SPF45 interacts with the UHM-ligand motif (ULM) of the U2 snRNP protein SF3b155 (SF3B1). We propose that splicing in a distinct subset of human short introns depends on SPF45 but not U2AF heterodimer.

Human pre-mRNA introns vary in size from under fifty to over a million nucleotides. We searched for essential factors specifically involved in the splicing of human short introns by screening siRNAs against 154 human nuclear proteins for activity on a model short 56-nucleotide intron-containing HNRNPH1 pre-mRNA. We identified a known alternative splicing regulator SPF45 (RBM17) as a general splicing factor that is essential to splice out this 56-nt intron. Whole-transcriptome sequencing of SPF45-deficient cells revealed that SPF45 is specifically required for the efficient splicing of many short introns. Our crosslinking and biochemical analyses demonstrate that SPF45 specifically replaces the U2AF heterodimer on the truncated poly-pyrimidine tracts in these short intron. To initiate splicing, the U2AF-homology motif (UHM) of the replaced SPF45 interacts with the UHM-ligand motif (ULM) of the U2 snRNP protein SF3b155 (SF3B1). We propose that splicing in a distinct subset of human short introns depends on SPF45 but not U2AF heterodimer.
There is a remarkable pattern in the distribution of higher eukaryotic pre-mRNA intron length; most introns fall either within a narrow peak under one hundred nucleotides or in a broad distribution peaking around several thousand nucleotides and extending to over a million nucleotides [1][2][3] . Pre-mRNA splicing is dependent upon a set of signal RNA elements recognized by essential factors that is a ubiquitous and essential part of eukaryotic gene expression. However, our understanding about specific and distinct mechanisms for the precise recognition of degenerated 5¢ and 3¢ splice site sequences within such extensively varied length of introns is fairly limited.
The canonical splicing mechanisms were studied and established using model pre-mRNAs with a single relatively short intron of a few hundred nucleotides, which are efficiently spliced in cells and in vitro 4,5 . According to such optimal systems, the essential splicing sequences in pre-mRNA, namely the 5′ splice site, the branch-site sequence, and the poly-pyrimidine tract (PPT) followed by the 3′ splice site, are initially recognized by the U1 snRNP, SF1, and the U2AF heterodimer (U2AF 65 /U2AF 35 , U2AF2/U2AF1 as HGNC approved symbol), respectively. Following the assembly of this early spliceosomal E complex, SF1 is replaced by the U2 snRNP in the A complex, which commits the intron for splicing reaction (reviewed in Ref. 6 ). The A-complex is an asymmetric globular particle (~26 × 20 × 19.5 nm) 7 that fully occupies 79-125 nucleotides (nt) of RNA 8 , and recent high-resolution cryo-electron microscopy structures of the A-complex have revealed molecular details of the overall architecture (reviewed in ref. 9 ). Interestingly, human ultra-short introns with much shorter lengths (43-65 nt) are nevertheless spliced 10,11 . This raises the question of how such ultrashort introns can be recognized and committed to splicing by an 'oversized' A complex without steric hindrance.
We postulate that splicing of short introns depend on distinct specific factors, which utilize alternative ways for early spliceosomal assembly. Here, we have shown that this is the case in a subset of human short introns with the truncated PPT, which is recognized by a novel constitutive splicing factor SPF45, but not by the authentic U2AF heterodimer.

Results
SPF45 is a novel essential splicing factor for a subset of short introns. To find potential factors involved in splicing of short introns, we screened an siRNA library targeting 154 human nuclear proteins for splicing activity of the HNRNPH1 pre-mRNA including 56-nt intron 7 10,11 . Many known RNA-binding proteins and splicing factors could be tested with this siRNA library (Table S1).
HeLa cells were transfected with each siRNA and recovered total RNAs were analyzed by RT-PCR to examine splicing activity of the endogenous HNRNPH1 pre-mRNA which contains a 56-nt intron (Fig. 1, left panel). The strongest splicing repression was markedly caused by knockdown of SPF45 (RBM17 as HGNC approved symbol; right panel) that indeed effectively depleted SPF45 protein (Fig. 1, middle panel). To test if SPF45 might have a general role in splicing of short introns, we assayed two other endogenous pre-mRNAs targeting the 70-nt (intron 9 of RFC4) and the 71-nt (intron 17 of EML3). Both introns were also significantly repressed in SPF45-depleted HeLa cells (Fig. 1, right 2 panels).
Splicing inhibition was proportional to SPF45-knockdown efficiency induced by independent siRNAs (Supplementary Fig. S1a). These SPF45 siRNA-induced splicing defects were also observed in HEK293 cells, testifying to the robustness of our results ( Supplementary Fig. S1b).
To test our hypothesis that SPF45 is specifically required to splice out short introns, we performed whole-transcriptome sequencing (RNA-Seq) with RNA from the SPF45-deficient HEK293 cells. The sequencing reads were mapped to the human genome reference sequence. We identified 517 changes in splicing from a total of 47,960 alternative splicing events (  Table S2 for the list of all 187 introns).
The analysis of these retained introns hinted at a potential mechanism for the role of SPF45. Remarkably, the length distribution of the retained-introns in SPF45-depleted cells is strongly biased towards shorter lengths compared to those in cells depleted of constitutive splicing factors, U2AF 65 and SF3b155 (SF3B1 as HGNC approved symbols), which show a distribution comparable to the whole set of introns (Fig. 2b).
We validated these RNA-Seq-based profiles by RT-PCR. As assumed, splicing of pre-mRNAs with two control introns were not affected by SPF45-knockdown, while in contrast, three arbitrarily chosen pre-mRNAs with short introns were repressed (Fig. 2c). These results demonstrate that SPF45 is required for the efficient splicing of a substantial population of pre-mRNAs with short introns.

SPF45 is required for splicing on intron with truncated poly-pyrimidine tract (PPT).
Next we searched for a potential cis-element in short introns through which SPF45 might act. From RNA-Seq data of SPF45-depleted cells, we found that strengths of the 5′/3′ splice sites and the branch sites of SPF45-dependent short introns are somewhat weaker than the average in RefGene (Supplementary Fig. S2a). Therefore, we first examined these cis-acting splicing signals using mini-gene splicing assays in SPF45-depleted HeLa cells. As expected, splicing of pre-mRNA containing HNRNPH1 56-nt intron 7 was repressed by depletion of SPF45, whereas splicing of the control adenovirus 2 major late (AdML) pre-mRNA (231-nt intron 1), which was used as a standard splicing substrate previously, was unaffected ( Fig. 3; top 2 panels). The SPF45-dependent splicing of the HNRNPH1 pre-mRNA was not altered even after replacement of either the 5′/3′ splice sites or the branch site by those of the AdML pre-mRNA ( Supplementary Fig. S2b). These results indicate that the requirement of SPF45 depend on neither the 5′/3′ splice sites nor the branch site.
We then examined whether the SPF45-dependency is attributed to the PPT. The PPT score (see Experimental Procedures) is one of the criteria to evaluate effective PPTs: PPT scores are 19 for the PPT (13 nt) in HNRNPH1-intron 7 and 52 for the PPT (25 nt) in AdML-intron 1 (Fig.  3, second panel). Remarkably, splicing of the SPF45-dependent HNRNPH1 pre-mRNA was altered toward that of an SPF45-independent pre-mRNA by replacement of the HNRNPH1-PPT with the conventional AdML-PPT (Fig. 3, 'AdML PPT25'). To determine whether SPF45 recognizes the strength or the length of a given PPT, we reduced the PPT score of AdML-PPT in two ways: one was transversion mutations in the PPT (C/UàG; score 52à32), and the other was truncation of the PPT (25 ntà13 nt; score 52à30). Notably, the transversion mutations in the PPT did not cause SPF45-dependency (Fig. 3, 'AdML PPT25mt') but the truncation of PPT did (Fig. 3, 'AdML PPT13'). Lastly, we expanded the distance between the 5′ splice site and the branch site in HNRNPH1 intron (27 nt) by replacement with the corresponding fragment in the AdML intron (192 nt). Interestingly, this chimeric pre-mRNA with the short HNRNPH1 PPT remained SPF45 dependent (Fig. 3, 'AdML 5′MT'). Taken together, these results demonstrate that short PPT per se in the HNRNPH1 intron 7 is the determinant for the SPF45-dependency in splicing.
These observations were further recapitulated and validated in the distinct SPF45-dependent EML3 pre-mRNA which contains a 71-nt intron ( Supplementary Fig. S3a). Moreover, our global PPT length analysis of the retained introns in SPF45-depleted cells showed that PPT lengths of SPF45-dependent short introns (<100 nt) were significantly shorter than those of the whole RefGene introns ( Supplementary Fig. S3b). Together, we conclude that it is the truncated PPTs in short introns that are crucial for the SPF45 function in splicing.
SPF45 replaces U2AF 65 on truncated PPTs in short introns to promote splicing. In an early transition of human spliceosome from the E to A complexes, the branch site, PPT and the 3′ splice site are bound cooperatively by SF1, U2AF 65 and U2AF 35 , respectively. A stable U2 snRNP-associated splicing complex is then formed by ATP-dependent displacement of SF1, where the p14 protein (a U2 snRNP-associated factor) contacts the branch site, and interaction of U2AF 65 with the U2 snRNP component SF3b155 (reviewed in ref. 6 ). Since the splicing activity of SPF45 depends on short PPTs, we hypothesized that insufficient U2AF 65 binding to a truncated PPT would allow its replacement by SPF45.
To examine whether SPF45 associates with short introns, mini-genes encoding HNRNPH1, EML3, MUS81, and control AdML introns, were expressed in HeLa cells. We confirmed that these mini-genes were spliced in an SPF45-dependent manner ( Fig. 3 and Supplementary Fig.  S4) as we observed in these endogenous genes (Figs. 1 and 2c), so we proceeded to analyze proteins associated with these ectopically expressed RNAs.
We used formaldehyde crosslinking to detect any indirect RNA association of SPF45 12 . HeLa cells were co-transfected with the four mini-genes and Flag-SPF45 expression plasmid, crosslinked with formaldehyde, immunoprecipitated with anti-FLAG antibody, and the coprecipitated RNAs were analyzed by RT-qPCR (Fig. 4a). SPF45 bound to all pre-mRNAs derived from these four mini-genes irrespective of the length of the introns. The binding of SPF45 to control AdML intron (231 nt) is consistent with previous proteomic detection of SPF45 in the human spliceosome 13 . SPF45 was also detected on MINX (131 nt) and PM5 (235 nt) introns 14,15 (reviewed in ref. 6 ).
We next examined the binding of U2AF 65 and SF3b155 to pre-mRNAs by UV crosslinkingimmunoprecipitation (CLIP).
Whole-cell extracts from crosslinked cells were immunoprecipitated with anti-U2AF 65 and anti-SF3b155 antibodies, and the precipitated RNAs were analyzed by RT-qPCR (Fig. 4b). SF3b155, as a component of the U2 snRNP, bound to all the pre-mRNAs, while significant U2AF 65 binding was observed only with control AdML pre-mRNA and it was very weak with the three SPF45-dependent short introns.
We biochemically verified the binding of U2AF 65 and SF3b155 to the HNRNPH1 intron and control AdML intron in HEK293 cell nuclear extracts by affinity pull-down assay using biotinylated pre-mRNAs (Fig. 4c). In agreement with our crosslinking experiments, SPF45 associated with both AdML and HNRNPH1 pre-mRNAs. However, importantly, U2AF 65 strongly bound only to AdML pre-mRNA and U2AF 65 was allowed to associate HNRNPH1 pre-mRNA only if SPF45 was depleted from nuclear extracts. These results together support our proposed hypothesis that SPF45 replaces U2AF 65 in the assembly of U2 snRNP complexes as U2AF 65 is poorly bound to truncated PPTs of short introns.
We noticed that endogenous U2AF 65 -knockdown barely repressed splicing of the SPF45dependent short intron (Supplementary Table S1, No. 142; Supplementary Fig. S5a). Therefore, we checked splicing efficiencies of these four mini-genes in U2AF 65 -knockdown HeLa cells (Fig. 5). This depletion of U2AF 65 also caused effective co-depletion of U2AF 35 (Fig. 5, left panel) that is consistent with previous reports 16,17 . In the control AdML mini-gene, spliced mRNA was reduced by the depletion of U2AF 65 , showing that U2AF heterodimer is essential for conventional AdML pre-mRNA splicing as expected. Remarkably, splicing of SPF45-dependent pre-mRNAs with short introns was rather activated by the depletion of U2AF 65 (Fig. 5, right 3 panels).
In endogenous SPF45-dependent pre-mRNAs ( Supplementary Fig. S5a), such marked activation was not observed that could be due to the almost saturated efficiency of splicing (see amounts of unspliced pre-mRNAs). Taken together, we conclude that SPF45 effectively competes out U2AF heterodimer on truncated PPTs and the newly installed SPF45 promotes splicing of pre-mRNAs with short introns.
SF3b155-U2AF 65 /U2AF 35 is displaced by SF3b155-SPF45 via ULM-UHM binding. The SPF45 protein contains a G-patch motif that may interact with nucleic acids and proteins 18,19 , and a C-terminal U2AF-homology motif (UHM) that binds the UHM-Ligand motifs (ULM) of its partner proteins. UHM-ULM interactions; e.g., U2AF 65 -UHM-SF1-ULM, U2AF 65 -UHM-SF3b155-ULM, and U2AF 35 -UHM-U2AF 65 -ULM, plays an essential role in the splicing reactions [20][21][22] (Reviewed in ref. 23 ). Remarkably, in vitro binding analyses using the purified recombinant proteins showed that the UHM of SPF45 can bind to the ULMs of SF3b155, U2AF 65 and SF1; on the other hand, the UHM and G-patch motif of SPF45 have not been shown to bind directly to RNA 22 . We therefore postulated that the SF3b155-U2AF 65 /U2AF 35 complex is remodeled to the SF3b155-SPF45 complex by switching of their ULM-UHM interactions and that SPF45 per se does not necessarily bind to the truncated PPT (see Fig. 8).
To test our hypothesis, we first examined the binding of SPF45 to SF3b155. We prepared E. coli recombinant glutathione S-transferase (GST)-fusion proteins of SPF45, its D319K mutant in the UHM (SPF45/UHMmt) that no longer binds any ULM, and G patch motif-deleted mutant (SPF45/∆G) that loses potential interaction with nucleic acids and proteins 22 ( Supplementary Fig. S6a). Our GST pull-down assays demonstrated that GST-SPF45 bound to SF3b155, but not to SF1, with crude nuclear extracts under physiological conditions ( Fig.  6a), even though both SF3b155 and SF1 contain ULM and they can interact with SPF45 in vitro 22 . As expected for the SPF45-SF3b155 interaction, the UHM of SPF45 is essential (Fig.  6a, 'GST-SPF45/UHMmt') while the G patch of SPF45 is not required (Fig. 6a, 'GST-SPF45/∆G'), confirming the ULM-UHM interaction in the SF3b155-SPF45 complex. GST-SPF45 also binds to two other previously suggested SPF45-partner proteins that lack ULMs: spliceosomal A complex protein, SF4 (SUGP1 as HGNC approved symbol) and DEAH helicase protein of the U2-related group, hPRP43 24 . However, we confirmed that these two SPF45 interacting factors are not relevant to splicing of short introns ( Supplementary Fig. S5b, c). Next, we tested whether SPF45 can bind to truncated PPT RNA from the SPF45-dependent introns of HNRNPH1 and EML3 in vitro by NMR. Our NMR titration experiments indicate that neither the G-patch motif nor the UHM domain of SPF45 show significant binding towards these two truncated PPT RNAs ( Supplementary Fig. S7).
The binding between U2AF 65 -ULM and U2AF 35 -UHM is extremely strong 20 . Remarkably, GST-SPF45 did not pull-down U2AF 65 and U2AF 35 in crude nuclear extracts (Fig. 6a). These data together suggest that the U2AF 65 -ULM does not interact with the SPF45-UHM in nuclear extracts, and thus that the SPF45-UHM and U2AF 65 -UHM compete for a functional binding toward the SF3b155-ULM (see Fig. 8). Therefore, we next investigated the competitive binding of U2AF 65 and SPF45 toward SF3b155 by titrating the dose of GST-SPF45 in the immunoprecipitation assays (Fig. 6b). Notably, GST-SPF45 interfered with the binding between SF3b155 and U2AF 65 in a dose-dependent manner, however, GST-SPF45/UHMmt did not disturb this binding. These results indicate that the SPF45-UHM competes with that of U2AF 65 -UHM for the SF3b155 binding.
Finally, we examined whether the SPF45-SF3b155 interaction and the G-patch of SPF45 are essential for the SPF45-dependent splicing on short introns. We performed functional rescue experiments with SPF45-depleted HeLa cells using three siRNA-resistant proteins; SPF45 (SPF45/siR), SPF45-UHM mutant (SPF45/UHMmt/siR), and a G-patch motif-deleted SPF45 (SPF45/∆G/siR; Supplementary Fig. S6a). We confirmed that the subcellular localization of these three mutant proteins did not change from that of endogenous SPF45 protein ( Supplementary Fig. S6b). Protein expression levels of endogenous SPF45 and ectopically expressed three SPF45 siRNA-resistant mutants were checked by Western blotting in SPF45depleted HeLa cells (Fig. 7, left panel). We analyzed splicing efficiencies of three mini-genes including short introns by RT-PCR (Fig. 7, right 3 panels). SPF45/siR rescued the splicing defects of all short introns in SPF45-depleted HeLa cells, however the SPF45/UHMmt/siR and SPF45/∆G/siR did not (compare with control 'vector' lane). Taken together, we conclude that it is SPF45 that competes out U2AF 65 and SPF45 is localized at the truncated PPT via proteinprotein interaction with the U2 snRNP component SF3b155 to promote splicing of pre-mRNAs with short introns (Fig. 8).

Discussion
Over a generation ago, two different splicing mechanisms termed 'intron-definition model' for short introns and 'exon-definition model' for long introns were proposed (reviewed in ref. 25 ). In the former model, the frequent lack of a canonical PPT in vertebrate short introns was noticed and an alternative mechanism that circumvents this problem were assumed. Here we provide answers to this puzzling question by demonstrating that a subset of human short introns, with significantly undersized pyrimidine tracts, is recognized by SPF45 but not by the U2AF heterodimer; implicating that SPF45 as a novel functional splicing factor in the spliceosomal complex A. This finding rationally explains why SPF45, which was previously considered just to be an alternative splicing factor, is essential for cell survival and maintenance in vivo 26 .
Our results discover the long-sought-after factor responsible for splicing on short introns, and provide mechanistic details of its function. We found that both SPF45 and U2AF 65 can bind on introns with SF3b155 irrespective of intron size presumably via interactions with the five ULMs in SF3b155, as previously shown in the simultaneous binding of U2AF 65 and PUF60 to SF3b155 27 . This observation can be explained by the previous mass spectrometry analysis using AdML, MINX and PM5 pre-mRNA with conventional introns; i.e., SPF45 is contained in E, A and B complexes as a U2 snRNP-related protein [13][14][15]28,29 (reviewed in ref. 30 ).
Our unique finding is that U2AF 65 needs to be expelled by SPF45 to promote splicing of a subset of short introns. Structural studies showed that U2AF 65 recognizes eight or nine nucleotides of pyrimidine tract 31,32 . The high affinity RNA binding and efficient U2AFdependent splicing requires at least eight pyrimidines 31 , which are likely to be part of a rather extended PPT 33 . We thus propose a mechanistic model that the weak and unstable U2AF 65 binding on the truncated PPT of short intron triggers the protein interaction of SPF45 with SF3b155, leading to the structural and functional replacement of the U2AF heterodimer by SPF45 (Fig. 8). We found that SPF45, either the UHM domain alone or in the presence of the G-patch, does not significantly bind the truncated PPT on its own in vitro. Since our results clearly implicate the critical role of the G-patch in SPF45-dependent splicing, we speculate that SPF45 is localized around the 3′ splice site by the possible involvement of other protein factor(s) in vivo. This will be investigated in our future work.
It was demonstrated that SPF45-depleted fruit fly (Drosophila melanogaster) S2 cells can be functionally rescued by human SPF45 34 . The SPF45-dependent splicing event on the shorter intron might be conserved in fruit fly. Interestingly, fruit fly spliceosomal B complex formed on Zeste pre-mRNA (with 62-nt intron including 14-nt PPT) contains SPF45, but that formed on Ftz pre-mRNA (with 147-nt intron including 33-nt PPT) does not 35 . Therefore, SPF45 is located exclusively in short introns in fruit fly, while in human, SPF45 exists in all introns on standby mode for short introns.
Our RNA-Seq analysis in the SPF45-knockdown cells detected the changes in alternative splicing (Fig. 2a). SPF45 was indeed identified and characterized as an alternative splicing regulator. In fruit fly, SPF45 interacts with Sex lethal (Sxl) protein and induces exon 3 skipping of Sxl pre-mRNA 34 . In mammals, SPF45 can cause exon 6 skipping in FAS pre-mRNA that produces soluble isoform of FAS inducing autoimmune phenotypes in mice 22 .
In the SPF45-induced regulation of alternative splicing, there was no competition between SPF45 and U2AF heterodimer on the Sxl pre-mRNA 34 . Whereas we found a competitive and mutually exclusive binding of SPF45 and U2AF heterodimer on the truncated PPT to splice out short intron. We speculate that the cooperative interaction of SPF45 and U2AF 65 with SF3b155 may be required for alternative splicing regulation, whereas, exclusive binding of SPF45 with SF3b155, but without the U2AF heterodimer, is critical for short intron-specific constitutive splicing (Fig. 8). Interestingly, the U2-related protein PUF60 and U2AF 65 cooperatively interact with SF3b155 to activate weak 3¢ splice sites 36 and the potential binding of both proteins to SF3b155 was demonstrated 27 . SPF45 interacts with other U2-related factors, U2SURP and CHERP, suggesting the role in alternative splicing 37 . However, our siRNA screening showed that knockdown of PUF60 and U2SURP have no effect on the splicing of hnRNPH1 short intron (Table S1, Nos. 89 and 126; unpublished data). Together, we conclude that the critical mechanism of SPF45 as a constitutive splicing factor is distinct from the mechanism of SPF45, together with other interactors, as an alternative splicing factor.
Here, we have just described one distinct subset of human SPF45-dependent short introns. Most recently, Smu1 and RED proteins were shown to activate spliceosomal B complexes assembled on human short introns 38 . The distance between the 5′ splice site and branch site need to be sufficiently short for Smu1/RED-dependent splicing, whereas in contrast, we clearly showed that this distance per se is not responsible for SPF45-dependent splicing (Fig. 3, 'AdML 5′MT'). Therefore, Smu1/Red-dependent and SPF45-dependent splicing mechanisms are essentially different, and thus they target two distinct subsets of human short introns.
The subset of SPF45-dependent short introns was identified by screening our siRNA library based on splicing activity on a 56-nt intron that contains conventional splice sites and branch site. We previously validated a list of ultra-short introns that includes remarkably atypical Grich introns with completely inefficient splice sites and branch sites, of which the 49-nt intron 12 in the NDOR1 gene and the 43-nt intron 6 in the ESRP2 gene were analyzed 10,11 . The mechanism of splicing involved in such atypical G-rich introns is enigmatic. We assume the existence of another exotic subset of human ultra-short G-rich introns.

Methods
Construction of expression plasmids. The mini-gene expression plasmids, pcDNA3-HNRNPH1, pcDNA3-EML3 and pcDNA3-MUS81, were constructed by subcloning the corresponding PCR-amplified fragment into pcDNA3 vector (Invitrogen-Thermo Fisher Scientific). The PCRs were performed using genomic DNA of HeLa cells and specific primer sets (Table S3). For the pcDNA3-AdML, the PCR was performed using the pBS-Ad2 plasmid 39 and specific primer sets (Table S3).
Splicing efficiency screening of siRNA library. HeLa cells were cultured in Dulbecco's modified Eagle's medium (Wako) supplemented with 10% fetal bovine serum. HeLa cells in 35-mm dishes were transfected with 100 pmol of each siRNA in the Stealth siRNA library targeting 154 human nuclear proteins (Invitrogen-Thermo Fisher Scientific) using Lipofectamine RNAiMax (Invitrogen-Thermo Fisher Scientific) according to the manufacturer's protocol.
At 48-96 h post-transfection, total RNAs were isolated from the siRNA-treated HeLa cells and splicing efficiency was analyzed by RT-PCR using a primer set targeting intron 7 of HNRNPH1 (Table S3). The PCR products were separated on 5% PAGE, visualized by imaging analyzer (ImageQuant LAS 500, GE Healthcare Life Sciences), and the unspliced pre-mRNA and spliced mRNA were quantified using NIH Image J software. See below for the detailed procedures.

siRNA knockdown and splicing assays. HeLa cells and HEK293 cells (in 35-mm dishes)
were transfected with 100 pmol siRNA using Lipofectamine RNAi max (Invitrogen-Thermo Fisher Scientific) according to manufacturer's protocol. At 72 h post-transfection, total RNAs were isolated from the siRNA-treated cells using a NucleoSpin RNA kit (Macherey-Nagel).
To analyze endogenous splicing products derived from the HNRNPH1, RFC4, EML3, DUSP1, NFKBIA, MUS81, RECQL4 and MTA1 genes, total RNAs from siRNA-treated cells were reverse transcribed by PrimeScript II reverse transcriptase (Takara Bio) with oligo-dT and random primers, and the obtained cDNAs were analyzed by PCR using the specific primer sets (Table S3). The PCR products were resolved by 6% PAGE. Splicing products were quantified using NIH Image J software. All the experiments were independently repeated three times and the means and standard errors of the splicing efficiencies were calculated.
To analyze splicing products derived from mini-genes, SPF45-and U2AF 65 -depleted HeLa cells were transfected at 48 h and 68 h post-transfection, respectively, with 0.5 µg of mini-gene plasmid (Table S3) using lipofectamine 2000 reagent (Invitrogen-Thermo Fisher Scientific). These cells were incubated for 24 h and 4 h, respectively, prior to the extraction of RNAs (described above). To analyze splicing products from mini-genes, RT-PCR was performed with T7 primer and a specific primer for each mini-gene (Table S3). All the PCR products were analyzed by 6% PAGE and quantified (described above).
In this study, all the oligonucleotide primers were purchased (Fasmac ; Table S3) and all the PCRs were performed with Blend Taq polymerase (Toyobo Life Science).
High-throughput RNA sequencing (RNA-Seq) analyses. Six independent total RNAs derived from HEK293 cells, treated with three control siRNAs and three SPF45-targeted siRNAs, were prepared by NucleoSpin RNA kit (Macherey-Nagel). Then rRNA depletion was performed with the RiboMinus Eukaryote System v2 (Invitrogen-Thermo Fisher Scientific). RNA libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs). These samples were sequenced on the high-throughput sequencing platform (HiSeq2500, Illumina) using a 100 bp single-end strategy.
The sequencing data was analyzed as previously described 41 . Obtained sequence reads were mapped onto the human genome reference sequences (hg19) using the TopHat version 2.1.1 (https://ccb.jhu.edu/software/tophat/index.shtml) and the mapped sequence reads, as BAM files, were assembled using Cufflinks version 2.2.1 (http://cufflinks.cbcb.umd.edu). Using the obtained Cufflinks GTF files as a reference, the BAM files were analyzed using rMATS version 3.2.0 (http://rnaseq-mats.sourceforge.net/) 42 to examine the changes of alternative splicing isoforms. Significant changes of splicing events were defined as when the false discovery rate (FDR) was calculated at less than 0.05.
To analyze the sets of retained introns in U2AF 65 -and SF3b155-depleted HeLa cells, these RNA-Seq data were obtained from the GEO database (Accession numbers GSE65644 and GSE61603) 17,45 .

Cellular formaldehyde-and UV-crosslinking followed by immunopreciptation assays.
To detect Flag-SPF45 association to a pre-mRNA expressed from a reporter mini-gene, we performed formaldehyde crosslinking followed by immunoprecipitation as previously described 12 . Briefly, HEK293 cells (in 60-mm dishes) were co-transfected with pcDNA3-Flag-SPF45 and a mini-gene plasmid (pcDNA3-hnRNPH1, pcDNA3-EML3, pcDNA3-MUS81 or pcDNA3-AdML) using Lipofectamine 2000 reagent (Invitrogen-Thermo Fisher Scientific). At 48 h post-transfection, cells were harvested, washed with cold PBS buffer, and fixed with 0.2% formaldehyde for 10 min. The fixation was quenched in 0.15 M glycine (pH 7.0) and cells were washed with PBS. Immunoprecipitations were performed using anti-Flag antibody-conjugated beads to analyze pre-mRNA, from the mini-gene, associated with Flag-SPF45.
To detect endogenous U2AF 65 -and SF3b155-association to a pre-mRNA expressed from a reporter mini-gene, we performed UV crosslinking followed by immunoprecipitation as previously described 46,47 . PBS-washed HEK293 cells were irradiated with 254-nm UV light on ice. The collected cells were lysed and immunoprecipitated with anti-U2AF 65 and anti-SF3b155 antibodies.
Immunoprecipitated RNAs were extracted with Trizol reagent (Invitrogen-Thermo Fisher Scientific). The isolated RNAs were reverse-transcribed using PrimeScript II reverse transcriptase (Takara Bio) with SP6 primer, and qPCRs were performed using specific primer sets (Table S3).
Biotinylated RNA pull-down assays. Nuclear extracts were prepared from HEK293 cells transfected with control siRNA or SPF45 siRNA according to the small-scale preparation procedure 48 . Biotin-labeled HNRNPH1 and AdML pre-mRNAs were transcribed with a MEGAscript T7 transcription kit (Invitrogen-Thermo Fisher Scientific) according to the manufacturer's instructions.
The biotinylated pre-mRNA (10 pmol) was immobilized with 5 µL of Dynabeads MyOne StreptavidinT1 magnetic beads (Invitrogen-Thermo Fisher Scientific) according to the manufacturer's instruction. The immobilized pre-mRNA beads were incubated at 30°C for 15 min in 30 µL reaction mixture containing 30% nuclear extract, RNase inhibitor (Takara Bio) and nuclease-free water. Then NET2 buffer [50mM Tris (pH7.5), 150 mM NaCl and 0.05% Nonidet P-40] was added to a final volume of 700 µL and incubated at 4°C for 1 h. The incubated beads were washed six times with cold NET2 buffer and boiled in SDS-PAGE sample buffer to analyze by Western blotting (described above).

GST pull-down assays.
GST-SPF45, GST-SPF45/UHMmt, or GST-SPF45/∆G were expressed in E. coli BL21(DE3) CodonPlus (DE3) competent cells (Stratagene-Agilent) and the GST-tagged recombinant proteins were checked by SDS-PAGE followed by Coomassie Blue staining. Induction was carried out at 37°C for 3 h. The GST-proteins were purified using Glutathione Sepharose 4B (GE Healthcare Life Sciences) according to the manufacturer's protocol.
The recombinant GST-SPF45 proteins (5 µg) were incubated at 30°C for 15 min in 100 µL mixture containing 30% HeLa cell nuclear extract. After RNase A treatment, NET2 buffer was added to a final volume of 1 mL with 20 µL of Glutathione Sepharose 4B or SF3b155aantibody conjugated with Protein G Sepharose (GE Healthcare Life Sciences) and incubated at 4°C for 3 h. The incubated beads were washed six times with cold NET2 buffer and boiled in SDS-PAGE sample buffer to analyze by Western blotting (described above).
Immunofluorescence microscopic assays. Immunofluorescence microscopic assays of ectopically expressed Flag-tagged SPF45 proteins were performed as essentially described previously 47 .
Preparation of recombinant proteins. Recombinant SPF45-G-patch-UHM (234-401) was expressed from pET9d vectors with His6-ProteinA TEV cleavable tag using E. coli BL21(DE3) in minimal M9 medium supplemented with 15 NH4Cl for [ 15 N]-labeled protein. Protein expression was induced at OD600 around 0.8-1.0 with 1 mM IPTG, followed by overnight expression at 18°C. Cells were resuspended in 30 mM Tris/HCl (pH 8.0), 500 mM NaCl, 10 mM imidazole with protease inhibitors and lysed using french press. After centrifugation, the cleared lysate was purified with Ni-NTA resin column. The protein sample was further purified by Size-exclusion chromatography on a HiLoad 16/60 Superdex 75 column (GE Healthcare Life Sciences) with 20 mM sodium phosphate (pH 6.5), 150 mM NaCl. The tag was cleaved with TEV protease and removed by Ni-NTA column.    The Flag-SPF45 was immunoprecipitated after formaldehyde crosslinking and then co-precipitated RNAs were quantified by RT-qPCR using specific primers. Means ± SE are given for three independent experiments (*P < 0.05, ***P < 0.0005). b, Cellular CLIP experiments shows strong U2AF 65 binding to control AdML pre-mRNA but not much to the three indicated short introns. Mini-genes containing these four introns were individually co-transfected into HEK293 cells and irradiated with UV light. The lysates were immunoprecipitated with anti-U2AF 65 and anti-SF3b155 antibodies and then immunoprecipitated RNAs were quantified by RT-qPCR using specific primers. Means ± SE are given for three independent experiments (*P < 0.05, **P < 0.005). c, Affinity pull-down experiments of biotinylated RNA indicates U2AF 65 binding to the short intron only if SPF45 was depleted. Biotinylated pre-mRNAs including short HNRNPH1 intron (56 nt) and control AdML intron (231 nt) were incubated with nuclear extracts from either control siRNA-or SPF45 siRNA-treated HEK293 cells. The biotinylated RNA-bound proteins were pulled down with streptavidin-coated beads and analyzed by Western blotting with antibodies against SF3b155, SPF45 and U2AF 65 .  . Proteins that associated with these GST-fusion proteins were detected by Western blotting using the indicated antibodies (right panel). b, Immunoprecipitation of SF3b155 with the indicated GST-fusion proteins shows the competitive binding between U2AF 65 and SPF45 to SF3b155 via UHM-ULM interactions. The same reaction mixtures in panel a were immunoprecipitated with anti-SF3b155 antibody and the associated proteins were analyzed by Western blotting with antibodies against SF3b155, U2AF 65 and SPF45. . Arrowheads indicate degraded siRNA-resistant (siR) proteins, but not endogenous SPF45 (see 'vector' lane for the depletion efficiency of endogenous SPF45). After the co-transfection of the indicated siRNA-resistant plasmids and three mini-gene plasmids, splicing efficiencies of the indicated three mini-genes were analyzed by RT-PCR (right 3 panels). Means ± SE are given for three independent experiments (*P < 0.05, **P < 0.005, n.s. P > 0.05).