Genetic Interaction Mapping Reveals a Role for the SWI/SNF Nucleosome Remodeler in Spliceosome Activation in Fission Yeast

Although numerous regulatory connections between pre-mRNA splicing and chromatin have been demonstrated, the precise mechanisms by which chromatin factors influence spliceosome assembly and/or catalysis remain unclear. To probe the genetic network of pre-mRNA splicing in the fission yeast Schizosaccharomyces pombe, we constructed an epistatic mini-array profile (E-MAP) and discovered many new connections between chromatin and splicing. Notably, the nucleosome remodeler SWI/SNF had strong genetic interactions with components of the U2 snRNP SF3 complex. Overexpression of SF3 components in ΔSWI/SNF cells led to inefficient splicing of many fission yeast introns, predominantly those with non-consensus splice sites. Deletion of SWI/SNF decreased recruitment of the splicing ATPase Prp2, suggesting that SWI/SNF promotes co-transcriptional spliceosome assembly prior to first step catalysis. Importantly, defects in SWI/SNF as well as SF3 overexpression each altered nucleosome occupancy along intron-containing genes, illustrating that the chromatin landscape both affects—and is affected by—co-transcriptional splicing.

Introduction throughput genetic interaction mapping, we uncovered multiple connections between the chromatin and splicing machineries. Importantly, we found that deletion of SWI/SNF nucleosome remodeling complex components left cells particularly sensitive to overexpression of the SF3 subcomplex of the U2 snRNP. We present evidence that SWI/SNF contributes to splicing catalysis by promoting recruitment of the Prp2 ATPase, which acts to destabilize SF3 immediately prior to first step catalysis. Our data provide a functional role for SWI/SNF in the splicing of introns with weak splice sites and promote the idea that nucleosomes, while serving as barriers to RNAPII elongation, may actually promote co-transcriptional splicing.

Results
Generation of a pre-mRNA splicing E-MAP in S. pombe In order to define the genetic interaction space of all pre-mRNA splicing factors in fission yeast, we created a set of mutants in which each gene involved in splicing was individually removed or perturbed. We first manually compiled a list of 81 genes whose products have either been experimentally implicated in fission yeast splicing or whose orthologs have been functionally characterized as splicing factors in other species (S2 Table). As was originally described in Roguev et al., (2008), for genes annotated as non-essential, we systematically replaced the open reading frame with a NAT resistance cassette to create deletion strains. For essential genes (which encode the majority of splicing factors in S. pombe), we created DAmP alleles [24] by replacing the 3'UTR with a NAT resistance cassette. DAmP alleles have been successfully used in the past to perturb mRNA levels of essential genes in order to shed light on their function in high-throughput genetic and drug screens [25]. Using the Pombe Epistasis Mapper (PEM) system our group developed [26], we screened 74 S. pombe splicing factors (several splicing mutants were too sick to be propagated through the E-MAP screen; see S2 Table) against a fission yeast mutant library containing more than 2,000 non-essential deletions (library as described in [27]). This collection represented virtually every major known biological process in the cell, creating a Splicing E-MAP with approximately 120,000 pairwise measurements. Positive genetic interactions between two mutants (> +2.0) represent epistasis or suppression, while negative genetic interactions (< -2.5) represent synthetic sickness, or in some cases, synthetic lethality. We present this data in S1 Table. Previous work (summarized in [28]) has shown that genes belonging to the same complex or pathway tend to have similar genetic interaction profiles. To gain an unbiased overview of the splicing factors in our E-MAP we clustered them according to the similarity of their genetic interaction profiles, i.e. based on how individual splicing factor mutants genetically interact with all factors represented in the deletion library. Several clusters (correlation coefficient > 0.3, see methods) emerged containing splicing factors from multiple steps of the splicing cycle ( Fig. 1A and S2 Table). Consistent with our understanding of early spliceosome assembly, our E-MAP revealed a high degree of correlation between a pair of splicing factors, prp11 (PRP5; where appropriate, S. cerevisiae orthologs will be listed parenthetically) and the branchpoint binding protein bpb1 (MSL5/BBP1), which are both known to function at the step of intron recognition (Fig. 1A, Cluster 9). Interestingly, cwf23 (CWC23) is also part of Cluster 9, perhaps indicating a role for this DNAJ domain-containing protein at an early step of spliceosome assembly.
Generally, splicing genetic interaction profile clusters did not breakdown into U snRNPs, the small nuclear ribonuclear protein complexes that assemble stepwise to form a functional spliceosome. Interpretation of the E-MAP genetic interaction profile correlation scores for splicing factors is complicated by the fact that most of them are essential and thus are represented in the E-MAP as DAmP alleles. Integration of the DAmP cassette is not expected to affect expression of all genes equally. Indeed, as was the case in the S. cerevisiae RNA processing E-MAP [29], many of the S. pombe DAmP strains showed little growth defect and had weak genetic interaction profiles, making clustering uninformative in some cases and resulting in clusters comprised mostly of mutants with strong genetic interaction profiles. Despite these caveats, our observation that certain "early" splicing factors and "late" splicing factors had similar genetic interaction profiles reinforces the dynamic nature and functional complexity of the spliceosome.

Genetics reveal myriad connections between splicing and chromatin machineries
We next asked what cellular pathways were enriched for positive or negative genetic interactions with splicing as a whole. Positive genetic interactions can identify genes whose products act in the same biological pathway, or in some cases, form a protein complex. Negative genetic interactions are often observed between two genes whose products act in parallel pathways that lead to a common biological outcome [28]. Strikingly, of all the pathways represented in the E-MAP [27], splicing was specifically enriched for negative genetic interactions with factors related to "transcription," "regulation of transcription," "chromatin modifications" and "chromosome organization" (Fig. 1B, methods). Looking next at negative genetic interactions between splicing and known protein complexes, we saw additional evidence for significant crosstalk between splicing and chromatin: S. pombe splicing mutants had strong negative genetic interactions with chromatin remodeling factors, such as the histone exchange complex SWR1, and histone modifying factors, including the histone demethylase Lid2 and the histone deacetylase complex Rpd3L (Fig. 1B). Interestingly, there appears to be some specificity regarding which splicing factors have genetic interactions with particular chromatin complexes. For example, splicing factors in Cluster 8 exhibited strong negative genetic interactions with Rpd3L factors, supporting a role for histone deacetylation in promoting splicing, while Cluster 10 factors did not (Fig. 1C).
Although the composition of U snRNPs was not immediately evident from the clustering, we still wanted to know whether there was any specificity in how different U snRNPs interacted genetically with chromatin complexes. We found that several chromatin complexes were enriched for negative genetic interactions with multiple U snRNPs (e.g. Set1/COMPASS, Lid2, SWI/SNF, and SWR1) (Fig. 1D). However, a smaller subset of chromatin complexes showed enrichment for negative genetic interactions with a limited set of U snRNPs-for example, tri-snRNP (U4/U5/U6) showed strong negative interactions with two ubiquitin ligase complexes (CLRC and Cul4B-RING) and the U1 snRNP was strongly negative with the SWR1 histone exchange complex. Notably, the nucleosome remodelers SWI/SNF and Ino80 were particularly enriched for negative interactions with the U2 snRNP (Fig. 1D), as well as with downstream tri-snRNP components. SWI/SNF and Ino80 are ATP-dependent chromatin remodeling complexes whose activities include ordering and reorganizing nucleosomes on the DNA template in order to regulate RNA polymerase II elongation. A subunit of SWI/SNF has been previously blue box indicates color of significance cut-off. Processes were defined using the S. pombe GO Slim database and complexes were taken from the S. pombe cellular component Gene Ontology list (PomBase). Bonferonni corrected p value <0.0001 for processes and 0.0005 for complexes. (C.) Genetic interactions of Splicing Cluster 8 and 10 with the chromatin machinery. Blue indicates negative genetic interactions; yellow indicates positive; grey indicates no data. (D.) Splicing U snRNPs (blue) enriched for negative genetic interactions with chromatin complexes (yellow). U snRNPs were defined as listed in S2 shown to influence alternative splicing in mammals [7], although its mechanism for doing so remains unclear. We therefore set out to elucidate the precise interplay between SWI/SNF and splicing in fission yeast.
We identified two splicing factors with strong negative genetic interactions with components of SWI/SNF: sap145 and sap62. Sap145 (CUS1) is a component of SF3b, a subcomplex of U2 that associates with the pre-mRNA branchpoint region [30]. Sap62 (PRP11) is a component of the SF3a complex, a group of three conserved proteins that binds to SF3b and also makes contacts with the intron branchpoint. Because both sap145 and sap62 are essential genes in S. pombe, they were represented in the E-MAP as DAmP alleles. In order to confirm the E-MAP results, as well as to assess temperature sensitivities, we crossed these DAmP alleles with two non-essential SWI/SNF deletions, Δsnf5 and Δsol1. Snf5 is a core component of SWI/SNF while Sol1 (SWI1) is an ARID (AT-rich interaction domain)-containing protein that makes specific contacts with DNA [31]. As predicted by the negative genetic interaction score in the EMAP, combining the SF3 DAmP alleles with Δsnf5 or Δsol1 caused synthetic sickness at all temperatures, most severely at 37°C and 16°C (Fig. 1E). Results from the E-MAP also showed negative genetic interactions between SWI/SNF and other components of SF3, namely sap114 and prp10 (PRP21 and HSH155, respectively) (S1 Fig), further suggesting there might be specific interplay between the SWI/SNF nucleosome remodeling complex and the U2 snRNP. We thus set out to understand the molecular phenotype behind the synthetic sickness in our SWI/SNF-SF3 double mutants.

SWI/SNF-SF3 double mutants have a broad yet intron-specific splicing defect
In order to ascertain whether defective splicing was responsible for the growth defect in the SWI/SNF-SF3 double mutant strains, we performed genome-wide splicing microarray analysis. These arrays contain oligonucleotide probes specific for the exon, intron, and exon-exon junction for nearly every canonically spliced intron in fission yeast-allowing us to compare the levels of total, pre-mRNA and mature mRNA, respectively, for any given mutant strain against an isogenic wild-type ( Fig. 2A). The experiment was performed as a competitive two-color hybridization and results are reported as the log 2 fold change (logFC) in mutant RNA compared to wild-type RNA (Lipp et al., in preparation), with yellow in the Fig. 2B heatmap signifying accumulation of pre-mRNA relative to mature mRNA. We observe little accumulation of pre-mRNA in any of the single mutants (sap145 DAmP , sap62 DAmP , Δsnf5, Δsol1, Fig. 2B), which we interpret to mean there is little to no change in splicing efficiency in these strains. However, upon combining mutations in splicing factors and SWI/SNF, we see a broad yet intron-specific accumulation of pre-mRNA (Fig. 2B). Average hierarchical linkage clustering reveals a high degree of correlation between sets of double mutants: SWI/SNF-SF3a double mutants have a correlation of 0.85 and the SWI/SNF-SF3b double mutants have a correlation of 0.76 (S2 Fig). Together, the SWI/SNF-SF3 strains correlate to a relatively high degree (0.65), indicating that similar introns are affected in all SWI/SNF-SF3 double mutant backgrounds. To quantify the exacerbated splicing defect in the double mutants, we imposed a significance cut-off of a logFC of 0.5 and calculated the number of introns that were spliced poorly (in yellow) or spliced better (in blue) for each of the single and double mutant strains (S2 Fig). Again, we observed few changes in the number of introns retained in the single mutants (<300 in each case), but many more retained introns in the SWI/SNF-SF3 double mutants (>1000 in most cases), and of those, the double mutants had more introns retained to higher degrees (logFC > 1.5, in darker yellow). While these results are consistent with a splicing defect, we cannot rule out that We validated our microarray results by reverse transcriptase quantitative PCR (RT-qPCR), choosing two representative multi-intron containing genes, spo14 and nda3, using primers that amplify the intron-exon junction (intron) and the exon (total) (Fig. 2C). Again, we only saw intron accumulation in the double mutant strains, relative to wild type, providing further evidence that the splicing defect is only seen when both SF3 and SWI/SNF are mutated. Interestingly, when we looked at a multi-intron containing gene like nda3, we saw that intron 4 accumulated to much higher levels in the double mutants than did other introns in the same gene, indicating that the splicing defect is intron-specific, even when affected introns are very close to other introns in the gene.
The specificity of intron retention suggested to us that there might be cis-splicing signals that make an intron more or less sensitive to the SWI/SNF-SF3 mutations. To interrogate this, we computationally assessed the contribution of different intron features to the splicing defect. While the cis-splicing signals of S. pombe introns are less constrained then those of S. cerevisiae, there are consensus sequences maintained by the majority of S. pombe introns. Analysis of the microarray data revealed that deviation from consensus in those sequences (the 5' splice site (5'SS), the branchpoint (BP), and the 3' splice site (3'SS)) increased the likelihood that an intron would be retained in the Δsol1sap145 DAmP double mutant strain (Fig. 2D). Interestingly, we observed that splicing in the double mutant is most sensitive to changes to the 5'SS, which may reflect cooperativity between SWI/SNF and the U6 snRNA's ability to engage the 5'SS, perhaps shedding light on the negative genetic interactions between SWI/SNF and the tri-snRNP from the E-MAP (Fig. 1D). Looking back to the introns that accumulated in our RT-qPCR experiments, we saw that intron 3 of spo14 has a rare 5'SS (GTATGC, used by less than 200 introns in S. pombe) and intron 4 of nda3 contains a disfavored AAG 3'SS (YAG is preferred, where Y = pyrimidine). No intron features significantly increased or decreased probability of retention in any of the single mutant strains, which is consistent with the overall lack of splicing phenotype in those strains. Several of the mutants had a population of introns whose splicing was improved (in blue on S2 Fig), but there were insufficient numbers for statistical analysis of enriched cis-splicing signals. Taken together, this analysis suggests that nonconsensus introns have a unique requirement for interplay between the SWI/SNF nucleosome remodeler and SF3, and surprisingly, suggests these two machineries may play a concerted role in splicing.

Overexpression of SF3 is synthetic sick with SWI/SNF
In order to look more closely at the mechanistic causes of the splicing defect in the SWI/ SNF-SF3 double mutant strains, we needed to understand the effect of the DAmP cassette on SF3 transcript levels. Generally, the DAmP cassette is thought to compromise mRNA processing and stability by removing a gene's 3'UTR, thus creating hypomorphic alleles. To interrogate the expression level of the sap145 and sap62 DAmP alleles, we performed RT-qPCR on total RNA isolated from each mutant strain. Unexpectedly, we found significantly higher levels of mRNA for both sap145 and sap62, indicating that these are in fact, overexpression alleles (Fig. 3A). While this result contradicts the paradigm of 3'UTR disruption creating increase likelihood of intron retention in Δsol1sap145 DAmP double mutant. Logos represent consensus site sequences for the 5'SS, BP and 3'SS. Probability of a specific nucleotide change increasing the likelihood of retention in the double mutant is represented as log odds (positive log odds at a given position indicates that non-consensus nucleotides correlate with intron retention). Affected introns were defined as having a corrected p value of < 0.05. Dark nucleotides in the consensus site logos are GCs, lighter letters are AUs.
doi:10.1371/journal.pgen.1005074.g002 hypomorphic alleles (DAmP = decreased abundance by mRNA perturbation), we propose that DAmP allele overexpression may be a general phenomenon for genes encoding RNA binding proteins, many of which are known to negatively regulate their own expression through binding to their 3'UTRs [32]. Removal of snf5 and sol1 alone did not affect sap145 or sap62 mRNA levels, although the overexpression appeared slightly exacerbated in the double mutant background. Importantly, overexpression of the sap145 ortholog CUS1 in S. cerevisiae has previously been purported to promote a tighter association between SF3b, SF3a, and Prp5 [33].
As measuring protein levels would require tagging the proteins, most simply at either the N-or C-terminus, and would likely unintentionally re-create the DAmP alleles, we set out to ectopically overexpress sap145 and sap62 and ask whether this recapitulated the synthetic sick phenotype seen with the DAmPs when combined with Δsnf5. Sap145 and sap62, along with sap49 and sap114, other components of SF3b and SF3a, respectively, were cloned into the pREP4x expression vector, which is under the control of the thiamine-repressible nmt1 promoter [34] Overexpression of any of the SF3 factors in WT cells did not confer any growth defect while overexpression in the Δsnf5 strain background resulted in synthetic sickness (Fig. 3B). We took this as confirmation that the DAmP cassette caused overexpression of sap145 and sap62 and that overexpression of these splicing factors causes synthetic sickness and intron retention when combined with SWI/SNF mutants. S3B Fig shows levels of SF3 factor overexpression in the pREP4x strains, which are comparable to levels induced by the DAmP cassette. As controls, we also overexpressed a non-SF3 U2 factor, uap2 (CUS2), and a U1 snRNP factor, usp104 (PRP40), in WT and Δsnf5 cells. We did not observe any specific genetic interaction between Δsnf5 and overexpression of uap2 or usp104, indicating that merely overexpressing any early splicing factor is not sufficient to induce a growth defect in Δsnf5 cells (Fig. 3C). We also overexpressed sap145 in other chromatin mutants (Δswr1(SWR1), Δpht1 (HTZ1) and Δset1(SET1)) and observed no synthetic sickness (S3 Fig), consistent with a specific interaction between deletion of SWI/SNF and SF3 overexpression. Because we only see intron accumulation when the two mutations are combined, we reasoned that SWI/SNF might contribute to an SF3-dependent step of splicing, possibly by promoting spliceosome recruitment and/or pre-catalytic spliceosome rearrangements. Because we observed no synthetic sickness between Δsnf5 and mutants of the Prp5 RNA-dependent ATPase (Prp11 in S. pombe) (S3 Fig) [35], we hypothesized that SWI/SNF may act at a step of splicing that follows spliceosome assembly.

SWI/SNF influences splicing at the Prp2 step
The microarray results and synthetic sick phenotype suggested that SWI/SNF and SF3 contribute to splicing in a concerted fashion, perhaps both influencing a particular step in the splicing cycle. Having confirmed that our SF3 DAmP alleles were being overexpressed, we wanted to look at specific steps of splicing that might be affected by an overabundance of SF3 components. Importantly, SF3 needs to be destabilized by the ATPase Prp2 prior to the first step of splicing in order for the branchpoint adenosine on the pre-mRNA intron to be available for nucleophilic attack (Fig. 4A) [36][37][38]. We therefore hypothesized that having an excess of SF3 factors could disfavor first step catalysis and would exacerbate a splicing defect occurring at the Prp2 step.
To follow spliceosome rearrangements in the context of chromatin, we performed splicing factor chromatin immunoprecipitation (ChIP) (Fig. 4B), which has been used successfully in the past as a read-out for the relative timing and efficiency of splicing factor recruitment [39,40]. We chose several intron-containing genes to ChIP based on the following criteria: 1.) The gene had one or more introns whose splicing was defective in a SWI/SNF-SF3 double mutant by microarray, RT-qPCR or both, 2.) The gene was highly expressed, 3.) The gene had introns long enough (>200nt) to provide enough spatial resolution for us to detect cotranscriptional loading and rearrangements of the individual spliceosome factors.
To test the hypothesis that the Prp2 step is affected in the double mutant, we HA-tagged cdc28 (PRP2) and performed ChIP qPCR on at set of representative genes. Strikingly, we found levels of Cdc28 to be significantly lower in the Δsnf5/sap145 DAmP strain compared to wild type at three different genes with distinct intron architectures: rps3, dbp2 and SPBC660.16 ( Fig. 4C and S4 Fig), indicating that defective splicing in the double mutant may be caused by a lack of Cdc28 (PRP2) recruitment or stabilization.
If reduced Cdc28 (PRP2) levels are due to decreased recruitment, we would expect a defect in Cdc28-dependent remodeling of the spliceosome. One way such a defect would manifest itself is by retention of the proteins released or destabilized by Cdc28 (PRP2), including the Nineteen Complex (NTC)-related protein Cwf24 (CWC24) or the RES (REtention and Splicing) factor Cwf26 (BUD13) [37]. We chose to tag and ChIP Cwf24 (CWC24). As expected, in the Δsnf5sap145 DAmP cells where we saw less Cdc28 (PRP2), we observed much higher levels of Cwc24 in all three genes we looked at, especially at primer sets at or downstream of affected introns ( Fig. 4C and S4 Fig). We interpret this to mean that in wild-type cells, Cwf24 is displaced by the Prp2 ATPase and is thus not in the vicinity of chromatin to be crosslinked. In the absence of optimal Cdc28 (PRP2) recruitment, the likelihood of Cwf24 (CWC24) remaining part of the pre-catalytic spliceosome increases, and this is borne out by higher Cwf24 ChIP signal in the Δsnf5sap145 DAmP double mutants.
We chose not to look at SF3 by ChIP directly for two reasons: first, because sap145 overexpression may change the local concentration of binding partners and complicate ChIP interpretation and second, because SF3 is known to be destabilized but not completely released from the spliceosome [37] and such nuanced rearrangements are likely not detectable by ChIP.
Notably, we do see a modest increase in Cdc28 (PRP2) ChIP in the sap145 DAmP single mutant cells and a decrease in Cdc28 ChIP in the Δsnf5 single mutant cells, similar to the levels in Δsnf5sap145 DAmP (S4C Fig), providing some mechanistic insight into how each of these mutations individually affects pre-spliceosome activation. We believe that the overabundance of Sap145 results in a greater requirement for Prp2 remodeling, leading to a higher local concentration of the enzyme, but in the absence of Snf5, Prp2 is not recruited as well. When the two mutations are combined, it creates a situation where the cell needs more Prp2 but has less, which leads to poor splicing outcome.
To provide further evidence that the splicing defect in the Δsnf5sap145 DAmP double mutants occurs at the step of spliceosome activation, we performed ChIP on a HA-tagged U1 snRNP factor, Usp105 (PRP39) as a control. Compared to wild type, we observed little change in U1 snRNP ChIP along the single intron-containing rps3 gene ( Fig. 4D and S4A Fig). The overall lack of U1 ChIP defect is consistent with our genetic data (showing no synthetic sickness between U1 overexpression and Δsnf5) and supports a model in which SWI/SNF contributes downstream of early spliceosome assembly.
Our ChIP results support a model whereby SWI/SNF contributes to splicing at the Prp2 step of splicing activation. To provide additional lines of evidence to these ends, we again turned to genetics. If our hypothesis that SWI/SNF helps recruit Cdc28 (PRP2) is correct, we would expect that abrogation of Cdc28 (PRP2) function via an alternative method would phenocopy the synthetic sickness seen between SWI/SNF and sap145 overexpression. We therefore mutated the conserved Gly553 to an Aspartate in S. pombe cdc28 (PRP2), based on the well-characterized prp2-1 temperature-sensitive allele in S. cerevisiae, which has been shown to stall splicing after assembly but before first-step catalysis [41,42]. S. pombe cdc28-1 (G553D) was synthetic sick when combined with Δsnf5, especially at the non-permissive temperature (37°C) (Fig. 5A), consistent with an increased requirement for SWI/SNF activity at the Prp2 ATPase step.
In S. cerevisiae, Prp2 is known to act with the cofactor Spp2, an essential G-patch protein that binds to Prp2 and promotes the first step of splicing [43]. Looking back at the S. pombe splicing E-MAP, we saw that the cwf28 DAmP (SPP2 in S. cerevisiae) allele had a strong positive genetic interaction with SWI/SNF (+2.0 and +2.6 with Δsnf5 and Δsol1, respectively; Fig. 1C), suggesting epistasis or suppression. To confirm that cwf28 and SWI/SNF interacted genetically, we re-created the double mutant by mating, and observed that the Δsol1/cwf28 DAmP double mutant grew better at 16°C than did the Δsol1 strain alone, confirming the E-MAP positive genetic interaction (Fig. 5B). In examining the cellular levels of cwf28 mRNA in the DAmP background, we saw that the DAmP cassette again led to overexpression of cwf28 (S5 Fig). Interestingly, SPP2 was originally identified as a high-copy suppressor of prp2-1 [44]. Because SWI/SNF appears to recruit Prp2 and overexpression of Spp2 can overcome the cold-sensitivity of Δsol1, we think Spp2 is also acting as a high-copy suppressor of ΔSWI/SNF.
To further implicate SWI/SNF in promoting Prp2-dependent spliceosome remodeling, we overexpressed several non-SF3 proteins that are released by Prp2, predicting that their overexpression-like that of SF3-would result in synthetic sickness with Δsnf5. Indeed, using the pREP4x vector, ectopic overexpression of cwf26 (BUD13) and cwf24 (CWC24) caused a severe growth defect in Δsnf5 but not wild-type cells (Fig. 5C), consistent with a role for SWI/SNF in promoting Prp2-recruitment and the subsequent release of Cwf26 (BUD13) and Cwf24 (CWC24) from the spliceosome. In summary, we have provided several lines of evidence that SWI/SNF contributes to the Prp2 step of splicing: Prp2 recruitment is defective in Δsnf5sap145-DAmP cells; SWI/SNF has opposite genetic interactions with cdc28-1 (PRP2-1) and overexpression of cwf28 (SPP2); and overexpression of cwf26 and cwf24 is synthetic sick with Δsnf5, presumably due to exacerbating the block in splicing imposed by defective Cdc28 (PRP2) ATPase recruitment.
Nucleosome positioning is altered in ΔSWI/SNF strains SWI/SNF is generally thought to deposit and remodel nucleosomes on a DNA template in order to regulate RNAPII transcription. To test whether nucleosome positioning and/or occupancy could be altered in ΔSWI/SNF cells, we performed a nucleosome-scanning assay, as described in Infante et al., 2012 [45]. Following micrococcal nuclease (MNase) treatment, mononucleosome-protected DNA fragments (~150bp) were excised from an agarose gel and tiling qPCR was performed along several fission yeast genomic loci (Fig. 6A). At all expressed genes examined, we saw a striking decrease in signal from protected DNA in Δsnf5 and Δsnf5sap145 DAmP cells, implicating Snf5 in depositing or maintaining the position of nucleosomes along coding regions (Fig. 6B, replicate experiment in S6B, C, D Fig). Both the Δsnf5  single mutant and the Δsnf5sap145 DAmP double mutant displayed lower nucleosome occupancy at intron-containing genes, although certain nucleosomes were differentially protected; generally, nucleosomes downstream of introns exhibited the biggest changes in nucleosome occupancy relative to wild type (Fig. 6B, bold circles). We propose that a certain level/distribution of nucleosomes is required for optimal recruitment of Prp2 and that SWI/SNF contributes to creation and/or maintenance of this nucleosome environment at intron-containing genes.
In order to show that the nucleosome occupancy phenotypes we report in Δsnf5 and Δsnf5sap145 DAmP cells were not simply due to differences in mononucleosome recovery or MNase accessibility we amplified two regions that we did not expect to be under the control of SWI/SNF: the silenced Mat3-Mm ORF in the mating type locus and an untranscribed genepoor region. At these two silent regions, we observed no significant change in overall abundance of nucleosomes in the different mutant strains ( Fig. 6C and S6B Fig). In our assay, nucleosomes appeared to be less strongly positioned overall (leading to lower signal relative to undigested genomic DNA) at Mat3_Mm and the untranscribed region. This is consistent with published genome-wide data, which show weak nucleosome positioning in these regions during mitotic growth [46].
To correlate the lower nucleosome signal we report at rps3 and dbp2 in the Δsnf5 single mutant and the Δsnf5/sap145 DAmP double mutant with RNAPII transcription, we examined the ChIP profile of RNA polymerase II, using an antibody against the unmodified RNAPII CTD, antibody (8GW16). RNAPII levels were consistently lower in Δsnf5/sap145 DAmP strains compared to wild type at almost all points throughout the gene bodies we looked at, even though signal at the promoter was comparable between the two strains ( Fig. 6D and S6E Fig). Because this phenotype was strongest in the double mutant, we propose that RNAPII is subject to a level of regulation that relies on both SWI/SNF and splicing being intact. ΔSnf5 single mutants generally displayed wild-type RNAPII density in ChIP using the 8GW16 antibody (Fig. 6D), although interestingly, additional ChIP experiments using an antibody directed against Ser5 phosphorylated CTD residues (4H8, Abcam) showed reproducibly lower levels of Ser5P CTD in Δsnf5 and Δsnf5sap145 DAmP strains, consistent with results from Batsche et al., which correlate Ser5P with the SWI/SNF subunit Brm at alternative exons [7]. While it is difficult to garner mechanistic understanding from these results alone, lower RNAPII levels are consistent with lower nucleosome occupancy in the Δsnf5 and Δsnf5/sap145 DAmP strains: if nucleosomes act as a barrier to transcription [47,48], then lower nucleosome density should allow for enhanced RNAPII elongation and thus decrease the probability of crosslinking the polymerase molecule to any one point in a gene, though we can not rule out the possibility of changes to RNAPII initiation or processivity.
Surprisingly, we also observed a dramatic increase in nucleosome occupancy in the sap145-DAmP single mutant cells at intron-containing genes (Fig. 6E). Importantly, when we looked at an intronless gene, act1, we did not observe higher nucleosome occupancy in the sap145 DAmP single mutant, suggesting that sap145 overexpression is only inducing chromatin changes in the context of splicing at intron-containing genes. These results support a model wherein splicing itself may influence chromatin dynamics. We propose that overexpression of sap145 inhibits spliceosome activation and that higher levels of nucleosomes serve to compensate for this defect by slowing RNAPII to promote co-transcriptional splicing. Notably, this increase in nucleosome occupancy presumably requires the integrity of the SWI/SNF complex, since, as we RNAPII CTD antibody (8WG16) in WT, Δsnf5, and Δsnf5sap145 DAmP strains. Gene diagram below indicates approximate location of qPCR amplicons and basepairs are shown. Capped error bars are ±SEM of n = 3 biological replicates. (E.) qPCR analysis of nucleosome occupancy at rps3, dbp2 and act1 (nonintron containing) for WT and sap145 DAmP strains.
doi:10.1371/journal.pgen.1005074.g006 describe above, nucleosome occupancy decreased relative to WT when we combined sap145 overexpression with deletion of snf5 in our double mutant.
Data from humans has shown that in addition to higher nucleosome abundance in exons vs. introns, nucleosome occupancy further peaks at exons that are flanked by long introns or weak splice sites [49], consistent with our data and with the idea that SWI/SNF-dependent nucleosome positioning is important for pre-mRNA splicing. Taken together, these results point to a role for SWI/SNF in maintaining nucleosome occupancy at expressed genes in S. pombe and hint at an additional role for splicing factors, or the act of splicing itself, in further promoting nucleosome occupancy.

Discussion
While a role for chromatin remodeling and histone modifying proteins in regulating alternative splicing has been appreciated for some time, the idea that constitutive splicing decisions also require input from chromatin proteins and RNAPII transcription itself, is just starting to gain traction. Indeed, our Splicing E-MAP is the first high-throughput genetic study to provide evidence that chromatin and transcription have profound connections to pre-mRNA splicing in fission yeast. Here, we provide several lines of evidence to show that the SWI/SNF nucleosome-remodeling complex is important for splicing introns with weak splice sites and contributes to Prp2-dependent spliceosome activation. Importantly, these results demonstrate a role for chromatin in regulating a specific ATP-dependent step of the splicing cycle and provide mechanistic insight into how SWI/SNF may influence RNAPII transcription and splicing through positioning nucleosomes at intron-containing genes.

A role for chromatin in promoting splicing of introns with weak splices sites
One reason we set out to create a Splicing E-MAP in fission yeast was because of its metazoanlike intron/exon architecture. Over half of intron-containing genes in fission yeast contain more than one intron and the cis-splicing signals in S. pombe introns are highly variable, in contrast to S. cerevisiae but mirroring the situation in metazoans. For example, the branch-site consensus sequence in S. cerevisiae is almost always UACUAAC, whereas the S. pombe branchsite consensus CURAY (where R is a purine and Y is a pyrimidine), is more similar to that found in mammals (reviewed in [50]). Metazoan splicing decisions are aided by an abundance of SR and hnRNP proteins that regulate exon inclusion or skipping by binding to specific sequences in pre-mRNAs [51]. While the fission yeast genome encodes a small number of SRlike proteins, their roles in intron recognition remain unclear [52,53]. The abundance of degenerate cis-splicing signals combined with the dearth of canonical regulatory proteins in S. pombe prompts questions of how introns with degenerate splicing signals are recognized in fission yeast and whether there are other pathways to promote co-transcriptional intron recognition and spliceosome assembly. Our results indicate that multiple cellular pathways contribute to efficient splicing in fission yeast, most notably chromatin and RNAPII transcription.
If we take a closer look at the particular chromatin complexes enriched for negative genetic interactions with splicing, we can infer some mechanistic insights. Both SWI/SNF and Ino80 are known to deposit and position nucleosomes in an ATP-dependent manner. Although to date there are no reports implicating Ino80 in splicing, there is solid evidence linking the SWI/SNF catalytic subunit Brm to variant exon inclusion in several human genes. Overexpression of the Brm subunit in human breast cancer cells led to increased exon inclusion in a handful of genes examined and siRNA knockdown of Brm abolished RNAPII accumulation on the same variant exons [7]. The authors of this report proposed SWI/SNF contributes to splicing by decreasing RNAPII elongation rate, thus facilitating co-transcriptional spliceosome recruitment and allowing for recognition of introns with suboptimal spice sites. The idea that a "slow" RNA polymerase promotes spliceosome assembly and recognition of weak splice sitesoften referred to as the "kinetic model" of co-transcriptional splicing-is supported by numerous lines of evidence, and seems to be conserved from budding yeast [18] to humans (reviewed in [10]).
Intriguingly, most of the chromatin complexes that were synthetic sick with splicing factors in the E-MAP have been implicated in negative regulation of RNAPII transcription. For example, Set1/Compass, which deposits H3K4 trimethylation has recently been shown, in coordination with H3K4 dimethylation, to repress transcription at coding genes [54]. Similarly, Rpd3L is involved in downregulating transcription at gene promoters by deacetylating histones [55,56]. The cycle of histone acetylation and deacetylation has already been shown to be important for spliceosome recruitment and rearrangements in S. cerevisiae, wherein deletion of histone deacetylases led to retention of U2 factors and a decrease in recruitment of later splicing factors [16,17]. Taken together, our SWI/SNF data and the Splicing E-MAP as a whole support the general premise of the kinetic model (i.e. repressive chromatin complexes should promote splicing) and provide strong evidence that constitutive splicing decisions also rely on transcriptional inputs.

Evidence for crosstalk between nucleosome occupancy and splicing across intron-containing genes
We propose a model whereby SWI/SNF contributes to the deposition and/or maintenance of nucleosomes in the coding regions of expressed genes in fission yeast (Fig. 7). These nucleosomes may influence splicing in several ways: by influencing polymerase speed and/or promoting pausing, by recruiting splicing factors directly or via adapter proteins, or by directing changes to the CTD modification status. Our data correlate lower nucleosome occupancy with a decrease in RNAPII ChIP and we hypothesize that this reflects changes to RNAPII elongation along coding regions. We propose that these SWI/SNF-dependent changes to nucleosomes and consequently to RNAPII, lead to decreased Prp2 recruitment and defective co-transcriptional spliceosome activation.
Our report that deletion of snf5 leads to a decrease in nucleosome occupancy across spliced genes is consistent with recent findings from Tolstorukov et al., who saw a decrease in nucleosome occupancy at promoters in snf5-deficient mammalian cells [57]. Both these results are in some ways at odds with the general view that SWI/SNF removes nucleosomes from DNA to allow RNAPII to elongate, thus activating gene expression. However, the function and composition of SWI/SNF is remarkably complex and there is good evidence for SWI/SNF also having a role in transcriptional silencing, likely through nucleosome positioning [58]. Our results may necessitate a reworking of how we think about activation of gene expression: to date, SWI/SNF was thought to activate gene expression by removing nucleosomes, but perhaps in some cases it does the opposite-by promoting deposition of nucleosomes that downregulate RNAPII elongation and encourage co-transcriptional splicing, leading to greater levels of steady-state processed mRNAs.
Our observations that sap145 overexpression led to higher level of nucleosomes at introncontaining genes was at first perplexing in light of the fact that we did not see a splicing defect in this strain. However, upon further contemplation, we contend that this nucleosome phenotype is actually compensation for the defect conferred by SF3 overexpression. Overexpression of SF3 alone is deleterious to the cell, in that it disrupts dynamics between SF3b, SF3a and Prp5 interactions [30] and likely interferes with Prp2-dependent spliceosome activation (this work).
We argue that this defect is sensed and, through an unknown feedback mechanism, leads to an increase in nucleosome deposition or stabilization of existing nucleosomes in order to inhibit RNAPII elongation and re-establish the timing and balance of co-transcriptional splicing. We believe that this "feedback" is dependent on SWI/SNF and that nucleosome occupancy cannot be restored in the absence of SWI/SNF (accounting for the low nucleosome signal observed when you combine the sap145 overexpression with deletion of snf5 (i.e. the double mutant phenocopies the Δsnf5 single mutant). These data fit nicely with reports from the Beggs and Neugebauer labs, both of which report RNAPII accumulation downstream of introns in S. cerevisiae [59,60]. The Beggs lab went on to show, quite elegantly, that mutating the branch point sequence in a reporter gene intron led to an accumulation of RNAPII around the 3'SS [60]. They refer to this accumulation as an RNAPII "pause" and propose that splicing itself acts as a checkpoint to stall RNAPII and promote co-transcriptional splicing. We propose that the presence of a nucleosome at or near the 3'SS could contribute to the RNAPII pause reported by the Beggs lab and also believe that splicing itself provides this checkpoint.
One outstanding question is how SWI/SNF and nucleosome occupancy directly impact RNA polymerase II elongation through open reading frames. While our ChIP data are consistent with a faster moving RNAPII decreasing ChIP signal, it is difficult to provide direct evidence of these dynamics in vivo. One line of experimentation that could address this question is combining SWI/SNF-SF3 double mutants with RNAPII "fast" and "slow" mutants that have been made and characterized in S. cerevisiae [18,61]. We would predict that if a slow polymerase mutant could phenocopy a paused polymerase at spliced genes, it would also alleviate the splicing defects seen in the double mutant. However, these mutants have yet to be made and validated in fission yeast.
The multiple connections between the splicing and chromatin machineries, and the specific regulatory connection we show here, support the exciting hypothesis that each of the ATPdependent steps of the splicing cycle could be regulated by a separate input-which may come from different chromatin remodeling or histone modifying complexes. Implications for such interplay are of great consequence, as mutations in both splicing and chromatin factors-SWI/SNF in particular-are known to promote tumorigenesis in metazoans [62]. Understanding connections between these pathways will be essential not only to elucidate pre-mRNA splicing mechanism, but also to create a full picture of tumor suppression.

Strains
A list of strains used in and made for this study are available in S2 Table. E-MAP analysis Analysis and generation of the E-MAP data was performed as previously described , Roguev et al., 2008). To identify splicing clusters, we performed a thresholded version of UPGMA hierarchical clustering using Pearson's correlation as the similarity metric. Initially each cluster contains single gene. Clusters were merged as long as the average Pearson correlation between their members was >0.3. Clusters generated using this approach are presented in Fig. 1 and S1 Table. To evaluate links between splicing factors and processes or complexes (Fig. 1B,D) we used the permutation method of Bandyopadhyay et al [64]. A p-value was calculated by comparing the mean observed interaction score between two groups to that expected by drawing 10 6 equal-sized random samples of interactions from the E-MAP. All p-values were corrected using the Bonferonni method.

RNA isolation and RT-qPCR
Cultures were grown overnight in YES5 to saturation and were diluted to 0.005 OD 600 to allow for overnight growth to mid log-phase. Strains were harvested by centrifugation between 0.5-0.8 OD 600 . Pellets were either snap-frozen in liquid nitrogen or immediately processed for RNA isolation. Total cellular RNA was isolated using hot acid phenol followed by isopropanol precipitation as described in (Bergkessel et al., 2011). Capped error bars represent Standard Error of the Mean (SEM) of 2-4 biological replicates. Uncapped error bars represent 3 technical replicates of a single biological sample.

Microarrays
S. pombe splicing-specific microarrays were designed by, and were a generous gift from, Jeffrey Pleiss. The arrays were designed to measure the relative (mutant vs. wild type) amount of every intron and constitutive junction in fission yeast. A single exon probe is used as a measure of total expression for each intron-containing gene. Microarray experiments were performed as outlined in Inada and Pleiss, 2010. Briefly, RNA was isolated as described above. cDNA synthesis was performed according to the manufacturer's protocol (Invitrogen, SuperScript III) but with slight modifications: 5 mg/ml dN9 random primers were added to 20 ug RNA per synthesis reaction. Reactions were left at 42°C for at least 4 hours. RNA was removed from samples by NaOH treatment and cDNA was labeled with Cy3 (wild-type) and Cy5 (mutant) dyes. Labeled cDNAs were hybridzed to Agilent arrays. Agilent protocols were followed for hybridization and washing conditions.

Microarray analysis
Splicing changes were expressed as (intron Ã exon)/junction, to take into account changes to overall gene expression for a given transcript. The R package limma was used to analyze the arrays and the data was normalized by loess normalization. To identify changes to cis-splicing signals that increased the likelihood of an intron being retained in our mutants, intron features were compiled and nucleotides were classified into common (>50% of introns in the fission yeast genome contain that nucleotide at that position) or rare (all other bases). Log odds and their standard errors were calculated based on a logistic regression model for each nucleotide position.

Chromatin immunoprecipitation (ChIP)
S. pombe strains were diluted to 0.005 OD 600 in 50-100 mL of YES5 media and allowed to grow overnight. Cultures were crosslinked in 1% formaldehyde for 15 minutes when they had reached mid-long phase OD 600 0.3-0.8. Formaldehyde was quenched with 125 mM glycine for 5 min. Crosslinked cells were spun down at 300 rpm and washed twice with cold dH 2 O. Pellets were flash frozen in liquid nitrogen and stored at -80°C or were immediately processed for ChIP. ChIP was performed as described in Kress et al., 2008, with certain changes for S. pombe: protease inhibitors were increased (1 mM PMSF, 20 μg/mL leupeptin), bead beating was increased to 1.5' x 10 cycles and sonication was increased to 6 cycles on high (10 min, 30 sec ON, 1 min OFF). Total and IPed DNA was analyzed by qPCR as described above. Capped error bars represent Standard Error of the Mean (SEM) of 2-4 biological replicates. Uncapped error bars represent 3 technical replicates of a single biological sample.

Nucleosome scanning assay
Mononucleosomes were prepared as described in Infante et al., 2012, except for the following changes. MNase treatment was performed for 5 minutes using 150U MNase, which resulted in >80% mononucleosome-sized fragments as assayed by agarose gel electrophoresis. Crosslinking was reversed in 1% SDS at 65°C overnight in the presence of Proteinase K (30μl of 30mg/ml). RNase treatment (10μl of 10mg/ml) followed. Total and mononucleosome DNA was analyzed by qPCR as described above using overlapping tiling primer sets. ) Average linkage clustering (using Cluster 3.0) of affected introns as measured by their splicing score ((intron Ã exon)/junction). Yellow indicates worse splicing (intron retention); blue indicates enhanced splicing. Numbers above heatmap indicate correlation coefficient. (B.) Number of introns whose splicing is defective (yellow) or enhanced (blue) for the single (Δsnf5, Δsol1, sap145 DAmP , sap62 DAmP ) and double mutants (Δsnf5 sap62 DAmP , Δsol1 sap62 DAmP , Δsnf5 sap145 DAmP , Δsol1 sap145 DAmP . Data are expressed as the average log fold change (logFC) log2 ratio of ((intron Ã exon)/junction) from 2-4 biological replicates. A significance threshold of logFC > 0. 5  RT-qPCR of expression levels conferred by the pREP4x vector of various SF3 factors in WT and Δsnf5 strain backgrounds. Bars are SF3 factor mRNA signal in pREP4x relative to level in a matched strain background expressing an empty vector, normalized to act1 mRNA levels. Severity of the growth defect prevented us from harvesting RNA from overexpression strains grown in inducing (minus thiamine) conditions. Therefore, experiment was conducted using yeast grown with thiamine (essentially giving a measure of the "leakiness" of the nmt1 promoter). n = average of 3 technical replicates. (C.) RT-qPCR of uap2(CUS2) expression levels in DAmP strain. Uap2 mRNA levels in WT and uap2 DAmP strains were normalized to act1 mRNA levels. (D.) Growth of SWI/SNF-Prp5-DPLD double mutants. Prp5-DPLD (Prp11 in S. pombe) mutants from the Query lab are deficient in Prp5's ability to bind SF3b. Each amino acid in the motif was changed to alanine, creating mutants of varying severity (DALD, DPAD). DPLD mutants were combined with Δsnf5 and Δsol1 strains and single and double mutants were grown at 30°C. (TIF)  Fig. 6. Data are represented as mononucleosome signal/undigested DNA signal at each primer set. Gene representation below is aligned by basepair. Circles represent approximate nucleosome location. Bold circles indicate nucleosomes downstream of introns. Colored background rectangles are used to group panels looking at the same gene locus (light purple = rps3; pink = gene poor region; beige = dbp2; light blue = act1). (E.) ChIP using antibodies against RNAPII CTD (antibody [8WG16], Neoclone) in WT, Δsnf5 and Δsnf5sap145 DAmP cells at the dbp2 gene. Error bars are ±SEM, n = 3 biological replicates (F.) ChIP using antibodies against RNAPII Ser5P CTD (antibody [4H8], Abcam) in WT, Δsnf5 and Δsnf5sap145 DAmP cells at dbp2 and rps3 genes. Data are shown as the change relative to WT (which is shown as 1). Error bars are ±SEM, n = 3 biological replicates. (G.) Biological replicates of experiment shown in Fig. 6E at rps3, dbp2 and act1 genes. Basepair alignments are indicated at the top of each graph. (TIF) S1 Table. Pairwise genetic interaction scores for splicing E-MAP. Splicing factor strains are on y-axis; library deletion strains are on x-axis. (ZIP) S2 Table. List of all splicing strains generated for splicing E-MAP, organized by snRNP. Clusters, as introduced in Fig. 1a are also detailed. S. cerevisiae orthologs are provided. (ZIP) S3 Table. Strain list. Detailed list of all strains described in this paper. (ZIP)