Type I PRMTs and PRMT5 Inversely Regulate Post-Transcriptional Intron Detention

Protein arginine methyltransferases (PRMTs) are required for the regulation of RNA processing factors. Type I enzymes catalyze mono- and asymmetric dimethylation; Type II enzymes catalyze mono- and symmetric dimethylation. To understand the specific mechanisms of PRMT activity in splicing regulation, we inhibited Type I and II PRMTs and probed their transcriptomic consequences. Using the newly developed SKaTER-seq method, analysis of co-transcriptional splicing revealed that PRMT inhibition resulted in slower splicing rates. Surprisingly, altered co-transcriptional splicing kinetics correlated poorly with ultimate changes in alternative splicing of polyadenylated RNA—particularly intron retention (RI). Investigation of RI following inhibition of nascent transcription demonstrated that PRMTs inversely regulate RI post-transcriptionally. Subsequent proteomic analysis of chromatin-associated polyadenylated RNA identified aberrant binding of the Type I substrate, CHTOP, and the Type II substrate, SmB. Targeted mutagenesis of all methylarginine sites in SmD3, SmB, and SmD1 recapitulated splicing changes seen with Type II PRMT inhibition. Conversely, mutagenesis of all methylarginine sites in CHTOP recapitulated the splicing changes seen with Type I PRMT inhibition. Closer examination of subcellular fractions indicated that RI were isolated to the nucleoplasm and chromatin. Together, these data demonstrate that PRMTs regulate the post-transcriptional processing of nuclear, detained introns through Sm and CHTOP arginine methylation.

ulated the splicing changes seen with Type I PRMT inhibition. Closer examination of subcellular 23 fractions indicated that RI were isolated to the nucleoplasm and chromatin. Together, these data 24 demonstrate that PRMTs regulate the post-transcriptional processing of nuclear, detained in- 25 trons through Sm and CHTOP arginine methylation.

INTRODUCTION 27
The mammalian genome encodes nine protein arginine methyltransferases (PRMTs 1-9; 28 PRMT4 is also known as CARM1). Arginine methylation is critical in regulating signal transduc- anisms that govern RI and the connection to PRMTs is of great interest to further advance our 45 understanding of this disease. 46 RNA splicing can occur during transcriptional elongation or after the transcript has been 47 process is an attractive means to control RI (Luo and Reed 1999). Despite this, the role of nu- 72 clear export factors and the impact of Type I PRMTs on regulating RI remains unexplored. 73 Here, we test both the co-and post-transcriptional consequences of PRMT5 or Type I PRMT 74 inhibition (PRMTi). We also examine which factors are responsible for PRMTi-mediated splicing 75 consequences. As both Type I PRMTs and PRMT5 have been extensively reported to be re-

83
Type I PRMT or PRMT5 inhibition promotes changes in alternative splicing 84 PRMTs consume S-adenosyl methionine (SAM) and produce S-adenosyl homocysteine 85 (SAH) to catalyze the post-translational methylation of either one or both terminal nitrogen atoms 86 of the guanidino group of arginine (Figure 1a) (Gary and Clarke 1998). All PRMTs can generate 87 monomethyl arginine (Rme1). Type I PRMTs further catalyze the formation of asymmetric 88 N G ,N G -dimethylarginine (Rme2a); Type II PRMTs (PRMT5 and 9) form symmetric N G ,N 'G -dime-89 thylarginine (Rme2s). PRMT5 is the primary Type II methyltransferase (Yang et al. 2015). As 90 previous reports indicated that lengthy treatment with PRMTi promotes aberrant RNA splicing, 91 we wanted to determine whether alternative splicing differences with PRMTi occurred as early 92 as day two and, if so, how they evolved over time (Bezzi et al. 2013;Fong et al. 2019; To confirm that these RI were specific to PRMTi and not a consequence of off-target effects 117 of GSK591 or MS023 we generated A549 cells expressing dCas9-KRAB-MeCP2 (Yeo et al. 118 2018). Independent transduction of two unique guide RNAs (gRNA) for four days targeting ei- 119 ther the major Type I methyltransferases-PRMT1 and PRMT4-or the Type II methyltransfer-120 ase, PRMT5 achieved efficient knockdown (Suppl. Figure 1a). Furthermore, we recapitulated 121 increased RI in GAS5 intron 9, ANKZF1 intron 9, and NOP2 intron 14 by RT-qPCR following 122 knockdown of PRMT5 with both unique gRNAs (Suppl. Figure 1b). We also observed re-123 duced RI in GAS5 intron 9, ANKZF1 intron 9, but not NOP2 intron 14 when analyzing RI fol- 124 lowing transduction with the stronger PRMT1 gRNA (Suppl. Figure 1b). Knockdown of 125 PRMT4 only reduced RI in GAS5 intron 9 with one gRNA, but not in ANKZF1 intron 9 or NOP2 126 intron 14. Together, these data support that the RI seen following Type I or PRMT5i are spe- 127 cific, yet the consequences of MS023 treatment on RI are likely more robust than CRISPR in-128 terference due to the inhibition of multiple Type I enzymes and the ability of PRMTs to scav-129 enge each other's substrates (Dhar et al. 2013;Eram et al. 2016). 130 PRMTi-mediated RI are conserved across cell types and share common characteristics 131 We next asked whether the RI in our data were common to other datasets in which PRMT 132 activity was perturbed. To accomplish this, we used rMATS on publicly available data (Braun et  PRMTi. As demonstrated by the high odds ratio (log2(OR) > 6) between all the datasets, we 138 showed that there was a highly significant (Fisher's exact adjusted P < 1e-05) overlap in RI (Suppl. Figure 2a). This further supports that the consequences of PRMTi on RI are not off-140 target effects of the inhibitors and are also common to multiple cell types. 141 RI have previously been reported to have common characteristics such as being shorter, 142 closer to transcription end site (TES), and having reduced splice site strength (Braunschweig et 143 al. 2014). To determine the common characteristics of the PRMT-regulated RI, we analyzed 144 their length, distances to the TES, and sequences. We found that the RI were significantly closer 145 to the TES and shorter when compared to the genomic distribution of A549 expressed introns 146 (P < 2.2e-16) (Figure 1f and 1g). Moreover, in analyzing the probability of nucleotide distribution 147 at the 5' and 3' splice sites we noted both a preference for guanine three nucleotides downstream 148 of the 5' splice site and increased frequency of cytosine in the polypyrimidine tract (Suppl. Fig-149 ure 2b). This is consistent with previous literature demonstrating that RI  The inverse relationship on RI by Type I or PRMT5i in addition to their TES-proximal location, 154 shorter length, and weaker splice sites led us to investigate the kinetics of co-transcriptional 155 splicing. To accomplish this, we used Splicing Kinetics and Transcript Elongation Rates by Se-156 quencing (SKaTER seq) (Casill et al. 2021). As splicing changes are present as early as two 157 days following GSK591 or MS023 treatment, we used this time point of PRMTi for our analysis. 158 Briefly, SKaTER seq uses a three-hour 5,6-dichloro-1-β-D-ribofuranosylbenzimidazole (DRB) 159 treatment to synchronize transcription, followed by a rapid wash-out to allow productive elonga-160 tion to commence. Once RNA pol II begins elongating, starting at 10 minutes nascent RNA is 161 collected every five minutes until 35 minutes post-DRB washout (Figure 2a). Nascent RNA isolated via a 1 M urea wash of chromatin and an additional poly(A)-RNA depletion is then se-163 quenced. The rate of nascent RNA formation-including: (1) RNA pol II initiation and pause-164 release (spawn) rate, (2) elongation rate, (3) splicing rate, and (4) transcript cleavage rate-is 165 then calculated by using a comprehensive model that determines the rates that best fit the se-166 quencing coverage (Casill et al. 2021). 167 To assess the accuracy of the rates determined by the SKaTER model, we used the spawn, 168 elongation, splicing, and cleavage rates to simulate a predicted poly(A)-RNA cassette exon Ψ 169 and compared these results with our poly(A)-RNA sequencing. We successfully predicted cas-170 sette exon Ψ detected in poly(A) RNA in DMSO (Spearman's correlation coefficient (ρ) = 0.54), 171 GSK591 (ρ = 0.61), and MS023 (ρ = 0.55) (P < 2.2e-16) treatment conditions (Figure 2b). Next, 172 we compared spawn rate to poly(A)-RNA sequencing transcripts per million (TPM). As predicted, 173 we observed a strongly significant (P < 2.2e-16) correlation between RNA pol II spawn rate and 174 TPM in DMSO-, GSK591-, and MS023-treated cells (ρ = 0.50, 0.51, 0.51, respectively) (Suppl. As poly(A)-RNA sequencing revealed opposing alternative splicing changes-specifically in-179 creased RI with GSK591 and decreased RI with MS023-we next asked how PRMTi affected 180 the global distribution of splicing rates relative to DMSO. Surprisingly, GSK591 treatment did not 181 significantly change the median of this distribution (Wilcoxon rank-sum test, P > 0.05), while 182 MS023 treatment resulted in significantly slower global splicing rates relative to DMSO (P < 2.2e-183 16) (Figure 2c). As our primary interest was in RI, we compared the splicing rates of introns that 184 were retained with PRMTi to the same introns in DMSO-treated cells. We found no significant difference in the median of GSK591-treated cells compared to DMSO (P = 0.53), while MS023 186 treatment similarly led to slower splicing rates (P = 0.07) (Figure 2d). This result was incon-187 sistent with the changes seen in poly(A) RNA as a slower splicing rate should increase RI. 188 Therefore, the co-transcriptional splicing rate suggested that changes in RI following PRMTi 189 were not due to co-transcriptional splicing. 190 RI share unique characteristics that limit their co-transcriptional removal 191 We next asked how RI splicing rates compared to non-RI. We observed that introns that 192 tended to be retained in GSK591-or MS023-treated cells were slower to splice (P = 0.01 and P 193 = 0.04, respectively) when compared to the genomic distribution ( Figure 2e). As we previously 194 found that RI were more likely to be closer to the TES and had slower splicing rates ( Figure 1f 195 and 1g), we asked whether intron position correlated with splicing rate. Indeed, we determined 196 that introns located closer to the TES had slower splicing rates (Suppl. Figure 4a).  amining additional characteristics of RI, we noted that they had a significantly higher GC fraction 198 in both GSK591-(0.52, P = 1.63e-15) and MS023-treated cells (0.50, P = 3.89e-6) relative to 199 the global median (0.41) ( Figure 2f). As GC content influences polymerase elongation, we an-200 alyzed the intronic elongation rate of RI in comparison to non-RI. Consistent with an increased 201 GC content, we found that RI had slower elongation rates relative to non-RI (P = 0.07 and 0.02 202 for GSK591 and MS023, respectively) ( Figure 2g) pol II took to transcribe RI compared to non-RI. We found that despite their slower elongation 207 rate, RI transcription was completed significantly faster compared to non-RI in both GSK591-and MS023-treated cells (P = 6.22e-17 and P = 8.12e-6, respectively) (Figure 2h) . Given the   209   proximity of RI to the TES, this likely further decreased their ability to be removed co-transcrip-210 tionally. Taken together, these results support the conclusion that RI in GSK591-and MS023-211 treated cells share unique characteristics that increase their probability of post-transcriptional 212 processing. 213 PRMTs regulate RI post-transcriptionally 214 The paradox that RI are decreased with Type I PRMTi-despite their slower co-transcriptional 215 splicing-led us to hypothesize that PRMTs exert their control over splicing post-transcription- 216 ally. To test this hypothesis, we used the elongation, splicing, and cleavage rates determined by 217 SKaTER seq to calculate the probability that a transcript will be cleaved from RNA pol II prior to 218 a given intron being spliced. Consistent with most splicing being co-transcriptional, we found 219 that the median probability of cleavage prior to splicing was 9.7% in DMSO-treated cells and 220 8.7% in GSK591-treated cells, yet the probability with MS023 treatment was significantly higher 221 at 18.1% (Figure 3a). The global distribution was significantly reduced with GSK591 treatment 222 (P = 0.002) and increased with MS023 treatment (P < 2.2e-16) when compared to DMSO (Fig-223 ure 3a). We next analyzed the probability of cleavage prior to splicing for RI. Consistent with the 224 hypothesis that PRMTs regulate splicing of RI post-transcriptionally, the median probability of 225 transcript cleavage prior to splicing was significantly higher for RI (50.6% and 63.2%, respec-226 tively) when compared to non-RI (11.7% and 23.1%, respectively; P = 5.74e-10 and 0.0002) 227 ( Figure 3b). Furthermore, intron position was strongly predictive of whether transcript cleavage 228 was likely to occur prior to splicing: TES proximal introns had a higher probability of cleavage 229 prior to their being spliced (Suppl. Figure 4b). All told, the proximity of RI to the TES, their shorter length, and their slower splicing rate likely drives their decreased probability of being 231 spliced co-transcriptionally. 232 The slowing of co-transcriptional splicing following PRMTi suggests that there would be an  249 We next validated these changes in RI following treatment with ActD using RT-qPCR for our 250 three candidate introns described above (Figure 1e). GSK591 treatment resulted in significantly 251 slower intron decay relative to DMSO for GAS5 intron 9 and NOP2 intron 14 but not ANKZF1 252 intron 9 after 60 minutes of ActD treatment (P = 0.06, 0.0005, 0.74, respectively) ( Figure 3f).
Conversely, intron decay was faster with MS023 treatment relative to DMSO for GAS5 intron 9 254 but not ANKZF1 intron 9 or NOP2 intron 14 (P = 0.0043, 0.27, 0.98, respectively) ( Figure 3f). 255 As the changes in RI following transcriptional inhibition with ActD reflect those of PRMTi alone-256 increased RI with GSK591 and decreased RI with MS023-these results support that PRMTs 257 regulate RI post-transcriptionally. 258 PRMTi alters binding of RNA processing factors to chromatin-associated poly(A) RNA 259 We next sought to identify the factors responsible for mediating the post-transcriptional con-  Table 1). Following GSK591 treatment, 118 had significantly altered abundance with 70 in-273 creased and 48 decreased (P < 0.05). In the MS023-treated input chromatin, we observed 255 274 differentially enriched proteins with 150 increasing in abundance and 105 decreasing (P < 0.05). 275 The poly(A) enriched fraction-after background subtraction-contained 1,251 unique proteins.
Of these, 32 were differentially bound in the GSK591-treated samples with 24 increased and 8 277 decreased (P < 0.05). In the MS023-treated cells, there were 55 proteins with altered binding-278 37 increased and 18 decreased (P < 0.05). 279 To identify the biological processes most affected by these inhibitors, we performed over rep-  Figures 5a and 5b). In the chro-281 matin fraction of both GSK591-and MS023-treated cells-consistent with the gross aberrations 282 in splicing following treatment with these inhibitors-we observed a significant enrichment of 283 gene ontology terms for RNA splicing and ribonucleoprotein complex biogenesis (Padj < 0.05) 284 (Suppl. Figure 5a). Interestingly, in GSK591-treated cells, there was also a unique enrichment 285 for the regulation of protein localization to Cajal bodies (Padj < 0.05). Conversely, in MS023-286 treated cells we observed a highly significant enrichment for nuclear RNA export and the regu-287 lation of RNA catabolism (Padj < 0.05). When looking specifically at the poly(A)-enriched frac-288 tion, we noted similar themes as those in the input chromatin (Suppl. Figure 5b). 289 PRMTi perturbs binding of nuclear export factors and snRNPs to poly(A) RNA 290 Of the proteins that were differentially bound to the input chromatin and poly(A) RNA following 291 treatment with PRMTi, we were interested in candidates that were most likely to mediate post-292 transcriptional RI. To accomplish this, we took into consideration the enriched gene ontology 293 categories described above and highlighted the most significant factors involved in RNA splicing, 294 transport, and degradation relative to their log2 fold change (Figure 4b and 4c). In the input 295 chromatin of GSK591-treated cells, we observed increased signal in the snRNP chaperone,

342
To address the question of whether CHTOP, ALYREF, or SNRPB were involved in regulation 343 of RI, we first checked whether there was any publicly available RNA-seq data in which these 344 proteins were perturbed. We found two independent datasets where SNRPB was knocked down in U251 glioblastoma cells or HeLa cells ( forming rMATS on these datasets, we observed that both SNRPB knockdown (SNRPBkd) ex-349 periments strongly recapitulated the increase in RI seen with GSK591 treatment (Figure 5a). 350 Conversely, although ALYREF knockdown did not significantly affect RI levels (Suppl. Figure   351 Figure 6b). 380 However, likely owing to the meticulous control of total Sm levels by the cell, we were unable to 381 achieve ample overexpression with the wildtype vector (Suppl. Figure 6c) (Prusty et al. 2017). 382 We observed a migratory shift toward a lower molecular weight in the R-to-A mutants when 383 compared to the wildtype or R-to-K mutants (Suppl. Figure 6c). We also observed a strong 384 Rme2s signal on the FLAG-SmB wildtype protein that was absent on the R-to-A and R-to-K 385 mutants (Suppl. Figure 6c). Furthermore, and consistent with a role for Sm methylarginine in 386 regulating RI levels, following Sm R-to-A mutagenesis, we observed increased inclusion in all 387 three RI candidates (P < 0.05) (Figure 5f). We also detected a significant increase in GAS5 388 intron 9 (P < 0.001) with the Sm R-to-K mutants and a trend toward increased RI in ANKZF1 389 intron 9 (P = 0.23) and NOP2 intron 14 (P = 0.16). Thus, mutagenesis of Sm methylarginine sites increased RI in our candidate transcripts, yet the greater effect of the R-to-A mutants sug-391 gests that the charge of arginine itself may also play an important role in regulating RI levels. 392 CHTOP arginine mutants decrease RI 393 We performed similar experiments with CHTOP whereby we mutated 30 arginines present 394 within the centrally located "GAR" motif-the preferred PRMT1 substrate recognition motif-to 395 either alanine or lysine (Figure 5g). This region of CHTOP has been previously shown to be 396 required for PRMT1-catalyzed methylarginine (Van Dijk et al. 2010). We transduced A549 cells 397 with the constructs and performed RT-qPCR for GAS5 intron 9, ANKZF1 intron 9, or NOP2 intron 398 14. We achieved similar expression of the mutant CHTOP proteins when compared to the 399 wildtype at the level of RNA and protein (Suppl. Figure 6c and 6d). Interestingly, when trans-400 ducing cells with the wildtype CHTOP, we noted a gross downregulation of the endogenous 401 transcript (Suppl. Figure 6e). CHTOP has been previously reported to control its own expres-402 sion as part of an autoregulatory loop (Izumikawa et al. 2016). We did not observe this compen-403 sation with either the R-to-A or R-to-K mutants. Consistent with the gel shift in CHTOP seen 404 following MS023 treatment, we observed a similar change in the R-to-A mutant when compared 405 to the wildtype or R-to-K mutants. Moreover, supporting a role for CHTOP methylarginine in 406 Type I PRMT-dependent RI inclusion, in both R-to-A and R-to-K mutants we observed de-407 creased inclusion in GAS5 intron 9 (P < 0.001) (Figure 5h). We also saw a trend toward de-408 creased inclusion of ANKZF1 intron 9 (P = 0.23 and 0.26) and significantly decreased inclusion 409 of NOP2 intron 14 (P < 0.0001) in the R-to-K mutant, but not the R-to-A mutant. Taken together, 410 these experiments support that CHTOP methylarginine is involved in regulating RI levels. 411 PRMT-regulated RI are detained within the nucleoplasm and chromatin PRMT5 has been proposed to specifically regulate DI-introns that persist in poly(A) RNA but 413 remain nuclear (Braun et al. 2017). However, direct experimental evidence-namely fractiona-414 tion of subcellular compartments to identify the location of RI-has been lacking. To address the 415 question of whether the RI in our data are nuclear and therefore DI, we first ran rMATS on pub-416 licly available ENCODE poly(A)-RNA seq from cytoplasmic and nuclear fractions of A549 cells 417 (ENCSR000CTL and ENCSR000CTM, respectively). We observed an enrichment of RI within 418 the nuclear fraction-97% of significant RI events (FDR < 0.05) had a +ΔΨ, where ΔΨ is the 419 difference in Ψ between the nuclear and cytoplasmic compartments. We then intersected these 420 data with the PRMTi-mediated RI and noted that there was a significant overlap with both 421 GSK591-(log2 OR 8.89, P < 1e-300) and MS023-treatment (log2 OR 7.75, P < 1e-300) when 422 compared to all A549 expressed introns (Figure 6a). We also observed that GAS5 intron 9, 423 ANKZF1 intron 9, and NOP2 Intron 14 were significantly increased in nuclear poly(A) RNA 424 (GAS5 and ANKZF1 not shown) (Figure 6b). 425 To validate that these RI remain nuclear following PRMTi, and to further resolve their lo-426 calization within the nucleus, we treated A549 cells with PRMTi and isolated cytoplasmic, nucle-427 oplasmic, and chromatin fractions. We then quantified levels of GAS5 intron 9, ANKZF1 intron 428 9, and NOP2 intron 14 or non-RI within the same poly(A) transcripts (Figure 6c). In accordance 429 with the RI being localized to the nucleus, in DMSO-, GSK591-, and MS023-treated cells, we 430 observed a significant increase in nucleoplasmic and chromatin signal compared to the cyto-431 plasm (P < 0.05) (Figure 6c). Importantly, there was no significant difference between cytoplas-432 mic or nuclear enrichment of the non-RI tested. Together, these data support that these RI are The peptide solution was pooled, spun at 1,000 x g for 30 sec and dried in a vacuum centrifuge. 615 Prior to mass spectrometry analysis, samples were desalted using a 96-well plate filter for publicly available RNA-seq data used in this study can be found in Supplemental Table 3.      T  T  T  T T T  T  T  T  T T   A A A  A  A  A  A A A   Bead   T  T  T  T T T  T  T  T  T

PRMT5
Productive splicing and export  RNA transport nucleic acid transport protein targeting to ER mRNA catabolic process viral transcription rRNA processing regulation of mRNA metabolic process ribonucleoprotein complex subunit organization rRNA metabolic process ribonucleoprotein complex assembly protein−containing complex localization nucleocytoplasmic transport ribonucleoprotein complex localization translational initiation cotranslational protein targeting to membrane RNA localization nucleobase−containing compound transport nuclear−transcribed mRNA catabolic process, NMD viral gene expression ribonucleoprotein complex biogenesis RNA splicing, via transesterification reactions mRNA splicing, via spliceosome RNA splicing, via transesterification reactions with bulged adenosine nuclear export regulation of nucleobase−containing compound transport translational initiation protein localization to endoplasmic reticulum RNA export from nucleus viral gene expression mRNA catabolic process RNA catabolic process regulation of mRNA metabolic process RNA localization protein localization to Cajal body protein localization to nuclear body positive regulation of protein localization to Cajal body regulation of protein localization to Cajal body rRNA metabolic process cell−cell junction assembly rRNA processing telomere organization telomere maintenance protein localization to plasma membrane regulation of RNA splicing ribonucleoprotein complex subunit organization ribonucleoprotein complex assembly ribonucleoprotein complex biogenesis RNA splicing