Identification of RBPMS as a smooth muscle master splicing regulator via proximity of its gene with super-enhancers

Alternative splicing (AS) programs are primarily controlled by regulatory RNA binding proteins (RBPs). It has been proposed that a small number of master splicing regulators might control cell-specific splicing networks and that these RBPs could be identified by proximity of their genes to transcriptional super-enhancers. Using this approach we identified RBPMS as a critical splicing regulator in differentiated vascular smooth muscle cells (SMCs). RBPMS is highly down-regulated during phenotypic switching of SMCs from a contractile to a motile and proliferative phenotype and is responsible for 20% of the AS changes during this transition. RBPMS directly regulates AS of numerous components of the actin cytoskeleton and focal adhesion machineries whose activity is critical for SMC function in both phenotypes. RBPMS also regulates splicing of other splicing, post-transcriptional and transcription regulators including the key SMC transcription factor Myocardin, thereby matching many of the criteria of a master regulator of AS in SMCs.


Introduction
Alternative splicing (AS) is an important component of regulated gene expression programmes during cell development and differentiation, usually focusing on different sets of genes than transcriptional control (Blencowe, 2006). AS programs re-wire protein-protein interaction networks (Buljan et al., 2012;J. D. Ellis et al., 2012;5 Yang et al., 2016), as well as allowing quantitative regulation by generating mRNA isoforms that are differentially regulated by translation or mRNA decay (McGlincy & Smith, 2008;Mockenhaupt & Makeyev, 2015). Coordinated cell-specific splicing programs are determined by a combination of cis-acting transcript features and transacting factors that compose "splicing codes" (Barash et al., 2010;Chen & Manley, 10 2009; X. D. Fu & Ares, 2014). The interactions between the trans component RNA binding proteins (RBPs) and the cis component regulatory elements in target RNAs coordinate the activation and repression of specific splicing events. Many regulatory proteins, including members of the SR and hnRNP protein families, are quite widely expressed, while others are expressed in a narrower range of cell types (David & 15 Manley, 2008;X. D. Fu & Ares, 2014). A further conceptual development of combinatorial models for splicing regulation has been the suggestion that a subset of RBPs act as master regulators of cell-type specific AS networks (Jangi & Sharp, 2014).
The criteria expected of such master regulators include that: i) they are essential for cell-type specification or maintenance, ii) their direct and indirect targets are important 20 for cell-type function, iii) they are likely to regulate the activity of other splicing regulators, iv) they exhibit a wide dynamic range of activity, which is not limited by autoregulation, and v) they are regulated externally from the splicing network, for example by transcriptional control or post-translational modifications. It was further suggested that expression of such splicing master regulators would be driven by 25 transcriptional super-enhancers, providing a possible route to their identification (Jangi & Sharp, 2014). Super-enhancers are extended clusters of enhancers that are more cell-type specific than classical enhancers and that drive expression of genes that are essential for cell-type identity, including key transcription factors (Hnisz et al., 2013).
By extension, RBPs whose expression is driven by super-enhancers are expected to 30 be critical for cell-type identity and might include master regulators of tissue-specific AS networks (Jangi & Sharp, 2014).
The transcriptional control of SMC phenotypic switching has been intensely studied, 40 but the role of post-transcriptional regulation has been relatively neglected (Fisher, 2010). For example, some markers of the contractile state, such as h-Caldesmon and meta-Vinculin, arise via AS (Owens et al., 2004), but nothing is known about the regulation of these events. A number of known splicing regulators, including PTBP1, CELF, MBNL, QKI, TRA2B, and SRSF1, have been implicated in the regulation of 45 individual SMC-specific ASEs, but these proteins are not restricted to differentiated SMCs and most act primarily in the de-differentiated state (Gooding et al., 2013;Gooding, Roberts, & Smith, 1998;Gromak, Matlin, Cooper, & Smith, 2003;Shukla & Fisher, 2008;van der Veer et al., 2013;Xie et al., 2017). Indeed, global profiling confirmed a widespread role of PTBP1 in repressing exons that are used in 50 differentiated mouse aorta SMCs (Llorian et al., 2016), but did not identify RBPs that act as direct regulators of the differentiated state. Biochemical identification of such RBPs is hampered by the fact that SMCs rapidly dedifferentiate in cell culture conditions.
Here, we used the approach suggested by Jangi and Sharp, to identify 55 candidate AS master regulators as RBP-encoding genes whose cell-specific expression is driven by super-enhancers (Jangi & Sharp, 2014). We identified RNA Binding Protein with Multiple Splicing (RBPMS), a protein not previously known to regulate splicing, as a critical regulator of numerous AS events in SMCs. RBPMS is highly expressed in differentiated SMCs, where it promotes AS of genes that are 60 important for SMC function. These include many components of the actin cytoskeleton and focal adhesion machineries, modulation of whose function is key to the transition from contractile to motile phenotypes. RBPMS also targets other splicing regulators, post-transcriptional regulators and the key SMC transcription factor Myocardin where RBPMS promotes inclusion of an exon that is essential for maximal SMC-specific 65 activity. RBPMS therefore meets many of the criteria expected of an AS master regulator in SMCs, and its identification validates the approach of identifying key cellspecific regulators via the super-enhancer-proximity of their genes.

RBPMS is a SMC splicing regulator
To identify potential SMC master splicing factors we used a catalog of 1542 human RBPs (Gerstberger, Hafner, & Tuschl, 2014) and data-sets of super-enhancers from three human SMC-rich tissues (aorta, bladder, stomach smooth muscle) and skeletal 75 muscle (Hnisz et al., 2013)(Supplementary File 1). Nine RBP genes were associated with super-enhancers in all SMC tissues but not skeletal muscle (Fig 1B). Using a set of super-enhancer-associated genes in the dbSUPER database (Khan & Zhang, 2016), we identified two candidates, of which only RBPMS was shared with our original 9 candidates. Examination of RBPMS expression in human tissues from the 80 Genotype-Tissue Expression (GTEX) project (Consortium, 2013) showed the top 8 expressing tissues to be SMC rich, including three arteries (Fig. S1A). RNA-Seq data from rat aorta SMCs showed that Rbpms levels decreased 3.8 fold during dedifferentiation from tissue to passage 9 cell culture , in parallel with known SMC transcriptional markers (Acta2, Cnn1, Smtn, (Fig. S1C,D) and AS events (Tpm1 85 and Actn1, Fig. S1E,F). Moreover, of the starting candidate RBPs (Fig 1B) Rbpms was the most highly expressed of those that were down-regulated from tissue to culture (Fig. S1C green labels). Expression of the Rbpms2 paralog also decreased upon dedifferentiation, but its absolute level of expression was >10-fold lower than Rbpms ( Fig. S1C,D). Other RBPs implicated in AS regulation in SMCs (PTBP1, MBNL1, QKI) 90 either showed more modest changes or higher expression in de-differentiated cells (Fig. S1C,D). In the PAC1 rat SMC line Rbpms mRNA levels decreased by ~10-fold between differentiated and proliferative states in parallel with SMC marker AS events ( Fig. 1D) and genes (Fig. 1E). Rbpms2 was expressed at much lower levels than Rbpms, and did not alter expression between PAC1 cell states ( Fig 1E). RBPMS 95 protein decreased to undetectable levels in proliferative PAC1 cells in parallel with smooth muscle actin (ACTA2) (Fig 1F). Immunofluorescence microscopy also showed higher levels of RBPMS in differentiated PAC1 cells where it was predominantly nuclear (Fig 1G), consistent with the hypothesis that it regulates splicing.
To further investigate RBPMS expression we cloned cDNAs from PAC1 cells, 100 representing seven distinct mRNA isoforms. These encoded two major protein isoforms (RBPMSA and RBPMSB) sharing a common N-terminus and RNA Recognition Motif (RRM) domain. RBPMSA and B differed by short C-terminal tails encoded by alternative 3' end exon 7 and exon 8 (Fig. 1C), and corresponded in size to the two protein bands seen in western blots (Fig. 1F). Other mRNA isoforms differed 105 by inclusion or skipping of exon 6 and by alternative 3' UTR exon inclusion. RBPMS and RBPMS2, which are 70% identical, have a single RRM domain that is responsible for both RNA binding and dimerization (Sagnol et al., 2014;Teplova, Farazi, Tuschl, & Patel, 2016). Optimal RBPMS binding sites consist of tandem CACs separated by a spacer of ~1-12 nt (Farazi et al., 2014;Soufari & Mackereth, 2017). We found 110 significant enrichment of CACN1-12CAC motifs within and upstream of exons that are less included in differentiated compared to cultured rat aorta SMCs (Fig. S1H). These are locations at which many splicing regulators mediate exon skipping (X. D. Fu & Ares, 2014;Witten & Ule, 2011). Consistent with this, the SMC-specific mutually exclusive exon pairs in Actn1 and Tpm1 (Gooding & Smith, 2008;Southby, Gooding, 115 & Smith, 1999) both have conserved clusters of CAC motifs upstream of the exon that is skipped in differentiated SMCs (see below).
In summary, the presence of super-enhancers at the RBPMS gene in SMC-rich tissues, its wide dynamic range of expression between differentiated SMCs and other tissues and proliferative SMCs (Fig. 1, S1), the nuclear localization of RBPMS, and the 120 presence of potential RBPMS binding sites adjacent to known SMC-regulated exons are all consistent with the hypothesis that RBPMS might act as a master regulator of AS in differentiated SMCs.

RBPMS promotes a differentiated SMC splicing program
To investigate the roles of RBPMS in shaping SMC transcriptomes we 125 manipulated levels of RBPMS expression in differentiated and proliferative PAC1 cells ( Fig. 2A,B). We used siRNAs to knockdown all Rbpms isoforms in differentiated PAC1 cells, achieving ~75% depletion with no effect on the SMC marker ACTA2 (Fig. 2B). In parallel, proliferative PAC1 cells were transduced with pINDUCER lentiviral vectors (Meerbrey et al., 2011) to allow Doxycycline inducible over-expression of RBPMSA. 130 No basal RBPMS expression was observed in proliferative cells, but upon induction substantial expression was observed from the FLAG-RBPMSA vector but not from the empty lentiviral vector (LV) (Fig. 2B). The effects of manipulating RBP levels can sometimes be compensated by related family members (Mockenhaupt & Makeyev, 2015). However, Rbpms2 levels were not affected by any of the treatments and 135 Rbpms2 knockdown, either alone or in combination with Rbpms, had no effects upon tested AS events (data not shown).
RNA samples from Rbpms knockdown and overexpression experiments were prepared for Illumina poly(A) RNAseq. Data was analysed for changes in mRNA abundance using DESEq2 (Love, Huber, & Anders, 2014) (Fig. 2C, Supplementary  140 File 2), and for changes in AS using rMATS (Shen et al., 2014) Fig. S2A). This suggests that at the level of mRNA abundance the differences between differentiated and proliferative PAC1 cells far outweigh any effects of manipulating RBPMS levels. Consistent with this, RBPMS knockdown was associated with ~10-fold fewer changes at the transcript abundance level (110 increased, 82 decreased, padj < 0.05, fold change > 2-fold) compared to the 150 differentiated vs proliferative control comparison (830 increased, 1444 decreased, Fig.   2C). RBPMS overexpression led to an intermediate number of changes in mRNA abundance levels, but only 29 genes were affected by both overexpression and knockdown, and of these only 4 genes other than Rbpms were regulated reciprocally.

RBPMS knockdown and overexpression had substantial effects at the level of 155
AS affecting all types of AS event (Figs. 2D, S2B). RBPMS knockdown led to changes in 318 AS events (FDR < 0.05, |DPSI| > 0.1, where PSI is Percent Spliced In), which was only 2-fold less than the number of events regulated between control differentiated and proliferative cells. Cassette exons were the largest group of events, with roughly equal numbers of up-and down-regulated exons (Fig. 2D). RBPMS overexpression 160 led to a larger number of AS changes (4934 regulated events), probably resulting from the combination of RBPMS expression in excess of levels usually present in differentiated PAC1 cells (Fig. 2C) and also because we expressed the more active RBPMSA isoform (see below). Cassette exons affected by RBPMS overexpression were strongly skewed (80%) towards greater exon skipping. A subset of ASEs 165 observed in the RNA-Seq experiments, encompassing RBPMS activated and repressed cassette exons and mutually exclusive exons were validated by RT-PCR ( Fig. 2E,F, Fig. S2E,F). DPSI values determined by RT-PCR and RNA-Seq were in good agreement (Fig. S2G). As an additional negative control, doxycycline induction of empty lentiviral transduced cells had no effect on RBPMS regulated AS events (Fig.  170   S2D).
In each of the three comparisons, there was little overlap between genes regulated at the levels of splicing and mRNA abundance (Fig. S2C). However, there were substantial overlaps of ASEs regulated by RBPMS knockdown, overexpression and PAC1 phenotype ( Fig 3A). Twenty percent of ASEs regulated in PAC1 cell 175 differentiation were congruently regulated by RBPMS knockdown, representing 40% of ASEs affected by RBPMS knockdown (Fig 3A, 3B left panel). The high correlation (R 2 = 0.95) suggests that for these 127 events changes in RBPMS expression are sufficient to explain their differentiation-specific splicing changes. Similarly RBPMS overexpression shared 180 events in common with PAC1 differentiation status (28% 180 of differentiation specific events). The correlation between DPSI for AS events coregulated in PAC1 differentiation and RBPMS overexpression (R 2 = 0.86) was lower than for RBPMS knockdown and represented only 3.7% of events regulated by RBPMS overexpression (Fig 3A,3B right panel). Hierarchical clustering of cassette exons regulated between PAC1 phenotypes across all 12 samples also revealed two 185 clusters of RBPMS-responsive events where knockdown and overexpression were sufficient to reciprocally convert the splicing pattern to that of the other cellular phenotype (Fig. 3E, clusters 1 and 4, containing RBPMS activated and repressed exons respectively).
Many AS events in differentiated PAC1 cells do not reach the fully differentiated 190 splicing pattern characteristic of tissue SMCs. However, for some events overexpression of RBPMS led to splicing patterns similar to tissue SMCs. For example the Actn1 SM exon is included to 69% in differentiated PAC1 cells, but to 93-94% in RBPMS overexpressing cells ( Fig 2E) and aorta tissue (Fig. S1F). We therefore hypothesized that some ASEs regulated by RBPMS overexpression but not knock-195 down, might reflect tissue SMC AS patterns that are not usually observed in cultured SMCs. To address this possibility we used RNA-Seq data monitoring de-differentiation of rat aorta SMCs (Fig. S1B,C). Of 1714 ASES regulated between tissue and passage 9 SMCs, 265 (15%) were also regulated by RBPMS overexpression in PAC1 cells (Fig.   3C,D, R 2 = 0.68). Strikingly, hierarchical clustering of cassette exons regulated 200 between aorta tissue and passage 9 cultured SMCs showed that the RBPMS overexpression sample clustered together with tissue, away from all other samples exons that are skipped in tissue and upon RBPMS overexpression (e.g. Tsc2, Fig. 3G), the majority of which are not regulated in PAC1 differentiated cells. Likewise Tpm1 mutually exclusive exon 3 is skipped nearly completely upon RBPMS overexpression in a similar pattern to tissue (Fig. S3B). Upregulation of RBPMS expression therefore appears to be sufficient to promote a subset of splicing patterns usually only observed 215 in differentiated SMCs in vivo.

RBPMS directly regulates exons with associated CAC motifs
To address whether RBPMS directly regulates target exons we looked for enrichment of its binding motif (Farazi et al., 2014) adjacent to cassette exons regulated by RBPMS 220 knockdown or overexpression. CACN1-12CAC motifs were significantly enriched around exons that were activated or repressed by RBPMS, with a similar positiondependent activity as other splicing regulators ( Fig 4A). Exons repressed by RBPMS showed strong enrichment of motifs within the exon and the immediate ~80 nt upstream intron flank, while exons activated by RBPMS showed motif enrichment 225 within the downstream intron flank. Consistent with the contribution of RBPMS to the AS changes between PAC1 differentiation states, CACN1-12CAC motifs were also enriched upstream of and within exons that are more skipped in differentiated cells, and downstream of exons that are more included in differentiated cells ( Fig 4A).
Moreover, in the set of exons activated by RBPMS overexpression, CACN1-12CAC 230 motifs were not only enriched downstream, but also significantly depleted in the repressive locations within and upstream of the exon (Fig. 4A,B). This suggests that binding in repressive locations might be dominant over activation.
To test whether RBPMS regulates AS events by directly binding to CAC motifs we cotransfected HEK293 cells with RBPMS expression vectors and minigenes of 235 representative activated (Flnb) and repressed (Tpm1, Actn1) exons, with potential RBPMS binding sites in expected locations for activation or repression ( Fig. S4C-E).
We initially established that transient expression of RBPMSA in HEK293 cells was sufficient to switch AS of endogenous FLNB, TPM1, MPRIP and ACTN1 towards the SMC splicing pattern (Fig 4C,G, Fig. S4A). For comparison, we also transfected 240 expression constructs for the RBPMSB isoform and the paralog RBPMS2. We found that RBPMSB had lower activity for some events (ACTN1, Fig. S4A), but that transfected RBPMS2 had similar activity to RBPMSA in all cases. RBPMSA and RBPMS2 also strongly activated inclusion of the Flnb H1 exon in a minigene context while RBPMSB activated to a lower extent ( Fig 4D). Tpm1 exon 3 is the regulated 245 member of a pair of mutually exclusive exons and is repressed in SMCs (P. D. Ellis, Smith, & Kemp, 2004;Gooding, Roberts, Moreau, Nadal-Ginard, & Smith, 1994).
MBNL and PTBP proteins promote this repression but are not sufficient to switch splicing (Gooding et al., 2013;Gooding et al., 1998). In contrast, RBPMSA expression was sufficient to cause a near complete switch from exon inclusion to skipping (Fig  250   4H). RBPMS2 had lower activity, but RBPMSB was by far the least active protein.
Likewise, RBPMSA and RBPMS2 completely switched splicing of Actn1 constructs (Gromak et al., 2003;Southby et al., 1999) from the NM to the SM mutually exclusive exon, while RBPMSB was nearly inactive (Fig. S4B). A construct containing only the Actn1 SM exon was unresponsive to cotransfection, while a construct containing only 255 the NM exon, which has 3 upstream CAC clusters (Fig. S4C) showed a complete switch from inclusion to skipping upon cotransfection of RBPMSA or RBPMS2, but not RBPMSB. Thus, for two mutually exclusive events RBPMSA is able to switch splicing to the SMC pattern by repressing the exon that is usually used in non-SMCs, while RBPMSB is less active. 260 We mutated CAC motifs in suspected binding sites to CCC, which disrupts RBPMS binding (Farazi et al., 2014). Mutation of 12 CACs downstream of the FlnB H1 exon had no effect on basal splicing, but the H1 exon was completely resistant to activation by RBPMSA ( Fig 4E). Likewise, mutation of 9 CAC motifs upstream of Tpm1 exon 3 265 had no effect on exon inclusion in the absence of RBPMSA, but completely prevented exon skipping in response to RBPMSA (Fig 4I, Fig. S4E), while mutations of individual clusters had intermediate effects (Fig. S4E). The response of both Flnb H1 exon and Tpm1 exon 3 therefore depends on nearby CAC motifs. To test whether these are binding sites for RBPMS, we used in vitro transcribed RNAs and recombinant 270 RBPMSA and B (Fig. 4F, J, Fig. S4F,G). Using both electrophoretic mobility shift assay (EMSA) and UV crosslinking, both RBPMSA and B were found to bind to the FlnB wild type RNA, but not to the mutant RNA (Fig. 4F). With Tpm1, RBPMSA and B bound to the WT RNA as indicated by EMSA assays (Fig. 4J). Binding was reduced by the mutations that abrogated RBPMSA repression of Tpm1 exon 3. These data therefore 275 show that RBPMS can inhibit splicing by binding to sites upstream of exons (Tpm1) and activate splicing by downstream binding (Flnb).

RBPMS regulated splicing targets the cytoskeleton and cell adhesion
To investigate the functional importance of RBPMS regulated AS we carried out Gene 280 Ontology (GO) analysis of the genes whose splicing or expression levels were regulated by RBPMS (Fig5A, Fig. S5A, Tables S4, S5), or between SMC differentiation states ( Fig. S5B,C). Splicing events affected by RBPMS-knockdown affected genes involved in processes, components and functions important for SMC biology, such as cytoskeleton, cell projection, cell junction organization and GTPase regulation (Fig  285   5A). These categories were very similar to those affected by AS during PAC1 cell dedifferentiation (Fig. S5B). Events regulated by RBPMS knockdown were also enriched within genes that are associated with super-enhancers in SMC tissues ( Fig.   5B) and therefore inferred to be important for SMC identity. In contrast, the relevance of GO terms associated with RBPMSA overexpression to SMC biology was less clear 290 ( Fig. S5A), and overexpression AS targets were not enriched for aorta super-enhancer associated genes (p = 0.37). Likewise, at the RNA abundance level enriched GO terms associated with RBPMS knockdown or overexpression did not align with those associated with differentiation, mainly being associated with stress responses (Supplementary File 5). 295 To further explore the consequences of RBPMS regulation we carried out network analysis using STRING (Szklarczyk et al., 2017) (Fig. 5C). We used only target AS events that are coregulated by RBPMS knockdown and PAC1 differentiation state (Fig. 3A,E) or by RBPMS overexpression and in aorta tissue (Fig. 3C,F), and restricted the output network to high confidence interactions. RBPMS targets comprised a 300 network (Fig. 5C) focused on functions associated with cell-substrate adhesion (yellow nodes) and the actin cytoskeleton (blue). Six of the network proteins (SORBS1, CALD1, PDLIM5, PDLIM7, ACTN1 and ARHGEF7) are also components of the consensus integrin adhesome (Horton et al., 2015), which mechanically connects, and mediates signalling between, the actin cytoskeleton and the extracellular matrix. 305 Underlining the importance of this network to SMC function, many of the genes are themselves super-enhancer-associated in one or more smooth muscle tissues (bold text, Fig. 5C). Actomyosin activity, and its mechanical connection to the extracellular matrix are important for both the contractile and motile states of SMCs (Min et al., 2012). It appears that RBPMS plays an important role in modulating the activity of this 310 protein network to suit the needs of the two cell states. This control of a functionally coherent set of targets that are important for SMC function is consistent with the hypothesis that RBPMS is a master regulator of AS in SMCs.

RBPMS controls post-transcriptional regulators 315
Among the direct targets of master splicing regulators are expected to be events controlling the activity of other AS regulators leading to further indirect AS changes.
Consistent with this, we noted that RBPMS affected splicing of Mbnl1 and Mbnl2.
RBPMS promoted skipping of the 36 nt and 95 nt alternative exons of Mbnl1 (here referred to as exons 7 and 8), as indicated by both knockdown and overexpression 320 ( Fig. 6A,B). In Mbnl2 the 36 nt exon was fully skipped in all conditions but the 95 nt exon was repressed by RBPMS (Fig. 6C). Consistent with the RNA-Seq data, we obtained Mbnl1 and Mbnl2 cDNAs from PAC1 cells that varied by inclusion of the 36 and 95 nt exons. Corresponding shifts in MBNL1, but not MBNL2, protein isoforms could be observed by western blot (Fig 6D). These events affect the unstructured C-325 termini of MBNL proteins and have been shown to affect their splicing activity (Sznajder et al., 2016;Tabaglio et al., 2018;Tran et al., 2011), suggesting that RBPMS might indirectly affect some AS events by modulating MBNL activity. To test this, we made expression constructs of rat MBNL1 isoforms with and without exons 7 and 8 (FL, D7, D8, D7D8) and MBNL2 with and without exon 8 (FL and D8), obtained as cDNAs from 330 PAC1 cells. When transfected into HEK293 cells, full length MBNL1 and 2 caused a shift from use of the downstream to an upstream 5' splice site (5'SS) on NCOR2 exon 47 (Fig 6E), an event that is differentially regulated by MBNL isoforms (Sznajder et al., 2016;Tran et al., 2011). The shorter MBNL isoforms showed lower activity in shifting towards the upstream 5'SS, although the difference was not statistically significant for 335 the MBNL1 shorter isoforms (Fig. 6E). In proliferative PAC1 cells, NCOR2 exon 47 mainly uses the upstream 5'SS, and knockdown of MBNL1 and 2 caused a significant shift to the downstream 5'SS ( Fig. 6F). Overexpression of RBPMSA, also caused a small shift to the downstream 5'SS ( Fig. 6F, also detected by rMATs, DPSI = 13%, FDR = 2 x 10 -6 ), but had no effect when MBNL1 and MBNL2 were knocked down. 340 These results are consistent with MBNL1 and 2 being direct regulators of Ncor2 splicing, with RBPMS acting indirectly by promoting production of less active MBNL isoforms.
Another post-transcriptional regulator affected by RBPMS is LSM14B, which is involved in cytoplasmic regulation of mRNA stability and translation (Brandmann et al., 345 2018) but also shuttles to the nucleus (Kirli et al., 2015). RBPMSA overexpression promoted skipping of Lsm14b exon 6, a pattern which is also seen in tissue SMCs (Fig.   6G,H), and knockdown was also seen to affect this event ( Fig 6G,H), with corresponding changes in LSM14B protein (Fig. 6I). Exon 6 contains the only predicted nuclear localization signal (RPPRRR) in LSM14B (Fig. 6H) lying between the LSM and 350 FDF domains. This suggests that RBPMS mediated AS might prevent nuclear shuttling and function of LSM14B in mRNA turnover.

RBPMS controls the SMC transcription factor Myocardin
Myocardin (MYOCD) is a key transcription factor in SMCs and cardiac muscle (S. Li, 355 Wang, Wang, Richardson, & Olson, 2003). Skipping of Myocd exon 2a in cardiac muscle and proliferative SMCs produces a canonical mRNA encoding full length MYOCD ( Fig. 7A) (Creemers, Sutherland, Oh, Barbosa, & Olson, 2006;van der Veer et al., 2013). Inclusion of exon 2a in differentiated SMCs introduces an in frame stop codon, and the N-terminally truncated Myocd isoform produced using a downstream 360 AUG codon lacks the MEF2 interacting domain and is more potent in activating SMCspecific promoters and SMC differentiation (Creemers et al., 2006;Imamura, Long, Nanda, & Miano, 2010;van der Veer et al., 2013). Significant changes in Myocd exon 2a splicing were not detected by rMATS, but manual inspection of RNA-Seq data and RT-PCR confirmed that Myocd exon 2a is more included in differentiated than 365 proliferative PAC1 SMCs ( Myocd exon 2a is repressed by binding of the RBP QKI to the 5' end of the exon (van der Veer et al., 2013). QKI is expressed more highly in proliferative SMCs (Llorian et al., 2016;van der Veer et al., 2013) suggesting that RBPMS and QKI could act antagonistically on AS events during phenotypic switching. To test for functional 395 antagonism we co-transfected the Myocd minigene with the two RBPs in HEK293 cells ( Fig. 7H). QKI strongly antagonized RBPMS activitation of exon 2a inclusion, even at low concentrations, showing it to be the dominant regulator ( Fig 7H). Thus, splicing of Myocd exon 2a, and thereby the transcriptional activity of Myocd, is under the antagonistic control of RBPs that are preferentially expressed in differentiated 400 (RBPMS) or proliferative (QKI) SMCs.

Using super-enhancers to identify master splicing regulators
By focusing on RBPs whose expression is driven by super-enhancers (Jangi & Sharp,405 2014) we identified RBPMS as a key regulator of the differentiated SMC AS program, with many of the criteria expected of a master regulator: i) it is highly up-regulated in differentiated SMCs (Fig. 1, Fig. S1). Indeed, single cell RNA-Seq identified Rbpms as part of a transcriptome signature of contractile mouse aorta SMCs cells (Dobnikar et al., 2018); ii) changes in RBPMS activity appear to be solely responsible for 20% of 410 the AS changes between differentiated and proliferative PAC1 cells ( RBPMS is reported to be a transcriptional co-activator (J. Fu et al., 2015;Sun et al., 2006). However, we did not observe changes in expression levels of SMC marker genes upon RBPMS knockdown or overexpression and changes in RNA 420 abundance were outnumbered by regulated AS events (Fig. 2). As a splicing regulator RBPMS can only affect actively transcribed genes, so it is unlikely to be sufficient to initially drive SMC differentiation. Notably, we identified RBPMS using superenhancers mapped in adult human tissues (Hnisz et al., 2013). Combined with the ability of RBPMS to promote AS patterns characteristic of fully differentiated tissue 425 SMCs, such as in Cald1, Fermt2 and Tns1 (Fig. 3F,G), this suggests that RBPMS plays a key role in maintaining a mature adult SMC phenotype. Consistent with a role in promoting a fully differentiated state, in other cell types RBPMS has anti-proliferative tumor-suppressive activity (J. Fu et al., 2015;Hou et al., 2018;Rastgoo, Pourabdollah, Abdi, Reece, & Chang, 2018). 430

RBPMS is an alternative splicing regulator
RBPMS and RBPMS2 (referred to as Hermes in Xenopus) are present across vertebrates, while the related proteins Drosophila Couch Potato and C. elegans MEC-8 bind to similar RNA sequences (Soufari & Mackereth, 2017). RBPMS and RBPMS2 can localize to the cytoplasm and nucleus, but apart from transcriptional co-regulation 435 (J. Fu et al., 2015;Sun et al., 2006) most attention has been paid to cytoplasmic roles in mRNA stability (Rambout et al., 2016), transport (Hornberg et al., 2013) and localization in cytoplasmic granules (Farazi et al., 2014;Furukawa, Sakamoto, & Inoue, 2015;Hornberg et al., 2013). RBPMS2 interacts with eukaryote elongation factor-2 (eEF2) in gastrointestinal SMCs, suggesting translational control (Sagnol et 440 al., 2014). MEC-8 was reported to regulate splicing of Unc-52 in C. elegans (Lundquist et al., 1996), but otherwise RBPMS family members have not been reported to regulate splicing. PAR-CLIP with over-expressed RBPMS in HEK293 cells revealed its preferred binding site, but accompanying mRNA-Seq did not identify regulated AS events associated with PAR-CLIP peaks (Farazi et al., 2014). Cell-specific splicing programs are usually driven by more than one regulatory RBPs. A number of splicing regulatory RBPs are known to promote the proliferative SMC phenotype, including QKI, PTBP1, and SRSF1 (Llorian et al., 2016;van der Veer 460 et al., 2013;Xie et al., 2017). The extent to which these proteins coordinately regulate the same target ASEs remains to be established. RBPMS and QKI antagonistically regulate at least two targets in addition to Myocd (Fig. 7). The Flnb H1 exon is activated by RBPMS (Fig. S2, Fig. 4) and repressed by QKI (J. Li et al., 2018), while the penultimate exon of Smtn is repressed by RBPMS ( Fig. S2E) but activated by QKI 465 (Llorian et al., 2016). Moreover, QKI binding motifs are significantly associated with exons regulated during SMC dedifferentiation, and with exons directly regulated by RBPMS overexpression or knockdown (data not shown). It is therefore an interesting possibility that RBPMS and QKI might target a common set of ASEs perhaps acting as antagonistic master regulators of differentiated and proliferative SMC phenotypes. 470 The two RBPs show reciprocal regulation of expression levels during SMC dedifferentiation ( Fig. S1D), which combined with antagonistic activities could lead to switch like changes in many ASEs. For Myocd splicing, QKI appears to have dominant activity, driving skipping of exon 2a even when RBPMS is present at higher levels ( Fig.   7G), so full inclusion of exon 2a requires the presence of RBPMS and absence of QKI. 475 The logic of this regulatory input can explain inclusion of Myocd exon 2a in SMC and skipping in cardiac muscle and is consistent with the observation that RBPMS is superenhancer associated both in vascular SMCs and heart left ventricle, while QKI is super-enhancer associated in left ventricle but not differentiated SMCs. Notably, QKI also controls a significant fraction of ASEs regulated during myogenic differentiation 480 of skeletal muscle cells, but it promotes differentiated myotube AS patterns (Hall et al., 2013), in contrast to its promotion of dedifferentiated splicing patterns in SMCs.

Activity of RBPMS isoforms
RBPMS was originally characterized as a human gene that encoded multiple isoforms of an RBP (Shimamoto et al., 1996), but differential activity of common RBPMS 485 isoforms has not previously been reported. Human RBPMSA and B have similar activity for co-regulation of AP1 transcriptional activity (J. Fu et al., 2015) although a third human-specific isoform had lower activity. We found differential activity of RBPMSA and B upon some ASEs (Figs. 4,7). In general, RBPMSA has higher activity particularly for repressed targets (Fig. 4F, Fig. S4A, B). The differential activity was 490 seen with similar levels of overall expression, although we could not rule out the possibility of variation in nuclear levels as GFP-tagged RBPMS was predominantly cytoplasmic. Nevertheless, some ASEs were differentially responsive to RBPMSA or B while others responded equally (e.g. compare MPRIP with ACTN1, Fig. S4A). The differential activity upon Tpm1 splicing could not be accounted for by differences in 495 RNA binding by RBPMSA or B (Fig. 4F,H). Therefore, it is possible that RBPMS repressive function depends on other interactions mediated by the 20 amino acid RBPMSA C-terminal ( Fig 1C). While the RRM domain is sufficient for RNA binding and dimerization in vitro (Sagnol et al., 2014;Teplova et al., 2016), previous studies have shown the extended C terminal region downstream of the RRM domain to be 500 involved in several aspects of RBPMS/RBPMS2 function including granular localization in retinal ganglion cells (Hornberg et al., 2013) and interaction with cFos in HEK293 cells (J. Fu et al., 2015). In vitro assays suggested that the C-terminal region increases RNA binding affinity and possibly the oligomeric state of RBPMSA (Farazi et al., 2014) and the C-terminal 34 amino acids of Xenopus RBPMS2 is 505 required for binding to Nanos1 RNA in vivo (Aguero et al., 2016). In preliminary studies we have also found the C-terminal region to be essential for splicing regulation (data not shown). Future work will aim to address the mechanisms of RBPMS splicing activation and repression as well as isoform-specific differential activity.

RBPMS targets numerous mRNAs important for SMC function
SMC phenotypic switching involves interconversion between a contractile phenotype and a motile, secretory, proliferative state (Owens et al., 2004). The actin cytoskeleton and its connections to the extracellular matrix (ECM) via focal adhesions are central to the function of both cell states, but with contrasting outcomes: tissue-wide contraction 515 or independent movement of individual cells. RBPMS-mediated AS plays a major role in remodelling the actin cytoskeleton, the integrin adhesome (Horton et al., 2015) and ECM components in the two cell states. This is reflected in the similar GO terms shared by RBPMS regulated AS events and the entire PAC1 cell AS program (Fig. 5, Fig. S5).
The importance of many of the RBPMS target genes to SMC function is further 520 indicated by their proximity to super-enhancers in SMC tissues (Fig. 5B,C). Indeed, three targets -ACTN1, FLNB and TNS1 -are all super-enhancer associated, interact directly with both actin and integrins and are components of distinct axes of the consensus integin adhesome (Horton et al., 2015). ACTN1 is a major hub in the network of proteins affected by RBPMS (Fig. 5C). The AS event in ACTN1 produces a 525 functional Ca 2+ -binding domain in the motile isoform, but lack of Ca 2+ binding in the differentiated isoform stabilizes ACTN1 containing structures in contractile cells (Waites et al., 1992). Many other RBPMS regulated AS events have not previously been characterized. Focal adhesion complexes and the integrin adhesome are mechanosensitive complexes that connect the cytoskeleton and ECM, and are hubs 530 of regulatory tyrosine phosphorylation signalling. We found a small network of RBPMS regulated AS events in the receptor tyrosine phosphatase PTPRF ( Fig 2E) and two interacting proteins PPFIA1 and PPFIBP1 (Fig. S2E). Another focal adhesion associated target is PIEZO1 (Fig. 2E), a mechanosensitive ion channel that is important in SMCs during arterial remodelling in hypertension (Retailleau et al., 2015). 535 An RBPMS repressed exon lies within the conserved Piezo domain immediately adjacent to the mechanosensing "beam" (Liang & Howard, 2018). ECM components affected by RBPMS include fibronectin (FN1), which interacts directly with integrins, and HSPG2 (Fig. S2E). Notably, HSPG2 ( also known as Perlecan) is the basement membrane heparan sulfate proteoglycan that is the identified splicing target of MEC-8 540 in C. elegans (Lundquist et al., 1996).
In addition to direct regulation of numerous functionally related targets, by directly targeting transcriptional and post-transcriptional regulators RBPMS has the potential for more widespread action (Figs. 6,7). RBPMS regulated events in MBNL1 and 2 modulate their splicing regulatory activity (Sznajder et al., 2016;Tabaglio et al., 545 2018;Tran et al., 2011). Changes in secondary AS targets are challenging to observe in a short duration overexpression experiment. Nevertheless, the modest change in NCOR2 splicing appears to be attributable to an RBPMS-induced switch to less active MBNL isoforms (Fig 6F). The regulated event in LSM14B (Fig 6G-I) also has the potential to regulate mRNA stability, or an as yet uncharacterized nuclear role of 550 LSM14B (Kirli et al., 2015).
We also identified the SMC transcription factor as a direct target of RBPMS (Fig. 7). A similar role has been shown for the proposed myogenic AS master regulator RBM24 (Jangi & Sharp, 2014), which stabilizes mRNA of the transcription factor Myogenin by binding to its 3'UTR (Jin, Hidaka, Shirai, & Morisaki, 2010). Similarly, by activating 555 inclusion of Myocd exon 2a (Fig. 7), RBPMS directs production of a Myocardin isoform that more potently promotes the differentiated SMC phenotype (van der Veer et al., 2013). Additional effects upon transcription could also be conferred by RBPMS activation of the FLNB H1 exon (Fig. S2E, Fig. 4). FLNB is primarily an actin binding and adhesion protein, but inclusion of the H1 hinge domain allows nuclear localization 560 and antagonism of the transcription factor FOXC1 in epithelial cells (J. Li et al., 2018). FOXC1 and FOXC2 are expressed at higher levels in adult arteries than any other human tissue (Consortium, 2013). The modulation by RBPMS of FLNB isoforms therefore provides another route for indirect transcriptome regulation. The importance of Filamin RNA processing in SMCs by adenosine to inosine editing of FLNA was also 565 recently highlighted by the cardiovascular phenotypes arising from disruption of this editing (Jain et al., 2018). A number of splicing regulators also influence miRNA processing (Michlewski & Caceres, 2019), so it is an interesting possibility that RBPMS might affect processing of SMC miRNAs such as miR143-145 (Boettger et al., 2009;Cordes et al., 2009). 570 RBPMS2 plays an important role in SMCs of the digestive tract. The RBPMS2 gene is associated with super-enhancers in stomach smooth muscle (Supplementary File 1) and is expressed early in visceral SMC development and at lower levels in mature cells (Notarnicola et al., 2012). Ectopic RBPMS2 overexpression led to loss of differentiated contractile function via translational upregulation of Noggin (Notarnicola 575 et al., 2012;Sagnol et al., 2014). This contrasts with our observations that RBPMS exclusively promotes differentiated SMC AS patterns. RBPMS2 is expressed at low levels in PAC1 and primary aorta SMCs (Fig 1) and its knockdown was without effect.
Nevertheless, ectopic expression of RBPMS2 in PAC1 or HEK293 cells promoted differentiated SMC AS patterns in a similar manner to RBPMSA (Fig. 4, Fig. S4), 580 suggesting that RBPMSA and RBPMS2 have intrinsically similar molecular activities.
The reasons for the apparent discrepancy between the promotion by RBPMS2 of differentiated SMC splicing patterns, but de-differentiated visceral SMC phenotypes, remain to be resolved. Possible explanations include variations in cell-specific signalling pathways, pre-mRNA and mRNA targets, interacting protein partners, post-585 translational modifications and subcellular localization, all of which could differentially modulate RBPMS and RBPMS2 activity in different SMC types.
In conclusion, our data vindicate the proposal that tissue-specific AS master regulators might be identified by the association of their genes with superenhancers (Jangi & Sharp, 2014), paving the way for the identification of further such regulators 590 in other tissues. While our data suggest that RBPMS has a critical role in SMCs, it is likely to play important roles in other cell types where its expression is also superenhancer driven, including cardiac muscle and embryonic stem cells. Our approach aimed to identify AS master regulators common to diverse smooth muscle types (vascular, bladder, stomach). However, SMCs show a great deal of diversity (Fisher, 595 2010), even within single blood vessels (Cheung, Bernardo, Trotter, Pedersen, & Sinha, 2012). The splicing regulator Tra2 b is responsible for some splicing differences between tonic and phasic SMCs (Shukla & Fisher, 2008), and it is possible that other RBPs might act as master regulators of some of these specialized SMC types. Our future studies aim to understand the mechanisms of splicing regulation by RBPMS, 600 the role of the RBPMS regulated splicing program in controlling different aspects of SMC phenotype and the potential role of subversion of this program in cardiovascular diseases.

Identification of potential master AS regulators
Locations of human super-enhancers (genome build Hg19) were taken from the data sets UCSD_Aorta, UCSD_Bladder, BI_Stomach_Smooth_Muscle and BI_Skeletal_Muscle in (Hnisz et al., 2013). Associated genes were obtained using the UCSD Table Browser (

DNA constructs
Coding sequences of rat Rbpms isoforms were PCR amplified from differentiated PAC1 cell cDNA and cloned into XhoI/EcoRI sites of the pEGFP-C1 vector (Clontech) and into EcoRI/XhoI sites of the pCI-neo-3x-FLAG vector (Rideau et al., 2006) to generate N-terminal Venus and 3xFLAG tagged in vivo overexpression constructs.

Cell culture, transfection and inducible lentiviral cell line
Rat PAC1 pulmonary artery SMCs (Rothman et al., 1992) were grown to a more differentiated or proliferative state as described in (Llorian et al., 2016). HEK293T cells were cultured following standard procedures. Rbpms siRNA mediated knockdown in PAC1 cells was performed as in (Llorian et al., 2016). Briefly, 10 5 differentiated PAC1 cells were seeded in a 6 well plate. After 24 hours, cells were transfected using oligofectamine reagent (Invitrogen) and 90 pmols of Stealth siRNAs from Thermo Fisher Scientific (siRNA1: RSS363828, GGCGGCAAAGCCGAGAAGGAGAACA). A second treatment was performed after 24 hours using lipofectamine2000 (Thermo To verify knockdown and transfection efficiency, total cell lysates were obtained by directly adding protein loading buffer to the cells. Lysates were run on a SDS-PAGE, followed by western blot against RBPMS and loading controls. See Supplementary File 6 for information on the antibodies used in this study. To monitor changes in splicing and mRNA abundance, RNA was extracted using TRI reagent (Sigma) according to manufacturer's instructions, DNase treated with Turbo DNA-free kit (Thermo Fisher) and cDNA synthesized, as described below, followed by PCR and QIAxcel or qRT-PCR analysis.

qRT-PCR and RT-PCR
cDNA was prepared using 1 μg total RNA, oligo(dT) or gene-specific oligonucleotides Genes were considered to be differential expressed with p-adj less than 0.05 in the For visualization of differentially spliced exons, sashimi plots were generated using rmats2sashimiplot (Gohr & Irimia, 2018). The sashimi plots show the RNAseq coverage reads mapping to the exon-exon junctions and Psi values from rMATS.

Twenty eight ASE identified by rMATS in the RBPMS knockdown or overexpression
were also validated by RT-PCR in the same manner as described in the qRT-PCR and RT-PCR section. DPsi predicted from the RNAseq analysis and the DPSI observed in the RT-PCR were then tested for a Pearson correlation in RStudio (http://www.rstudio.com/).
For comparison and visualization of the overlap between the genes with different mRNA abundance and the genes differentially spliced in the RBPMS knockdown, RBPMS overexpression and the PAC1 dedifferentiation, proportional Venn diagrams were made using BioVenn (Hulsen, de Vlieg, & Alkema, 2008). Venn diagrams were also generated for the visualization of the common AS events across RBPMS knockdown, overexpression and the PAC1 or aorta tissue dedifferentiation datasets.

Gene ontology and PPI analysis
Enrichment for gene ontology terms in the differentially abundant and spliced genes were obtained from Gorilla (Eden, Navon, Steinfeld, Lipson, & Yakhini, 2009). Two unranked lists of genes, target and background lists, were used for the GO analysis.
The target list contained either the significant differential abundant genes (p-adj < 0.05 and log2 Fold Change greater than 1 or less than -1) or the significant differentially spliced genes (FDR < 0.05 and DPsi threshold of 10%). Background gene lists were created for the PAC1 experiments and aorta tissue by selecting the genes whose expression were higher than 1 TPM in either of the conditions analyzed. For visualization, only the top 5 enriched GO terms of each category (biological process, cell component and molecular function) were shown in Figure 5A and Figure S5. The complete list of enriched terms in differentially spliced and abundant genes can be found in Tables S4 and S5. A Protein-protein interaction network for the genes differentially spliced by RBPMS was constructed using the STRING v10.5 database (Szklarczyk et al., 2017). RBPMS regulated genes were obtained by merging two lists: i) overlap of genes concordantly differentially spliced in the RBPMS knockdown and PAC1 experiments and, ii) overlap of genes concordantly differentially spliced in the RBPMSA overexpression and the aorta tissue datasets that were shown to be regulated in the same range in both conditions. A cut off of DPsi greater than 10% was also applied to the RBPMS regulated gene list, similar to the GO analysis. The human database was chosen for the analysis and the following parameters applied to the PPI network: confidence as the meaning of the network edges, experiments and database as the interaction sources and high confidence (0.700) as the minimum required interaction score.
STRING functional enrichments, using the whole genome as statistical background, were also included for visualization. Human super-enhancer associated genes from Supplementary File 1 were highlighted.

Recombinant protein
Rat RBPMS A and B with a 3xFLAG N terminal tag were cloned into the BamHI/XhoI sites of the expression vector pET21d, for expression of recombinant RBPMS containing a T7 N-terminal tag and a His6 C-terminal tag in E. coli. Recombinant RBPMS A protein was purified using Blue Sepharose 6 and HisTrap HP columns whereas RBPMS B was purified only through the latter, since low binding was observed to Blue Sepharose 6. The identity of purified recombinant proteins was confirmed by western blot (Fig. S4G) and mass mapping by mass spectrometry.

In vitro transcription and binding
a 32 P-UTP labelled RNA probes were in vitro transcribed using SP6 RNA polymerase.
After incubation, samples were run on a 4% polyacrylamide gel. For UV-crosslinking experiments, the same binding incubation was performed followed by UV-crosslink on ice in a Stratalinker with 1920 mJ. Binding reactions were then incubated with RNase A1 and T1 at 0.28 mg/ml and 0.8 U/ml respectively, for 10 minutes at 37°C. Prior to loading the samples into a 20% denaturing polyacrylamide gel, SDS buffer was added to the samples which were then heated for 5 minutes at 90°C.

Statistical analysis
Analysis

Competing interests
The authors declare that they have no conflicts of interest.  (E) qRT-PCR analysis of Rbpms (all isoforms), Rbpms2 and SMC differentiation markers Acta2, Cnn1 and Smtn, in PAC1 cells D (green) and P (blue). Expression was normalized to the average of two housekeepers (Gapdh and Rpl32) and the mean of the relative expression is shown (n=3). Statistical significance was performed using Student's t-test (* P < 0.05, ** P < 0.01,*** P < 0.001).

References
(F) Western blots for RBPMS in D and P PAC1 cells. ACTA2 is a SMC differentiation marker and GAPDH a loading control. A and B indicates the two RBPMS isoforms.
(G) Immunofluorescence in D and P PAC1 cells for RBPMS. DAPI staining for nuclei.   In panels C-E and G-I, values are the mean Psi ± SD (n = 3). For Flnb cassette exon, the Psi is the sum of both short and long isoforms generated by a A5SS event. For Tpm1 MXE, the Sm exon Psi is shown (PsiSm). Schematics of the splicing isoforms indicate the PCR products, differentiated (green) and proliferative (blue). Statistical significance was calculated using Student's t-test (* P < 0.05, ** P < 0.01,*** P < 0.001).
Western blot anti-GFP and GAPDH, loading control, were carried out to assess RBPMS isoforms overexpression in HEK293. The western blot in (C) is also a representative of (G), since both RT-PCRs (FLNB and TPM1) are from the same overexpression experiment. Significance determined by hypergeometic P-value.
(C) PPI network of genes differentially spliced upon RBPMS knockdown and overexpression that overlap with the PAC1 and aorta tissue dedifferentiation datasets.
PPI network was generated in STRING using experiments and database as the sources of interactions. Network edges represent the interaction confidence. Enriched GO terms (BP, biological process, MF, molecular function and CC, cellular component) were also included in the analysis and are indicated in red, blue and yellow. Superenhancer associated gene names are in bold and are highlighted gray, light green or dark green shading according to whether they were super-enhancer associated in 1 , 2 or 3 SMC tissues. In all the western blots TUBULIN was used as a loading control.
(C) mRNA abundance of RBP genes in differentiated aorta tissue (T) and proliferative passage 9 (P9). RBPs whose genes were found associated with smooth muscle tissue super-enhancers (Fig. 1B)  Values indicate the motif enrichment. Statistical significance was calculated by the Matt tool (* P < 0.05, ** P < 0.01,*** P < 0.001).              L iv e r C e ll s -T r a n s fo r m e d fi b r o b la s ts P it u it a r y S k in -N o t S u n E x p o s e d ( S u p r a p u b ic ) P a n c r e a s M u s c le -S k e le ta l B r a in -S u b s ta n ti a n ig r a B r a in -C e r e b e ll u m B r a in -P u ta m e n ( b a s a l g a n g li a ) B r a in -S p in a l c o r d ( c e r v ic a l c -1 ) B r a in -C a u d a te ( b a s a l g a n g li a )