NG2/CSPG4, CD146/MCAM and VAP1/AOC3 are regulated by myocardin-related transcription factors in smooth muscle cells

The present work addressed the hypothesis that NG2/CSPG4, CD146/MCAM, and VAP1/AOC3 are target genes of myocardin-related transcription factors (MRTFs: myocardin/MYOCD, MRTF-A/MKL1, MRTF-B/MKL2) and serum response factor (SRF). Using a bioinformatics approach, we found that CSPG4, MCAM, and AOC3 correlate with MYOCD, MRTF-A/MKL1, and SRF across human tissues. No other transcription factor correlated as strongly with these transcripts as SRF. Overexpression of MRTFs increased both mRNA and protein levels of CSPG4, MCAM, and AOC3 in cultured human smooth muscle cells (SMCs). Imaging confirmed increased staining for CSPG4, MCAM, and AOC3 in MRTF-A/MKL1-transduced cells. MRTFs exert their effects through SRF, and the MCAM and AOC3 gene loci contained binding sites for SRF. SRF silencing reduced the transcript levels of these genes, and time-courses of induction paralleled the direct target ACTA2. MRTF-A/MKL1 increased the activity of promoter reporters for MCAM and AOC3, and transcriptional activation further depended on the chromatin remodeling enzyme KDM3A. CSPG4, MCAM, and AOC3 responded to the MRTF-SRF inhibitor CCG-1423, to actin dynamics, and to ternary complex factors. Coincidental detection of these proteins should reflect MRTF-SRF activity, and beyond SMCs, we observed co-expression of CD146/MCAM, NG2/CSPG4, and VAP1/AOC3 in pericytes and endothelial cells in the human brain. This work identifies highly responsive vascular target genes of MRTF-SRF signaling that are regulated via a mechanism involving KDM3A.

To test this, we transduced human coronary artery SMCs (hCASMCs) with control virus (Ad-CMV-Null) or with viruses encoding MRTFs (MYOCD, MRTF-A/MKL1, or MRTF-B/MKL2). We then assayed the CSPG4 mRNA level by reverse transcription quantitative polymerase chain reaction (RT-qPCR). All MRTFs increased the CSPG4 transcript (8-12-fold, Fig. 1E). In western blots with a specific NG2/CSPG4 antibody, we observed two high molecular weight bands (> 250 kDa) that increased following viral transduction of MRTFs ( Fig. 1F and summarized data in G). Imaging of cells by confocal fluorescence microscopy ( Fig. 1H) revealed that the NG2/CSPG4 signal (red) increased by about eightfold after MRTF-A/MKL1 transduction. This increase is comparable to that seen with the positive control synemin/SYNM, an intermediate filament protein targeted by all MRTFs 36 (green staining in Fig. 1H, summarized data in Fig. 1I). Taken together, these findings argue that the cell lineage marker NG2/CSPG4 is regulated by MRTFs in human SMCs. Membranes were cut horizontally in this, and the following, figures to allow for detection of multiple proteins on the same membrane. Full length blots (as full as possible) are found in Fig. S4. Panel (G) shows summarized data for the western blot experiments (N = 3). The bands migrating at ≥ 250 kDa were included in the analysis. In panel (H), cells were transduced with tagged MRTF-A/MKL1 (blue), fixed at 96 h, and stained for NG2/CSPG4 (red) and the intermediate filament synemin/SYNM (green), followed by confocal imaging. Red and green labels are shown in black and white below the colored panels for clarity. Summarized data from two independent experiments with three independent replicates each time (N = 6) is shown in panel (I). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, all versus Null.  Fig. 2A-D). Therefore, we also addressed if this gene is regulated by MRTFs. The MCAM mRNA (Fig. 2E) and protein levels (Fig. 2F,G) were increased by all MRTFs, with MRTF-A/MKL1 causing the greatest increase (≈250-fold at the mRNA level). Imaging demonstrated that viral transduction of MRTF-A/MKL1 also increased MCAM fluorescence (Fig. 2H, green, summarized data in 2I), which was higher than the positive control used in these experiments (smooth muscle α-actin/ACTA2, Fig. 2H and data not shown). Therefore, CD146/MCAM is also regulated by MRTFs. The last transcript in the CSPG4 co-expression module (in Fig. 1A) that we examined was AOC3 (a.k.a. VAP1). The AOC3 mRNA correlated with MYOCD (Fig. 3A,B) as well as with CSPG4 (Fig. 3C) and with MCAM (Fig. 3D). Overexpression of MRTFs increased the mRNA level of AOC3 (30-80-fold, Fig. 3E). Using a noncommercial antibody (TK10-79, kind gift from M. Salmi), and cells rapidly fixed in ice cold acetone, we could also demonstrate a modest increase of AOC3 fluorescence following MRTF-A/MKL1 transduction (Fig. 3F, and summarized data in G). We conclude that NG2/CSPG4, CD146/MCAM, and VAP1/AOC3 are regulated by MRTFs in human smooth muscle cells.
So far, our results suggest, but do not prove, that MRTF-A/MKL1 is a more effective coactivator of CSPG4, MCAM, and AOC3 than MYOCD and MRTF-B/MKL2. However, MRTF levels may differ after overexpression, and are difficult to compare due to lack of reliable myocardin antibodies 37 . Comparison of the cycle threshold (Ct) values after overexpression suggests that MRTF-B reaches a higher level after overexpression than MRTF-A, which in turn reaches a higher level after overexpression than MYOCD (not shown). MRTF-A is therefore likely a more efficacious transactivator of these genes than MRTF-B. However, the apparently smaller effect of MYOCD could be due to fewer copies of MYOCD mRNA and protein (compared to MRTF-A/MKL1). We have previously seen that MRTFs may antagonize each other when transactivation efficacy differs 38  Mechanistic studies. Genes targeted by MRTFs have so called CArG-boxes (CC-A/Tx6-GG) to which serum response factor (SRF) binds. We inspected chromatin immunoprecipitation-sequencing (ChIP-Seq) data for SRF (ENCODE) at the three gene loci using the UCSC Genome Browser (screenshots from these analyses are shown in Fig. S1A-C). No SRF binding was noted at the CSPG4 and AOC3 loci ( Fig. 4A shows diagrammatical representations of the gene loci), but several binding sites were present at the MCAM locus (dark red arrows in Fig. 4A middle, called S1-S4). The AOC3 locus harbored a computationally identified CArG 39 (Fig. 4A, lighter red arrow). For MCAM, two computationally identified CArGs mapped to two of the experimentally determined SRF binding regions (S2 and S4). The presence of CArGs coinciding with binding of SRF in the promoters suggested involvement of SRF. Thus, we examined the role of SRF using short hairpin silencing (shSRF). Silencing of SRF reduced the mRNA levels of MCAM, AOC3, and CSPG4 in MRTF-A/MKL1 transduced cells (Fig. 4B). A modest effect was seen on CSPG4, probably because knockdown of SRF was incomplete (− 43 ± 4%, P < 0.0001). Limited SRF knockdown is to be expected because it is important for the cell cycle, and SRF knockdown in culture should select for cells where it was less effectively reduced. SRF silencing was not done in the absence of MRTF-A/MKL1 transduction because MCAM and AOC3 levels were low under basal culture conditions. Overall, these findings support involvement of SRF in MRTF-driven expression of these genes.
We next aimed to verify a regulatory role of the CArGs identified. Because there is no documented SRF binding at the CSPG4 locus, we focused on MCAM and AOC3. Two DNA regions, each containing three out of four of the SRF binding sites in MCAM (S1-S3 and S2-S4), were used to create promoter reporters. For AOC3, we used a commercial reporter plasmid covering ≈1 kb of the proximal promoter and including the computationally identified CArG. We then examined if the respective constructs responded to MRTF-A/MKL1. As shown in Fig. 4C, both MCAM reporters (S1-S3 and S2-S4) were activated by MRTF-A/MKL1, and the AOC3 reporter was similarly activated (almost tenfold, Fig. 4C). However, reporter activation appeared smaller than effects seen at the mRNA level (compare the 80-fold increase of AOC3 mRNA with the tenfold increase of promoter activity). This difference suggested either that (1) DNA regions beyond those included in the reporters are involved in regulation, or (2) epigenetic effects requiring intact chromatin are involved, or (3) indirect effects play a role. To address the last possibility, we performed time-course studies for comparison with the direct target gene ACTA2 40 . We reasoned that indirect effects should be reflected in a delayed transcript elevation relative to the direct target ACTA2, whereas a direct effect should be at least as fast. Interestingly, CSPG4, MCAM (Fig. 4D) and AOC3 (Fig. 4E) were increased at least as fast as ACTA2 following overexpression of myocardin.
We also considered histone acetylation (see H3K27Ac track in Fig. S1A-C), but this was not helpful in terms of identifying additional regulatory mechanisms, and treatment of cells with the deacetylase inhibitor trichostatin A only modestly increased MRTF-A-driven MCAM levels, leaving the other transcripts unaffected (Fig. S2).
Histone methylation represents another epigenetic mechanism, and SRF is known to interact with the lysine demethylase KDM1A 41 , while MRTFs (MRTF-A > MYOCD > MRTF-B) interact with KDM3A to activate transcription 42 . We focused on KDM3A because KDM3A was reported in previous work to bind a genomic region flanking the MCAM transcription start site (between S3 and S4), and to regulate MCAM expression 43 . This raised the possibility that MRTFs exert their effects on MCAM in part via KDM3A. In keeping with this possibility, we found that MRTF-A/MKL1 increased the mRNA level of KDM3A (Fig. 4F), and that silencing of KDM3A using a short hairpin construct mitigated MRTF-driven induction of MCAM and AOC3 (Fig. 4G). A similar dependence on KDM3A was seen for classical SMC markers (not shown). MRTF-A/MKL1 also induced KDM3A at the protein level (Fig. 4H 4H). Taken together, these findings suggest that regulation of MCAM and AOC3 by MRTF-A/MKL1 involves, in part, a local and KDM3A-dependent epigenetic mechanism. Previous work implicated the transcription factors SOX10, ASCL1, and OLIG2 in regulation of CSPG4 transcription in the brain 44 . We therefore tested if these transcription factors were increased by MRTF-A/MKL1. Overexpression of MRTF-A/MKL1 had no effect on SOX10, ASCL1, and OLIG2, whereas the positive controls (ACTA2 and CSPG4) increased as expected (Fig. 4I). This argues against an indirect effect mediated by any of these transcription factors.
CSPG4, MCAM and AOC3 levels change during smooth muscle differentiation, after depolymerization of actin, and following knockout of ternary complex factors. We next examined if incubation of hCASMCs in medium with differentiation supplement (DS), increased mRNA levels of CSPG4, MCAM and AOC3 compared to the growth supplement (GS) used in the previous experiments. Clear increases were seen for AOC3 and MCAM (Fig. 5A). CSPG4 also tended to increase, but this difference did not reach statistical significance (Fig. 5A, P = 0.06).
An important property of MRTFs is that they are controlled by actin dynamics 4,45 . Therefore, we next tested how the CSPG4, MCAM and AOC3 transcripts respond to Latrunculin B (LatB), which depolymerizes actin 46 . In initial experiments using either GS or DS media, we did not see an effect of LatB (not shown). We reasoned that this could be due either to the low basal levels of these transcripts in cell culture, or to a preferential drive from MYOCD, which is unresponsive to actin. Consequently, we performed experiments with low-level overexpression of MRTF-A/MKL1, which should increase expression levels sufficiently to see LatB-driven reductions. Under these conditions, LatB reduced the mRNA levels of four positive controls (ACTA2, TAGLN, SYNPO2, and CAV1, It remained possible that MCAM, AOC3, and CSPG4 represent comparatively stable transcripts. Therefore, we next tried a cycling protocol over 4 days, in which cells recovered from LatB for 4 h every 24 h to maintain cell viability. In this protocol, CSPG4, MCAM, and AOC3 were all reduced by LatB (Fig. 5C, 96 h).
Our inhibitor experiments so far depended on prior overexpression of MRTF-A/MKL1 in vitro. To circumvent this potential caveat, we compared freshly dissected mouse caudal arteries with caudal arteries maintained in organ culture for 96 h in the presence of vehicle or CCG-1423. In organ culture, which should approximate in situ conditions and that does not rely on overexpression of MRTFs, CCG-1423 reduced Mcam and Aoc3 beyond the reduction caused by organ culture alone (Fig. 5F). This was not seen for Cspg4, which however declined more sharply with organ culture as such (Fig. 5F). MRTFs are thus likely essential drivers of Mcam and Aoc3 expression in the vascular wall of mice.
MRTFs are antagonized by ternary complex factors (Elk1, Elk3, and Elk4) due to competition for binding to SRF. To examine if this competition applies for Cspg4, Mcam and Aoc3, we next compared wild type (WT) mouse embryonic fibroblasts (MEFs) with MEFs lacking all three Elks (KO). Cspg4 and Mcam were higher in KO MEFs, as predicted, but Aoc3 was reduced (Fig. 5G). Similarly, overexpression of the atypical ternary complex factor FLI1 in human SMCs reduced MCAM and CSPG4 but increased AOC3 (Fig. S3). AOC3 thus deviates from MCAM and CSPG4 with respect to its ternary complex factor regulation.
MRTFs and cerebrovascular NG2/CSPG4 expression. Recent work examined the transcriptional control of NG2/Cspg4 in the central nervous system, and it was found that the transcription factors Sox10, Olig2 and Ascl1 activate Cspg4 expression in glial cells through an enhancer in the first intron (called int1-3b) 44 . This enhancer controlled Cspg4 expression in Sox10 + cells in vivo but failed to drive expression in pericytes and ECs 44 . Thus, the transcriptional drive on Cspg4 in pericytes/ECs differs from that in other brain cells. We compared MYOCD-CSPG4 correlations with SOX10-CSPG4, OLIG2-CSPG4 and ASCL1-CSPG4 correlations in the human nucleus accumbens (GTEx data) and found that SOX10 (P < 0.0001, R = 0.59), OLIG2 (P < 0.0001, R = 0.62), ASCL1 (P < 0.0001, R = 0.63), and MYOCD (P < 0.0001, R = 0.35) all correlated with CSPG4 in the nucleus accumbens. Similar results were obtained in all other brain regions examined (data not shown). Importantly, SOX10, OLIG2, and ASCL1 were not represented among transcripts in the extreme of the CSPG4 correlations in 20 peripheral tissues (see Fig. 1A). These observations, together with previous work 44 , suggest that MRTFs control CSPG4 expression in pericytes/ECs in the brain, and that SOX10, OLIG2, and ASCL1, while relevant in non-vascular brain cells, play little role in the cerebral vasculature and peripherally. The latter notion was also supported by our finding that SOX10, OLIG2 and ASCL1 were unresponsive to overexpression of MRTF-A/MKL1 (c.f. Fig. 4I).
MRTF-driven cerebrovascular transcription in man would be strongly supported if two or more MRTFregulated proteins are co-expressed in ECs and pericytes. To examine co-expression in cerebral vessels, we stained human brain sections (hippocampus) for NG2/CSPG4 and CD146/MCAM. Both antibodies labelled vascular structures, including ECs and pericytes (Fig. 6A-C). The MCAM antibody did not stain any other cells (Fig. 6D). Due to difficulties with AOC3 detection, we inspected the Human Protein Atlas 49 for this protein. AOC3 staining of vascular structures was seen in all brain regions examined (Fig. 6E shows four examples). ECs and pericytes www.nature.com/scientificreports/ (white arrows) were AOC3 positive (Fig. 6E), and SMCs in larger arterioles were also positive (Fig. 6E, right micrograph, bottom). We conclude that co-expression of NG2/CSPG4, CD146/MCAM and VAP1/AOC3 in ECs and pericytes support MRTF-driven transcription at the human blood-brain barrier with high probability (estimated P = 0.0005) and distinguishes pericytes and ECs from other NG2 + cells in the human brain.
Our results support a model in which MRTFs exert a transcriptional drive at the human blood brain barrier, regulating the expression of MCAM, CSPG4 and AOC3 (Fig. 6F). The utility of any model is best determined by its ability to make predictions that can be confirmed. Two predictions were explored. First, we tested whether other commonly used markers of pericytes in the brain are targeted by MRTFs. Indeed, RGS5 (regulator of G protein signaling 5) and PDGFRB (platelet derived growth factor receptor β), which are widely used markers of pericytes, were increased 8 days after MYOCD transduction as shown using RT-qPCR (Fig. 6G). Surprisingly, and unlike the gene trio examined above, MYOCD and MRTF-A/MKL1 regulated RGS5 in opposite directions (Fig. 6H). The second prediction that we made was that novel markers of ECs and pericytes in the brain can be identified by focusing on MRTF target genes. To gauge this possibility, we used a published RNA-seq dataset for MRTF-A 50 , and examined staining for highly regulated and poorly characterized targets in the Human Protein Atlas. This strategy identified TINAGL1 (tubulointerstitial nephritis antigen like 1), for which staining of cerebral ECs and pericytes was particularly clear (Fig. 6I, in brown). We conclude that two predictions based on our model (Fig. 6F) can be confirmed.

Discussion
The current work examined regulation of three transcripts that, from a computational point of view, are highly co-expressed across human tissues, forming a closely knit cluster. Correlation analyses implicate MRTF-SRF signaling in regulation of these genes, and we accordingly tested the hypothesis that the gene trio consisting of NG2/CSPG4, CD146/MCAM and VAP1/AOC3 is regulated by MRTF-SRF signaling in vitro and in situ. We find that (1) these mRNAs are increased following overexpression of MRTFs, (2) the transcripts are reduced by SRF depletion, and (3) they are affected by ternary complex factors and increased during SMC differentiation. Depolymerization of actin, a well-established control mechanism of MRTFs 4,45 , only slowly reduced CSPG4, MCAM and AOC3, but this was accounted for by slow mRNA degradation compared to other MRTF target genes. Importantly, the MRTF-SRF inhibitor CCG-1423 reduced Mcam and Aoc3 in mouse arteries without prior overexpression of MRTFs, arguing in favor of an essential transcriptional drive from MRTFs on these genes in the intact vascular wall. The human gene loci for MCAM and AOC3 bind SRF or have putative SRF binding sites, and reporter plasmids containing these regions conferred responsiveness to MRTF-A/MKL1. MCAM and AOC3 transcripts responded more strongly to overexpression of MRTFs than the classical target gene ACTA2. The large effect-size of MRTFs, along with high quality antibodies (for MCAM in particular), may explain why these gene products have emerged as cell lineage markers. CSPG4, MCAM, and AOC3 are expressed on the cell surface, allowing for FACS-based isolation approaches. In recent work, a mouse model for lineage tracing of CD146 + /NG2 + (MCAM + /CSPG4 +) cells was developed 51 . Based on our current findings, this model may tentatively be used to track cells in which MRTFs are particularly active.
A weakness of the present work is that we did not inactivate CArG-boxes responsible for MCAM and AOC3 regulation. However, it seems plausible that one (AOC3) or several (MCAM) CArGs included in the reporter constructs contribute to regulation. Lack of reporter data with CSPG4 may be taken to indicate an indirect effect. However, if one entertains the idea of an indirect effect, one will have to reconcile this with our finding that SRF is the most highly ranking transcription factor in the human CSPG4/MCAM/AOC3 co-expression module, computationally defined using > 1 M correlation analyses. Moreover, an explanation must be provided for the lack of delay in induction compared to the direct target gene ACTA2, when a delay by at least 24 h would be expected for an indirect effect. The large effect sizes at the mRNA and protein levels, which clearly match, or even outperform, that of known direct target genes, also need to be accommodated. Therefore, we favor the view that all the Figure 4. The MCAM and AOC3 gene loci have SRF binding motifs that confer responsiveness to MRTF-SRF signaling. Panel (A) shows graphical representations of the CSPG4, MCAM and AOC3 gene loci with known (red arrows, ENCODE data accessible via the UCSC genome browser) and predicted (lighter red/pink arrows) binding sites for MRTF-SRF 39 . Gene loci and binding sites are not drawn to scale. Panel (B) shows effects of SRF silencing (by 43 ± 4%, P < 0.0001, N = 6), using a short hairpin construct, on the indicated mRNA levels measured by RT-qPCR. Panel (C) shows reporter assays for MCAM (S1-S3 plasmid and S2-S4 plasmid, see panel A) and AOC3. We used HEK 293 cells for transfection of the reporter plasmids in (C), because these cells were more readily transfected compared to human coronary artery smooth muscle cells used elsewhere. Panels  www.nature.com/scientificreports/ genes studied are direct targets of MRTF-SRF-signaling. The smaller effects of MRTF-A on reporter activities compared to the effects at the mRNA levels do however allow for additional activation mechanisms involving for example DNA regions beyond those studied, and/or chromatin remodeling. We found that KDM3A, known to bind MRTFs 42 and to demethylate the repressive histone mark H3K9Me2 52 , is induced by MRTF-A suggesting chromatin remodeling. Moreover, KDM3A silencing reduced MCAM and AOC3, consistent with prior work showing involvement of KDM3A in MCAM regulation 43 . We therefore propose that the MRTF-SRF complex acts in part by binding to CArGs in the promoters of MCAM and AOC3, and in part by an indirect mechanism that involves increased KDM3A expression. CSPG4 stands out by responding less than MCAM and AOC3 to SRF depletion, depolymerization of actin, and KDM3A silencing. Further studies are therefore warranted to unravel the exact mechanisms underlying regulation of CSPG4 by MRTFs. The strong correlations between MRTF-SRF and CD146/MCAM, VAP1/AOC3, and NG2/CSPG4, suggest that MRTFs constitute an important, if not major, drive on these genes in many human organs. Our findings moreover support a critical MRTF-SRF drive on these genes in the mouse caudal artery in situ. NG2/CSPG4 expression in the healthy brain is an exception. In the brain, SoxE and bHLH factors appear to play a dominating role for CSPG4 expression in NG2 glia 44 . Importantly, our experiments argue against the possibility that SoxE and bHLH factors are indirectly involved in the effects of MRTFs in vascular cells, the reason being that the SoxE and bHLH factors in question were unresponsive to MRTF overexpression. The highly dynamic expression of the lineage markers studied here caution against their use for uncritical enumeration of cerebrovascular cells in e.g. neurodegenerative disease 53 . Indeed, if overall NG2/CSPG4 levels change with treatment, this could be due to a change in MRTF activity rather than a change in cell numbers. We have relied on NG2 for pericyte counting 54 , but only in combination with simultaneous assessment of cell shape and spatial cell distribution. Given the complex transcriptional influences on these genes, special caution is obviously needed.
Two of the genes studied here (AOC3 and MCAM) behaved as bona fide SMC differentiation markers. These markers were increased in an in vitro differentiation paradigm and reduced in organ cultured mouse arteries. A well-established model for SMC differentiation postulates that competition between ternary complex factors (TCFs), such as ELK1, and MRTFs, for binding to SRF is a key event in SMC differentiation 55 . That is, when growth is stimulated and mitogen-activated protein kinases are active, ELK1 binds to the same site on SRF as MRTFs but activates a distinct set of target genes important for growth 55,56 . According to this antagonistic model, the behavior of classical SMC differentiation markers is dictated as much by their MRTF-activation as by their TCF-repression. Our experiments on mouse cells lacking three TCFs showed higher levels of Cspg4 and Mcam, whereas Aoc3 was reduced. Particularly forceful repression by TCFs therefore does not explain why AOC3 increases more than MCAM and CSPG4 in our in vitro SMC differentiation paradigm.
While previous work has demonstrated an important role of MRTFs at the blood-brain barrier (BBB) in mice, our current findings are the first to demonstrate a role of MRTFs for lineage marker expression at the human BBB. Our findings therefore overcome an important obstacle in clinical translation. Interestingly, in mice, both MRTF-SRF signaling and MCAM control BBB integrity 9,57 . MCAM does this via effects in both ECs and pericytes. In ECs, loss of MCAM reduces Claudin5 expression and disrupts the structure of tight junctions, whereas MCAM loss from pericytes impairs PDGF receptor signaling and pericyte recruitment 57 . Like the loss of MCAM in ECs, inducible deletion of SRF in ECs reduces Claudin5 expression and tight junction integrity 9 . In further support of a function of MRTFs at the human BBB, we show that two widely used markers of brain pericytes, namely RGS5 and PDGFRB, are regulated by MYOCD.
To summarize, the present work establishes NG2/CSPG4, CD146/MCAM and VAP1/AOC3 as MRTF-SRF regulated genes. Furthermore, we demonstrate that MCAM and AOC3 activation by MRTF-A depends on promoter regions containing CArG motifs, and on the epigenetic modifier KDM3A. Finally, these genes respond to changes in actin dynamics and ternary complex factors, and to the MRTF-SRF inhibitor CCG-1423.

Materials and methods
Correlation analyses using GTEx data. The Genotype-Tissue Expression (GTEx) project is a database open to the scientific community to study relationships between genetic variation and gene expression in human tissues 28 . Summary statistics on the number of tissues for which RNA-Seq data is accessible, and the age, sex and race of donors can be found at https ://gtexp ortal .org/home/tissu eSumm aryPa ge. RNA-Seq data was downloaded in Oct 2018 and TMM-normalized as described 47,58 . Pearson analyses, where CSPG4 was correlated versus all other transcripts across 20 tissues, were made in Excel. Individual correlations of interest were subsequently tested using the Spearman method in GraphPad Prism. The tissues included were tibial artery, aorta, coronary artery, sigmoid colon, the muscular layer of the esophagus and the gastroesophageal junction, tibial nerve, uterus, stomach subcutaneous adipose tissue, terminal ileum, lung, visceral adipose tissue, transverse colon, prostate, breast, skeletal muscle, amygdala, hypothalamus, and ovary. The number of individuals in each group ranged from 100 (amygdala) to 564 (skeletal muscle).
For the analyses described above, we found that CSPG4xMYOCD correlations in the brain were weaker than elsewhere in the human body. We therefore compared MYOCD with three transcription factors previously shown to regulate CSPG4 (SOX10xCSPG4, OLIG2xCSPG4, and ASCL1xCSPG4) 44 using six different brain regions. The regions included were cerebral cortex, nucleus accumbens, nucleus caudatus, putamen, hippocampus, and amygdala. MYOCDxCSPG4 correlations were highly significant across the brain regions examined in this analysis (P < 0.0001 throughout).
ChIP-Seq data for SRF (c.f. arrows in Fig. 4A) 61 . Written informed consent for research use of brain tissue and clinical data was obtained from patients or next of kin in accordance with the International Declaration of Helsinki 62 . The medical ethical evaluation committee of VU Medical Centre, Amsterdam, approved methods for brain tissue collection. Approval of the study was obtained from the regional ethical review panel in Lund.
We examined immunohistochemical staining for AOC3 using the human protein atlas 49 . Two antibodies were represented (HPA000980 and CAB025797) showing similar overall staining patterns. Staining for CAB025797 was stronger and crisper and was therefore used for illustrating AOC3 positive pericytes and endothelial cells in different brain regions.
Organ cultured mouse caudal arteries and ethics statement. 8-12-week-old wild type C57Bl/6 mice of both sexes from ongoing breeding efforts were euthanized with carbon dioxide. The abdominal side of the tail was marked, and the tail was cut and transported to the lab in ice cold HEPES-buffered Krebs solution (135.5 mM NaCl, 5.9 mM KCl, 1.2 mM MgCl 2 , 2.5 mM CaCl 2 , 11.6 mM glucose, 11.6 mM HEPES, pH 7.4). The skin and the underlying fascia on the abdominal side of the tail were cut under a dissection microscope. As much as possible of the tail artery was then removed and cleaned. Each artery from 6 animals was cut in three pieces. One piece was immediately frozen while the remaining two pieces were blindly assigned to organ culture with vehicle (DMSO) or the MRTF-SRF inhibitor CCG-1423 (10 µM). DMSO or CCG-1423 was added to the culture medium before immersion of the arterial segments. The sample size of n = 6 in each of the three groups was based on empirical observations, no values were excluded, and each CCG-1423 treated segment had a vehicle control from the same animal. Following organ culture, tail arteries were frozen, RNA was isolated, and mRNA levels of Mcam, Aoc3, and Cspg4 were determined by RT-qPCR. All animal experiments were approved by the Ethical Review Board at Lund University (approval number M4-26) and adhere to guidelines (ARRIVE 2.0) and regulations for animal experimentation.
Statistics. Statistical tests were made using Log2-transformed expression data. For pairwise comparisons we used student t-test. Multiple comparisons were made using one-way ANOVAs followed by the Bonferroni or Dunnett post-hoc tests. For individual correlation analyses we used the Spearman method. All statistical calculations were made in GraphPad Prism.