RhoA, Rac1, and Cdc42 differentially regulate αSMA and collagen I expression in mesenchymal stem cells

Mesenchymal stem cells (MSC) are suggested to be important progenitors of myofibroblasts in fibrosis. To understand the role of Rho GTPase signaling in TGFβ-induced myofibroblast differentiation of MSC, we generated a novel MSC line and its descendants lacking functional Rho GTPases and Rho GTPase signaling components. Unexpectedly, our data revealed that Rho GTPase signaling is required for TGFβ-induced expression of α-smooth muscle actin (αSMA) but not of collagen I α1 (col1a1). Whereas loss of RhoA and Cdc42 reduced αSMA expression, ablation of the Rac1 gene had the opposite effect. Although actin polymerization and MRTFa were crucial for TGFβ-induced αSMA expression, neither Arp2/3-dependent actin polymerization nor cofilin-dependent severing and depolymerization of F-actin were required. Instead, F-actin levels were dependent on cell contraction, and TGFβ-induced actin polymerization correlated with increased cell contraction mediated by RhoA and Cdc42. Finally, we observed impaired collagen I secretion in MSC lacking RhoA or Cdc42. These data give novel molecular insights into the role of Rho GTPases in TGFβ signaling and have implications for our understanding of MSC function in fibrosis.

Fibrosis, the excessive production of collagen I and other extracellular matrix (ECM) 2 proteins, is an important clinical problem with few treatment options (1,2). In the liver, fibrosis caused by viral hepatitis, alcohol, or obesity can lead to liver failure and increase the likelihood for liver cancer (3). However, fibrosis occurs also in many other tissues such as kidney, heart, and lungs and leads to organ failure due to functional impairment. The major cell type responsible for pathological ECM production is the myofibroblast, which can be described as a contractile, collagen I-producing, fibroblastoid cell, characterized mostly by high expression of ␣-smooth muscle actin (␣SMA; gene name, ACTA2) (4). The biology of myofibroblasts is complex, because apparently many different precursor populations are able to differentiate into myofibroblasts, including tissue-resident fibroblasts and epithelial cells, which may result in different subtypes of myofibroblasts with different biological properties. The extent to which each progenitor population contributes to myofibroblasts in disease is debated and might be dependent on the tissue as well as on the stimulus. Recently, vascular mesenchymal stem cell-like cells (MSC) were described as critically important myofibroblast precursors in different murine fibrosis models (5). This originally small population, present in all vascularized tissues, expands strongly during disease and gives rise to a large share of the differentiated myofibroblasts. Moreover, the elimination of this population strongly reduces the development of kidney and heart fibrosis. It is important therefore to understand the mechanisms that trigger the differentiation of MSC into myofibroblasts. In vitro, TGF␤ can induce myofibroblast differentiation of MSC derived from adipose or prostate tissue, as characterized by ␣SMA and collagen I expression (6,7). TGF␤ signaling is complex and can be divided into a canonical part that is dependent on the Smad2/3/4 transcription factors and a less characterized noncanonical part, which includes activation of the small Rho GTPase RhoA (8). The latter has been shown to regulate actin polymerization via ROCK (9), triggering nuclear translocation of the transcription co-factor MRTFa, which then binds to the transcription factor SRF triggering the expression of ␣SMA and collagen I in fibroblasts (10 -12). How TGF␤ signaling activates RhoA, how ROCK is controlling F-actin formation, and how SRF-MRTF synergizes with Smad2/3/4 canonical signaling is not well-understood. It is also not clear to what extent other major regulators of actin polymerization such as the small GTPases Rac1 and Cdc42 might play a role in MRTFa activation during myofibroblast differentiation.
To increase the understanding of the molecular pathways underlying fibrotic disease, we set out to investigate how TGF␤-induced myofibroblast differentiation is regulated in the pathophysiologically relevant precursor population of MSC. Importantly, we tested how the Rho GTPases RhoA, Rac1, and Cdc42 affect MSC differentiation into the myofibroblast and which signal transduction pathways are involved downstream of Rho GTPases. Our results indicate important roles for RhoA, Rac1, and Cdc42 in controlling TGF␤-induced expression of the myofibroblast differentiation marker ␣SMA in MSC but showed surprisingly little effect on the regulation of col1a1 mRNA. TGF␤-induced regulation of ␣SMA expression depended on contraction, but independently of cofilin, in contrast to previous suggestions. These data give important insights into myofibroblast differentiation of a disease-relevant progenitor population, which may help to find novel treatments preventing fibrosis in different tissues.

Establishment of a spontaneously immortalized MSC line
Standard preparations of primary mouse MSC are contaminated with other cell types at low passage numbers and prone to senescence upon longer passaging, leaving only a small window for experimentation. We therefore established a spontaneously immortalized murine MSC line from bone-derived MSC. After 58 passages, the MSC line showed high expression of the murine MSC surface markers CD105, CD44, and Sca1, whereas no contaminating hematopoietic cells expressing CD45 or macrophages expressing CD11b could be detected by FACS (Fig. 1A). Light microscopy revealed a fibroblast-like triangular cell morphology similar to primary MSC (Fig. 1B). Moreover, the MSC showed efficient differentiation into adipocytes and osteoblasts in the respective differentiation media (Fig. 1C).
Moreover, no differentiation into these cell types was detectable under normal growth conditions.
These data indicate that the established MSC line displays major hallmarks of primary MSC at least up to passage 58. Further experiments were carried out with cells from passage 42 to 47.

Canonical TGF␤ signaling triggers myofibroblast differentiation of MSC
TGF␤ promotes the differentiation of fibroblasts into myofibroblasts, characterized by expression of ␣SMA and collagen I. MSC are assumed to behave similarly but have to date been tested only for ␣SMA expression in response to TGF␤1. Treatment of our MSC line with TGF␤1 indeed resulted in effective up-regulation of both ␣SMA and col1a1 mRNA after 24 h (Fig.  2, A and B). To investigate the dependence of this differentiation on canonical Smad-related TGF␤ signaling, we established MSC lines lacking Smad2, Smad3, or Smad4 using lentiviral CRISPR/Cas-mediated genome editing and confirmed the absence of respective proteins by Western blotting (Fig. 2C). Deletion of Smad2 or Smad3 alone showed little effect, but ablation of the Smad4 gene effectively abrogated TGF␤-induced expression of ␣SMA or col1a1 (Fig. 2, D-F). Basal expression of ␣SMA and col1a1 was not significantly altered by the loss of Smad4.
These results demonstrate an essential role for canonical Smad4-dependent TGF␤ signaling for myofibroblast differentiation of MSC, whereas Smad2 or Smad3 are not required and

Rho GTPase regulation of myofibroblast differentiation
probably have redundant functions, in line with earlier observations in other cell types.

Rho GTPases control myofibroblast differentiation of MSC
Treatment of MSC with TGF␤ resulted in increased phosphorylation of Smad2 and the Rho GTPase downstream effectors MLC2 (in particular RhoA but also Rac1 and Cdc42), Pak1 (Rac1 and Cdc42), and Pak4 (Cdc42), suggesting that TGF␤ induces Rho GTPase activation in parallel with canonical TGF␤ signaling (Fig. S7). Kinetic analysis indicated a high and significant activation of Smad2, PAK1, and Pak4 after 24 h and of MLC2 after 2 h. To analyze the functional role of basal and TGF␤-induced Rho GTPase signaling in the TGF␤-induced myofibroblast differentiation of MSC, we established MSC lines lacking RhoA, Rac1, or Cdc42 by CRISPR/Cas9-mediated genome editing.
Although RhoA and Cdc42 were efficiently deleted, the loss of Rac1 protein was not complete, indicating the presence of a minor amount of nonrecombined cells (Fig. 3A). This was most likely because of the severely reduced growth of Rac1-null MSC, giving nonrecombined cells a competitive advantage in the polyclonal cell mixture (data not shown). MSC suppressed for RhoA, Rac1, or Cdc42 expression showed morphologies similar to fibroblastoid cells lacking these Rho GTPase genes ( Fig. 3B). Time-lapse migration assays revealed that Rac1-null MSC are hardly motile, whereas the long, thin extensions of Cdc42-null cells seen in still images (Fig. 3B) were observed to rupture during migration, leaving behind small patches of cytoplasm on cell tracks (Movies 1-4). RhoA-null MSC were also more elongated than WT cells, and at least a subpopulation of cells displayed extended rear edges. With respect to their differentiation potential, we found that the loss of Rho GTPases did not prevent or induce the differentiation of MSC to adipocytes or osteoblasts (Fig. S6). Furthermore, none of the Rho GTPase KO MSC lines showed altered phosphorylation of Smad2 after 24-h treatment with TGF␤ (Fig. 3, C and D), suggesting normal activation of TGF␤R kinase activity, crucial for both canonical and noncanonical signaling.
TGF␤-induced expression of ␣SMA was significantly decreased in RhoA KO and Cdc42 KO MSC after 24-h stimulation, whereas the loss of Rac1 increased basal ␣SMA expression strongly (Fig. 4A). With respect to -fold change, only Cdc42 KO resulted in a significantly reduced induction (Table S1). After 3-day TGF␤ treatment, the protein levels of ␣SMA were significantly lower in RhoA KO and Cdc42 KO cells compared with WT MSC, whereas no clear difference was detectable in Rac1 KO MSC (Fig. 4, B and C). To check whether constitutively

Rho GTPase regulation of myofibroblast differentiation
active Rho GTPase mutants show effects opposite to the KO, we overexpressed fast-cycling mutants of RhoA, Rac1, and Cdc42 (35) in MSC and tested the effect on TGF␤-induced ␣SMA expression. Fast-cycling mutants (fcRho GTPases) are able to cycle between a GDP-bound and a GTP-bound form but are mostly in the GTP-bound form. Lentiviral transduction of MSC resulted in a high percentage of fast-cycling mutant expressing cells, as determined by co-expressed EGFP (Fig. S9, A and B). fcRhoA increased TGF␤-induced ␣SMA (Fig. S9, E and F), thus showing opposite effects than KO of RhoA, while fcRac1 prevented TGF␤-induced ␣SMA expression. fcCdc42 MSC behaved similar to control cells, suggesting that although a basal Cdc42 activity is required for myofibroblast activation, superactivation of Cdc42 is not having an extra effect.
Surprisingly, the deletion of RhoA in MSC showed no influence on TGF␤-induced col1a1 mRNA expression (Fig. 4D), although RhoA is reportedly required for collagen 1 expression in fibroblasts (12). The loss of Rac1 or Cdc42 resulted in slightly reduced col1a1 levels after 1-day stimulation (Fig. 4D). However, after 3-day stimulation, the deletion of Cdc42 did not cause any change in col1a1 mRNA expression (Fig. 4E).
Following analysis of the cellular and secreted collagen 1 protein, both RhoA KO and Cdc42 KO MSC displayed an intracellular accumulation of cellular collagen 1 (Fig. 4, F and G). Moreover, secreted collagen 1 was significantly reduced in the supernatant of Cdc42-null MSC (Fig. 4, F and H).
These data indicate that both RhoA and Cdc42 are required for TGF␤-induced ␣SMA expression, whereas Rac1 plays an inhibitory role. They furthermore suggest a role for RhoA and Cdc42 in collagen I secretion but no or only a minor role in the regulation of col1a1 mRNA.

Defective ␣SMA induction in Rho GTPase KO correlates with changes in F-actin and pMLC2
To identify changes in downstream components of Rho GTPase signaling that might correlate with the defective induction of ␣SMA, we first analyzed F-actin levels by quantifying cell staining with fluorescently labeled phalloidin. TGF␤ treatment of control MSC induced a clear increase in F-actin in control but not in in RhoA KO and Cdc42 KO MSC ( Fig. 5A and Fig. S1). Rac1 KO MSC, on the other hand, showed increased levels of F-actin in both the presence and absence of TGF␤ (Fig.  S1). These changes in F-actin correlated with changes in pMLC2, which is a marker for cell contraction (Fig. 5B). MSC expressing fcRhoA showed increased TGF␤-induced pMLC2, while fcRac1 and fcCdc42 MSC showed no significant changes (Fig. S9, C and D).
ADF/cofilin family members are important regulators of F-actin by severing it and promoting depolymerization with cofilin-1 (cofilin) as the major form in nonmuscle tissues. Cofilin is known to be inactivated by phosphorylation downstream of RhoA, Rac1, and Cdc42. However, we could not detect any significant changes in cofilin phosphorylation in MSC lacking these Rho GTPases after 24 h of treatment with TGF␤ ( Fig. 5C and Fig. S2). p38, ERK, and JNK are described as phosphorylated and activated by noncanonical TGF␤ signaling (36). Testing pERK and pp38 after 24 h of TGF␤ treatment, no significant TGF␤-dependent stimulation and no Rho GTPase-dependent

TGF␤-induced actin polymerization controls ␣SMA but not col1a1 expression
In fibroblasts, actin polymerization promotes nuclear translocation of the transcription factor MRTFa, which induces expression of both ␣SMA and collagen 1 (10,11). In MSC, TGF␤ treatment increased the formation of F-actin stress fibers over 24 h, visualized by fluorescent LifeAct (Fig. S8). We therefore investigated whether this pathway plays a related role in MSC. Inhibition of actin polymerization in MSC by latrunculin abrogated basal ␣SMA as well as induced ␣SMA expression and caused a partial reduction of TGF␤-induced col1a1 expression (Fig. 6, A and B). Moreover, jasplakinolide, which stabilizes F-actin, dramatically increased basal ␣SMA levels (Fig. 6A), but TGF␤ treatment did not result in a further increase in ␣SMA. By contrast, jasplakinolide did not affect basal col1a1 expression and even inhibited the TGF␤1-induced increase of col1a1 (Fig. 6B).
Actin polymerization downstream of Rac1 and Cdc42 is prominently promoted by the Arp2/3 complex and down-stream of RhoA by formins of the mDia family. To probe the role of Arp2/3-mediated F-actin formation during myofibroblast differentiation, we deleted Arpc2, which is an essential component of the Arp2/3 complex (13,14). Although Arpc2 was efficiently deleted (Fig. 6C), we did not observe an alteration in TGF␤-induced expression of ␣SMA or col1a1 (Fig. 6, D and E).
CRISPR/Cas9-mediated cofilin targeting strongly increased stress fiber formation and ␣SMA levels but did not affect inducibility of ␣SMA by TGF␤1 (Fig. 6, F and G), confirming the notion that cofilin is not required for the TGF␤-induced increase of ␣SMA (Figs. 5C and 6G). CRISPR/Cas9-mediated deletion of the cofilin gene however did not affect col1a1 expression (Fig. 6H). Downstream of RhoA, Rac1, and Cdc42, LIMK inhibits cofilin by phosphorylation. Loss of LIMK should therefore decrease cofilin phosphorylation, increase cofilin activity, decrease F-actin, and decrease ␣SMA, at least if one assumes that cofilin expression is unaltered. LIMK KO were made, and frameshifts in the coding region of the respective genes (LIMK1 and LIMK2) were confirmed by sequencing of the genomic PCR fragments (Fig. S3A). Paradoxically, loss of LIMK1 increased basal and TGF␤-induced expression of ␣SMA, whereas LIMK2 KO had no influence (Fig. 6, I and J). Interestingly, only LIMK2 KO reduced pCofilin levels, indicat-

Rho GTPase regulation of myofibroblast differentiation
ing that LIMK2, but not LIMK1, is crucial for cofilin phosphorylation in MSC (Fig. 6L). LIMK1 KO, on the other hand, resulted in an increase of pCofilin, implying increased activation of LIMK2. Indeed, double knockout of LIMK1 and LIMK2 restored ␣SMA levels to that of WT cells (Fig. 6K). Conceivably, the deletion of LIMK1 was overcompensated by LIMK2, which was rescued by the additional deletion of LIMK2. Correspondingly, LIMK1 KO increased and LIMK2 or LIMK1/2 KO decreased F-actin in MSC (Fig. S3, B and C).
LIMK deletion did not influence TGF␤-induced col1a1 expression (Fig. 6M). Finally, deletion of the MRTFa gene, confirmed by genomic sequencing of the targeted genomic region (Fig. S3), blocked TGF␤ induction of ␣SMA but had no influence on col1a1 mRNA (Fig. 6, N and O). Deletion of MRTFa did not interfere with the differentiation of MSC to adipocytes or osteoblasts (Fig. S6).
These data demonstrate that Arp2/3 complex-independent but F-actin/MRTFa-dependent signaling is crucial for TGF␤induced ␣SMA expression in MSC. However, F-actin/MRTFa signaling is again dispensable for the stimulation of col1a1 expression by TGF␤. Moreover, Rho-GTPase-dependent regulation of ADF/cofilin activity is not essential for the TGF␤induced expression of ␣SMA.

TGF␤-induced cell contraction is required for increased actin polymerization
To understand the relationship between TGF␤-induced cell contraction, F-actin distribution, and ␣SMA expression, we

Rho GTPase regulation of myofibroblast differentiation
inhibited cell contraction directly by blebbistatin and indirectly by the ROCK inhibitor Y27632 or KO of contraction-promoting Rho GTPase effectors. Blebbistatin strongly reduced F-actin levels and prevented TGF␤-induced ␣SMA expression but did not affect col1a1 expression (Fig. 7, A, B, and E). Similarly, inhibition of the RhoA effectors ROCK1 and ROCK2 by Y27632 inhibited stress fiber formation and effectively reduced TGF␤-induced ␣SMA expression but not col1a1 expression (Fig. 7, C, D, and E). Individual knockout of ROCK1 and ROCK2 showed a partial reduction of pMLC2, F-actin, and ␣SMA, whereas the combined deletion of ROCK1 and ROCK2 had a stronger effect, indicating redundancy between ROCK1 and ROCK2 (Fig. 7, F, G, and I-K, and Fig. S4A). The combined deletion of ROCK1 and ROCK2 did not interfere with the differentiation of MSC to adipocytes or osteoblasts (Fig. S6).
Aside from ROCK, the Cdc42 effectors MRCK␣ and MRCK␤ are also reported to regulate cell contraction in a fashion similar to ROCK. However, deletion of these genes showed little influence on pMLC2, F-actin, or ␣SMA (Fig. 7, F, H, and L-N, and Fig. S5A), indicating that Cdc42 affects cell contraction in MSC in a nonclassical manner. Importantly, none of the manipulations described above led to a significant change in basal or TGF␤-induced col1a1 mRNA, corroborating that the regulation of ␣SMA and col1a1 expression in MSC is distinct (Figs. S4B and S5B).

Discussion
This study on the TGF␤-induced differentiation of MSC to myofibroblasts revealed several surprising findings (Fig. S10). First, it shows that TGF␤-induced expression of ␣SMA and collagen 1 are differentially regulated in MSC, in contrast to fibroblasts. Secondly, it indicates that Rac1 and Cdc42 control TGF␤-induced ␣SMA expression in an opposite manner and independent of Arp2/3 complex-mediated actin polymerization. Thirdly, it demonstrates that the regulation of cofilin activity is not important for TGF␤-induced ␣SMA expression, contrary to earlier expectations. Fourthly, the data suggest that RhoA, Rac1, and Cdc42 control ␣SMA expression by a contraction-dependent pathway. Finally, this study reveals that Cdc42 and RhoA affect cytoplasmic retention and secretion of collagen I in MSC.
Although MRTFa is required for collagen I expression in lung fibroblasts and in myofibroblasts derived from scleroderma (10, 11), collagen I expression in our MSC line was clearly independent of MRTFa, although TGF␤-induced ␣SMA expression was highly dependent on MRTFa. On the other hand, canonical Smad4-dependent TGF␤ signaling was essential for both collagen I as well as ␣SMA expression. This fits with an earlier report on mesangial cells showing that Smad4 is required for TGF␤-induced collagen I expression (15). It is well-known that the regulation of TGF␤-dependent genes is strongly context-dependent and cell type-specific (8). Synergistic interaction of MRTFa with Smad4 signaling at the collagen I promoter is required in fibroblasts and myofibroblasts (10, 11) but, as our data reveal, not in MSC. WT MSC will therefore be resistant to fibrosis-inhibiting drugs targeting the RhoA/ROCK/MRTFa pathway. It will be interesting to investigate why MSC do not require MRTFa activation for col1a1 expression. It is possible that MRTFa removes an inhibitory transcription factor or epigenetic mark from the col1a1 gene locus in fibroblasts but not in MSC. Such a mechanism was found in genes for master regulators of embryonic stem cell differentiation, where the Smad3-binding factor TRIM33 disables repressive histone marks (16).
Introduction of this col1a1 gene inhibitory signal to MSC will make their collagen I expression dependent on the activation of RhoA/ROCK/MRTFa. This could increase the efficiency of ROCK inhibitors as antifibrotic drugs. ROCK inhibitors such as fasudil show some efficacy in mouse and rat models of fibrosis (17), but no reports on clinical efficacy in humans have been published up to now.
Polymerization and depolymerization of F-actin and thus MRTFa activation is not only regulated by RhoA but also by Rac1 and Cdc42 (18). It was therefore expected that deletion of Rac1 or Cdc42 might decrease MRTFa-dependent ␣SMA expression by reducing Arp2/3-dependent actin polymerization or by increasing actin depolymerization via reduced phosphorylation of cofilin. Unexpectedly, our data revealed that Rac1 and Cdc42 modulate F-actin amounts and ␣SMA expression but in an Arp2/3-and cofilin-independent manner. Moreover, Cdc42 and Rac1 had an opposite effect on actin polymerization, suggesting that they regulate F-actin in MSC in a rather unconventional manner, probably dependent on cell contraction. MRCK␣ and MRCK␤ are ROCK-like effectors of Cdc42 that mediate contraction (19). In MSC, however, MRCK␣ and MRCK␤ are apparently not involved in the regulation of myofibroblast differentiation. Although the molecular details of the regulation of contraction by Cdc42 and Rac1 in MSC remain to be elucidated, our data clearly identify Cdc42 as a potential novel drug target for fibrosis therapy.
We observed increased JNK phosphorylation in Cdc42 KO MSC after 24 h of TGF␤ stimulation. However, control cells or cells lacking RhoA or Rac1 showed no effect, although a transient, earlier increase cannot be excluded. JNK activation leads, among other pathways, to phosphorylation and activation of c-Jun, reported to inhibit canonical TGF␤ signaling (20,21). Indeed, Cdc42-null MSC showed a certain reduction of collagen I expression, which may be related to the increased JNK activation. On the other hand, the JNK inhibitor CC-401 significantly inhibited renal fibrosis in a kidney fibrosis model (22), suggesting a profibrotic function of JNK. Moreover, JNK activation is a well-known part of noncanonical TGF␤ signaling and has been described as crucial for TGF␤-induced myofibroblast development in vitro (23). The molecular mechanism of Cdc42 loss-of-function-dependent JNK activation is not clear, as the latter is thus far considered a downstream effector of Cdc42 (24). JNK inhibitors employed in Cdc42-null MSC may reveal the relevance of JNK for myofibroblast differentiation.
Cofilin inhibition by phosphorylation commonly promotes F-actin formation (25), expected to result in increased MRTFa activation and increased MRTFa-dependent gene expression. Indeed, deletion of cofilin (cofilin-1) in MSC increased F-actin and ␣SMA expression, suggesting that MRTFa is able to stimulate ␣SMA expression also in the absence of exogenous TGF␤. However, the addition of TGF␤ resulted in a similar -fold change in ␣SMA expression as in WT cells, indicating that the TGF␤ induction of ␣SMA expression is largely independent of cofilin. Normal ␣SMA expression after deletion of both LIMK genes, negative regulators of cofilin downstream of RhoA, Rac1, and Cdc42, supported this notion. Interestingly, the deletion of RhoA, Rac1, or Cdc42 did not lead to changes in the basal levels of TGF␤-induced phosphocofilin, neither in the presence or absence of TGF␤ (at 2 h), although all three Rho GTPases affected the F-actin levels. These data suggest that the importance of Rho GTPase-dependent cofilin regulation is highly cell type-dependent.

Rho GTPase regulation of myofibroblast differentiation
A possible model explaining our data is that RhoA, Rac1, and Cdc42 regulate ␣SMA expression in MSC via a RhoA/ROCK/ pMLC2/F-actin/MRTFa pathway. Apparently, increased contraction of F-actin prevents its depolymerization, as already suggested previously (26). Because Arp2/3 complex-dependent actin polymerization is not required for TGF␤-induced ␣SMA expression, formin-mediated stress fiber formation could be important for regulating MRTFa in MSC, which would fit the correlation between stress fiber formation and ␣SMA expression observed here and the well-known correlation between contraction and stress fiber formation (27). Because Cdc42-and Rac1-dependent actin assembly operating in TGF␤-induced differentiation of MSC is not controlled through conventional regulation by Arp2/3, cofilin, or MRCKdependent regulation of contraction, it is possible that they act upstream of RhoA. Cross-talk between Rho GTPases is wellestablished in various conditions and cell types and may involve competition for GDIs, GDFs, and GAPs (28,29).
Finally, we observed an accumulation of intracellular collagen I in Cdc42 KO and RhoA KO cells. In Cdc42-null MSC, this correlated with a decreased amount of collagen I in the cell supernatant, but this was not obvious in RhoA-null cells. It was reported previously that in smooth muscle cells Cdc42 regulates collagen I secretion via PKC␦ (30). It appears that a similar mechanism might take place in MSC. Cdc42 is known to regulate transport between the endoplasmic reticulum and Golgi (31), whereas RhoA has not been reported to be involved in constitutive secretion. In any case, the secretion defect in Cdc42-null MSC further increases the interest in Cdc42 as a potential drug target for fibrosis therapy.
In conclusion, not only RhoA but also Rac1 and Cdc42 are important for noncanonical TGF␤ signaling, at least in MSC differentiation into myofibroblasts. Noncanonical TGF␤ signaling differentially regulates expression of ␣SMA and col1a1 in MSC, in contrast to fibroblasts and myofibroblasts, which may be relevant for fibrosis therapies. Finally, Cdc42 emerges as an interesting drug target for fibrotic diseases, as it may interfere with myofibroblast differentiation as well as collagen I secretion.
Primary MSCs were split 1:3 (primary) or 1:4 (MSC line) after reaching 90% confluence. For detachment, MSC were washed once in a 10-cm tissue culture dish (Greiner Cellstar P7612) with 2 ml of warm PBS and then incubated with 1 ml of 5% trypsin-EDTA (Gibco 25200-056) for exactly 2 min at 37°C. Trypsin digestion was stopped by adding 1 ml of MSC growth medium. Because contaminating macrophages mostly will not detach during this time, MSCs would be enriched by passaging.

MSC differentiation
The differentiation potential of MSC to adipocytes or osteoblasts was measured using corresponding kits from STEMCELL Technologies (STEMCELL catalog no. 05503 and 05504). Myofibroblast differentiation of MSC was stimulated by 5 ng/ml human TGF␤1 (Chinese hamster ovary cell linederived, Ala 279 -Ser 390 ; R&D Systems 240-B-002) for 24 or 72 h.
Lentivirus was produced by transfection of subconfluent HEK293 cells with 0.8 g of lentiCRISPRv2, as described above, or 0.8 g of pLenti.PGK.LifeAct-GFP (addgene 51010), 0.6 PAX2 (addgene 12260), and 0.6 g of VSVG plasmid (addgene 8454) in the presence of 8 g of polyethylenimine (Polyplus) in 200 l of optiMEM (Thermo Fisher Scientific 11058021). Virus-containing supernatant was collected after 24 and 48 h, filtered through a 0.45-m cellulose acetate membrane filter (VWR 28145-481) to remove cells, and used directly for the transduction of MSC. Supernatant of 40 -60% confluent MSC seeded the previous day was aspirated, and 1-2 ml of filtrated virus supernatant and 4 l of 4 g/ml hexadimethrine bromide (Polybrene; Sigma 107689) in PBS were added to each well of a 6-well plate. After 1 day, the supernatant was replaced with MSC growth medium. 48 h after transduction, antibiotic selection was started with 4 g/ml puromycin for 2 days or 8 g/ml Rho GTPase regulation of myofibroblast differentiation blasticidin for 5 days. MSC expressing fluorescent LifeAct were not selected by antibiotics but were identified by fluorescence.

F-actin staining
Sterile coverslips were placed into the wells of a 24-well plate with 500 l of 50 g/ml poly-L-lysine solution. After 1 h of incubation at 37°C, the coating solution was removed, and the surface was allowed to dry. 5000 MSC were plated on each of the glass slides in 1 ml of MSC growth medium. The next day, the MSC were stimulated with 5 ng/ml TGF␤1 (R&D Systems 240-B-002) for 24 h, and then MSC were fixed with 4% paraformaldehyde in PBS for 10 min at room temperature and stored at 4°C. To stain for F-actin, MSC on coverslips were incubated with 1:200 diluted phalloidin-Atto 568 (Thermo Fisher Scientific A12380) in PBS with 2% BSA and 0.5% Triton for 1 h at room temperature. Coverslips were washed with 1 ml of PBS for 10 s and then mounted on a glass slide using mounting buffer (Sigma F6057). For each sample, five randomly chosen nonoverlapping images were taken. For a given experiment, all images were taken with the same exposure time using a fluorescent channel appropriate for detection of Alexa Fluor 568 by conventional fluorescence microscopy (Axio Imager 2, Zeiss). Total F-actin signal was measured on each slide using ImageJ; each cell was selected manually, and the integrated density in each cell on the red channel was measured and quantified.

Movies
15,000 WT, KO, and LifeAct-expressing MSC were plated in each well of 4-well glass-bottom dishes (Thermo Fisher Scientific 154461PK) coated with 10 g/ml human fibronectin (Roche Life Science) 1 day before live cell imaging. Time-lapse microscopy was performed on an Axio Observer (Zeiss) equipped with an automated stage, a VIS-LED light source for phase-contrast optics, and light source for GFP, and a Coolsnap-HQ2 camera (Photometrics) driven by VisiView software (Visitron Systems). The microscope was equipped additionally with a 37°C incubator and CO 2 -aerated lid and switched on at least 1 h before the start of the experiment for equilibration and to avoid focus drifts during the experiment. Two channels of phase-contrast and GFP fluorescence movies were acquired using a ϫ10/0, 3NA objective on several randomly chosen positions per condition, at a frame rate of 6/h (100 images in total per region). Image stacks were then transformed into AVI movie files using ImageJ software.

Statistics
Data were presented as mean Ϯ S.E. Unpaired, two-tailed Student's t tests were used for comparisons between two groups. For multiple comparisons, two-way analysis of variance with Tukey's multiple comparisons was applied. Statistical analysis was done using GraphPad Prism. The following indications of significance were used throughout this study (Figs. 2-7): *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001.