YAP controls cell migration and invasion through a Rho-GTPase switch

Understanding the mechanisms controlling the invasive spread of normal and transformed cells is central to understanding diverse processes including cancer progression. Here, we report that Yes- associated protein (YAP), a central transcriptional regulator implicated in controlling organ and body size, modulates a Rho-GTPase switch that drives cellular migration by directly transactivating the Rac1-GEF protein TRIO. Additionally, YAP and TRIO activate the Rac1-STAT3 axis to promote invasive behavior. While we find this YAP-dependent infiltrative program in many cell types, it is particularly enhanced in a patient-specific way in the most common malignant brain tumor, glioblastoma (GBM), where hyperactivation of the YAP-mediated TRIO and STAT3 network also confers poor patient outcome and up-regulation of genes associated with the Mesenchymal subtype of GBM. Our analysis suggests that the YAP-TRIO-STAT3 signaling network identified in this study is a ubiquitous regulator of invasive cell spread in both normal and pathological contexts.

Shah, SR et al cell lines and non-neoplastic NHA and MCF10A cells (Fig. 1b-c, Fig. S1f-g, Movie S1). Conversely, stable overexpression of a constitutively active YAP (YAP OE) versus control vector (CONT) in GBM (JHGBM612 and JHGBM1A) and NHA cells showed that YAP up-regulation was sufficient to increase migration speed (Fig. 1d, Fig. S1h-i). To determine whether YAP affects infiltrative cell migration and spread in vivo, we used a murine intracranial xenograft tumor model of human brain cancer (25,26), which recapitulated the cellular infiltration along the corpus callosum seen in the donor patient tumor (Fig. S1j).
To do this, we injected 150,000 shCtrl or shYAP JHGBM612 cells into the striatum of the right hemisphere (at pre-determined coordinates from the bregma; see Materials and Methods) of immunocompromised mice and quantified the number of human GBM cells infiltrating along the corpus callosum into the left hemisphere 5 weeks post-injection. Firstly, we confirmed the presence of tumor after the xenograft transplantation of GBM cells by performing a standard H&E staining. The images indeed display features commonly seen in murine xenograft models of human GBM (Fig. S1k). More importantly, we observed significantly fewer shYAP cells infiltrating the contralateral hemisphere than shCtrl JHGBM612 cells ( Fig. 1e-f, Fig. S1l). These findings, coupled with our in vitro observations, suggest that YAP can enhance infiltrative cell migration in different micro-environmental contexts.
We then sought to identify the mechanism by which YAP modulates migratory behavior in nonneoplastic and cancer cells. To do so, we examined the regulation of small Rho GTPases, which are implicated as central regulators of cell motility(1). Interestingly, we found that YAP knockdown increased the number and area of focal adhesions, strongly affecting the shape of the cells (Fig. 1g-h, Fig. S1m-n) in a pattern reminiscent of the focal adhesion assembly phenotype observed upon RhoA hyperactivation (27). Consistent with pronounced focal adhesion assembly, an increase in absolute levels and ratio of phosphorylated and total LIM-kinase, a downstream target of Rho-associated protein kinase (ROCK), a major effector of RhoA(1), was observed in shYAP JHGBM651, JHGBM612, JHGBM1A, NHA, and MCF10A cells than their corresponding shCtrl cells (Fig. 1i, Fig. S2a). From these results, we Shah, SR et al conclude that YAP regulates cytoskeletal organization and dynamics to potentiate cellular motility in vitro and in vivo.

YAP regulates migration through a Rho-GTPase switch
To better understand the regulation of RhoA by YAP, we measured its GTP load through G-LISA assays. We observed a significant increase in RhoA-GTP levels after YAP knockdown in JHGBM651 and NHA cells (Fig. 2a, Fig. S2b). Based on this data, we sought to test the relevance of RhoA-ROCK activity on cell speed. We found that pharmacologic inhibition of ROCK I/II using either Y27632 or H1152 restored the migration speed of shYAP-1 and -2 cells in a dose-dependent manner while having no effect on shCtrl cells (Fig. 2b, Fig. S2c, Movie S2). Furthermore, treatment with Y12732 or H1152 decreased the phosphorylation of Nonmuscle Myosin II (NMII), confirming the inhibition of ROCK activity (Fig.   S2d). These results demonstrate that YAP can modulate RhoA-ROCK activity and that ROCK inhibition is sufficient to rescue the migratory speed of YAP knockdown cells. These results suggest that YAP promotes migration at least in part by suppressing RhoA-ROCK activity.
YAP can modulate the activity of RhoA either directly or indirectly. For instance, prior studies suggested that another small Rho-GTPase, Rac1, can antagonistically regulate RhoA (28)(29)(30), raising the possibility that modulation of RhoA activity by YAP might involve Rac1. We indeed found that Rac1-GTP levels were reduced after YAP knockdown (Fig. 2c, Fig. S3a). We also examined a downstream target of Rac1, p21-activated kinase (PAK), and found a pronounced decrease in phosphorylated PAK (p-PAK) level after normalizing to the total PAK protein level in shYAP cells compared to shCtrl cells (Fig.   S3b). Thus, we inquired whether changes in Rac1 activity accounted for the decreased migratory speed of shYAP cells. We found that a pharmacological inhibitor of Rac1-GEFs, NSC23766, significantly decreased the migration speed of shCtrl cells in a dose-dependent manner, but not that of shYAP-1 or -2 cells (Fig 2d, Fig. S3c, Movie S3). Collectively, these findings demonstrate that YAP regulates the small Shah, SR et al Rho-GTPase network governing cellular polarity and migration, leading to the activation of Rac1 and inhibition of RhoA in normal and cancer cells (Fig. S3d).
We next sought to clarify the precise regulatory mechanisms of YAP's interaction with Rac1 and RhoA given multiple scenarios of mutual regulation between the components. Our results suggest three possible hypotheses: 1) YAP activates Rac1 to inhibit RhoA, 2) YAP inhibits RhoA to activate Rac1, or 3) YAP independently modulates both Rac1 and RhoA. To test these hypotheses, we applied the logic of epistatic analysis by using pharmacological inhibitors of these small Rho-GTPases in lieu of dominant negative mutants. More specifically, we used sequential perturbation of the network components to deduce the regulatory hierarchy (further details and logic of the experiment are presented in Fig. S4a). Unlike the experiments described in Fig. 2a-d which explored the dose dependence of the inhibitors on the shCtrl and shYAP cells, our experiment here explores the time dependence of cell responses to either single inhibitor or sequential inhibitor application at fixed concentrations, high enough to achieve the maximal effect (10M Y27632 or 250M NSC23766). In addition, the time of the initiation of the experiment is when the first or only inhibitor used is already present long enough for the inhibition effect to be achieved (1 hour). Hence, the response of the cells can only change if the application of the second inhibitor has an added effect. Otherwise, the effects are expected to result in an unchanging response over time. Using this logic, we first verified that the migration speeds of shCtrl and shYAP-1or -2 cells were stable over a 6hour period with or without treatment using Y27632 or NSC23766 (Fig. 2e, Panels I-III, Fig. S4b). These results are consistent with our prior observations in Fig. 2a-d. Next, we used one or both pharmacological inhibitors of Rac1 and ROCK to determine how these signaling intermediates regulate each other in these cells. Specifically, we treated shCtrl and shYAP-1 or -2 cells with either the Rac1-GEF inhibitor or the ROCK inhibitor for 3 hours, followed by the addition of the other inhibitor for the next 3 hours (Fig. 2e,   Panels IV-V, Fig. S4b). The order of inhibitor treatments was indispensable in this analysis to ascertain which one of the aforementioned hypotheses is valid (Fig. S4a). Remarkably, we found that in cells treated to inhibit ROCK, subsequent inhibition of Rac1-GEF did not further alter the migration speed in either Shah, SR et al shCtrl or shYAP cells (Fig. 2e, Panel IV, Fig. S4b). However, ROCK inhibition after the initial Rac1-GEF inhibition rescued the cell migration speed of both shCtrl and shYAP cells (Fig. 2e, Panel V, Fig.   S4b). Altogether, these results strongly favored the model in which YAP activates Rac1 to inhibit ROCK signaling, resulting in the enhancement of cell migration.
We then investigated the effect of Rac1-GEF or ROCK inhibition on the activity of RhoA and Rac1 in shCtrl and shYAP cells. Our model predicts that Rac1 activity is not dependent on RhoA-ROCK signaling, and indeed we observed no change in Rac1-GTP or RhoA-GTP levels upon ROCK inhibition ( Fig. 2f, Fig. S4c, Fig S4d). Again, consistent with our model, we observed a significant increase in RhoA-GTP levels in both shCtrl and shYAP cells following Rac1 inhibition with NSC23766 (Fig. 2g).
As expected, Rac1 inhibition also significantly reduced Rac1-GTP levels in both shCtrl and shYAP cells (Fig. S4e). Finally, we tested whether RhoA perturbation was sufficient to restore the migratory speed of shYAP cells. Treatment with two RhoA siRNAs or a dominant-negative RhoA (DN-RhoA) rescued the migratory speed of shYAP cells, with no significant effect on the migration speed of shCtrl cells (Fig. 2hj). Likewise, introducing a constitutively active Rac1 construct yields similar results (Fig. 2k). These findings demonstrate that YAP activates Rac1, resulting in the inhibition of RhoA-ROCK signaling, which altogether enhances cellular migration (Fig. S4f). Thus, YAP simultaneously and sequentially tunes Rac1 and RhoA activities to potentiate cell motility.

YAP directly transactivates the Rac1-GEF TRIO to modulate Rac1 activity
Next, we sought to determine how the transcriptional co-activator YAP regulates the activity of Rac1. Small GTPases can be regulated by GTPase activating proteins (GAPs) and guanine nucleotide exchange factors (GEFs) (31). Since NSC23766 disrupts the interaction of Rac1 with two Rac1-GEFs, TRIO, and TIAM1 (32), we sought to determine if YAP regulates one or both of these Rac1-GEFs. We observed a decrease in TRIO protein and gene expression after YAP knockdown in all cell lines tested using two independent YAP-targeting shRNA sequences ( Fig. 3a-b, Fig. S5a, b). Moreover, TRIO Shah, SR et al expression increased following YAP overexpression both in GBM and MCF10A cells (Fig. 3c, Fig. S5c).
However, we did not observe any changes in TIAM1 gene expression in shYAP cells compared to shCtrl cells (Fig. S5d). Given that YAP, a transcriptional co-activator, cooperates with TEAD family transcription factors to directly modulate gene expression (6), we examined whether TRIO is transcriptionally regulated by YAP and whether YAP is found at known enhancer and promoter regions of TRIO. Using published chromatin immunoprecipitation (ChIP) sequencing data for YAP and TEAD4 in MDA-MB-231 cells (33), we found that both YAP and TEAD4 bind an intronic element of TRIO (Site 1: 14265617 to 14266192 base pairs on chromosome 5) that bears histone modification patterns indicative of active enhancers (34) (Fig. 3d). Specifically, this locus bears enhancer-associated H3K4 monomethylation (H3K4me1) and H3K27 acetylation (H3K27ac), with less promoter-associated H3K4 trimethylation (H3K4me3) (35). Notably, this enhancer region was independently reported to physically interact with TRIO's promoter (36). In addition, we found a weaker enrichment of YAP-TEAD4 binding at a second, intronic binding site (Site 2: 14268225 to 14268865 bp on chromosome 5) (Fig. 3d). By contrast, we found no evidence of YAP or TEAD4 binding at TIAM1 enhancer and promoter elements.
We tested the MDA-MB-231 YAP binding sites by ChIP-PCR on both intronic TRIO enhancer sites in GBM cells. Indeed, significant YAP binding was observed at both enhancer sites of TRIO (Fig. 3e, Fig.   S5e). To evaluate the functional consequence of this binding, we performed an enhancer luciferase assay using a 150 bp sequence (centered on the peak of YAP binding) paired with the promoter region of TRIO in HEK293 cells, with or without overexpression of YAP. We found that this sequence increased reporter activity compared to negative control and showed increased activity upon YAP overexpression (Fig. 3f).
To further confirm that YAP only interacts with the enhancer and not the promoter sequence of TRIO, we performed an additional control experiment using a 1470 bp promoter sequence of TRIO paired with a luciferase reporter, with or without overexpression of YAP. Since Zanconato et al, have previously shown that YAP's transcriptional regulation is largely restricted to direct enhancer binding with a few exceptions like CTGF where YAP occupies its promoter region, we also included a 612 bp sequence of CTGF's YAP co-activates migration and invasion Page 10 of 56 Shah, SR et al minimal promoter linked to a luciferase reporter as a positive control. Unlike CTGF's promoter, TRIO's full-length promoter sequence did not increase reporter activity and showed no enhancement in activity upon YAP overexpression (Fig. S5f). All these results suggest that YAP directly activates TRIO transcription by binding to its enhancer region.
Next, we examined whether YAP-TRIO signaling was sufficient to regulate Rac1 activity. We found an increase in Rac1-GTP levels in YAP OE cells compared to CONT cells (Fig. 3g, Fig. S5g).
Furthermore, stable overexpression of wildtype TRIO (TRIO OE) was sufficient to increase Rac1-GTP levels (Fig. 3h, Fig. S5h). Consistent with these findings, shRNA-mediated stable knockdown of TRIO (shTRIO) decreased Rac1-GTP levels (Fig. 3i, Fig. S5i). Given the antagonistic relationship between Rac1 and RhoA signaling, we assessed the activity of RhoA-GTP in these cells. We observed decreased RhoA-GTP levels in YAP OE and TRIO OE cells vs. control cells (Fig. S5j), and increased RhoA-GTP levels in shTRIO cells vs. shCtrl cells, paralleling our observations in shYAP cells (Fig. S5l). Collectively, these results demonstrate that YAP directly transactivates the Rac1-GEF TRIO and that TRIO is necessary and sufficient to activate Rac1 and inhibit RhoA.
Given that there has been evidence suggesting that TRIO-Rac1 signaling can inhibit YAP phosphorylation in HEK293 cells (37), we explored this potential feedback regulation in GBM cells.
Interestingly, we did not observe any significant change in either phospho-YAP or YAP protein levels when we knocked down TRIO (Fig. S5k), indicating that no potential direct feedback regulation between YAP and TRIO-Rac1 signaling exists in the context of GBM cells.
To test whether our in vitro observations were consistent with data from clinical samples, we assessed TRIO expression in intraoperatively obtained primary GBM tissues from patients (Fig. S5m).
We found that the full-length TRIO protein was indeed expressed in a majority of the GBM tissues analyzed. More importantly, its expression positively correlated with YAP expression in these clinical samples (Fig. 3j). Next, we surveyed gene expression profiles from the Repository for Molecular Brain Neoplasia Data (REMBRANDT) and The Cancer Genome Atlas (TCGA). Specifically, we explored YAP co-activates migration and invasion  Table 2). We refer to these YAP-regulated genes as the GB-YAP signature. Gene signatures allow a more robust estimation of YAP activity from gene expression datasets than mRNA alone, since many transcriptional regulators, including YAP, undergo significant post-transcriptional regulation. We found that the GB-YAP signature of GBM cells correlated with YAP signatures derived from other cell types (38) ( Fig. S5n). Furthermore, the REMBRANT data suggested that the patients with elevated GB-YAP signature genes also have significantly higher expression of TRIO, but not TIAM1 (Fig. 3k). Indeed, patients with TRIO overexpression were significantly overrepresented among patients with high GB-YAP signature expression in this dataset (comparing all observed patient group sizes to random expectation using Fisher's exact test (Fig. 3l). Taken together, these results derived from patient-derived tumor tissues support our in vitro observations that YAP modulates TRIO expression.

TRIO is necessary and sufficient for YAP-driven migration
Given our previous results, we sought to test whether TRIO was necessary and/or sufficient for YAP's control of cell migration. Consistent with its putative functional role, we found that silencing the expression of TRIO decreased the migration speed of GBM, NHA, and MCF10A cells (Fig. 4a, Fig. S6ab, Movie S5). As a preliminary test, we used a specific pharmacological inhibitor of TRIO, ITX3 (39). We confirmed ITX3 specificity by treating shTRIO cells with this compound and observed no further change in migration speed or Rac1-GTP levels ( Fig. S6c-d). We then found that ITX3 reduced the migration speed of shCtrl cells in a dose-dependent manner without affecting shYAP-1 or -2 cells (Fig. S6d, Fig.   S7a, Movie S4). Furthermore, higher doses of ITX3 also decreased the migration speed of YAP OE GBM cells to match the speed of CONT cells (Fig. S7b). Furthermore, ITX3 treatment significantly reduced Rac1-GTP levels in shCtrl and shYAP cells (Fig. S7c). Conversely, upon treatment with ITX3, a Shah, SR et al significant increase in RhoA-GTP level was observed in shCtrl, but not in shYAP cells (Fig. S7d), demonstrating negative regulation of RhoA activity by TRIO-Rac1 signaling downstream of YAP. These results suggest that the effects of YAP on RhoA-GTPases and cell migration can be entirely explained by TRIO expression.
We further explored whether YAP-mediated Rac1 activation was largely driven by TRIO. Given that we found that, in contrast to TRIO, the gene expression of TIAM1 was not reduced in shYAP cells, combined TIAM1/TRIO inhibition using NSC23766 did not significantly alter the migration speed or Rac1-GTP level of shTRIO-1 or -2 cells (Fig. S7e-f). On the other hand, the introduction of constitutively active Rac1 (CA-Rac1) rescued the migration speed in shTRIO cells but not shCtrl cells (Fig. S7g).
Furthermore, we found no significant change in RhoA-GTP level in shTRIO cells treated with NSC23766, which contrasted with the increase seen in shCtrl cells treated with the inhibitor compared to vehicle (Fig.   S7h). These results confirmed the specific role of TRIO but not TIAM1 signaling in mediating YAP's control of cell migration.
We then explored the effect of perturbations of TRIO on RhoA-dependent signaling and its phenotypic consequences. We found that ITX3 treatment significantly increased the RhoA-GTP level of shCtrl, but not shTRIO cells (Fig. S8a), providing additional evidence of the inhibitory effect of TRIOdependent signaling on RhoA. Furthermore, the introduction of a dominant-negative RhoA construct (DN-RhoA) rescued the migratory phenotype of shTRIO cells but not of shCtrl cells (Fig. S8b). Suppression of the downstream effector of RhoA with the ROCK inhibitors, Y27632 or H1152, rescued the migration speed of shTRIO cells, like our observations with shYAP cells (Fig. S8c, Movie S6). Consistent with our prior results, ROCK inhibitors did not significantly affect Rac1-GTP or RhoA-GTP levels (Fig. S8d), suggesting no feedback from ROCK on these Rho-GTPases. These results further indicate that TRIO inhibits RhoA activity to enhance cellular migration.
Next, we inquired whether TRIO was sufficient to rescue the migration speed of shYAP cells.
First, we confirmed that overexpression of TRIO increases migratory speed (Fig. 4b, Fig. S8e). Shah, SR et al Remarkably, overexpression of a knockdown-resistant TRIO in shYAP-1 or -2 cells (shYAP-1 or -2 + TRIO OE) completely rescued migration speed to levels of shCtrl cells (Fig. 4c, Fig. S8f). To further explore the molecular mechanisms of TRIO activity, we focused on its GEF function. TRIO has two distinct GEF domains, GEF1 and GEF2 that control Rac1 and RhoA activation, respectively. However, our findings indicate that TRIO activates Rac1 signaling while inhibiting RhoA to potentiate migration.
Thus, we sought to determine whether one or both GEF domains are relevant for this TRIO-mediated migratory behavior in GBM cells. We monitored migration speed upon stable overexpression of wildtype TRIO (TRIO OE), TRIO with inactivating GEF1 (TRIO iGEF1 OE) or GEF2 (TRIO iGEF2 OE) mutations (Fig. 4d). Notably, we observed an increase in migration speed of TRIO iGEF2 cells, phenocopying the migratory behavior of TRIO OE cells (Fig. 4d). However, we observed no change in migratory speed of TRIO iGEF1 cells (Fig. 4d). These results demonstrate that the GEF1 domain of TRIO, responsible for activation of Rac1, is indispensable for TRIO's pro-migratory effect. Moreover, TRIO's RhoA-activating GEF2 domain appears dormant and is dispensable for YAP-TRIO-driven migration in GBM cells. Our results demonstrate that TRIO is necessary and sufficient to account for Rac1 activation by YAP, resulting in RhoA inhibition and increased cell migration.

YAP promotes invasion by activating STAT3
In addition to modulation of migration speed, infiltration of the surrounding stroma tissue requires enhanced invasive capacity, enabling cell navigation in structurally complex tissue environments.
Interestingly, in both the TCGA and REMBRANDT tumor gene expression datasets, the GB-YAP signature is associated with higher expression of an invasive gene signature (IGS) (40) (Fig. 5a, Fig. S9a).
In a reciprocal fashion, patients with elevated expression of the IGS were significantly overrepresented among patients with high GB-YAP signature expression (Fig. S9b). Consistent with these observations, we observed that fewer shYAP and more YAP OE cells invaded through Matrigel-coated Boyden chambers than their respective control cells in vitro (Fig. 5b-c, Fig. S9c). This led us to explore the YAP co-activates migration and invasion Page 14 of 56 Shah, SR et al mechanisms of YAP-mediated regulation of invasion by surveying genes implicated in ECM remodeling, degradation, and proteolysis (41,42). We found that MMP2 and MMP9 (as well as the YAP-target control CTGF) mRNA expression was decreased in shYAP GBM cells (Fig. S9d). Given that these genes are transactivated by STAT3(43-45), a transcription factor known to promote invasion and metastasis(46), we hypothesized that YAP may modulate STAT3 activity. Indeed, we found evidence of an association between elevated expression of the GB-YAP signature and a published GB-STAT3 signature(47), with significant overrepresentation of patients with elevated expression of the two signatures in GBM samples from both clinical datasets ( Fig. 5d-e, Fig. S9e). This prompted us to further investigate this correlation between YAP and STAT3 activity. In vitro, we observed decreased phosphorylation of STAT3 on Tyr705 in shYAP-1 or -2 cells (Fig. 5f, Fig. S9f). Moreover, YAP OE cells expressed higher levels of phosphorylated STAT3 than CONT cells (Fig. 5g, Fig. S9f). Functionally, pharmacological inhibition of STAT3 using LLL12 impaired both the invasive and migratory capacity of shCtrl cells more than that of shYAP cells (Fig. 5h, Fig. S9g-h, Movie S7). These results suggest that YAP activates STAT3 to drive both migration and invasion.
We then surveyed primary GBM tissues for expression and phosphorylation of STAT3 to assess a possible clinical significance of this factor (Fig. 5i). We indeed observed both expression and phosphorylation of STAT3 in most patient-derived tumor samples and the majority of specimens with elevated YAP expression also exhibited high STAT3 phosphorylation levels ( Fig. 5i-j). Conversely, only a few tumor samples with low YAP levels had high STAT3 phosphorylation (Fig. 5i-j). Taken together, these clinical results lend support to our observations in vitro that YAP increases STAT3 activity to potentiate invasion.

YAP potentiates invasion by activating the TRIO-mediated Rac1-STAT3 axis
What is the mechanism of YAP-mediated STAT3 regulation? We explored the possibility that it might depend on the TRIO-mediated activation of Rac1 because there is evidence suggesting that Rac1 Shah, SR et al can activate STAT3(48). It is reported that Rac1 can directly bind to and regulate STAT3 activity(49, 50).
Rac1 can also indirectly induce STAT3 activity through the autocrine production of cytokine IL-6(51, 52). Various normal biological processes involve activation of the Rac1-STAT3 signaling such as gland involution(53), chondrogenesis(54), and primordial follicle pool production(55). Additionally, studies have shown that the Rac1-STAT3 axis is important in promoting invasion and epithelial-mesenchymal transition in colorectal cancer(50) and gastric cancer(56). However, few efforts have been put into exploring this signaling axis in GBM. Given its presentation in clinical specimens and datasets, we sought to investigate its connection to the YAP-TRIO-Rac1 network. Firstly, we observed a decrease in STAT3 phosphorylation and invasion following the silencing of TRIO expression vs. shCtrl cells, phenocopying the decreased invasive capacity of shYAP cells ( Fig. 6a-b, Fig. S10a-b). Similarly, applying TRIO inhibitor ITX3 decreased STAT3 phosphorylation levels in a dose-dependent manner (Fig. S10c). In addition, we observed reduced invasion of shCtrl cells treated with this TRIO inhibitor, but not in shYAP and shTRIO cells after treatment (Fig. S10d). Furthermore, TRIO OE GBM cells exhibited an increase in STAT3 phosphorylation and invasion compared to CONT cells ( Fig. 6c-d). Remarkably, stable overexpression of TRIO in shYAP cells (shYAP + TRIO OE) recovered the phosphorylation levels of STAT3 to shCtrl cell levels (Fig. 6e). To explore the link between Rac1 and STAT3, we treated the cells with the Rac1-GEF inhibitor, and we observed decreased phospho-STAT3 expression in a dose-dependent manner, mirroring our findings with TRIO inhibition (Fig. S10c). We also tested the involvement of RhoA in this YAP-STAT3-signaling cascade by applying the ROCK inhibitors, Y26732 and H1152, in shYAP and shTRIO cells. Not surprisingly, ROCK inhibition rescued the invasive capacity of shYAP and shTRIO cells without significantly altering the invasion of shCtrl cells (Fig. S10e). These results suggest that not only Rac1 activation but also RhoA inhibition is involved in this YAP-driven regulation of cell invasion.
Finally, we confirmed that there was a significant decrease in MMP2 protein level after knocking down YAP, TRIO, or STAT3 compared to shCtrl cells indicating that MMP2 acts as a candidate downstream target of STAT3 to promote GBM cell invasion (Fig. S10f). Collectively, these findings suggest that there Shah, SR et al exists a crosstalk between STAT3 and YAP-TRIO driven Rho-GTPase switch to modulate the invasive capacity of GBM cells.
Given the critical role of YAP-TRIO signaling in driving the migratory and invasive capacity of GBM cells, we sought to evaluate the relevance of this cascade in vivo using our murine intracranial xenograft model of GBM infiltration. Specifically, we sought to determine whether TRIO is necessary for YAP-driven cellular infiltration in vivo. To that end, we established control (CONT) and YAP overexpressing (YAP OE) JHGBM612 cells with or without TRIO knockdown. In vitro, we found decreased STAT3 phosphorylation in both CONT + shTRIO and YAP OE + shTRIO cells, compared to their respective controls, reinforcing the relevance of TRIO signaling in YAP-driven STAT3 activation ( Fig. 6f). Next, we injected 50,000 CONT + shCtrl, CONT + shTRIO, YAP OE + shCtrl, or YAP OE + shTRIO JHGBM612 cells into the striatum (at pre-determined coordinates from the bregma; see Materials and Methods) of immunocompromised mice and quantified the number of human GBM cells infiltrating along the corpus callosum 5-weeks post-injection. We observed that overexpression of YAP was sufficient to increase invasion of JHGBM612 cells via the corpus callosum into the contralateral hemisphere (CONT + shCtrl vs YAP OE + shCtrl; Fig. 6g-h). Moreover, knocking down TRIO significantly attenuated the invasive potential of these GBM cells (CONT + shCtrl vs CONT + shTRIO; Fig. 6g-h). In addition, we found a significant reduction in the number of YAP OE + shTRIO cells vs. YAP OE + shCtrl cells migrating through the corpus callosum ( Fig. 6g-h). Together, these results indicated that while YAP is sufficient to increase the invasive capacity of GBM cells, TRIO is necessary to promote YAP-driven cell dispersion in vivo. In summary, our in vitro and in vivo findings identify TRIO as a critical switch through which YAP activates Rac1 and STAT3 signaling to potentiate cellular infiltration in GBMs (Fig. 6i).

YAP-TRIO-STAT3 signaling predicts poor clinical outcome in glioblastoma
Our results suggest YAP-driven migration and invasion are critical for the biology of aggressive cancers. We, therefore, explored the clinical implications of these findings for gliomas, of which GBM is Shah, SR et al the highest grade (Grade 4) tumor with a median survival of 14 months. We found that GB-YAP signature expression increased with glioma grade in the REMBRANDT dataset (Fig. S11a); notably, STAT3 and TRIO have also been reported to increase with glioma grade (57). Additionally, the percentage of glioma patients overexpressing the GB-YAP signature compared to non-cancer cortex increased with glioma grade (Fig. 7a). Having found that YAP is hyperactive in GBMs, we focused on its prognostic value for these patients as determined by Kaplan-Meier analyses. Strikingly, higher expression of the GB-YAP signature predicted poor prognosis of GBM patients from both TCGA (Fig. 7b) and REMBRANDT cohorts (Fig. S11b). In addition, the YAP gene signature was more predictive than 91.5% of 1 million simulated, size-matched gene sets (empirical p = 0.085), suggesting it is one of the most predictive gene sets of its size.
Due to the molecular heterogeneity of GBMs, we explored if YAP activity was associated with one or more of the GBM subtypes including proneural, classical, and mesenchymal subtypes recently described by Wang et al (58). Notably, Mesenchymal GBMs are distinguished by the invasion of brain parenchyma coupled with pronounced angiogenesis (59). Moreover, the transcription factors STAT3 and C/EBPβ have been proposed to synergistically drive Mesenchymal GBMs(47). Of interest, TAZ, a paralog of YAP, was also found to potentiate Mesenchymal differentiation and transformation, but in a fashion that is independent of STAT3and C/EBPβ, through direct transcriptional activation (60). Consistently, Bhat et al (60) observed no modulation of STAT3 upon TAZ knockdown in GBM cells. These results raised the possibility that YAP might modulate the Mesenchymal phenotype associated with the eponymous sub-class in a way distinct from TAZ, through the activation of STAT3. Indeed, there was a significant over-representation of GB-YAP-high patients among the Mesenchymal GBMs, but not among other subclasses (Fig. 7c). Furthermore, our findings that YAP-TRIO signaling regulated the activity of STAT3 by promoting its phosphorylation (Fig. 5, Fig. 6, Fig. S9, Fig. S10) suggest that YAP may be a driver of mesenchymal reprogramming in GBMs. Consistent with this hypothesis, higher expression of the GB-YAP signature corresponded to higher expression of Mesenchymal subclass genes in both GBM Shah, SR et al datasets (Fig. 7d). In vitro, we observed decreased expression levels of several Mesenchymal subclass genes after YAP knockdown in all three GBM cell lines (Fig. 7e). Taken together, these results demonstrate that YAP can drive Mesenchymal subclass-related genes in GBMs at least in part by modulating STAT3 activity.
Interestingly, we also found that YAP-low patients were significantly overrepresented among GBMs in the glioma CpG island methylator phenotype (G-CIMP) subclass, which exhibits hypermethylation at CpG islands and longer progression-free survival (Fig. 7c). This result suggested that patients with lower YAP activity would present a less aggressive subtype of GBM, and most patients in this group might not benefit from any potential therapies targeting YAP. To explore this idea further, we inquired whether the prognostic power associated with enhanced YAP activity was restricted to a particular subclass of patients. Using Kaplan-Meier analyses, we evaluated progression-free survival along with YAP signature expression in each GBM subclass individually. Strikingly, higher expression of the GB-YAP signature predicted rapid recurrence in all the GBM subclasses except the G-CIMP subtype ( Fig. S11c-e). This suggests that YAP-targeted therapies may be effective for patients with non-G-CIMP tumors, and especially the Mesenchymal subclass.
We then investigated whether the YAP-TRIO-STAT3 signaling axis is clinically relevant in GBMs. First, using the TCGA dataset, we determined the distribution of patient groups with hyperactivation of either one, two, or all three of the network components (i.e., YAP, TRIO, and STAT3).
Hyperactivation was determined using the GB-YAP signature, TRIO transcript, and GB-STAT3 signature ( Fig. 7f). Indeed, we found a significant over-representation of GBM patients with hyperactivation of all the three members of the YAP-TRIO-STAT3 molecular network. Next, we stratified patients into 8 groups representing every possible combination of YAP, TRIO, and STAT3 co-activation, and performed Kaplan-Meier survival analyses. Notably, the groups that exhibited YAP co-activation with either TRIO or STAT3 had a significantly worse prognosis than the group without hyperactivation of any of these proteins (YAP LOW , TRIO LOW , STAT3 LOW ) ( Fig. 7g-i, Fig. S11f-g, Fig. S12). Furthermore, the results Shah, SR et al showed that YAP+TRIO or YAP+STAT3 co-activation was associated with a similarly poor prognosis as hyperactivation of all three regulators (i.e. YAP HIGH , TRIO HIGH , STAT3 HIGH ). Importantly, hyperactivation of TRIO and STAT3 (i.e., TRIO+STAT3) without high YAP activity did not appear to be associated with poor patient prognosis in this dataset (Fig. S12), emphasizing the prognostic power of YAP-dependent signaling. Overall, these analyses emphasize the co-dependence of TRIO and STAT3 on YAP to confer poor patient prognosis. These results should be viewed as preliminary, as the tumor purity, mRNA degradation, and cohort size might be confounding factors in this analysis. Larger patient cohorts with matched expression data, which are currently unavailable, would help to validate these results.
Nevertheless, collectively, our study provides important initial evidence of the clinical value of the YAPdriven pro-migratory and invasive genetic circuit in predicting patient outcomes (Fig. 7j).

DISCUSSION:
This study suggests a central and widespread role for YAP in controlling the migratory speed and invasiveness of several cancer and normal cell types through the modulation of a Rho-GTPase switch.
Previous work has illustrated the function of cytoplasmic YAP in regulating endothelial cell migration (61). Our work uncovers a YAP-mediated signaling circuit focusing on the role of nuclear YAP as a transcriptional coactivator. Specifically, we demonstrate that YAP increases the expression of the Rac1-GEF TRIO, leading to activation of Rac1 and inhibition of RhoA small GTPases, and subsequently resulting in STAT3 activation. This intricate cascade simultaneously promotes cell motility and invasion, using both transcriptional and post-transcriptional mechanisms (Fig. 7j). Furthermore, we find that, although TRIO harbors two GEF domains, its YAP-driven functions are mediated by the GEF1 domain's activation of Rac1. This agrees with studies in different systems that demonstrate Rac1 is the major target of TRIO (20). One known example of RhoA activation by TRIO is Gaq signaling, where Gaq binds to the PH2 module within the GEF2 domain, releasing PH2-mediated inhibition of DH2, thereby stimulating GEFD2 activity on RhoA (62). Nevertheless, the activity of both Rac1 and RhoA was found Shah, SR et al to be important in mediating the YAP-driven TRIO-dependent control of cell migration and invasion, suggesting the importance of the crosstalk between Rac1 and RhoA, downstream of TRIO activation.
Intriguingly, it has been reported that TRIO-mediated Rho-GTPase signaling can regulate YAP (63)(64)(65)(66)(67)(68) which, coupled with our results, suggests that regulation between YAP and Rho-GTPases can be mutually enhancing, constituting a feedback loop that can stabilize migratory and invasive phenotypes. Recently, it has been shown that YAP regulates actin dynamics through the modulation of ARHGAP29 to promote metastasis of gastric cancer cells (69). Interestingly, unlike in GBM, YAP modulates Rac1 via transactivation of the Rac1-GEF, TIAM1, through an enhancer-mediated mechanism, during invadopodia formation in breast cancer, supporting the argument for disease-specific regulation (70). Since YAP expression and activity are controlled by a diverse range of stimuli and genetic alterations (3,5,7,14,15,64,67,71), the evidence of crosstalk between YAP, Rho-GTPases, and STAT3 establishes a wide spectrum of inputs and mechanisms that regulate cell locomotion. Furthermore, our results implicate YAP and STAT3 as key transcriptional regulators of cell migration and invasion, which could aid the development of metastatic gene expression signatures for cancer. A recent study reported the role of YAP signaling in the activation of the JAK-STAT3 pathway to regulate endothelial cell proliferation during angiogenesis (72), further establishing the significance of YAP-mediated activation of STAT3 in different phenotypic and cellular contexts. Similar to our findings, others have demonstrated the regulation of STAT3 by Rac1, further reinforcing the relationship between these signaling cascades.
Likewise, additional studies have shown that STAT3 activation by YAP through other intermediatory regulators can be facilitated by direct interaction. Altogether, these findings support a model of crosstalk between multiple shared signaling components in a cell-type and context-specific manner.
Together with our previous study by Park et al (73), our results suggest that YAP-controlled cell migration and invasion are fundamental functions present in many cell types. This pathway may be particularly important in developmental and physiological contexts involving episodes of large-scale migration, such as wound healing and progenitor cell navigation during neural tissue development. In this Shah, SR et al context, our study suggests that the central role of YAP in controlling cell and tissue growth is balanced by regulating cell dispersion required by many morphogenetic events.
Given the widespread hyperactivation of YAP in many cancers (11), our study further clarifies the mechanisms controlling invasive cancer cell phenotypes. One of the hallmarks of tumor malignancy and progression to higher pathological grades is the ability of cancer cells to invade their surrounding parenchyma, intravasate surrounding blood, and lymphatic vessels, and eventually, seed distant tissues as metastatic tumors (74). Tumor recurrence and lethality can be greatly aided by the migratory and invasive capacity of cells (75). While the YAP-TEAD interaction has been shown to promote metastasis in breast cancer(6), the downstream signals responsible for increased metastatic potential have remained poorly understood. As with metastatic cancers, GBMs often evade eradication because individual cells spread from the primary bulk tumor, thus making complete resection and localized radiation treatment virtually impossible. Clinically, more than half of metastatic tumors display pronounced local infiltration (24) and can recur in as little as 3 months despite radical surgery, chemo-and radiotherapy (21)(22)(23). Thus, understanding the molecular basis for cell dispersal can inform patient prognosis and facilitate the development of improved treatment modalities (75)(76)(77)(78).
Our finding that YAP-TRIO signaling can promote STAT3 activity through activation of Rac1 establishes a novel pro-invasive signaling axis with possible therapeutic implications in glioblastoma and other cancers, where YAP and STAT3 can independently and synergistically confer malignancy and predict poor prognosis(47). Moreover, this YAP-TRIO-STAT3 network is found in multiple cell types and thus forms a general molecular framework for predicting clinical outcomes in aggressive cancers such as glioblastoma. Importantly, our work highlights the dependency of the downstream effects exercised by YAP on TRIO and/or STAT3 activation in conferring poor GBM patient prognosis. Therefore, our study suggests that a network, systems perspective on the etiology of aggressive cancers can benefit from explicit analysis of molecular cascades that integrate numerous signaling pathways, inform the design of         Table   1. GBM cell line derivation and culture protocols were as previously described (26), without the addition of N2 supplement and cultured on laminin-coated (1μg/cm 2 ) plates. Specifically, DMEM/F12 (Invitrogen) was supplemented with B27 supplement (Invitrogen), antibiotic/antimycotic (Invitrogen), human EGF (Peprotech), and FGF (Peprotech). MCF10A cells were purchased from ATCC and maintained in complete MEGM media (Lonza) supplemented with 100 ng/mL cholera toxin (Sigma) as recommended.
GBMA172 cells were purchased from ATCC and maintained in DMEM (Gibco) supplemented with 10% FBS (Gibco). NHA cells were purchased from Lonza and cultured as recommended. All cell lines were tested and confirmed to be mycoplasm free using a PCR-based MycoDtect TM kit from Greiner Bio-One.   were cut into 10-micron coronal sections using a cryostat microtome.
Brain sections spaced every 500μm were stained with hematoxylin and eosin to identify landmark neuroanatomical structures and therefore identify comparable slides from each mouse for further analysis.
Slides immediately adjacent to the identified slides were then selected for immunohistochemistry and quantification. Antigen retrieval was performed using sodium citrate (10 mM Immunoreactive bands were visualized using the appropriate horseradish peroxidase-conjugated anti-IgG antibodies (Pierce). Bands were detected using enhanced chemiluminescence or prime detection reagent (GE Healthcare) whenever appropriate. Densitometric analysis was done with ImageJ and results are listed in Suppl For focal adhesion analysis, high-resolution images were captured using a fluorescence microscope at 20-40X. Quantification of area or size was conducted using ImageJ.
Antibodies used for these procedures are listed at the end of this section. high-power fields at 10x using an Olympus 1X81 microscope system, and calculating the mean per well.

Pull-down of GTP-bound
The assays were run in triplicates and at least three independent experiments were performed. See also Illumina beadchip data, signature probesets (GB-STAT3 and GB-C/EBPβ datasets (47)) were required to map to a single Entrez Gene ID. Redundant probesets mapping to the same gene were eliminated after differential expression analysis.
To derive gene signatures, each cell line was analyzed for differential expression independently (GB-YAP, GB-STAT3, and GB-C/EBPβ datasets). Genes with q < .05 and absolute fold-change > 1.5 were considered as differentially expressed genes, although q < 0.1 was used when no genes passed q < 0.05.
This relaxed FDR was excused because the final signatures were taken as the genes significantly differentially expressed in both cell lines, at least one of which required q < .05.
To stratify patients into groups of low, intermediate, and high expression of a given gene signature, we used the gene set z-score(82). Briefly, RMA normalized data were quantile normalized, variance filtered, and log2 transformed. Gene sets were filtered to remove genes not present in the normalized, filtered data for each analysis. Each gene was converted into an expression z-score relative to other samples of the same class (e.g., other GBMs but not low-grade gliomas), and the gene set score (GSS) was computed as the sum of the member genes' z-scores divided by the square root of the size of the gene set. Samples with

YAP co-activates migration and invasion
Page 45 of 56 Shah, SR et al GSS < -0.5 were categorized as low, samples with GSS > 0.5 were high, and those between were intermediate. Since microarray data to derive a gene signature of TRIO was unavailable, patients with a TRIO gene z-score < -0.5 were called low and those with a TRIO gene z-score > 0.5 were called high.

Target Protein
Company Note YAP p-S127 Cell Signaling

Rac1
Cell Biolabs Included in kit

RhoA
Santa Cruz Clone 26C4
Supplemental Table 3. Densitometric quantification of all immunoblotting results included in this paper.