Receptor tyrosine kinase activation of RhoA is mediated by AKT phosphorylation of DLC1

The RhoA GTPase (RhoA) is an essential, widely expressed, membrane-associated, guanine nucleotide–binding protein that contributes to various physiologic processes, including cell proliferation, cytoskeletal dynamics, cell migration, cell metabolism, cytokinesis, and vesicle trafficking. It is frequently activated in advanced cancer and has also been implicated in cardiovascular and other diseases (Zhou and Zheng, 2013; Loirand, 2015; Ricker et al., 2016; Shimokawa et al., 2016; Wu and Xu, 2016).

RhoA acts as a molecular switch that is active when bound to GTP and inactive when bound to GDP. Regulation of RhoA by ligands for G protein–coupled receptors, especially those for lysophosphatidic acid (LPA), has been recognized for many years (Xiang et al., 2013; Yu and Brown, 2015). RhoA can also be regulated by adhesion and mechanical factors (Marjoram et al., 2014). In addition, receptor tyrosine kinases (RTKs) may up-regulate RhoA under some conditions, secondary to RTK-dependent activation of Rho guanine nucleotide exchange factors (Rho-GEFs), which catalyze replacement of GDP-bound RhoA with GTP-bound RhoA (Schiller, 2006).

RhoA can also be negatively regulated by Rho guanine nucleotide dissociation inhibitors, which sequester RhoA from the membrane (Garcia-Mata et al., 2011), and Rho GTPase-activating proteins (RhoGAPs), which inactivate RhoA by catalyzing the hydrolysis of GTP-bound RhoA to GDP-bound RhoA. However, their role in ligand-dependent RhoA signaling is not well established. Here, we report that ligand-dependent stimulation of RTKs in epithelial cells and fibroblasts can stimulate the activation of RhoA, and we determined that the activation was attributable to a previously unknown mechanism, down-regulated activity of a specific, widely expressed RhoGAP DLC1 by a process that involves its phosphorylation by the serine/threonine kinase AKT.

We observed that stimulation of the EGF RTK, with its cognate ligand EGF, could activate RhoA in two nontransformed epithelial cell lines, a fibroblast line, and a subset of cancer cell lines. Analysis of the lines unexpectedly found an excellent correlation between the ability of EGF to increase RhoA-GTP and the expression of endogenous DLC1, which is a tumor-suppressor gene that encodes a 1091 amino acid protein containing a highly conserved RhoGAP domain and is required for its tumor-suppressor function (Durkin et al., 2005; Lukasik et al., 2011). The nontransformed lines H2071 (skin epithelial cells), FHL124 (lens epithelial cells), and H1634 (fibroblasts) all express DLC1, and EGF increased RhoA-GTP in each of them (Fig. 1 A), as did all four DLC1-positive cancer lines tested: two breast cancer lines, BT549 and MCF10Ca1h, and two non–small cell lung cancer (NSCLC) lines, H1703 and H157 (Fig. 1 B and Fig. S1 A). However, EGF did not increase RhoA-GTP in the DLC1-negative lines examined: two breast cancer lines, T47D and MDA-MB-468, and two NSCLC lines, H358 and A549 (Fig. 1 C and Fig S1 B).

To determine whether the EGF-induced increase in RhoA-GTP depended on DLC1, the effect of DLC1 knockdown by siRNA was tested in DLC1-positive lines: two nontransformed, H2071 and H1634, and two transformed lines, H1703 and BT549. The siRNAs efficiently suppressed DLC1 expression in each line, leading, as expected, to an increase in basal RhoA-GTP levels (Fig. 1, D and E; and Fig. S1, C and D). Although EGF activated the EGF receptor (EGFR), as measured by EGFR-Y845 phosphorylation, whether or not the cells had been treated with DLC1 siRNAs, EGF increased RhoA-GTP only under conditions with continued expression of DLC1 (Fig. 1, D and E; Fig. S1, C and D; and Fig. S1 E, which shows the relative DLC1 expression of all lines used in this study). Thus, the EGF-induced increase in RhoA-GTP depended on the presence of DLC1.

In a preliminary exploration of the mechanism by which EGF might be regulating DLC1, the level of serine phosphorylation of DLC1 increased after EGF treatment (Fig. S1 F), whereas tyrosine phosphorylation was unchanged (unpublished data). One serine/threonine kinase known to be activated by EGF is AKT (Fig. S1 G; Garay et al., 2015; Nishimura et al., 2015). A previous study identified one serine in DLC1 (S567) as being phosphorylated by AKT (Ko et al., 2010), which decreased the growth inhibitory activity of DLC1 by a mechanism the authors concluded was not associated with a change in RhoA-GTP. Two additional serines in DLC1 are possible AKT substrates (S298 and S329), but Ko et al. (2010) concluded they were not phosphorylated.

We evaluated whether AKT activation might be associated with the EGF-induced stimulation of RhoA. As expected, EGF treatment led to increased AKT activity, as measured by AKT-S473 phosphorylation, which was not attenuated by siRNA knockdown of DLC1 (Fig. 1, D and E; and Fig. S1, C and D). To explore the temporal relationship between EGF-induced changes in EGFR activation, AKT activation, and RhoA-GTP level, we evaluated two nontransformed DLC1-positive lines (H2071 and H1634) with low AKT activity and three DLC1-positive cancer lines (BT549, H1703, and H157) with high AKT activity. After 15 min of EGF treatment, EGFR was activated in each line; it remained high at 30 min and then decreased at 60 and 120 min (Fig. S1 H). The kinetic profiles of AKT activity, as measured by pAKT-S473 and RhoA-GTP, paralleled that of EGFR activity (Fig. S1 H).

To test whether AKT activity was mechanistically linked to the observed changes in RhoA-GTP, two DLC1-positive lines (BT549 and H1703) and two DLC1-negative lines (T47D and H358) were treated with an AKT inhibitor, MK-2206 (Hirai et al., 2010). Although AKT activity was efficiently inhibited in all lines, MK-2206 substantially reduced RhoA-GTP in the DLC1-positive lines (Fig. 1 F, two left columns), but not in the DLC1-negative lines (Fig. 1 F, two right columns). To verify that the observed effect of AKT activity on RhoA-GTP depended on DLC1, H1703 cells in which DLC1 had been knocked down by siRNAs were treated with MK-2206. Under those conditions of low DLC1, MK-2206 did not affect RhoA-GTP (Fig. 1 G). We also examined MK-2206 on an H358-derived cell line that, although the parental line does not express DLC1, had been stably transfected with DLC1. MK-2206 had no effect on RhoA-GTP in the parental cells, but it did reduce RhoA-GTP in the DLC1 transfectant (Fig. 1 H). Thus, AKT-regulated RhoA-GTP was dependent on DLC1. AKT also negatively regulated the RhoGAP activity of other DLC family members, DLC2 and DLC3 (Lukasik et al., 2011; Braun and Olayioye, 2015), but the effect of AKT inhibition on DLC2 and DLC3 on RhoA-GTP was less than that on DLC1 when cells transfected with the DLC genes were treated with MK-2206 (Fig. S1 I).

Because EGFR is an RTK, we explored whether ligands that activate other RTKs and AKT might also increase RhoA-GTP. Analogous to EGF, insulin (Fig. S2 A) and IGF-1 (Fig. S2 B) activated AKT in all lines tested, but they increased RhoA-GTP only in the DLC1-positive lines.

We also examined LPA, a ligand frequently used to increase RhoA-GTP, to see whether DLC1 was involved in that process. However, LPA, which increases RhoA-GTP through G protein–coupled receptors, rather than through RTKs (Yu and Brown, 2015), did not activate AKT and induced RhoA-GTP in a DLC1-independent manner in all lines examined (Fig. S2, C–E). Thus, LPA and RTK ligands induce RhoA-GTP by distinct mechanisms.

LPA treatment also enabled us to verify that the observed inability of AKT to increase RhoA-GTP in cells that were deficient for DLC1 expression (Figs 1, D and E; and Fig. S1, C and D) was not because they were incapable of further increasing their RhoA-GTP. Indeed, LPA increased RhoA-GTP in cells that were DLC1-negative (Fig. S2 C, right) and in cells whose DLC1 had been subjected to knockdown by siRNAs (Fig. S2 E).

The aforementioned results suggest that AKT might bind DLC1. To determine whether the two proteins form a complex in cells, we performed coimmunoprecipitation (coIP) experiments in cells expressing both proteins: two nontransformed lines (H2071 and H1634) and three cancer lines (BT549, H1703, and H157). DLC1 and AKT formed a complex in each line, whether cell lysates were immunoprecipitated (IP), first with a DLC1 antibody, and then immunoblotted (IB) for AKT (Fig. 2, A and B; and Fig. S3, A, C, and D) or by the reciprocal coIP (Fig. 2, C and D; and Fig S3 B), implying the interaction is physiologically relevant.

Some AKT associated with DLC1 was inferred to be enzymatically active because positive coIP results were obtained from all three lines when the lysates were IP with a phospho-specific antibody against pAKT-S473, followed by IB for DLC1 (Fig. 2, E and F; and Fig. S3 E). However, MK-2206 did not reduce the amount of the AKT/DLC1 complex (Fig. S3, F–J), and AKT activity stimulation by EGF did not increase the AKT/DLC1 complex (Fig. S3, K–N), implying complex formation does not depend on the enzymatic activity of AKT. AKT also formed a complex with DLC2 and DLC3, but less efficiently than it did with DLC1 (Fig. 2, G and H). Colocalization of AKT and DLC1 was confirmed by confocal microscopy in the two NSCLC lines, with colocalization coefficients of ∼0.60 (Fig. 2, I and J), and by proximity ligation assay (PLA; Fig. 2, K and L).

DLC1 has four recognized regions: an N-terminal SAM domain, a linker region, a RhoGAP domain, and a C-terminal START domain (Fig. 3 A; Tripathi et al., 2014). To map the regions of DLC1 required for AKT binding, lysates from human embryonic kidney (HEK) 293T cells expressing various GFP-tagged DLC1 fragments, which were numbered by their respective N-terminal– and C-terminal–encoded amino acids (Fig. 3 A), were IP with AKT antibody, followed by IB with a GFP antibody. Interestingly, AKT bound to some nonoverlapping N-terminal and C-terminal DLC1 fragments, implying that AKT interacts with more than one region of DLC1 (Fig. 3, B and C). To more precisely map the DLC1 regions required for AKT binding, smaller N-terminal and C-terminal DLC1 fragments were used. N-terminal AKT binding required at least some DLC1 amino acids 300–400 because DLC1 (80–400) was positive, whereas DLC1 (80–300) was negative (Fig. 3, D and E). The C-terminal DLC1 fragment that bound AKT was mapped to the RhoGAP domain because AKT interacted with DLC1 (609–850) but not with DLC1 (850–1091; Fig. 3 F).

Although a previous study (Ko et al., 2010) concluded only S567 in DLC1 was phosphorylated by AKT, we evaluated whether S298 and/or S329 (Fig. 3, A and I) might be phosphorylated under our growth conditions using ³²P labeling. First, we confirmed that most of the serine phosphorylation of DLC1 was dependent on AKT by treating the cells with MK-2206, which greatly reduced both AKT activity and serine phosphorylation of DLC1 (Fig. 3 G). Using partially purified, full-length DLC1, we observed DLC1 was strongly phosphorylated by AKT kinase in vitro, unlike the GFP control (Fig. 3 H, left, lanes 1 and 2). Thus, DLC1 is a direct substrate for AKT. DLC2 and DLC3 were also phosphorylated in vitro but less strongly than DLC1 (Fig. 3 H, left; and Fig. S4 A; see Fig. 3 I for consensus motifs).

To determine whether S298, S329, and S567 are the major AKT phosphorylation sites in DLC1, all three were mutated to alanine (DLC1-3A), resulting in a drastically reduced in vitro phospho signal (Fig. 3 H, middle, lane 4). When serine-to-alanine double mutants were analyzed, the phospho signal was reduced but was greater than the DLC1-3A mutant; single serine-to-alanine mutants retained a strong phospho signal (Fig. 3 H, middle and right; S329A mutant not depicted). Liquid chromatography–mass spectrometry confirmed those results, suggesting that all three serines were phosphorylated under these growth conditions because the analysis of the partially purified DLC1 from cells transfected with DLC1-WT or the DLC1-3A mutant detected phosphorylation in S298, S329, and S567 in the relevant DLC1-WT peptides, but not in the DLC1-3A mutant (Fig. 3 J and Fig. S4 B).

Our observation that AKT could increase RhoA-GTP in a DLC1-dependent manner suggested that AKT phosphorylation of the three DLC1 serines would attenuate its RhoGAP and tumor suppressor functions. To evaluate those possibilities, we compared the activities of DLC1-WT with the nonphosphorylatable, triple alanine mutant (DLC1-3A). We also constructed and analyzed the phosphomimetic, triple serine-to-aspartate mutant (DLC1-3D) and included a “GAP-dead” mutant (DLC1-R718A) as a control.

In H1703 cells that stably expressed similar levels of the constructs, the RhoA-GTP level in the DLC1-3D transfectant was similar to the GAP-dead DLC1 (DLC1-R718A; Fig. 4, A and B). However, the RhoA-GTP in the DLC1-3A mutant was at least as low as that induced by DLC1-WT. Analogous results were seen when the DLC1 mutants were analyzed for Rho kinase (ROCK) activity (Fig. 4 C), in vitro RhoGAP activity (Fig. 4 D), and phosphorylation of myosin regulatory light chain (pMRLC; Fig. 5 A), a major downstream effector of the RhoA-ROCK pathway. By immunofluorescence, cells transfected with GFP, DLC1-3D, or DLC1-R718A showed similar strong staining of pMRLC (Fig. 4 E), with well-formed stress fibers (Fig. 4 F). In contrast, cells transfected with DLC1-WT or DLC1-3A showed less pMRLC staining, fewer stress fibers, and few, if any, concave boundaries, consistent with reduced RhoA-ROCK signaling and reduced cell contraction.

When less-drastic individual or combined mutants were analyzed in DLC1-negative A549 cells, RhoA-GTP was inversely related to the number of phosphomimetic serine-to-aspartate (S-to-D) mutations in DLC1 (Fig. 5, B and C). These results suggest the effects of each serine phosphorylation on RhoA-GTP are additive.

We examined MK-2206 treatment of stable DLC1 transfectants in DLC1-negative H358 cells. The findings verified that MK-2206 did not affect RhoA-GTP in DLC1-3A or DLC1-3D mutants, although, consistent with earlier results (Fig. 1 H), it did decrease RhoA-GTP for DLC1-WT (Fig. 5, D and E).

We had previously found that CDK5 phosphorylation of four other serines in the linker region (S120, S205, S422, and S509) increased the RhoGAP activity of DLC1 (Tripathi et al., 2014), which is the opposite of the effects of AKT phosphorylation of DLC1. To determine whether AKT phosphorylation of DLC1 was phenotypically dominant over CDK5 phosphorylation of DLC1, we studied A549 cells, in which both endogenous CDK5 and endogenous AKT are active and stably transfected with mutants of all four “CDK5” serines in DLC1 (DLC1-4A-CDK5 and DLC1-4D-CDK5) and those transfected with the AKT serine mutants DLC1-3A and DLC1-3D mutants described here (for clarity, they are designated DLC1-3A-AKT and DLC1-3D-AKT in this paragraph), as well as with WT DLC1. The effects of AKT inhibition or CDK5 inhibition on the phosphorylation of the AKT-serines and the CDK5-serines and on RhoA-GTP were evaluated. Consistent with AKT activity being phenotypically dominant over CKD5 activity, CKD5 inhibition with roscovitine did not increase RhoA-GTP for the DLC1-3A-AKT mutant, but AKT inhibition with MK-2206 did reduce RhoA-GTP for the DLC1-4A-CDK5 mutant (Fig. S5, A and B).

To assess the biological significance of DLC1 phosphorylation by AKT, we evaluated the DLC1-3A and DLC1-3D mutants in several bioassays. Stable DLC1-WT transfectants in the H358 line reduced the following functional bioassays: anchorage-independent cell growth (Fig. 6, A and B), RhoA-GTP level (Fig. 6, C and D), transwell cell migration (Fig. 6, E and F), and xenograft tumors in immunodeficient mice (Fig. 6, G and H). The DLC1-3D mutant was as deficient as the GAP-dead DLC1-R718A mutant in those bioassays, whereas the DLC1-3A mutant was even more active than DLC1-WT.

Given the location of the three AKT serine phosphorylation sites in the linker region (Fig. 3 A), we speculated their phosphorylation might attenuate the RhoGAP activity of DLC1 by increasing the binding of the linker region to the RhoGAP domain, placing the protein in a closed conformation. Such a mechanism would be the opposite of the one by which CDK5 activates DLC1 (Tripathi et al., 2014).

To test that hypothesis, we generated three isogenic versions of a GST-tagged DLC1 linker region encoding amino acids 80–600, which contains the three serines phosphorylatable by AKT: a WT fragment [GST-DLC1 (80–600)–WT], its nonphosphorylatable 3A mutant [GST-DLC1 (80–600)–3A], and its phosphomimetic 3D mutant [GST-DLC1 (80–600)–3D]. A549 cells, which have high AKT activity and are DLC1 negative, were cotransfected with a fragment encoding the GFP-tagged RhoGAP domain [GFP-DLC1 (609–878)] and each of the linker region constructs. Consistent with our hypothesis, GST pull-downs indicated the RhoGAP domain efficiently formed a complex with GST-DLC1 (80–600)–WT and with the GST-DLC1 (80–600)–3D mutant, whereas the GST-DLC1 (80–600)–3A mutant had weaker binding (Fig. 7 A). Binding of the constructs in A549 cells was also analyzed by immobilization-free, microscale thermophoresis (MST), which provides quantitative in vitro measurements of protein interactions in close-to-native conditions. Similar to the GST pull-down assays, binding between the RhoGAP domain and DLC1 (80–600)–WT was similar to that of the phosphomimetic DLC1 (80–600)–3D mutant, whereas the nonphosphorylatable DLC1 (80–600)–3A mutant did not bind the RhoGAP domain even at very high concentrations (Fig. 7 B).

The linker region of DLC1 also binds tensin (Liao et al., 2007; Qian et al., 2007) and talin (Li et al., 2011), two ligands implicated in integrin signaling, whose binding contributes to the tumor-suppressor function of DLC1. We speculated the closed conformation induced by AKT phosphorylation might also reduce DLC1 binding of tensin and talin. Indeed, they interacted less efficiently with the phosphomimetic full-length DLC1-3D mutant than they did with DLC1-WT (Fig. 7, C and D) in HEK 293T cells, which have lower AKT activity than A549 cells. The decreased binding was attributable to the closed conformation, rather than to a putative, intrinsic decrease in binding to the linker region because the N-terminal 3D fragment [GST-DLC1 (80–600)–3D] bound tensin and talin as efficiently as the N-terminal WT fragment [GST-DLC1 (80–600)–WT] or the N-terminal 3A fragment [GST-DLC1 (80–600)–3A; Fig. S5, C and D].

The results also suggested the closed conformation of DLC1 would reduce the efficiency of the interaction between the RhoGAP domain and RhoA-GTP. Consistent with that hypothesis, the DLC1-3D mutant, which places the protein in the closed conformation, bound RhoA-GTPγS in vitro almost as weakly as the GAP-dead DLC1-R677A mutant, which is known to have reduced RhoA-GTPγS binding (Fig. 7, E and F; Jaiswal et al., 2014). In contrast, the binding efficiency of RhoA-GTPγS to the DLC1-3A mutant, which is in the open conformation, was at least as strong as that of DLC1-WT.

Because active DLC1 is a dimer (Ko et al., 2013), we asked whether DLC1 might be a monomer when in the closed conformation induced by AKT. To evaluate that possibility, we cotransfected GST-tagged, full-length DLC1 together with GFP-tagged, full-length DLC1 into A549 cells, performed GST pull-downs, and IB the pulled-down proteins for GFP (Fig. 7, G and H). Consistent with the Ko et al. (2013) findings, there was a strong GFP band in cells cotransfected with the DLC1-3A mutant, which has the open conformation. However, the GFP band was much weaker with the DLC1-3D mutant, which is closed, implying most of that mutant protein is monomeric. The intensity of the GFP band with DLC1-WT, which is mainly closed, was reduced compared with DLC1-3A, consistent with the closed conformation being a monomer. When the cells were treated with MK-2206, the intensity of the GFP band from the DLC1-WT transfectants increased, consistent with them becoming active dimers. We conclude that the closed conformation induced by AKT arises via an intramolecular interaction because that form is mainly monomeric.

Part of the linker region has been referred to as the FA targeting domain (Liao et al., 2007) because it participates in DLC1 localization to FAs. We used H1703 cells to test the hypothesis that FA localization would be affected by the closed conformation induced by AKT (Fig. 8). Consistent with that prediction, the phosphomimetic DLC1-3D mutant colocalized poorly with the FA protein vinculin (Fig. 8 D; colocalization coefficient = 0.12), whereas the nonphosphorylatable DLC1-3A mutant was strongly colocalized (Fig. 8 C; colocalization coefficient = 0.78). Colocalization was intermediate for the DLC1-WT (Fig. 8 B; colocalization coefficient = 0.55), suggesting that some DLC1-WT is phosphorylated by AKT in H1703, which implies that MK-2206 treatment could further increase its FA colocalization. Consistent with that hypothesis, MK-2206 treatment increased FA colocalization of the DLC1-WT (Fig. 8 F; colocalization coefficient = 0.79) but did not affect colocalization of GFP or the DLC1-3A or DLC1-3D mutants (Fig. 8, E, G, and H).

One potential implication of these findings is that inhibition of AKT, a candidate approach for cancer treatment (Nitulescu et al., 2016), could reverse the attenuation of DLC1 activity that depends on AKT. Although AKT has many targets, if DLC1 is a critical one, AKT inhibition might have more potent activity against tumors that express WT DLC1 compared with those that do not. To test that possibility, we evaluated MK-2206 in the MMTV-PyMT transgenic breast cancer model, in which the MMTV promoter drives the polyomavirus middle T antigen, which stimulates AKT (Meili et al., 1998; Summers et al., 1998). First, we confirmed that, compared with mammary epithelial tissue from pregnant mice, MMTV-PyMT tumors have greater AKT activity (Fig. 9 A and Fig. S5 E). Consistent with the cell culture findings that linked AKT, DLC1, and RhoA-GTP, AKT activity in the tumors was associated with higher levels of RhoA-GTP, greater downstream effector phospho-MRLC, and greater phosphorylation of DLC1 serine residues (Fig. 9 A and Fig. S5, E and F). We then treated randomized tumor-bearing mice with oral MK-2206 for 5 d, which induced a threefold reduction in mean tumor weight (Fig. 9 B) and was correlated with decreased AKT activity and reduced RhoA-GTP, phospho-MRLC, and DLC1 serine phosphorylation by AKT (Fig. 9 A and Fig. S6, A and B).

We speculated the DLC1 reactivation from MK-2206 treatment had contributed to the observed antitumor activity. To evaluate that hypothesis more directly, we used two isogenic tumor cell lines, from H358, which differed only in their DLC1 expression (GFP versus GFP-DLC1-WT), and evaluated the ability of MK-2206 to affect their growth in agar and tumor growth in nonobese diabetic–severe combined immunodeficiency (NOD-SCID) mice. In both bioassays, MK-2206 potently reduced the growth of GFP-DLC1-WT cells (Fig. 9 C, for mouse tumors; and Fig. S6, C and D, for agar growth), whereas its effects were less pronounced in GFP control cells. These results were correlated with MK-2206 reducing RhoA-GTP and its RhoA-dependent signaling (pMRLC level) in the GFP-DLC1-WT cells but not in the GFP control cells, although the AKT activity was similarly reduced in both isogenic lines (Fig. 9 D). A separate experiment, which included H358 cells stably expressing DLC1-3A and DLC1-3D in addition to those expressing DLC1-WT, confirmed the specificity of the Fig. 9 C results. MK-2206 treatment for 5 d decreased the size of DLC1-WT tumors by 69%, compared with decreases of 18–20% for tumors expressing DLC1-3A, DLC1-3D, or vector control (Fig. S6, E and F).

This study identified increased RhoA-GTP by RTK ligands (EGF, IGF-1, and insulin) as a previously unrecognized, but common, physiological determinant of RhoA activation. We found the kinase activity of AKT, which is stimulated by the cognate RTKs these ligands activate, is required for the increased RhoA-GTP. DLC1, a tumor-suppressor gene that encodes a RhoGAP, is the key downstream target for the AKT kinase in this phenomenon, which also occurs in tumor cell lines that express DLC1 but not in those with down-regulated DLC1.

In DLC1-positive cells, AKT phosphorylated three serines in DLC1 (S298, S329, and S567, the AKT serines), which greatly attenuated its RhoGAP and tumor-suppressor activities in several bioassays. Analysis of single, double, and triple serine-to-alanine (nonphosphorylatable) and serine-to-aspartate (phosphomimetic) mutants of the three AKT serines indicated that each phosphorylation contributed to the reduction in those activities, suggesting an electrostatic mechanism. In cells expressing WT DLC1, AKT inhibition by MK-2206 greatly reduced serine phosphorylation of DLC1 and RhoA-GTP, specific changes that were not seen in cells expressing either of the triple mutants of the AKT serines treated with MK-2206. In contrast to the RTK ligands, increased RhoA-GTP induced by LPA, which is mediated by G protein–coupled receptors, occurs by a distinct mechanism that is not associated with AKT activation or DLC1.

A notable feature of the AKT serines in DLC1 is that they are located in the linker region of DLC1, rather than in the RhoGAP domain, which is necessary and sufficient for the RhoGAP activity of DLC1 (Healy et al., 2008). To explore the mechanism by which those phosphorylations outside the RhoGAP domain can reduce the RhoGAP activity, we studied complex formation between the RhoGAP domain and WT or serine mutants of the linker region after their cotransfection in cells and in vitro by immobilization-free MST to compare relative binding efficiencies of those polypeptides. The data led us to conclude the phosphorylations of the AKT serines induce strong binding between the linker region and the RhoGAP domain, which places DLC1 in a closed, inactive conformation that is mainly monomeric (Fig. 9 E) and reduces RhoA-GTP binding to DLC1, tensin and talin binding to DLC1, and DLC1 colocalization to FAs. That mechanism is essentially the opposite of what we found from phosphorylation of four serines in the DLC1 linker region by CDK5 (Tripathi et al., 2014), which placed DLC1 in an open, active conformation by decreasing an autoinhibitory interaction between the linker region and the RhoGAP domain.

Our results differ from, and extend in important ways, previous studies that DLC1 is an AKT substrate. Hers et al. (2006) found that insulin induces phosphorylation in the rat version of DLC1 at S322 (which corresponds to S329 in human DLC1) and showed that phosphorylation was mediated by AKT, but they did not examine the possible downstream consequences. Ko et al. (2010) made the important observation that AKT reduces the tumor-suppressor activity of DLC1 and concluded that phosphorylation of S567 by AKT was responsible for that reduction because they did not detect phosphorylation of S298 or S329. Under our growth conditions, however, all three AKT serines are phosphorylated in cells, as documented with the DLC1 mutants and mass spectrometry results. Consistent with these findings, our analysis of phosphoproteomic data in the National Cancer Institute’s Clinical Proteomic Tumor Analysis Consortium (CPTAC) database indicates a significant correlation in human breast cancer (Mertins et al., 2016) between AKT activity, as measured by phosphorylation of AKT T308 and S473, and phosphorylation of DLC1 S298 and S329 (P = 0.03; the CPTAC screen did not identify the peptide with the S567 residue).

Ko et al. (2010) concluded the reduced tumor-suppressor activity induced by AKT was not associated with a decrease in the RhoGAP activity of DLC1 because they did not detect an increase in RhoA-GTP. However, we developed strong evidence that AKT-dependent phosphorylation of DLC1 is associated with an increase in RhoA-GTP in nontransformed lines, tumor-derived lines, and the MMTV-PyMT transgenic breast cancer model. One possible way to reconcile the RhoA-GTP conclusions of Ko et al. (2010) with ours is that they focused on the RhoA-GTP levels of the single DLC1-S567D mutant; in agreement with their results, we found the RhoGAP activity in that mutant was close to that of WT DLC1. In our experiments, the serine-to-aspartate mutation of at least two of the three AKT serines was needed to detect a decrease in RhoGAP activity.

EGF treatment of some cell lines has been reported to have somewhat different effects on DLC1. Cao et al. (2012) reported that EGF can activate DLC1 in MCF10A breast epithelial cells via a mechanism that includes replacement of tensin-3 binding to DLC1 with CTEN (C-terminal tensin-like) binding, a process that takes much longer than the one described here. MCF10A is the only nontransformed line we have examined in which endogenous DLC1 is inactive (Tripathi et al., 2014), so it may not be surprising that it behaves differently. Ravi et al. (2015) examined the effects of EGF on a DLC1-negative, HeLa-derived cancer line. In agreement with our data, EGF treatment of the parental line had no effect on RhoA-GTP. However, they found that, although transient transfection with WT DLC1 RhoGAP did not alter RhoA-GTP, EGF treatment of those transfectants resulted in reduced RhoA-GTP, in contrast to the increased RhoA-GTP we observed for endogenous DLC1 and the stable transfectants in several cell lines we tested. The two-step mechanism that they concluded was responsible for their observation is unrelated to the one described here.

Our observations may have therapeutic implications in cancer. The AKT-induced attenuation of the tumor-suppressor function of DLC1 is potentially reversible because AKT inhibition can decrease phosphorylation of the three AKT serines and reactivate DLC1. In the transgenic MMTV-PyMT cancer model, which we determined has constitutive activation of AKT and high RhoA-GTP, pharmacologic AKT inhibition by MK-2206 decreased phosphorylation of the AKT serines in DLC1, decreased RhoA-GTP, and reduced tumor size. We extended those results by testing an isogenic version of a tumor line with constitutively active AKT that either expressed DLC1 or did not express DLC1. Similar to the MMTV-PyMT, MK-2206 reduced RhoA-GTP and the size of xenograft tumors from the DLC1-positive subline. However, in the DLC1-negative subline, AKT inhibition by MK-2206 did not change RhoA-GTP and had only marginal antitumor activity. The results strongly imply the antitumor activity observed in the DLC1-positive line was attributable, at least in part, to reactivation of DLC1 by MK-2206.

These findings highlight the potential utility of monitoring the reactivation of tumor-suppressor genes when inhibiting an oncoprotein, such as AKT, which has a clinical potential that is still being explored (Nitulescu et al., 2016). Such reactivation may also be relevant for inhibition of other oncoproteins that inactivate tumor-suppressor genes indirectly or directly, as occurs with CDK4/6 and pRB via phosphorylation (Hamilton and Infante, 2016) or MDM2 and p53 via the ubiquitin system (Wasylishen and Lozano, 2016).

GFP-tagged DLC1 WT (GFP-DLC1-WT); DLC1 GAP-dead mutant (GFP-DLC1-R718A); GFP-tagged DLC1 fragments encoding DLC1 residues 1–492, 500–1091, 1–110, 80–200, 80–300, 80–400, 400–500, 609–850, 850–1091, and 609–878; and GFP-DLC3 were constructed by PCR and subcloned into a modified pEGFP-C1 vector (Takara Bio Inc.) through Kpn1-NotI sites, as described (Qian et al., 2007). GST-tagged tensin and talin fragments were described previously (Qian et al., 2007; Li et al., 2011). GFP-DLC2 was a gift from M. Mowat (Manitoba Institute of Cell Biology, Winnipeg, Manitoba, Canada). A series of individual and combined serine-to-Alanine (S-to-A) and serine-to-aspartate (S-to-D) mutations were introduced into full-length DLC1-WT and into DLC1 fragments encoding residues 80–600, using a site-directed mutagenesis kit (Agilent Technologies). Table S1 lists the primers used. Full-length DLC1 and DLC1 (80–600) linker region fragments, with or without 3A or 3D mutations, were engineered into the PEBG vector with BamHI and NotI, resulting in GST-tagged DLC1 constructs. All PCR regions were confirmed by sequencing.

The following antibodies were purchased from Cell Signaling Technology: AKT mouse (2920), AKT rabbit (4691), phospho-AKT-pS473 rabbit (4060) and phospho-AKT-pT308 rabbit (13038), phospho-Akt substrate (RXXS*/T*) rabbit (9614), phosphorylated Thr18/Ser19 pMRLC rabbit (3674), and GAPDH (2118) rabbit. Two DLC1 antibodies, which gave similar results, were used: one, generated in our laboratory (DLC1 antibody; clone 428; 24), and the other, DLC1 mouse mAb (612021), purchased from BD. pEGFR-Y845 rabbit (ab5636), MRLC mouse (ab11082), GFP mouse (ab1218), and GFP rabbit (ab290) antibodies were purchased from Abcam. RhoA mouse (ARH04) and pMRLC goat (sc-12896) antibodies were obtained from Cytoskeleton, Inc., and Santa Cruz Biotechnology, Inc., respectively. The phospho-serine mouse (612547) antibody was purchased from BD. Anti–rabbit and anti–mouse IgG HRP-linked secondary antibodies were obtained from GE Healthcare. Alexa Fluor 568 anti–rabbit IgG, Alexa Fluor 488 anti–mouse IgG, Alexa Fluor 488 phalloidin, and DAPI were purchased from Thermo Fisher Scientific.

HEK 293T, human skin epithelial H2071, human lens epithelial FHL124, and human fibroblastic H1634 cells were cultured in DMEM supplemented with 10% FBS. Human breast cancer lines (BT549, MCF10A1Ca1h, MDA-MB-468, and T47D) and NSCLC lines (H1703, H157, A549, and H358, provided by C. Harris, National Cancer Institute, Bethesda, MD) were cultured in RPMI-1640 supplemented with 10% FBS. Transient transfections were performed with Lipofectamine 3000 (Thermo Fisher Scientific) and cultured for 48 h. Stable clones expressing GFP or DLC1 mutants were made by transfecting H1703 or H358 cells with Lipofectamine 3000, followed by G418 selection (0.9 µg/ml).

To suppress DLC1 expression, cells were transfected with 160 nM of DLC1 siRNAs or with scrambled control siRNAs, and harvested 48 h later. Suppression of protein expression, with two different siRNAs, was confirmed by immunoblotting. Validated siRNAs for human DLC1 (Hs_DLC1 siRNA_5, SI03219909, and Hs_DLC1 siRNA_11, SI04952213) were from QIAGEN, as were negative control siRNAs (control siRNA 1, 1027280; and control siRNA 2, 1027310). The sequence for each DLC1 siRNA was as follows: Hs_DLC1 siRNA _5 sense sequence: 5′-CGAUGUCGUAAUUCCUAUATT-3′; Hs_DLC1_5 antisense sequence: 3′-CGGCUACAGCAUUAAGGAUAU-5′; Hs_DLC1_11 sense sequence: 5′-GGAGUGUAGGAAUUGACUATT-3′; Hs_DLC1_11 antisense sequence: 3′-gaCCUCACAUCCUUAACUGAU-5′.

The final concentration of ligands was as follow: EGF (100 ng/ml), insulin (100 nM), IGF-1 (100 ng/ml), and LPA (1.0 µM). All four ligands were procured from Sigma-Aldrich. AKT inhibitor MK-2206 (used at 10 µM) was from Selleck Chemicals. After overnight incubation in serum-free medium, cells were treated with the indicated ligands or inhibitors for 15–120 min.

Lysates from transfected cells were IP with GFP antibody, and immunopellets were sequentially washed once with high-salt HNTG buffer (20 mM Hepes, 500 mM NaCl, 0.1% Triton X-100, and 10% glycerol), twice with low-salt HNTG buffer (20 mM Hepes, 150 mM NaCl, 0.1% Triton X-100, and 10% glycerol), and once with kinase reaction buffer (35 mM Hepes, pH 7.4, 10 mM MgCl₂, 1 mM EGTA, 1% Tween 20, 0.1 mM sodium vanadate, and 1 mM DTT). The kinase reaction was performed in 30 µl of reaction buffer containing 15 µM cold ATP, 2.5 µCi γ-[³²P]ATP, and 100 ng of recombinant active AKT (EMD Millipore) at 30°C for 45 min. The reaction was stopped by adding 10 µl of 4× Laemmli sample buffer and heating at 95°C for 5 min. Proteins were separated by gel electrophoresis and autoradiographed to detect ³²P incorporation.

A Rho activation assay kit (EMD Millipore) was used to measure GTP-bound RhoA, as described previously (Tripathi et al., 2014). In brief, equal amounts (1,000 µg) of each cell lysate were incubated with 30 µg GST-Rhotekin Rho-binding domain coupled to glutathione-agarose beads for 45 min. Beads were washed three times with washing buffer, samples were subjected to 4–12% SDS-PAGE, transferred onto nitrocellulose membranes (Thermo Fisher Scientific), and detected by IB, using RhoA antibody (ARH04 from Cytoskeleton, Inc., and 05–778 from EMD Millipore; clone 55).

Cells were fixed and harvested in 10% TCA containing 10 mM DDT. Pellets were dissolved in 10 µl of 1 M Tris base and mixed with 100 µl of extraction buffer (8 M urea, 2% SDS, 5% sucrose, and 5% 2-mercaptoethanol). Equal amounts of protein from each cell extract were subjected to 10% SDS-PAGE, transferred onto nitrocellulose membranes, and incubated with an antibody specific for phospho-myosin binding subunit (phospho-Thr853-MYPT1) or myosin binding subunit (MYPT1), and bands were visualized by enhanced chemiluminescence. ROCK activity was expressed as the ratio of phospho-MYPT1 to total MYPT1.

GFP-tagged DLC1 constructs were purified by IP using GFP rabbit polyclonal (ab290) antibody from transfected cells using a high-stringency buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 5 mM MgCl₂, NP-40 [0.5%], 1 mM DTT, and protease and phosphatase inhibitor). Highly purified Rho was bound to γ-labeled [P³²]GTP. The GTPase-accelerating activity (RhoGAP activity) of various DLC1 mutants at each time point was determined by incubating with GTP-bound RhoA at 18°C with shaking and removing samples at the indicated time points. The guanidine nucleotides were separated by chromatography on cellulose filter paper dissolved in buffer. The γ-P³² signal was then determined, and the net GTP hydrolysis was calculated.

Cells were transiently cotransfected with plasmids expressing GST or the indicated GST fusion constructs together with GFP or the indicated GFP-DLC1 constructs. 48 h after transfection, cells were lysed with golden lysis buffer. The cleared supernatants were collected, and a small portion of supernatants was taken to determine the protein concentration using the DC protein assay (Bio-Rad Laboratories). For the pull-down assay, 1.0 mg of total protein from each cell extract was used, to which 30 µl of glutathione sepharose-4B slurry (GE Healthcare) was added, with continuous rotation for 3 h at 4°C. The pellets were sequentially washed once with golden lysis buffer, once with high-salt HNTG buffer, and twice with low-salt HNTG buffer. The beads were incubated with 30 µl Laemmli sample buffer, subjected to 10% SDS-PAGE, transferred onto nitrocellulose membranes (Thermo Fisher Scientific), and detected by IB using specific antibodies. A portion of the cell extracts was used as a loading control to verify expression of the GFP fusion proteins and the GPF control. For coIP experiments, equal amounts of protein from each cell lysate were precleared with protein G slurry (Thermo Fisher Scientific) and incubated with the indicated antibodies or control IgG for 1 h at RT. After incubation, 30 µl of protein G slurry was added to each immune reaction and rotated at 4°C overnight. The immunopellets were washed three times with IP buffer. Coimmunoprecipitated proteins were eluted by boiling for 5 min in 30 µl Laemmli sample buffer containing 5% (vol/vol) 2-mercaptoethanol. Eluted proteins were resolved on a NuPage 4–12% BisTris gel and detected by immunoblotting using specific antibodies. Immunoreactive bands were detected by enhanced chemiluminescence (ECL-Plus; GE Healthcare) using HRP-linked anti–rabbit or anti–mouse secondary antibodies (1:5,000 dilutions).

Binding efficiency of DLC1 N terminus to its RhoGAP domain was studied by MST (Khavrutskii et al., 2013; Seidel et al., 2013) using a Monolith NT.115 instrument (NanoTemper). Serial 1:1 dilutions of the DLC1 N-terminal fragment (80–600) WT or its combined 3A or 3D mutant versions were mixed with each aliquot and a fixed volume of fluorescent GFP-tagged RhoGAP domain [GFP-DLC1 (609–878)], the resulting probes were placed in glass capillaries (NanoTemper), and assayed with “blue” light-emitting diode excitation (excitation, 460–480 nm, and emission, 515–530 nm), observing the fluorescence signal upon start of illumination with a pin-point infrared laser (1,480 nm). The MST plots for each capillary set were analyzed by NT.Analysis 1.5.41 software (NanoTemper), providing fluorescence intensity versus protein concentration plots for each binding pair. The inflection point of those plots corresponds to the dissociation constant of each interaction.

Transiently or stably transfected cells were seeded onto glass chambers, incubated for 24 h, and fixed with 4% PFA for 20 min. Fixed cells were permeabilized with 0.25% Triton X-100 in PBS and then blocked with 3% BSA in PBS for 2 h. The cells were incubated with a 1:200 dilution of the indicated primary antibodies at 4°C overnight. After being thoroughly washed in PBS, the cells were incubated with the appropriate 1:250 Alexa Fluor–conjugated secondary antibodies for 1 h. To visualize actin or nuclei, cells were incubated with phalloidin (1:50) or DAPI (1:2,500) for 1 h. After staining, the cells were thoroughly washed with PBS and mounted with gel-mounting solution (BIOMEDA).

Confocal microscopy of fluorescent-labeled cells was performed with a microscope (LSM 780; ZEISS) with an excitation wavelength of 488 nm to detect transfected GFP fusion proteins. Alexa Fluor probes were viewed with excitation wavelengths of 488 (Alexa Fluor 488) and 568 (Alexa Fluor 568) nm. Images were made at RT using photomultiplier tubes with a Plan-Apochromat 63×/1.4 NA oil-differential, interference-contrast, objective lens with a 2× magnifier to produce a 125× magnification. The colocalization of two proteins was analyzed by confocal software (ZEN 2012; ZEISS). For quantification of representative morphology in each group, ∼15 cells per condition, randomly selected from several fields, were analyzed. The Mander’s overlapping colocalization coefficient means ± SD were calculated and are shown below each figure panel. The overlapping colocalization coefficients can range from 0 to 1, where 0 means no colocalization and 1 means full colocalization of the two proteins. The images were minimally processed for levels/contrast adjustment in DAPI panels, and the adjustment was performed on entire images using Photoshop CC software (Adobe). The adjustments do not enhance, erase, or misrepresent any information present in the original images.

PLA was used to visualize proximity colocalization (<40 nm) of DLC1 and AKT in NSCLC lines using the Duolink Detection kit (Olink Proteomics). The cells were fixed with 4% PFA for 20 min at RT and then incubated with 0.25% Triton-X-100 for 5 min. After blocking with 3% BSA, cells were incubated overnight at 4°C with rabbit anti–DLC1 (1:100) plus mouse anti–AKT (1:500) antibodies. After washing, cells were incubated with secondary antibodies with PLA probes (MINUS probe–conjugated anti–rabbit IgGplus PLUS probe–conjugated anti–mouse IgG (Olink Proteomics). Circularization and ligation of the oligonucleotides in the probes were followed by an amplification step. A complementary fluorescent-labeled probe was used to detect the product of the rolling-circle amplification. Slides were mounted with Duolink II mounting medium containing DAPI. Images were obtained with an LSM 780 (ZEISS) using ZEN software (ZEISS). The colocalization PLA dots in cells were counted using ImageJ software (National Institutes of Health). Quantifications are given as means ± SD.

Lysates from HEK 293T cells expressing GFP-tagged DLC1 constructs, either untreated or treated with the MK-2206 AKT inhibitor, were IP with GFP antibody. The immunopellets were resolved on a NuPage 4–12% BisTris gel. The Coomassie-stained DLC1 gel band was destained, and proteins were reduced, alkylated, and digested with trypsin or LysC as described (Shevchenko et al., 2006). Digested peptides were further desalted using StageTip C18 columns (Rappsilber et al., 2007) and were analyzed on a Q-Exactive instrument (Thermo Fisher Scientific) equipped with a Proxeon EASY-nLC 1,000 UHPLC System. Liquid chromatography–tandem mass spectrometry data were searched against a human Refseq database using MaxQuant v1.3.0.5 (Cox and Mann, 2008; Cox et al., 2011) with carbamidomethylation as a fixed modification and the following variable modifications: oxidation of methionine, acetylation of protein N termini, deamidation of asparagine, and phosphorylation of serine, threonine, and tyrosine residues. For peptide identification, we applied a 1% false-discovery rate using a target–decoy search strategy (Elias and Gygi, 2010).

Cell migration was measured by 6.5-mm-diam Falcon cell culture inserts (8 µm pore size; BD). Transiently transfected cells or stable clones were trypsinized, resuspended in serum-free RPMI-1640 medium, and transferred to the upper chamber (7.5 × 10⁴ cells in 300 µl). 600 µl of 10% FBS in RPMI-1640 was placed in the lower chamber. After 18 h incubation, the cells remaining on the upper surface of the insert were removed five times with a cotton swab moistened in PBS. Migrated cells on the lower surface were fixed in methanol for 20 min at RT, followed by staining with 2% crystal violet (Sigma-Aldrich) in methanol for 30 min, destained, examined, and photographed by microscopy. For quantification, migrated cells were solubilized with 1% Triton X-100 and counted in a spectrophotometer at an OD of 590 nm.

For soft agar assays, a 0.6% agar (BD) base in RPMI-1640 medium was placed in 60-mm dishes for 1 h at RT. 1.0 × 10⁵ cells were mixed with complete medium containing 0.4% agar and placed over 0.6% basal agar in 60-mm dishes. Cells were grown for 3 wk, and colonies were photographed microscopically and quantified with a colony counter. For clonogenic assays, 0.6 × 10⁵ cells were seeded in six-well plates and cultured in 0.9 µg/ml G418 RPMI-1640 medium with 10% FBS for 3 wk. Colonies were fixed, stained with 4% crystal violet, and counted.

The mouse studies were approved by the National Cancer Institute Animal Care and Use Committee and were conducted in compliance with the approved protocols. For the tumor xenograft, H358 stable clones expressing GFP, GFP-DLC1-WT, GFP-DLC1-3A, GFP-DLC1-3D, and GFP-DLC1-R718A were trypsinized, washed with cold PBS, diluted to 10⁸ cells/ml with serum-free medium/Matrigel basement membrane matrix (BD) at a ratio of 3:1, and injected subcutaneously into NOD-SCID mice (1.0 × 10⁷ cells/injection). The animals were monitored for tumor growth, and tumor masses were weighed (in grams) 6 wk after injection.

For mice with xenograft tumors, H358 stable clones expressing various GFP-tagged DLC1 constructs were trypsinized, washed with cold PBS, diluted to 10⁸/ml with serum-free medium/Matrigel basement membrane matrix (BD) at a ratio of 3:1, and injected subcutaneously into NOD-SCID mice (1.0 × 10⁷ cells/injection). When tumors were 0.5–1.0 cm, mice were treated with MK-2206 or vehicle control for 5 d, and the remaining tumor tissues were then excised, weighed, and processed for biochemical assays. For MMTV-PyMT mice, tumors were randomly divided into two groups. Mice were treated orally with 50 mg/kg MK-2206 AKT inhibitor for five consecutive days or the vehicle control. The remaining tumor tissues were then excised, weighed, and processed for biochemical assays.

At least two independent experiments were performed for all in vitro analysis. Immunoblots were quantified by densitometric scanning using ImageQuant software. Results are expressed as density means ± SD from two or three experiments. All experiments were designed with matched control conditions within each experiment. Data distribution was assumed normal, but that was not tested formally. For statistical analysis, parametric two-tailed t tests were performed using PRISM software (version 7.0a; GraphPad Software), and P < 0.05 was considered statistically significant.

Fig. S1 shows that EGF-induced AKT activity, which increases RhoA-GTP through DLC1 and AKT, regulates all three DLC family members. Fig. S2 shows that insulin and ILGF-1 also induce AKT activity and increase RhoA-GTP through DLC1. However, LPA induces RhoA-GTP in a DLC1-independent manner and does not activate AKT. Fig. S3 shows DLC1 and AKT form a protein complex in human cell lines, and the complex formation does not require enzymatic activity of AKT. Fig. S4 shows all three members of the DLC gene family, DLC1, DLC2, and DLC3, are AKT substrates. The phosphorylation of the three AKT serines in DLC1 (S298, S329, and S567) was confirmed by mass spectrometry. Fig. S5 shows that the AKT phenotype was dominant over the CDK5 phenotype. PyMT-positive tumors have high AKT activity, high RhoA-GTP, and high MRLC phosphorylation. Fig. S6 shows that the MMTV-PyMT–positive tumors have high AKT activity, high RhoA-GTP, and high DLC1 phosphorylation. Treatment with MK-2206 inhibits colony formation, growth in soft agar, and tumor formation in mice of DLC1-WT transfected cells only, but does not alter anchorage-independent growth or tumor formation by cells transfected with the GFP control, DLC1-3A, DLC1-3D, or DLC1-R718A.

We thank Marian Durkin for critically reading the manuscript, the National Cancer Institute Center for Cancer Research Imaging Core Facility for confocal microscopy, Michael Mowat for the GFP-DLC2 construct, and Curt Harris for H1703 and H358 cell lines.

This research was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, and Center for Cancer Research.

The authors declare no competing financial interests.

Author contributions: B.K. Tripathi and D.R. Lowy conceived the project, designed the experiments, and analyzed the data. B.K. Tripathi performed most of the experiments with the help of T. Grant; X. Qian engineered the DLC1 plasmid constructs. M. Zhou, P. Mertins, and S.A. Carr performed mass spectrometric analyses. D. Wang performed bioinformatics analyses of proteomics data. S.G. Tarasov helped with the MST assay, and A.G. Papageorge helped with in vitro RhoA-GTP binding assay. K.W. Hunter provided the MMTV-PyMT transgenic mice. B.K. Tripathi and D.R. Lowy wrote the manuscript with input from all authors.