Gain-of-Function RHOA Mutations Promote Focal Adhesion Kinase Activation and Dependency in Diffuse Gastric Cancer

Gastric cancer, the third leading cause of cancer-related death worldwide (1), is classically divided into two histologic types, intestinal and diffuse (2). Histologically, diffuse gastric cancer (DGC) is notable for the frequent appearance of mucin-filled “signet-ring” cells, highly invasive and poorly differentiated cancer cells, lack of cellular cohesion (3), and an invasive growth pattern that contributes to rapid invasion and peritoneal metastases. Molecularly, DGCs largely fall into the genomically stable molecular group, with tumors typically lacking hypermutation and chromosomal instability (4). The absence of mutations in conventional oncoproteins and uncertainty over mechanisms of transformation in DGC have hindered therapeutic development. The most specific genomic aberration in sporadic DGC, whether through mutation (5) or methylation (6, 7), is somatic inactivation of the tumor suppressor gene CDH1, which encodes the adhesion protein E-cadherin. In hereditary DGC, CDH1 is inactivated in the germline (8, 9). More recently, genomic characterization by our group and others (3, 4, 10–12) identified missense mutations in the small GTPase RAS homolog family member A (RHOA) in 15% to 26% of DGC.

Like RAS, RHOA cycles between inactive GDP-bound and active GTP-bound conformations, the latter of which interacts with downstream effectors to regulate the actin cytoskeleton, cell migration, cytokinesis, and the cell cycle (13). Yet, RHOA missense mutations in DGC occur at residues distinct from conventional activating mutations found in RAS (Supplementary Fig. S1A). Neither the consequences of these mutations for RHOA activity nor their impacts on disease pathogenesis have been clearly established. Studies of RHOA mutations in DGC have reached conflicting conclusions. Kakiuchi and colleagues described recurrent RHOA mutations as gain-of-function; siRNA-mediated silencing of RHOA reduced proliferation in non-DGC cancer cells harboring RHOA mutations (3). In contrast, Wang and colleagues suggested that RHOA^Y42C is a loss-of-function mutant, as ectopic RHOA^Y42C attenuated GTP levels, inferred from cell-based pulldown analyses using the RHOA GTP binding domain (RBD) of Rhotekin (10).

In this study, we characterized the RHOA^Y42C mutant via extensive biochemical analyses and detailed investigation of its activity in the gastric epithelium using a genetically engineered mouse model. We demonstrate that recurrent genomic alterations found in DGC, CDH1 loss coupled with RHOA^Y42C, induces metastatic DGC in mice resembling the human disease. Using detailed biochemistry, we established that the Y42C mutation activates RHOA, impairing GTP hydrolysis and promoting RHOA interaction with ROCK, and enhancing actin rearrangements and focal adhesion formation. Furthermore, we demonstrate that Cdh1 loss and RHOA^Y42C induce DGC via activation of focal adhesion kinase (FAK), promoting activation of YAP–TAZ, PI3K–AKT, and β-catenin, thereby identifying therapeutic approaches for DGC. FAK inhibition abrogates tumor growth in our novel model and shows efficacy across a broader panel of patient-derived DGC cell lines, suggesting that FAK may serve as a potent therapeutic target for these cancers.

Given the lack of DGC cell lines harboring RHOA mutations, we chose to study RHOA mutation in the gastric lineage by establishing a murine model, LSL-RHOA^Y42C (using Y42C, the most recurrent RHOA mutation in DGC; Supplementary Fig. S1A), with RHOA^Y42C engineered into the Col1A1 locus where its expression is activated by Cre recombinase (Fig. 1A). We introduced the Mist1-CreERT2 allele to express tamoxifen-activated Cre in the Mist1 locus, a marker of gastric chief cells suggested to be expressed in isthmus stem cells (14–16). To represent the most common genomic aberration in DGC, loss of CDH1, we used a conditional Cdh1 allele, Cdh1^Flox/Flox. Finally, the R26-mTmG “Tomato-GFP” allele was introduced to mark Cre-recombined cells by conversion from Tomato (red) to GFP (green). We bred cohorts of mice where we could inducibly express RHOA^Y42C in the gastric epithelium and inducibly inactivate Cdh1, either alone or in combination.

While we aged mice following in vivo induction of Cre activity, we developed murine gastric organoids to evaluate RHOA^Y42C activity. Recombination was induced in the organoids in vitro via tamoxifen or adenoviral Cre-recombinase, and validated by conversion of Tomato to GFP expression (Fig. 1A), immunoblotting, and immunofluorescence (Supplementary Fig. S1B–S1E). Following induction, we observed dramatic morphologic changes and induction of mesenchymal markers (Fig. 1B and C; Supplementary Fig. S1D–S1F; Supplementary Video S1) in organoids expressing RHOA^Y42C in the absence of Cdh1 (Cdh1^−/−RHOA^Y42C/+: Mist1-CreERT2, Cdh1^Flox/Flox, LSL-RHOA^Y42C/+), but not with RHOA^Y42C/+ alone (RHOA^Y42C/+: Mist1-CreERT2, LSL-RHOA^Y42C/+) or Cdh1 loss alone (Cdh1^−/−: Mist1-CreERT2, Cdh1^Flox/Flox). Whereas Cdh1^−/− or RHOA^Y42C/+ organoids retained spherical forms with hollow interiors, Cdh1^−/−RHOA^Y42C/+ organoids exhibited abnormal morphology and central filling (Fig. 1B and C), a phenotype associated with transformation and consistent with morphologies of patient-derived DGC organoids (12, 17). Histologic review identified signet-ring cells, a characteristic feature of DGC, in Cdh1^−/−RHOA^Y42C/+ organoids (Fig. 1D; Supplementary Fig. S1G).

To determine whether these organoids were transformed, we transplanted them (5 × 10⁵ cells) via surgical injection into gastric walls of NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice (Fig. 1E). Mice implanted with Cdh1^−/−RHOA^Y42C/+ organoids formed tumors and exhibited peritoneal spread, ascites, and metastases to lung and liver (Fig. 1F and G). In contrast, no tumors formed with Cdh1^−/− or RHOA^Y42C/+ or Mist1Cre control organoids. Accordingly, survival analysis showed that mice implanted with Cdh1^−/−RHOA^Y42C/+ organoids succumbed rapidly, whereas survival was not decreased in other groups (Fig. 1H). We obtained analogous results following subcutaneous flank injections (Supplementary Fig. S1H and S1I).

We then evaluated autochthonous expression of RHOA^Y42C in mice aged for 14 months after in vivo tamoxifen induction. Tumors were identified in the stomachs of only Cdh1^−/−RHOA^Y42C/+ mice (5/8 mice, 62.5%; Supplementary Fig. S1J). These results recapitulate those showing that Mist1-CreERT2, Cdh1^Flox/Flox mice do not develop tumors unless infected with Helicobacter felis (16). Histologic analysis confirmed that Cdh1^−/−RHOA^Y42C/+ tumors were poorly differentiated with cells resembling signet-ring cells (Supplementary Fig. S1K). These results establish RHOA^Y42C as an oncogene that, with Cdh1 loss, induces tumors resembling human DGC.

We next characterized the consequences of the Y42C mutation for RHOA function by assessing its effect on RHOA-regulated cellular activities. Seminal studies establishing involvement of RHOA in regulating actin cytoskeletal organization, and cell adhesion and migration utilized lab-generated, constitutively activated RHOA mutants (G14V, Q63L; analogous to RAS residues G12 and Q61) expressed in NIH/3T3 mouse fibroblasts and related 3T3 cell lines (18, 19). To evaluate the activity of RHOA^Y42C relative to extensive literature evaluating RHOA variants in NIH/3T3 cells, we established NIH/3T3 cells stably expressing exogenous RHOA^Y42C protein at levels comparable with exogenous wild-type RHOA (RHOA^WT) or RHOA^Q63L (Supplementary Fig. S2A and S2B). NIH/3T3 fibroblasts are mesenchymal cells lacking expression of the epithelial cell–restricted Cdh1 gene.

RHOA promotes actin stress fibers, which are important for cell morphology and adhesion. As we described previously (20), RHOA^Q63L enhanced stress fiber formation compared with RHOA^WT (Fig. 2A and B). RHOA^Y42C also enhanced stress fiber formation, at a level intermediate between Q63L and WT, indicating that Y42C causes a gain-of-function phenotype with respect to this canonical RHOA function.

RHOA also stimulates assembly of focal adhesions (FA), protein complexes that connect the actin cytoskeleton with the extracellular matrix (21, 22). We investigated the ability of RHOA^Y42C to regulate FA assembly in NIH/3T3 cells, utilizing the Focal Adhesion Analysis Server. Both RHOA^Q63L and RHOA^Y42C but not RHOA^WT increased FA size (Fig. 2C, area). Conversely, RHOA^WT reduced the numbers of FA per cell, whereas RHOA^Y42C did not (Fig. 2D). RHOA^Q63L but not RHOA^WT enhanced FA eccentricity (deviation from a circular shape; Fig. 2E). Although the increase did not reach statistical significance, RHOA^Y42C also enhanced FA eccentricity (Fig. 2E). We conclude that like RHOA^Q63L, RHOA^Y42C exhibits a gain-of-function phenotype with respect to FA assembly.

We next evaluated the ability of RHOA^Y42C to regulate cell-matrix adhesion. As reported previously, RHOA^Q63L impaired adhesion to fibronectin (Fig. 2F). Surprisingly, RHOA^WT also impaired adhesion, whereas RHOA^Y42C did not. Thus, the effect of RHOA^Y42C on adhesion differs from both WT and Q63L. Actin stress fiber and FA organization regulate cell migration (22). As shown previously (23, 24), RHOA^Q63L impaired the velocity (Fig. 2G) and directionality of migration (Supplementary Fig. S2C). In contrast, RHOA^Y42C-expressing cells showed similar migration velocity and directionality as WT.

We then evaluated these findings in our murine gastric organoids, in which we could induce Cdh1 loss and/or RHOA^Y42C expression (overview of models used in this study in Supplementary Table S1). Immunofluorescence microscopy studies revealed increased F-actin levels in RHOA^Y42C/+ organoids (especially in Cdh1^−/−RHOA^Y42C/+ organoids) compared with Mist1Cre or Cdh1^−/− organoids (Fig. 2H), consistent with our NIH/3T3 results (Fig. 2A and B).

Overall, our studies of actin- and FA-mediated cell adhesion and migration show that RHOA^Y42C causes a gain-of-function phenotype that does not simply phenocopy lab-generated constitutively activated mutant RHOA^Q63L. It has been a puzzle as to why RHOA mutations analogous to RAS mutations are not found in cancer. Our results suggest a possible clue: The lab-generated RHOA^Q63L mutant may reduce actin dynamics by overly increasing stress fiber formation, thus decreasing rather than increasing motility. In contrast, RHOA^Y42C may optimally increase both stress fiber formation and motility to more effectively engage the actin cytoskeleton in a protransformation manner.

We next determined the mechanistic basis for the gain-of-function biochemical Y42C phenotype. We initially hypothesized that, like the cancer-associated RAC1B splice variant that is impaired in RhoGDI1 interaction (25), the Y42C mutation impairs RHOA interaction with RhoGDI1, a regulator of membrane association and subcellular localization (13). We ectopically co-overexpressed HA epitope–tagged RHOA and GFP-tagged RhoGDI1 in COS-7 cells, immunoprecipitated HA-RHOA, and immunoblotted for GFP. In agreement with previous studies (26, 27), we found that RHOA^WT but not active RHOA^Q63L or the RHOA^T19N dominant-negative mutant associated with RhoGDI1 (Supplementary Fig. S2D and S2E). However, RHOA^Y42C binding to RhoGDI1 was similar to that of RHOA^WT. Consistent with this, RHOA^Y42C and RHOA^WT displayed similar subcellular localization in NIH/3T3 cells (Supplementary Fig. S2F). We concluded that Y42C does not alter RHOA interaction with RhoGDI1.

We next hypothesized that Y42C imparts a fast-cycling phenotype, like the gain-of-function, cancer-associated P29S mutation in the small GTPase RAC1 (28). We purified E. coli–expressed recombinant RHOA proteins and measured intrinsic nucleotide exchange using a fluorescence-labeled nucleotide assay. However, activities of RHOA^WT and RHOA^Y42C were not significantly different (Fig. 3A). Thus, Y42C does not alter GDP–GTP cycling.

To evaluate Y42C interaction with GEFs, we evaluated data from The Cancer Genome Atlas for expression of RHO family GEFs and GAPs in gastric cancer (Supplementary Fig. S3A). The RHOA-selective GEF ECT2 is preferentially upregulated in gastric cancer compared with adjacent normal tissue (Supplementary Fig. S3B). ECT2 overexpression is associated with gastric cancer progression and poor prognosis (29). Therefore, we tested the ability of recombinant ECT2 C-terminal DH-PH domain (catalytic GEF fragment) to stimulate nucleotide exchange on WT and mutant RHOA. ECT2-catalyzed nucleotide exchange activities on RHOA^WT and RHOA^Y42C were similar to each other (k_cat = 23.3 × 10⁻⁴ s⁻¹ and 21.1 × 10⁻⁴ s⁻¹, respectively, Fig. 3B; Supplementary Fig. S3C), and to activities described for other RhoGEFs (30), indicating that Y42C does not alter GEF sensitivity.

Activating RAS mutations (e.g., Q61L) impair intrinsic and GAP-stimulated GTP hydrolysis, thereby favoring the active, GTP-bound form. We directly measured GTP bound to recombinant RHOA proteins in high-performance liquid chromatography (HPLC) assays. The intrinsic GTP hydrolysis rate, which was completely abolished by the Q63L mutation (Fig. 3C and D), was reduced 440-fold in RHOA^Y42C relative to RHOA^WT (0.05 × 10⁻⁵ s⁻¹ and 22.1 × 10⁻⁵ s⁻¹, respectively). To verify this striking GTPase deficiency, we also applied a fluorescence-based hydrolysis assay using a phosphate-binding-protein sensor, which confirmed that intrinsic GTP hydrolysis is greatly impaired in RHOA^Y42C (Fig. 3E and F). Finally, because Y42C is located in the switch I region of RHOA, which is also involved in GAP binding, we performed precise biochemical assays to determine whether Y42C alters GAP activation. Using a recombinant version (catalytic domain only) of the RHOA-specific GAP p190RhoGAP/ARHGAP35, expressed at high levels in gastric cancer (Supplementary Fig. S3A), we found that GAP-stimulated catalytic activity of RHOA^Y42C was reduced 4-fold relative to RHOA^WT (k_cat = 2.6 × 10⁻² s⁻¹ and 10.4 × 10⁻² s⁻¹, respectively; Fig. 3G; Supplementary Fig. S3D). These results indicate that reductions in both intrinsic and GAP-stimulated GTP hydrolysis favor increased levels of RHOA^Y42C-GTP.

To directly measure the guanine nucleotide bound to RHOA^Y42C in living cells, we adapted the classic method of ³²P-orthophosphate metabolic labeling developed originally for RAS. Unlike the standard pulldown assay that detects only relative levels of RHOA-GTP (21), the ³²P radiolabeling assay enables precise quantitation of the percentage of RHOA bound to GTP or GDP. As expected, RHOA^WT was predominantly GDP-bound (only 1.8% GTP-bound; Fig. 3H; Supplementary Fig. S3E), whereas RHOA^Q63L was predominantly (78%) GTP-bound. Surprisingly, despite its impaired GTP hydrolysis, RHOA^Y42C showed a reproducible but not statistically significant increase in GTP binding (3.2%) compared with RHOA^WT. This result suggests that additional alterations other than those tilting the balance of GTP to GDP contribute to the ability of RHOA^Y42C to stimulate canonical RHOA functions such as actin stress fiber and FA formation.

Y42 lies in RHOA's effector domain. Studies performed prior to discovery of RHOA mutations in DGC showed that a Y42C substitution generated in cis with RAS-like activating RHOA mutations (G14V or Q63L) altered RHOA effector interactions (31, 32). However, how Y42C alone affects effector interactions has not been determined. ROCK and mDia are RHOA effector proteins regulating F-actin dynamics (33), and Rhotekin-RBD is the standard pulldown reagent used to determine RHOA-GTP levels in cells (21). Utilizing a well-established fluorescence-based effector interaction assay, we found that the binding affinities between recombinant RHOA^WT and the isolated RBDs of Rhotekin (500 ± 40 nM), ROCK (380 ± 30 nM), and mDia (460 ± 60 nM) were similar to each other (Supplementary Fig. S3F) and comparable to previous studies (34, 35). In contrast, whereas RHOA^Y42C binding to mDia-RBD was comparable to that of RHOA^WT, its binding to ROCK-RBD was increased 12-fold (Fig. 3I; Supplementary Fig. S3F). Unexpectedly, RHOA^Y42C did not bind to Rhotekin-RBD, indicating that the standard Rhotekin-RBD pulldown assay does not accurately measure GTP levels of RHOA^Y42C. Mechanistically, because ROCK promotes actin stress fiber formation, whereas Rhotekin has been reported to antagonize it (36), the enhanced ROCK binding of RHOA^Y42C together with loss of Rhotekin binding provides further basis for its gain-of-function phenotype observed in stress fiber formation assays (Fig. 2A and B). In addition, because ROCK enhances FA formation (37), increased ROCK binding may contribute to the gain-of-function phenotype of RHOA^Y42C in our FA studies. Taken together with its impaired intrinsic and GAP-stimulated GTP hydrolysis, we demonstrated that RHOA^Y42C is a gain-of-function mutant due to alterations in both GDP–GTP regulation and effector interactions (Fig. 3J, see model). These results suggest a biochemical basis for RHOA^Y42C serving as a driver of DGC in our mouse model.

Having established that RHOA^Y42C is an oncogene, we next evaluated the signaling and cellular effects of RHOA^Y42C expression in the gastric lineage. We established isogenic organoids from the gastric epithelia of Cdh1^Flox/Flox mice (Supplementary Fig. S4A and S4B). We introduced ectopic EGFP-RHOA^Y42C, EGFP-RHOA^Q63L, EGFP-RHOA^WT, or EGFP control via lentiviral transduction and created either Cdh1-null or WT models using adenoviral Cre recombinase. These organoids were then evaluated by reverse-phase protein array (RPPA) analyses to evaluate the consequences of genetic alterations on global signaling. Analyses revealed increased phosphorylation of diverse signaling proteins in Cdh1-null organoids expressing RHOA^Y42C compared with RHOA^WT (Supplementary Fig. S4C; Supplementary Table S2). These differences were markedly increased in Cdh1-null compared with isogenic Cdh1^WT organoids. Consistent with our biochemical and cellular analyses, RPPA identified ROCK activation in Cdh1-null organoids expressing RHOA^Y42C compared with RHOA^WT (Fig. 4A; Supplementary Fig. S4D). Increased activation of PI3K–AKT–mTORC1 signaling in Cdh1-null RHOA^Y42C–expressing organoids was also observed (Fig. 4A; Supplementary Fig. S4E). S473 and T308 were two of the most significantly upregulated AKT phosphorylation sites in Cdh1-null organoids with exogenous RHOA^Y42C versus RHOA^WT (Fig. 4B). Many significant changes found in Cdh1-null organoids were not observed with Cdh1^WT organoids (Fig. 4C), suggesting that signaling required to induce tumor formation is dependent on both Cdh1 loss and RHOA^Y42C expression (Fig. 1).

Interestingly, the EGFP–RHOA^Q63L vector could not be expressed in Cdh1-intact organoids. Although Cdh1-null organoids expressing RHOA^Q63L showed the same increased AKT phosphorylation at S473 as with RHOA^Y42C, the overall signaling profile was distinct (Supplementary Fig. S4F and S4G). These findings together with our biochemical and cellular studies may provide evidence for why RHOA^Q63L is not found in cancer.

Because of the RPPA results, we focused on further evaluation of AKT signaling. Immunoblotting confirmed that AKT was not activated in Cdh1-null organoids unless RHOA^Y42C was also expressed (Fig. 4D). Under those conditions, we also observed elevated phosphorylation of AKT target GSK3β at S9 (Fig. 4D), which downregulates destruction of β-catenin by the GSK3β/Axin/APC complex (38–40). Immunoblotting also demonstrated that β-catenin abundance was markedly attenuated in Cdh1^−/− compared with Cdh1^+/+ organoids and was not altered upon expression of RHOA^WT (Fig. 4D). In contrast, RHOA^Y42C increased β-catenin expression in Cdh1-null organoids. E-cadherin tethers β-catenin to the cell membrane. Thus, we hypothesized that the combination of GSK3β repression and loss of Cdh1/E-cadherin enables nuclear translocation β-catenin in Cdh1^−/−RHOA^Y42C/+ organoids. Indeed, these organoids displayed elevated nuclear β-catenin in vitro and following orthotopic implantation (Fig. 4E; Supplementary Fig. S4H–S4J). We further validated that downstream targets of β-catenin, c-MYC, and cyclin D1 were upregulated by RHOA^Y42C expression (Supplementary Fig. S4K). To investigate the relationship of AKT activation and β-catenin induction, we ectopically expressed the constitutively activated mutant PI3Kα^H1074R (PIK3CA^H1074R) in Cdh1-null organoids and observed increased levels of phosphorylated AKT (pAKT) and active β-catenin (Supplementary Fig. S4L). Conversely, pharmacologic blockade of PI3K (with pictilisib) or AKT (with MK-2206) suppressed the TOP/FOP WNT/β-catenin reporter in Cdh1^−/−RHOA^Y42C/+ organoids (Supplementary Fig. S4M). These data demonstrate that, in the setting of Cdh1 loss, RHOA-mediated PI3K activation promotes β-catenin activation.

We next sought additional mediators of RHOA's oncogenic activity. RHOA mediates activation of YAP–TAZ (41, 42), which interacts with β-catenin (43). We found expressing RHOA^Y42C in Cdh1-null organoids increased YAP and TAZ expression (Fig. 4F) and active, nonphosphorylated form YAP (Fig. 4G), suggesting that YAP–TAZ signaling is activated in Cdh1^−/−RHOA^Y42C/+ organoids. As RHOA^Y42C activates both β-catenin and YAP–TAZ signaling, we evaluated whether activation of these pathways replaces RHOA^Y42C in mediating transformation of Cdh1-null gastric organoids. Although expression of either active YAP^S127A (Supplementary Fig. S4N) or β-catenin^S33Y alone failed to induce tumor growth, the combination of both YAP^S127A and β-catenin^S33Y induced robust tumor formation (Fig. 4H) with cells resembling signet ring cells both in vitro and following xenograft growth (Supplementary Fig. S4O and S4P).

We next investigated whether the YAP–TAZ and β-catenin pathways are necessary for RHOA^Y42C oncogenicity. We first targeted these pathways with genetic tools in Cdh1^−/−RHOA^Y42C/+ organoids, using the YAP dominant-negative mutant S94A (YAP-DN; Supplementary Fig. S5A) and a dominant-negative mutant of β-catenin cofactor TCF4 (TCF4-DN; aa 1-31 del; Supplementary Fig. S5B). Although only YAP-DN inhibited organoid growth in vitro, both attenuated the aberrant organoid morphology (Supplementary Fig. S5C and S5D). Importantly, either YAP-DN or TCF4-DN dramatically inhibited tumor growth in vivo (Fig. 5A and B; Supplementary Fig. S5E). Histologic analyses showed that small tumors formed following expression of YAP-DN or TCF4-DN displayed less resemblance to DGC and instead displayed greater gland formation and differentiation (Fig. 5C). To determine if YAP-DN or TCF4-DN also inhibits organoid-derived gastric cancer models that were not driven by Cdh1 loss and RHOA^Y42C, we utilized a model generated from Trp53^−/−Kras^G12D/+ mice (44). We found only modest effects on tumor growth and in vitro proliferation of Trp53^−/−Kras^G12D/+ organoids similarly engineered to express YAP-DN and/or TCF4-DN (Fig. 5D; Supplementary Fig. S5F), demonstrating that targeting these pathways did not induce nonspecific toxicity. Overall, these data suggest that the YAP–TAZ and β-catenin–TCF4 pathways are necessary and sufficient to drive tumor formation in the Cdh1^−/−RHOA^Y42C/+ gastric model.

We next explored the therapeutic potential of pharmacologically targeting these pathways in Cdh1^−/−RHOA^Y42C/+ organoids. We targeted the AKT/β-catenin axis with ICG-001, an antagonist of β-catenin cofactor TCF4, or with the AKT inhibitor MK-2206, both of which attenuated aberrant organoid morphology and had modest but significant effects on viability (Supplementary Fig. S5G-S5I). Similarly, the YAP pathway inhibitor verteporfin had modest effects on organoid proliferation (Supplementary Fig. S5I). Further testing suggested that pharmacologic inhibition of single pathways led to adaptive changes in other pathways, which may mitigate responses to single drug treatment (Supplementary Fig. S5J). Specifically, YAP–TAZ inhibition induced pAKT; conversely, AKT inhibition enhanced TAZ expression. Accordingly, the combination of verteporfin with either ICG-001 or MK-2206 markedly inhibited proliferation and induced apoptosis in the Cdh1^−/−RHOA^Y42C/+ model (Fig. 5E and F; Supplementary Fig. S5G), whereas the same treatments had only modest effects on organoids from Trp53^−/−Kras^G12D/+ mice (Fig. 5E and F; Supplementary Fig. S5I and S6A). We also tested these combinations in normal and Cdh1^−/− organoids, finding only a modest effect relative to their effects in the Cdh1^−/−RHOA^Y42C/+ model (Supplementary Fig. S5K and S5L). We next tested these combinations in vivo in the transplanted Cdh1^−/−RHOA^Y42C/+ model. Instead of ICG-001, we inhibited the PI3K–AKT pathway, given the availability of drugs in advanced clinical development. Our in vitro studies showed the PI3K inhibitor pictilisib, which inhibits AKT, to be superior to MK-2206 (Supplementary Fig. S6B and S6C). We therefore selected pictilisib for in vivo combination with verteporfin. Both verteporfin and pictilisib individually inhibited tumor growth, and we saw greater effects with the combination (Fig. 5G; Supplementary Fig. S6D), suggesting clinical potential for combined PI3K–AKT- and YAP–TAZ-directed therapy in RHOA-mutant DGC.

We next investigated the more proximal means by which RHOA^Y42C activates pathways including PI3K–AKT and YAP–TAZ. We hypothesized that FAK could mediate these activities because RHOA^Y42C induces FA assembly and actin rearrangements, and activates PI3K–AKT signaling. FAK signaling has been described to regulate FA dynamics, actin reorganization, and PI3K–AKT signaling to drive invasion and metastasis and to contribute to RHOA activity in multiple cancers (45, 46). Large-scale proteomic studies also implicated FAK-mediated pathways in human DGC (46, 47). Our RPPA analyses revealed increased phosphorylation of FAK substrates (as pFAK was not in the RPPA panel) such as the receptor tyrosine kinase RET (48) in Cdh1-null organoids expressing RHOA^Y42C (Fig. 4B).

By immunoblotting, we confirmed enhanced FAK activation upon RHOA^Y42C expression in Cdh1-null organoids (Fig. 6A) and in Cdh1^−/−RHOA^Y42C/+ organoids (Fig. 6B; Supplementary Fig. S7A). Similarly, ectopic expression of WT PTK2 (encoding FAK) in Cdh1-null organoids increased both pAKT and active YAP levels (Fig. 6C; Supplementary Fig. S7B). Furthermore, immunoprecipitation studies showed that FAK coimmunoprecipitated with PI3KCA in Cdh1^−/−RHOA^Y42C/+ organoids, indicating that FAK directly activates PI3K (Supplementary Fig. S7C). These results are consistent with studies on FAK-mediated PI3K–AKT activation controlling actin cytoskeletal remodeling and FA formation in multiple cancers (45, 46). In addition, FAK also induced aberrant morphology in these organoids (Fig. 6D), recapitulating features seen with RHOA^Y42C, including mucin production which contributes to signet ring cell formation (Fig. 6E).

We then evaluated genetic and pharmacologic targeting of FAK in Cdh1^−/−RHOA^Y42C/+ models. shRNA-mediated silencing of Ptk2 (FAK) reduced pAKT and active YAP (Fig. 6F; Supplementary Fig. S7D). Silencing of Ptk2 (FAK) reverted the morphology of Cdh1^−/−RHOA^Y42C/+ organoids, but had little effect on Trp53^−/−Kras^G12D/+ organoids (Supplementary Fig. S7E). We next evaluated pharmacologic FAK inhibition. PF-573228, a small-molecule FAK inhibitor, attenuated activation of AKT, YAP, and β-catenin (Fig. 6G; Supplementary Fig. S7F–S7H). PF-573228 treatment dramatically reversed the aberrant Cdh1^−/−RHOA^Y42C/+ morphology (Fig. 6H) in a dose-dependent manner (Supplementary Fig. S7I) and decreased staining with Ki-67 (Supplementary Fig. S7J) and Alcian Blue, a marker of mucin formation (Fig. 6I). Live-cell confocal imaging demonstrated that PF-573228 normalized the morphology of Cdh1^−/−RHOA^Y42C/+ organoids over approximately 30 hours (Fig. 6J; Supplementary Video S2).

We next further explored FAK as a DGC therapeutic target. The FAK inhibitor PF-573228 markedly decreased proliferation in Cdh1^−/−RHOA^Y42C/+ but not Trp53^−/−Kras^G12D/+ gastric organoids in vitro (Supplementary Fig. S7K and S7L), indicating FAK as specific target in our DGC model. We also found lack of efficacy in FAK inhibition in normal gastric organoids, but greater effects of FAK inhibition in Cdh1^−/− organoids (Supplementary Fig. S7M and S7N), consistent with our finding of modestly enhanced pFAK following Cdh1 loss (Supplementary Fig. S7O). Relatedly, PF-573228 abrogated tumor growth of Cdh1^−/−RHOA^Y42C/+ organoid xenografts (Fig. 6K). To validate our findings, we tested the clinical candidate FAK inhibitor defactinib and obtained similar results in vitro and in vivo (Fig. 6L; Supplementary Fig. S7P–S7R). To further characterize effects of FAK inhibition, we performed Ki-67 and TUNEL staining of Cdh1^−/−RHOA^Y42C/+ organoid xenografts, finding evidence of both diminished proliferation and apoptosis induction (Fig. 6M and N). Taken together, these results establish FAK as a mediator of RHOA^Y42C function in DGC pathogenesis and establish this kinase as a promising candidate target (Fig. 6O).

We next evaluated FAK inhibition in human DGC cell line models. Although there are no DGC cell lines with endogenous RHOA mutation, RHOA activity has been demonstrated to be elevated in existing DGC cell lines relative to intestinal gastric cancer (IGC) models (49). Utilizing immunoblotting, we verified enhanced FAK activation in DGC lines compared with IGC lines (Fig. 7A and B). We checked E-cadherin in these DGC lines and found them either lacking E-cadherin expression (SNU668 and FU97) or possessing CDH1 mutation (NUGC4 with CDH1, p.D257V) whereas IGC cell lines had markedly higher E-cadherin expression (Supplementary Fig. S8A). To further confirm our hypothesis that RHOA promotes FAK activation, we silenced RHOA using siRNA and observed attenuated pFAK levels (Supplementary Fig. S8B). We also found that the FAK inhibitors PF-573228 and defactinib attenuated pFAK and pAKT levels in the DGC cell lines FU97 and SNU668, but not in IGC line SNU719, which lacks evident pFAK (Fig. 7C). Furthermore, DGC lines were sensitive to PF-573228 treatment in vitro whereas IGC cells were insensitive (Fig. 7D and E). In DGC, PF-573228 induced G₂-M cell-cycle arrest (Supplementary Fig. S8C) in a dose-dependent manner (Supplementary Fig. S8D). We next compared in vivo growth of DGC SNU668 and IGC SNU719 cell lines. We found that DGC SNU668 tumor growth was inhibited by PF-573228 treatment, but had minimal response to 5-fluorouracil (5-FU), a commonly used agent in DGC (Fig. 7F and G). In contrast, IGC line SNU719 showed markedly greater sensitivity in vivo to 5-FU compared with PF-5732228 (Fig. 7H; Supplementary Fig. S8E). PF-573228 treatment not only attenuated proliferation (Supplementary Fig. S8F), but also induced DNA damage and apoptosis of SNU668 in vivo (Supplementary Fig. S8G and S8H), consistent with data from organoids (Fig. 6M and N).

To further investigate FAK in DGC, we evaluated FAK activation in DGC surgical samples using IHC, finding pFAK staining in 17 of 18 evaluated samples, with FAK staining in 70% to 100% of tumor cells compared with negative staining in surface epithelial cells and minimal staining in normal glandular epithelial cells (Fig. 7I; Supplementary Fig. S8I). We also evaluated the level of pFAK in 8 non-DGC samples and observed positive staining in only one case (Supplementary Fig. S8J). These data provide support for the potential relevance of FAK in DGC beyond that observed in our engineered murine model.

Genomic studies of DGC have found that the two most characteristic classes of alterations affect cellular adhesion (e.g., CDH1 and CLND18) and RHO signaling (e.g., RHOA and ARHGAP26; refs. 4, 12, 50, 51). However, these findings have yet to be translated therapeutically or to deeper functional understanding of disease pathogenesis. Whether RHOA mutations represent activating or inactivating alterations has remained uncertain. The specific residues recurrently mutated contributed to this uncertainty. RHOA^Y42C is analogous to Y40C in RAS, an effector domain mutant impairing interaction of RAS^G12V with RAF (52). Following these studies in RAS, RHOA^Y42C had been studied in cis with activating RHOA mutations (G14V or Q63L) as an experimental strategy to attribute RHOA functions to specific RHOA effectors (31, 32). Thus, when RHOA^Y42C mutations were identified in DGC, it was unexpected that a putative loss-of-effector-function mutant could serve as an oncoprotein, fueling discussion of whether RHOA serves as a tumor suppressor (3, 10). Our analyses demonstrate RHOA^Y42C to be an oncoprotein. Similar to RHOA^G14V/Q63L, RHOA^Y42C exhibits an activated phenotype, stimulating actin stress fiber formation and FA assembly, albeit at a reduced potency. Interestingly, RHOA^Y42C did not simply phenocopy RHOA^Q63L, but instead exhibited distinct activities in regulation of cell adhesion and migration, and signaling. That RHOA^Q63L suppressed, whereas RHOA^Y42C stimulated, migration is consistent with our finding that RHOA^Y42C promotes DGC that is invasive and metastatic.

By studying RHOA^Y42C without secondary mutations, we found impaired intrinsic and GAP-stimulated GTP hydrolysis activity, properties favoring formation of active GTP-bound RHOA. However, these biochemical defects did not drive significant steady-state accumulation of RHOA-GTP in cells. Our analyses of effector binding revealed additional alterations: increased affinity for ROCK, the key effector that drives RHOA stimulation of actin stress fibers and FA assembly, and loss of binding to Rhotekin, an effector that antagonizes actin stress fiber formation. Together, these altered effector interactions provide a mechanistic explanation for the RHOA^Y42C gain-of-function phenotype, stimulating FA assembly and activation of FAK. In addition, loss of Rhotekin binding explains why standard Rhotekin–RBD pulldown assays (21) would produce misleading results for RHOA^Y42C activation.

Our murine gastric model provides further validation that RHOA^Y42C is an oncoprotein. That oncogenic function was unmasked with loss of E-cadherin may explain why RHOA mutations are restricted to cancers arising from a very limited spectrum of tissue types. These results also suggest a functional interaction of RHOA and E-cadherin. Our data suggest gastric CDH1 loss alone does not promote invasive cancer (16, 53). Indeed, cells with CDH1 loss are more prone to undergo anoikis (54), a checkpoint whereby cells undergo apoptosis upon loss of attachment (55). In vivo, Cdh1^−/− gastric epithelial cells were lost over time (16). Cell–cell adhesion (promoted by E-cadherin) suppresses anoikis by activating PI3K–AKT and other pathways (56), a model consistent with our results that pAKT was attenuated in gastric organoids following Cdh1 inactivation (Fig. 4D). The ability of RHOA^Y42C to activate the PI3K pathway may enhance survival and tumorigenicity of Cdh1-null gastric cells, via activation of both pAKT and β-catenin. These data suggest that PI3K–AKT inhibition could be used to block WNT–β-catenin signaling. However, further studies would be needed to fully evaluate the effects of PI3K–AKT inhibition in DGC, given the potential for induction of FOXO3A with these agents to have tumor-promoting effects (57). It is notable that the other tumor type associated with germline CDH1 loss, lobular breast cancer, possesses recurrent co-occurrence of CDH1 and PIK3CA mutations (58). PIK3CA mutations are less common in DGC, whereas breast cancers lack recurrent RHO alterations. This unique selection for RHOA dysregulation in DGC suggests the importance of a PI3K- and β-catenin–independent pathway. In additon, it is the case that not all RHOA-mutant DGCs harbor detectable CDH1 mutations, raising the question of what other alterations may collaborate with RHOA mutation.

Our results showing RHOA/FAK-mediated activation of PI3K and YAP–TAZ could explain the DGC predilection for RHOA mutation. Many studies have implicated RHOA and the actin cytoskeleton as upstream activators of YAP–TAZ (41, 42, 59). We validated the activation of YAP–TAZ in gastric cells transformed by RHOA^Y42C and Cdh1 loss. Aberrations of the actin cytoskeleton are commonly seen in cancer (60), and enhanced F-actin rearrangement with RHOA^Y42C likely provides even greater mechanical stimulus for YAP–TAZ activation. YAP also confers resistance to chemotherapeutic agents in ovarian cancer (61) and oral squamous cell carcinoma (62), suggesting that YAP may contribute to the poor efficacy of cytotoxic therapy in DGC.

Humar and colleagues found that although pFAK was not present in early preneoplastic foci in patients with hereditary DGC, FAK became activated in more advanced lesions (53), consistent with our findings that activating FAK with Cdh1 loss promotes progression. Furthermore, our results establish a contribution of FAK signaling to DGC pathogenesis and demonstrate a mechanistic connection between RHOA and FAK with activation of PI3K–AKT, β-catenin, and YAP–TAZ. Whether FAK activation is more ubiquitously essential with CDH1 loss is not established. The sensitivity we observed of DGC cell lines to FAK inhibition implies that FAK may serve as a target for DGCs, including those without RHOA mutation. Indeed, CDH1 silencing was shown to sensitize mesothelioma cells to FAK inhibition (63), suggesting a potential role for FAK blockade more broadly in CDH1-null cancers.

In summary, these data address uncertainties following discovery of RHOA mutations in DGC. Our biochemical studies establish RHOA^Y42C as a gain-of-function mutant that modulates RHOA interaction with downstream effectors. These data demonstrate how aberrant RHOA activation collaborates with loss of tumor suppressor CDH1 to stimulate signaling networks that mediate transformation (Fig. 6O, see model), leading to features typical of the disease, including signet ring cell formation, peritoneal spread, and ascites. Furthermore, these models reveal how PI3K activation promotes nuclear β-catenin localization in CDH1-null gastric cells by attenuating GSK3β-mediated β-catenin destruction. Our data also demonstrate that FAK activation is a mediator of both PI3K–AKT-β-catenin and YAP activity, secondary to RHOA^Y42C mutations. These data and models provide a new foundation for mechanistic and translational inquiry into these deadly cancers where progress has languished, and where treatments continue to be reliant on minimally effective cytotoxic therapy.

We generated a mouse allele with inducible expression of RHOA^Y42C. The human RHOA cDNA coding region with the Kozak sequence (GCCGCCACC) was introduced into vector pGV at the EcoRI cloning site using blunt-end cloning. Sequencing-confirmed pGV-RHOA^Y42C vectors were coelectroporated with plasmid expressing FLP recombinase into mouse embryonic stem (ES) cells (MESC10, Mirimus) engineered with an FLP homing cassette at the Co1A1 locus, and positive clones were identified by PCR. Positive ES clones were injected into mouse blastocysts for chimera generation. Chimeric mice were crossed with WT mice to generate mice with germline mutations. A detailed strategy was described previously (64). The gene is expressed following Cre recombinase-mediated excision of a stop cassette flanked by LoxP sites [loxP-stop-loxP (LSL) RHOA^Y42C/+]. Mist1-CreERT2, Cdh1^Flox/Flox, R26-mTmG mice were developed as published previously (16). Mist1-CreERT2, Cdh1^{Flox Flox}, R26-mTmG mice were crossed with LSL-RHOA^Y42C/+ mice to generate Mist1-CreERT2, Cdh1^Flox/Flox, LSL-RHOA^Y42C/+, R26-mTmG mice; genotyping was confirmed with appropriate primers (see Supplementary Table S3). All animals were maintained and used in accordance with the guidelines of the Institutional Animal Care and Use Committee of the Dana-Farber Cancer Institute (Boston, MA).

NIH/3T3 mouse fibroblasts were obtained originally from Dr. Geoffrey M. Cooper (Dana-Farber Cancer Institute, Boston, MA) and COS-7 cells and HEK293T cells were obtained from ATCC. Cells were maintained in DMEM supplemented with 10% calf serum (NIH/3T3) or FBS (COS-7 and HEK293T), penicillin, and streptomycin. Cell lines were passaged for one month or 10 passages before a new aliquot was thawed. Cell lines were monitored monthly for Mycoplasma contamination using the Lonza MycoAlert Mycoplasma Detection Kit and not authenticated. SNU668, NUGC4, AGS, and SNU719 cells were from the Broad Institute (Cambridge, MA) and maintained in RPMI-1640 supplemented with 10% FBS. FU97 was from the Broad Institute and maintained in DMEM supplemented with 10% FBS and 10 mg/L insulin (Sigma). These cells were authenticated by the Broad Institute. Isogenic MCF10 cell lines (CDH1-WT/KO) were purchased from Sigma and maintained in Mammary Epithelial Cell Growth Medium (Lonza). The isogenic MCF10 cells were authenticated by Western blot analysis with E-cadherin antibody. All cell lines were maintained in a humidified chamber with 5% CO₂ at 37°C. Cells were monitored regularly for Mycoplasma contamination.

Prior to orthotopic transplantation, organoids were collected and dissociated to approximately single cells using TrypLE Express (Life Technologies). Approximately 5 × 10⁵ cells were resuspended in 50 μL of a mixture of Matrigel and media (1:1). NSG mice (6–8 weeks old, The Jackson Laboratory) were sedated using isoflurane inhalation anesthesia. The stomach was exteriorized through a midline abdominal incision and the 50 μL of cell suspension was surgically injected into the stomach wall with all care to avoid stomach puncture. The incision was immediately closed using a running 7-0 polypropylene suture (Prolene, Ethicon). The presence of tumors was evaluated by twice weekly abdominal/stomach palpation. All animal experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at the Dana-Farber Cancer Institute, in compliance with NIH guidelines.

To measure directly the intrinsic GTP hydrolysis, 70 μmol/L RHOA-GTP was incubated at 25°C in 50 mmol/L HEPES pH 7.5, 100 mmol/L NaCl, 5 mmol/L MgCl₂, 5% glycerol, and 5 mmol/L β-mercaptoethanol. Aliquots of 40 μL were flash-frozen in liquid nitrogen at indicated time points to stop the reaction and incubated for 2 minutes at 95°C. After centrifugation (14,000 × g, 1 minutes) of the denatured protein, the supernatant was applied to an HPLC column (Agilent 1100). GDP and GTP were separated on a C18column (Agilent) with 100 mmol/L potassium-phosphate pH 6.5, 10 mmol/L tert-butyl-ammonium-bromide, and 7.5% acetonitrile as mobile phase (65). The concentration of nonhydrolyzed GTP was plotted against time. Data were described by a monoexponential equation to determine the observed rate constant (k_obs) using GraphPad Prism. A similar protocol was used to measure the efficiency of RHOA nucleotide loading (25% acetonitrile for mant-nucleotides).

For details, see Supplementary Table S4.

For details, see Supplementary Table S5.

RPPA analysis was performed in triplicate on gastric organoids harboring various Cdh1 and RHOA genetic perturbations. Samples were lysed as described previously (66). Cell lysates from organoids were immobilized onto nitrocellulose-coated glass slides (Grace Bio-labs) using an Aushon 2470 arrayer (Aushon BioSystems) in biological triplicates along with reference standards for quality control. Selected arrays were stained with Sypro Ruby Protein Blot Stain following the manufacturer's instructions to quantify the amount of protein in each sample. Immunostaining was performed as described previously (66). Clustering by antibodies was performed using k-means clustering. Fold change values represented were calculated with respect to the corresponding RHOA-EV control median intensity value by Cdh1 status and subsequently log₂ changed, unless otherwise noted. Testing between specific comparisons utilized the Student t test with P values <0.05 denoting significant alterations. Testing results were represented by volcano plots and/or reported in the Supplementary Tables. Heat maps were generated in R using the ComplexHeatmap package from Bioconductor.

Mouse organoids expressing GFP (R26-mTmG after tamoxifen induction) were cultured in conditioned media in a Lab-Tek II Chambered Coverglass System dish. Images of organoids with different genotypes were captured on an LSM 510 Meta live-cell confocal microscope, with GFP autofluorescence images taken every 10 minutes for a total of 36 to 48 hours of culture. For PF-573228 treatment, Cdh1^−/−RHOA^Y42C/+ GFP organoids were collected and dissociated into single cells using TrypLE Express, then plated into the same coverglass system at a density of 40,000 cells/25 μL Matrigel, and cultured for 2 days prior to treatment with DMSO or PF-573228 (5 μmol/L). Live-cell imaging was started immediately after drug treatment.

Data are represented as mean ± SD or SEM as indicated in the figure legends. For each experiment, the number of independent biological experiments are as noted in the figure legends, with representative images shown of replicates with similar results. Statistical analysis was performed using Microsoft Office statistical tools or in Prism 7.0 (GraphPad). Pairwise comparisons between groups (that is, experimental vs. control) were performed using an unpaired two-tailed Student t test, one-way ANOVA Tukey multiple comparison test, or two-way ANOVA as appropriate. P < 0.05 is considered to be statistically significant. P values are denoted by *, P < 0.05; **, P < 0.01; ** P < 0.001; ****, P < 0.0001. For all experiments, the variance between comparison groups was found to be equivalent. Sample sizes and animal numbers were determined from pilot laboratory experiments and previously published literature. Animals were excluded from analysis if they were euthanized due to health reasons unrelated to tumor growth. For in vivo experiments, all mice were randomized before drug treatment.

H. Zhang is a co-founder of Signet Therapeutics. M. Pierobon has ownership interest (including patents) in Theranostics Health/Avant. G. Church's conflict-of-interest disclosures can be found at v.ht/PHNc. K.-K. Wong is a consultant at G1 Therapeutics, Janssen, Pfizer, Merck, Ono, and Array and has ownership interest (including patents) in G1 Therapeutics. E.F. Petricoin is a consultant for and has ownership interest (including patents) in Avant Diagnostics, Inc. C.J. Der is a consultant for Ribometrix, Eli Lilly, and Jazz Therapeutics and reports receiving commercial research grants from Mirati Therapeutics and Deciphera Pharmaceuticals. A.J. Bass reports receiving commercial research grants from Novartis, Merck, and Bayer and has ownership interest (including patents) in Signet Therapeutics, Earli, and HelixNano. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Zhang, A. Schaefer, K.-K. Wong, C.J. Der, A.J. Bass

Development of methodology: H. Zhang, A. Schaefer, Y. Wang, R.G. Hodge, A.G. Papageorge, J. Liao, E.F. Petricoin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Schaefer, Y. Wang, R.G. Hodge, D.R. Blake, A.G. Papageorge, M.D. Stachler, J. Liao, J. Zhou, Z. Wu, L.K. de Klerk, S. Derks, M. Pierobon, K.-K. Wong, E.F. Petricoin, D.R. Lowy

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Zhang, A. Schaefer, Y. Wang, R.G. Hodge, D.R. Blake, J.N. Diehl, A.G. Papageorge, M.D. Stachler, M. Pierobon, K.A. Hoadley, G. Church, E.F. Petricoin, D.R. Lowy

Writing, review, and/or revision of the manuscript: H. Zhang, A. Schaefer, Y. Wang, R.G. Hodge, J.N. Diehl, M.D. Stachler, K.A. Hoadley, T.C. Wang, G. Church, E.F. Petricoin, A.D. Cox, C.J. Der, A.J. Bass

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Schaefer, Y. Wang, J.N. Diehl, J. Liao, L.K. de Klerk, S. Derks, G. Church, C.J. Der

Study supervision: K.-K. Wong, C.J. Der, A.J. Bass

Other (evaluating the IHC slides under the microscope): F.G. Akarca

Other [supplied the vast majority of unstained slides from human for FAK IHC as displayed in Fig. 7I and Supplementary Fig. S8I and S8J; obtained the tissue blocks from the pathology department; verified diffuse vs. intestinal histology (aided by cytokeratin IHC); and confirmed absence of EBER staining and presence of MMR machinery (MSH2, MSH6 PMS2, MLH1 by IHC) in the diffuse gastric cancer samples]: L.K. de Klerk

This work was supported by grants from the NCI (R01 CA223775; to A.J. Bass, C.J. Der, and T.C. Wang), The DeGregorio Family Foundation (to A.J. Bass and T.C. Wang) and the Department of Defense Congressional Directed Medical Research Program (to H. Zhang). H. Zhang was supported by the 2017 Debbie's Dream Foundation-AACR Gastric Cancer Research Fellowship (grant number 17-40-41-ZHAN). R.G. Hodge was supported by the 2018 Debbie's Dream Foundation-AACR Gastric Cancer Research Fellowship, in Memory of Sally Mandel (grant number 18-40-41-HODG). D.R. Blake was supported by NCI fellowships (T32CA071341 and F31CA216965). J.N. Diehl was supported by NCI T32CA071341 and the Slomo and Cindy Silvian Foundation. A.J. Bass received support from a DFCI Medical Oncology pilot project grant and the Schottenstein Gastric Cancer Fund. Histology and confocal core services were supported by the Harvard Digestive Disease Center and NIH grant P30DK034854.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.