Promotion of Colorectal Cancer Invasion and Metastasis through Activation of NOTCH – DAB 1 – ABL – RHOGEF Protein TRIO

UNLABELLED
We have recently identified a metastasis suppressor gene for colorectal cancer: AES/Aes, which encodes an endogenous inhibitor of NOTCH signaling. When Aes is knocked out in the adenomatous epithelium of intestinal polyposis mice, their tumors become malignant, showing marked submucosal invasion and intravasation. Here, we show that one of the genes induced by NOTCH signaling in colorectal cancer is DAB1/Dab1. Genetic depletion of DAB1 suppresses cancer invasion and metastasis in the NOTCH signaling-activated mice. DAB1 is phosphorylated by ABL tyrosine kinase, which activates ABL reciprocally. Consistently, inhibition of ABL suppresses cancer invasion in mice. Furthermore, we show that one of the targets of ABL is the RAC/RHOGEF protein TRIO, and that phosphorylation at its Tyr residue 2681 (pY2681) causes RHO activation in colorectal cancer cells. Its unphosphorylatable mutation TRIO Y2681F reduces RHOGEF activity and inhibits invasion of colorectal cancer cells. Importantly, TRIO pY2681 correlates with significantly poorer prognosis of patients with colorectal cancer after surgery.


SIGNIFICANCE
These results indicate that TRIO pY2681 is one of the downstream effectors of NOTCH signaling activation in colorectal cancer, and can be a prognostic marker, helping to determine the therapeutic modality of patients with colorectal cancer.


INTRODUCTION
The direct cause of cancer death is often its metastasis to the vital organs.Metastasis is achieved through a multistep cascade of events, and therefore ineffi cient as a whole ( 1 ).Despite the efforts to fi nd mutations that are responsible for metastasis, relatively few such genetic changes have been found ( 2 ), leading to the speculation that metastasis is driven by the mechanisms for physiologic and/or pathologic body reactions.
The AES / Aes (Amino-terminal enhancer of split) gene encodes the colorectal cancer metastasis suppressor AES that functions as an endogenous inhibitor of NOTCH signaling ( 3 ), which plays a variety of roles in cancer in a contextdependent manner.In colorectal cancer cells, AES protein can block the NOTCH signaling transcription complex composed of the NOTCH intracellular domain (NICD), the transcription factor RBPJ, and the cofactor MAML.Upon introduction of homozygous Aes knockout mutation into the adenomatous epithelium of Apc +/Δ716 intestinal polyposis mice, their tumors become malignant due to NOTCH signaling activation, showing submucosal invasion and intravasation ( 3 ).Consistently, transendothelial migration (TEM) is increased signifi cantly, when colorectal cancer cells with activated NOTCH signaling are placed on an endothelial cell (EC) layer in culture.Thus, reduced level of AES and stimulation of NOTCH signaling are implicated in the invasion and intravasation of colorectal cancer cells during metastasis ( 3,4 ).
Although NOTCH signaling is involved in cell migration during nervous system development ( 5 ), the precise molecular mechanisms remain to be elucidated.In the present study, we have investigated how NOTCH signaling stimulates colorectal cancer metastasis.We have found coordinated activation of NOTCH signaling in colorectal cancer via RBPJ-dependent transcription of Dab1 , followed by ABL tyrosine kinase activa-

RESEARCH ARTICLE
Figure 1.Transcription factor RBPJ is essential for invasion of mouse intestinal tumors and induces expression of DAB1/Dab1 as one of the downstream genes.A and B, Rbpj gene knockout blocks invasion of intestinal tumors in Apc/Aes compound mutant mice.Note that after induction of intestinal epithelium-specifi c Cre recombinase-mediated mutagenesis, the tumor genotypes were Apc −/− Aes −/− and Apc −/− Aes −/− Rbpj −/− , respectively.The depth of tumor invasion was analyzed histopathologically by hematoxylin and eosin staining (A), and scored (B).Dotted lines in A, muscularis mucosae .Asterisk in A, a colorectal cancer gland invading into the submucosa.mu, intramucosal; sm, mp, and se, invasions reaching the submucosa, muscularis propria , and serosa, respectively ( n = 5 mice for each genotype).Apc/Aes/Rbpj mice showed extended life spans (6-7 months from ∼5 months).C, RBPJ target genes common to mouse and Drosophila .Mouse and ( fl y ) gene symbols.Data were extracted from ( 6,7 ).D, binding sequence motifs for the NOTCH signaling transcription factor Su(H)/RBPJ/CBF1 in the fl y, mouse, and human promoter regions for the dab/Dab1/DAB1 genes, respectively.A pair of horizontal triangles; primers for the chromatin immunoprecipitation (ChIP) analysis in F. TSS, transcription start site.E, recombinant DLL4 (rDLL4) ligand induces DAB1 in cultured colorectal cancer cells.F, binding of RBPJ and NICD to the DAB1 promoter.G, RBPJ-dependent expression of Dab1 in mouse intestinal tumors.Data are presented as the mean of three independent experiments with SD, unless otherwise stated.*, P < 0.01; #, P < 0.05.Scale bars, 100 μm.affecting tumor size or number ( Fig. 1A and B ).These results indicate that RBPJ-mediated transcription plays a key role in colorectal cancer progression in the Apc/Aes mouse.By comparing RBPJ target genes reported in the mouse ( 6 ) and fruit fl y ( 7 ), we noticed that only three genes were shared between the two species: Dab1 ( dab in Drosophila ), Notch1 ( Notch ), and Hes1/5 ( E(spl) ; Fig. 1C ).In this study, we focused on Dab1 because it enhances neuronal motility during mouse brain development ( 5 , 8 ).We excluded NOTCH1 for being in the direct NOTCH signaling feedback loop, whereas exogenously introduced Hes1 caused massive death of Colon26 mouse colorectal cancer cells (data not shown).First, we found that both human and mouse DAB1/Dab1 genes contain highaffi nity RBPJ-binding motifs in their proximal promoter regions as in Drosophila ( Fig. 1D ), suggesting evolutionally conserved transcription by RBPJ ( 9 ).Consistently, immobilized NOTCH ligand DLL4 induced DAB1 mRNA in cultured LS174T human colorectal cancer cells, whereas the γ-secretase inhibitor N-[N-(3,5-Difl uorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester (DAPT) inhibited this induction by blocking NOTCH receptor activation ( Fig. 1E ).In addition, RBPJ, as well as the activated NOTCH receptor fragment NICD, was preferentially bound to

RESEARCH ARTICLE
the DAB1 gene promoter in chromatin immunoprecipitation (ChIP) assays, suggesting the roles of the RBPJ transcription complex in DAB1 expression ( Fig. 1F ).We obtained similar results with another human colorectal cancer cell line, Colo205, as well (Supplementary Fig. S1A and S1B).These two cell lines expressed DAB1 mRNA at higher levels than normal colonic mucosa, whereas others did at lower levels (Supplementary Fig. S1C).Consistent with these cultured cell data, we found RBPJ-dependent induction of Dab1 mRNA and protein in the intestinal cancer of Apc/Aes mice ( Fig. 1G and Supplementary Fig. S1D).These results indicate that the DAB1/Dab1 gene is one of the NOTCH signaling transcription targets of RBPJ in colorectal cancer cells in both humans and mice.
Importantly, homozygous null mutation of Dab1 in the intestinal epithelium inhibited invasion and intravasation of the 4-HT-dosed Apc +/Δ716 ; Aes Flox/Flox ; Dab1 Flox/Flox ; TgvCre ERT2 mouse intestinal tumors (abbreviated as Apc −/− Aes −/− Dab1 −/− or Apc/ Aes/Dab1 ).Namely, the tumor intravasation incidence (i.e., the number of mice that showed at least one lesion of intravasat-ing tumor into the intestinal wall blood vessels per animal) was 80% (4 of 5) in Apc/Aes mice, whereas it was 0% (0 of 5) for Apc/ Aes/Dab1 mice.Refl ecting this difference, the depth of tumor invasion was signifi cantly reduced in the Apc/Aes/Dab1 mice as compared with the Apc/Aes mice ( Fig. 2A and B ).In addition, constitutive knockdown of the DAB1 gene by shRNA in LS174T cells signifi cantly inhibited lung metastasis (employed as the surrogate for liver metastasis; ref. 3 ) after their transplantation into the nude mouse rectum, without affecting the primary tumor size ( Fig. 2C-F ).On the other hand, exogenous expression of DAB1 in RKO, one of the low-expresser human colorectal cancer cell lines (Supplementary Fig. S1C), increased the metastasis frequency approximately 4 times upon rectal transplantation, without affecting the primary tumor size or the growth rate in culture ( Fig. 2G-J and data not shown).Although the primary tumors of DAB1 high-expresser LS174T cells showed lesions of intravasation to blood vessels (Supplementary Fig. S2A and S2B), such histology was not observed in the tumors of sh DAB1expressing LS174T cells (Supplementary Fig. S2C).Collectively, A and B, Dab1 is critical for progression of colorectal cancer in genetically altered mice.Note that after induction of intestinal epithelium-specifi c Cre recombinase-mediated mutagenesis, the tumor genotypes were Apc −/− Aes −/− and Apc −/− Aes −/− Dab1 −/− , respectively.The same methods and keys are employed as in Fig. 1A and B ; n = 5.Boxed areas are magnifi ed in the insets.Arrow in A, intravasating tumor cells inside the vascular lumen (*) with red blood cells.C-F, knockdown of DAB1 inhibits metastasis of colorectal cancer transplants.LS174T cells were transduced with an expression virus encoding EGFP and nonsilencing shRNA (Ns) or shRNA against DAB1 (sh DAB1 ; C).Two clonal lines (#1 and #2) were established for each transductant and injected into the rectal smooth muscle of nude mice.Their lung metastases were evaluated under a fl uorescent dissection microscope (D), and their numbers were scored (E).The primary tumor volumes were not affected by expression of either sh DAB1 (F).G-J, expression of DAB1 stimulates colorectal cancer metastasis in vivo .RKO cells were transduced with expression viruses for EGFP and DAB1, or EGFP alone (G), and the established clones were transplanted into the nude mouse rectum.Microscopic examination (H) and quantifi cation (I) were performed as in D and E, respectively.The primary tumor volume was not affected by expression of DAB1 in RKO (J).Data are presented as the mean of three independent experiments with SD, unless otherwise stated.*, P < 0.01.Scale bars, 100 μm (for A) and 500 μm (for D and H). Research. on

RESEARCH ARTICLE
these results suggest prometastatic roles for DAB1 in NOTCH signaling-dependent colorectal cancer progression.

DAB1 Protein Is Tyrosine-Phosphorylated in Colorectal Cancer Cells by ABL Kinase, and Phosphorylated DAB1 Helps ABL Autophosphorylation Reciprocally
DAB1 was identifi ed as one of the proteins that bound ABL as well as SRC family tyrosine kinases ( 10 ).In RKO colorectal cancer cells in which the endogenous level of DAB1 mRNA expression is low (Supplementary Fig. S1C), its exogenous expression more than doubled the number of Matrigel-invading cells ( Fig. 3A ).Notably, dasatinib, an ABL/ SRC dual inhibitor, suppressed Matrigel invasion down to 5% to 10%, either with or without additional expression of DAB1.However, the SRC family-specifi c inhibitor PP2 had little effect in the same assay ( Fig. 3A ), suggesting that the effect of dasatinib was caused mainly by inhibition of ABL.Consistently, we found dose-dependent suppression of Matrigel invasion by the ABL inhibitor imatinib ( Fig. 3B ).The invasion inhibition by these inhibitors was not affected substantially by their viability suppression which was <15% (see Methods).mRNA increased Matrigel invasion of RKO cells, and SRC/ABL dual inhibitor dasatinib (Das; at 0.01 μmol/L ) showed a tighter suppressive effect on Matrigel invasion of RKO cells than the SRC inhibitor PP2 (at 10 μmol/L), either with or without additional expression of DAB1.Imatinib (Ima; 10 μmol/L) also showed signifi cant inhibition of invasion (see B below).B, the ABL inhibitor imatinib blocks Matrigel invasion of RKO cells in a dose-dependent manner.C, ABL knockdown in human colorectal cancer cells inhibits Matrigel invasion.RKO cells were transfected with either of two independent siRNA sets (#1 or #2), each consisting of distinct siRNAs against ABL1 (si ABL1 -1 and -2) and ABL2 (si ABL2 -1 and -2).Ns, nonsilencing control.D, DAB1 increases ABL Tyr-kinase activity in human colorectal cancer cells.Lanes 1 to 3, HCT116 cells were transfected with the same amount of FLAG-tagged WT DAB1 (lane 2) or its 5YF mutant (lane 3), and their phospho-Tyr (pY) contents were determined by immunoprecipitation-Western blotting (IP-WB).Lanes 4 to 7, HCT116 cells were cotransfected with FLAG-tagged DAB1 (DAB1-FLAG) and HA-tagged WT ABL1b (ABL1b-HA).Note that the input amount of DAB1-FLAG cDNA for lanes 6 and 7 was the same as that for lanes 2 and 3. Lanes 8 to 11, HCT116 cells were cotransfected with the same set of DAB1 cDNA as in Lanes 4 to 7, and KD mutant of ABL1b.E, ABL is found in DAB1 immunoprecipitation (IP) of intestinal tumors in Apc/Aes mice as analyzed by Western blotting (left), whereas DAB1 is found in ABL IP, reciprocally (right).WB indicates antibodies used for detection of ABL or DAB1 by Western blotting.Data are presented as the mean of three independent experiments, with SD. *, P < 0.01.

RESEARCH ARTICLE
The ABL subfamily of tyrosine kinases is conserved evolutionarily, consisting of ABL1 and ABL2 (aka ARG) in mammals, and has pleiotropic roles in cell proliferation, migration, etc. ( 11 ).Knockdown of the ABL1 and ABL2 genes ( Fig. 3C , bottom) in RKO cells reduced the mRNA levels below 40%, and Matrigel invasion to 15% to 20%, without affecting cell viability ( Fig. 3C , bottom and top, respectively; see also Methods).These results show that ABL tyrosine kinase activity is essential for Matrigel invasion of human colorectal cancer cells, implicating its role in colorectal cancer progression.
Consistent with a previous report ( 10 ), ABL coprecipitated with DAB1 with anti-DAB1 Ab from the lysate of Apc/Aes mouse intestinal tumors, whereas DAB1 coprecipitated with ABL with anti-ABL Ab in a reciprocal experiment ( Fig. 3E ), suggesting their physical interaction in vivo either directly or through additional proteins.Importantly, imatinib signifi cantly inhibited colorectal cancer invasion in Apc/Aes mice without affecting the tumor size or number ( Fig. 4A-C ).Consistently, the tumor intravasation incidence was decreased from 100% (5 of 5) in vehicle-treated Apc/Aes mice to 0% (0 of 5) in imatinib-dosed mice.These results indicate that DAB1 is phosphorylated when coexpressed with ABL in colorectal cancer cell lines, and, reciprocally, pY-DAB1 stimulates ABL autophosphorylation, which suggests in vivo activation to promote colorectal cancer invasion as one of the downstream effects of NOTCH signaling.

One of the Targets Phosphorylated by Activated ABL in Colorectal Cancer Cells Is the RAC/RHOGEF Protein TRIO
In an attempt to determine the substrate(s) of ABL responsible for colorectal cancer cell invasion, we looked into Drosophila biology.As a downstream effecter of Drosophila ABL (dABL), Triple functional domain (dTRIO) plays key roles in fruit fl y neuronal migration ( 15 ).TRIO belongs to the DBL family of guanine nucleotide exchange factor (GEF) proteins and can activate the RHO family of small GTPases, wellcharacterized and essential regulators of cell motility ( 16,17 ).TRIO carries two GEF domains: one for RAC (GEF1) and the other for RHO (GEF2; ref. 18 ; Fig. 5C ; top).Therefore, we speculated that RAC and/or RHO was also activated by TRIO in human colorectal cancer cells to promote their invasion.
Interestingly, RHO was activated signifi cantly as the GTPbound form in invasive ( Apc/Aes ) mouse colon cancer compared with benign ( Apc ) adenomas, whereas RAC-GTP was below the detection level in Apc/Aes tumors, although it showed a weak band in Apc adenomas ( Fig. 5A ) and a strong band in a control experiment (Supplementary Fig. S3A).Consistently, inhibition of RHO by C3T ( 19 ), or its immediate downstream effector ROCK by Y-27632 ( 20 ), blocked the Matrigel invasion and TEM of RKO human colorectal cancer cells in a dose-dependent manner ( Fig. 5B and Supplementary Fig. S3B, respectively).This inhibition of invasion appears to be mostly due to direct effects, although a small fraction may be attributable to inhibition of viability (up to 20% at higher doses; see Methods).On the other hand, RAC inhibition by NSC23766 or RAC Inhibitor II ( 21 ) did not affect Matrigel invasion (Supplementary Fig. S3C and S3D).These results suggest that the NOTCH signaling-dependent Matrigel invasion and TEM of colorectal cancer cells in culture ( 3 ) are mediated by RHO activation.Furthermore, knockdown of TRIO in HCT116 cells (Supplementary Fig. S3E and S3F) reduced their Matrigel invasion ( Fig. 5D ) and lung metastasis rate upon rectal transplantation ( Fig. 5E ) without affecting the primary tumor size or growth rate in culture (Supplementary Fig. S3G, and data not shown), suggesting that TRIO RHOGEF activity plays a key role in the invasion of NOTCH receptor-activated colorectal cancer cells.

In Colorectal Cancer Cells, TRIO Y2681 Is One of the Major Phosphorylation Sites Activated by DAB1-ABL
In Drosophila S2 cells, dTRIO is Tyr-phosphorylated in the presence of dABL ( 15 ).Also, in HCT116 colorectal cancer cells, we found that some Tyr residues of TRIO were heavily phosphorylated when simultaneously expressed with DAB1 and ABL1b ( Fig. 5F , lane 4), but not with DAB1 5YF mutant (lane 5) nor with ABL1b KD mutant (lane 7).These results suggested that Tyr-phosphorylation of TRIO was dependent on ABL kinase activity.
To determine the particular phospho-Tyr residues in TRIO, we next screened its primary amino acid sequence with NetPhos2.0software ( 22 ).It predicted the probability of phosphorylation for all 61 Tyr residues in TRIO, and identifi ed 30 as likely candidates with scores above the threshold of 0.5.Among a series of TRIO tyrosine-phenylalanine (YF) mutants ( Fig. 5C ), we found signifi cant reduction in the pY level of TRIO 18YF compared with TRIO WT when expressed in RKO cells ( Fig. 5G , lanes 1 and 2).Although the colorectal cancer cells expressing TRIO 7YFs-a or TRIO 7YFs-b contained pY-TRIO at a similar level to control TRIO WT, those expressing TRIO 11YFs-b that carried an additional four YF mutations at residues 1990, 2562, 2681, and 2757 had a markedly reduced level of pY-TRIO ( Fig. 5G , lane 5).As anticipated, the pY level was similarly reduced in cells expressing the TRIO 4YFs mutant in which only the additional four tyrosine residues were mutated ( Fig. 5G , lane 6).Among these four Tyr residues, we identifi ed Y1990 and Y2681 as the key targets of phosphorylation by DAB1-ABL, because expression of either Y1990F or Y2681F mutant (Fig. 5H, lanes 2 and 4), but not Y2562F or Y2757F (lanes 3 and 5), markedly reduced the pY levels in the colorectal cancer cells.Notably, the aa sequence adjoining TRIO Y2681 has a strong similarity to the ABL target site consensus ( 11 ).It is also known that dTRIO protein binds to dABL ( 15 ).Accordingly, it is likely that ABL directly phosphorylates TRIO Y2681 in colorectal cancer cells.
To determine the direct effects of introduced proteins, we employed the TetON system to conditionally express TRIO, together with DAB1 and ABL1b, and confi rmed that TRIO Y2681F failed to promote Matrigel invasion of RKO cells, although TRIO WT and TRIO Y1990F could do so ( Fig. 5I ).Because induction of TRIO WT alone in RKO cells did not increase Matrigel invasion (Supplementary Fig. S3H), these results indicate that TRIO is the effector of DAB1-ABLinduced invasion.In RHOGEF GTP-exchange assays performed in vitro using purifi ed proteins, TRIO WT enhanced the exchange of GDP in RHOA-GDP with GTP (i.e., activation to RHOA-GTP), as reported (ref.18 ; Fig. 5J ).Importantly, TRIO WT protein purifi ed from imatinib-treated transductant cells and TRIO Y2681F mutant protein showed signifi cantly decreased RHOGEF activity (40%-50%; Fig. 5K ) without affecting RACGEF activity (Supplementary Fig. S3I), suggesting that phosphorylation at Y2681 by ABL is essential for its maximal RHO activation.Together, these results indicate that ABL causes phosphorylation of TRIO at Y2681 and increases the level of RHO-GTP, stimulating invasion of colorectal cancer cells in culture.

TRIO pY2681 Signifi cantly Correlates with Poor Prognosis of Patients with Colorectal Cancer
To determine the clinical relevance of the TRIO phosphorylation in colorectal cancer, we raised polyclonal antibodies specifi c for TRIO pY1990 and TRIO pY2681 peptides.Namely, antibodies were raised against phospho-Tyrcontaining peptides surrounding Y1990 and Y2681, respectively, and absorbed with the same sequence peptides lacking Tyr-phosphorylation.We fi rst confi rmed their specifi cities using HEK293T cells that expressed TRIO WT, Y1990F, or Y2681F mutant (Supplementary Fig. S4A and S4B).We also tested them on human colorectal cancer cell line mouse xenograft tumors by immunohistochemistry (IHC).The anti-TRIO pY2681 cytoplasmic staining disappeared upon expression of sh DAB1 or sh TRIO (Supplementary Figs.S2B and  S2C; S4C and S4D), verifying their specifi cities.
We then analyzed surgical specimens of colorectal cancer by IHC with these antibodies.Consistent with the above results that TRIO pY1990 hardly affected Matrigel invasion of RKO cells, the disease-specifi c survival rates (DSS) were similar between the TRIO pY1990-low and -high patients ( P = 0.9, n = 129; data not shown).In contrast, the patients with high levels of TRIO pY2681 in their primary colorectal cancer showed a statistically signifi cant reduction in their DSS, as compared with those who had TRIO pY2681-low colorectal cancer (all stages included, n = 337, P < 0.001 in the log-rank test; Fig. 6A , left).When the stage II colorectal cancer subpopulation was examined, the 5-year DSS was 100% for TRIO pY2681-low patients, whereas it was approximately 80% for the TRIO pY2681-high patients ( n = 115, P = 0.015; Fig. 6A , center).Even within the stage IV subpopulation, the 5-year DSS was approximately 50% for TRIO pY2681low patients, in contrast to only approximately 10% for the Accordingly, the phosphorylation status of TRIO Y2681 may help future prognostic studies, together with the currently available biomarkers.
In the TRIO pY2681-high specimens, we found strong immunoreactivity in the cytoplasm of colorectal cancer cells compared with the adjacent normal mucosa ( Fig. 6B ).Notably, TRIO pY2681 staining was found also in the isolated, budding, or invading colorectal cancer cells in the stroma (arrowheads in Fig. 6C-E ), supporting the thesis that TRIO pY2681 is critical for human colorectal cancer invasion.Interestingly, these budding colorectal cancer cells expressed some stem cell markers such as LGR5 ( 23 ) and SOX9 ( 24 ) and showed signs of partial epithelial-mesenchymal transition (EMT), such as decreased expression of cytokeratin 18 and E-cadherin, consistent with earlier reports (ref.25 ; Fig. 6F-I , and Supplementary Fig. S4E).Notably, Colon26 mouse colorectal cancer cells showed reduced levels of

RESEARCH ARTICLE
the EMT-related transcription factors SLUG and TWIST (but not SNAIL) when AES expression was induced and NOTCH signaling was inhibited (Supplementary Fig. S4F).These results show that TRIO pY2681 and partial EMT can be found at the invasion front of colorectal cancer where signs of increased stemness are detected.They suggest that activation of TRIO RHOGEF may play a role in colorectal cancer stem cell functions associated with partial EMT.Because EMTrelated molecular changes are induced even in adenomas of Min mice (i.e., in an early stage of tumorigenesis; ref. 26 ), however, the relationship of NOTCH signaling with stemness about TRIO pY2681 will require further study.
To further confi rm that TRIO phosphorylation at Y2681 was caused by upstream activation of NOTCH signaling in human colorectal cancer, we immunostained the cleaved and activated NOTCH receptor NICD and TRIO pY2681 in serial pathologic sections.The immunostaining of nuclear NICD positively correlated with that of cytoplasmic TRIO pY2681 ( n = 40, P < 0.01 in χ 2 tests; Fig. 6J ).Consistent with these human clinical data, we found a much higher pY2681 content in the TRIO protein immunoprecipitated from the Apc/Aes mouse intestinal cancer than that from the Apc adenoma (Supplementary Fig. S5A).These results support our interpretation that TRIO pY2681 is caused by activation of NOTCH signaling in both human and mouse colorectal cancers, although pathways other than DAB1-ABL phosphorylation may also induce TRIO pY2681.
Bound with TLE1, AES suppresses colorectal cancer progression by inhibiting NOTCH signaling ( 3 ).Notably, nuclear expression of AES was inversely correlated with TRIO pY2681 levels in human colorectal cancer specimens as determined by IHC ( n = 102, P < 0.01; representative photographs in Supplementary Fig. S5B).Furthermore, exogenous expression of AES and TLE1 inhibited ABL activity as determined by Tyr-phosphorylation of DAB1 in RKO cells (Supplementary Fig. S5C).
We have already shown that immobilized NOTCH ligand DLL4 activates NOTCH signaling in colorectal cancer cells and promotes their migration, which is suppressed by AES ( 3 ).As anticipated, expression of AES in mouse Colon26 cancer cells reduced the level of RHO-GTP after scratchinduced migration (see Methods) in a DLL4-coated dish (Supplementary Fig. S5D).These results collectively indicate that suppression of NOTCH signaling-induced transcription inhibits TRIO Tyr-phosphorylation, resulting in a decreased RHO-GTP level, which reduces invasion and metastasis of colorectal cancer cells ( Fig. 7 ; see Discussion).

DISCUSSION
In the cancer invasion-metastasis cascade ( 1 ), events such as local invasion, intravasation, and extravasation involve active cell motility.Tumor cells move either as individual cells or en masse via collective migration ( 27 ).Individual tumor cells have two different modes of movement: mesenchymal mode and amoeboid mode ( 28 ).Although the former is characterized by an elongated morphology that requires extracellular proteolysis localized at cellular protrusions, the latter is independent of proteases, and cells have rounded morphology with no obvious polarity ( 29 ).Members of the RHO family of GTPases are key regulators of cell movement through their actions on actin assembly, actomyosin contractility, and microtubules ( 30 ).The amoeboid and mesenchymal modes of movement are distinguished by their different usage of signaling pathways.Namely, the amoeboid mode involves signaling through RHOA-ROCK, whereas the mesenchymal mode requires extracellular proteolysis ( 31 ) for RAC-dependent actin protrusions to be pushed through channels in the extracellular matrix ( 28 ).Notably, these two modes of motility are observed in invading tumor cells in vivo ( 29 ), and tumor cells may switch between the elongated and rounded modes of movement in an adaptive response to the microenvironment ( 28 , 31 ).Importantly, RAC and RHO activities are mutually antagonistic; active RAC represses RHO activity and vice versa ( 29,30 ).Although the ability to switch between modes of movement gives cells adaptability to a variety of conditions, some tumor cells appear to be hardwired to stick with a particular mode ( 32 ).
In contrast to single-cell migration, the mechanics of collective cell migration has been investigated only recently, and the roles of RHOA and RAC1 have been explored in more collective processes such as tumor invasion ( 33 ).For example, RHO, rather than RAC, plays the key role in collective migration of epithelial cells ( 34 ), by which colorectal cancer very often invades into the stroma ( 3 ).Consistently, our present results show that TRIO Y2681F mutation reduces its RHOGEF activity ( Fig. 5K ) without affecting RACGEF activity (Supplementary Fig. S3I).Some colorectal cancer cell lines, including LS174T, appear to migrate by the amoeboid migration mode

RESEARCH ARTICLE
in three-dimensional (3D) Matrigel, whereas others migrate in an elongated morphology ( 31 ).Yet, these cells can switch among these migration modes.In the two-dimensional (2D) scratch assay, colorectal cancer cells migrate as collectives ( 35 ).Our results above have been obtained using invasive mouse tumors in vivo , as well as cultured human colorectal cancer cells in 3D Matrigel invasion and 2D scratch assays.Accordingly, the data may indicate that invasive mouse tumors and human colorectal cancer cells in culture share common mechanisms in terms of utilizing the RHO GTPases.
In the intestinal crypt, NOTCH signaling has been implicated in tissue stem cell functions.For example, activation of NOTCH signaling in the intestinal epithelium causes expansion of the proliferating cell population ( 36 ), whereas its suppression by a γ-secretase inhibitor reduces proliferation and causes goblet cell metaplasia ( 37 ).NOTCH signaling is crucial for the maintenance of intestinal crypt bottom columnar stem cells ( 38 ).However, the downstream molecules directly involved in stem cell homeostasis remain to be investigated.At the same time, EMT can cause stem cell-like changes in breast cancer ( 25 ), although in colorectal cancer, some molecular changes of EMT are detected even in adenomas that lack overt EMT phenotypes ( 26 ).We have found that the budding colorectal cancer cells at the invasion front often show TRIO pY2681, and simultaneously express some stem cell markers, such as LGR5 and SOX9, with partial signs of EMT ( Fig. 6F-I and Supplementary Fig. S4E).Accordingly, it is conceivable that TRIO pY2681 induced by NOTCH signaling activation is related to the stem cell-like functions of colorectal cancer, although further investigations are needed.
The above results that imatinib can block DAB1-ABL phosphorylation, and the subsequent activation of TRIO-RHO and invasive motility of colorectal cancer cells, suggest the possibility that ABL tyrosine kinase inhibitors, such as imatinib and dasatinib, may be useful for metastasis prevention as neoadjuvant chemotherapy and/or post-operative chemotherapy.Notably, these inhibitors have been studied and proposed for colorectal cancer chemotherapy because of their additional activities against platelet-derived growth factor receptor ( 39,40 ) and SRC ( 41 ).The present results can add a novel mechanistic rationale for the use of these inhibitors against colorectal cancer.
Collectively, we have revealed a novel signaling mechanism downstream of NOTCH that stimulates colorectal cancer progression, suggesting the following scenario: Loss of AES activates NOTCH signaling transcription, and as one of the downstream effector pathways, induces DAB1 expression, which potentiates ABL autophosphorylation and activation, causing phosphorylation of TRIO Y2681 to promote colorectal cancer cell invasion and TEM through its RHOGEF activity ( Fig. 7 ).We propose that TRIO pY2681 may be a useful prognostic marker, and that the ABL activity and its downstream effectors, such as TRIO, RHO, and ROCK, may be novel therapeutic targets against malignant progression of colorectal cancer.

Mutant Mice, Colorectal Cancer Cells, and Transplantation-Metastasis Model
Apc +/Δ716 mutant mice were generated and maintained as described previously ( 42 ).Apc/Aes mice were constructed and induced for Cre activity according to previous reports ( 3 , 43 ).To derive triple-mutant mice, we crossed the Apc/Aes double-mutant mice with Rbpj -null ( 44 ) or Dab1 -null mice (H.Imai and colleagues, unpublished data).For ABL inhibition in vivo , Apc/Aes mice were treated with 50 mg/kg/day (ip) of imatinib (LC Laboratories; ref. 45 ) for 9 weeks from 3 weeks of age.At 12 weeks of age, a total of approximately 100 intestinal tumors (2 mm < diameter) from 5 mice were scored for the invasion depth in each mutant genotype or treatment.
We have confi rmed the identity of each colorectal cancer line (ATCC) recently (July-August 2014) by short tandem repeat analysis (Takara Bio CDM Center).Colorectal cancer transplantation experiments were performed as described previously ( 3 ).All animal experiments were conducted according to the protocol approved by the Animal Care and Use Committee of Kyoto University.

Matrigel Invasion and TEM Assays
Matrigel invasion and TEM assays were performed as described previously ( 3 )

Knockdown/Overexpression Experiments
Forty-eight hours after transfection of specifi c siRNA oligonucleotides (QIAGEN), colorectal cancer cells were subjected to Matrigel invasion, gene expression, or MTS assays in which we did not fi nd any signifi cant changes in cell viabilities between transfection groups.To establish stable knockdown cells for transplantation experiments, we prepared lentiviral particles using sequences sh DAB1 #1 (AAGGATTAAGTAGGATGTCAA) and sh DAB1 #2 (CCG GTACAAAGCCAAATTGAT) or sh TRIO #1 (CCGGGAATGTATGGA TACGTA) and sh TRIO #2 (CAACGGAGAGTCCATGTTAAA), and pLB lentiviral vector (Addgene).

dab/Dab1/DAB1 Promoter Analysis
Fruit fl y dab , mouse Dab1 , and human DAB1 gene promoters were extracted from the UCSC Genome Browser Database and determined for high-and low-affi nity Su(H)/RBPJ-binding sites, as described previously ( 50 ).

rDLL4 Experiments
Culture plate surfaces were coated with recombinant DLL4 ligand (rDLL4; R&D) at 4°C overnight, and then washed once with PBS and seeded with colorectal cancer cells ( 3 ).Four hours after plating, mRNA was extracted and analyzed for DAB1 mRNA by qRT-PCR.qPCR TaqMan primers/probes were purchased from ABI. ChIP analyses were performed on colorectal cancer cell lysates using anti-RBPJ (Institute of Immunology) and anti-NICD (Cell Signaling Technology) antibodies and hDAB1ChIP.F2 (CAAGCTCTGTGCTT-GTCTCA) and hDAB1ChIP.R2 (GTAGCTGTGTGGTCTTATCA) primers ( Fig. 1D , paired triangles) according to a published protocol ( 51 )

IHC
Frozen and paraffi n-embedded tissues of mice were prepared according to the standard procedures.Human colorectal cancer tissues ( n = 337, stages 0-IV) had been resected from patients in Kyoto University Hospital who had undergone operations with informed consent between 2005 and 2008, with the protocol approved by the Ethics Committee of Kyoto University.Sections were stained with hematoxylin and eosin or for DAB1 (SIGMA), AES ( 3 ), NICD (Cell Signaling Technology), TRIO pY1990, or TRIO pY2681 (prepared as described below).For evaluation of IHC on clinical specimens, 2 to 3 examiners scored independently, followed by a consensus discussion with advice from a board-certifi ed clinical pathologist (H.Haga).
Endogenous DAB1, ABL, and TRIO were immunoprecipitated with anti-DAB1, anti-ABL, and anti-TRIO antibodies (Santa Cruz Biotechnology), respectively, from lysates of the mouse intestines.

Immunocytochemistry
HEK293T cells were transfected with expression plasmids for T7-TRIO, ABL1b-HA, and DAB1-FLAG, stained for T7 and Tyr-phosphorylated TRIO pY1990 or TRIO pY2681, and mounted under coverslips with VECTASHIELD Mounting Medium with DAPI (Vector Laboratories).

Active RHO GTPases Pull-Down Assays
RHO-GTP or RAC-GTP was detected using a RHO activation assay kit (Thermo) or a RAC1 activation assay kit (Millipore), respectively.Colon26 TetON-AES-FLAG mouse cancer cells ( 3 ) in which AES was induced for 24 hours by doxycycline were cultured in monolayers, scratched in multiple parallel lines, incubated further for 16 hours, and subjected to the RHO/RAC pull-down assays.

TRIO Mutagenesis
Possible Tyr phosphorylation sites in human TRIO were predicted by NetPhos2.0online software ( 22 ).To create unphosphorylatable YF mutants, we mutagenized 18 Tyr residues in TRIO WT with high pY probabilities (>0.5) to Phe.

Preparation of Specifi c Antibodies for TRIO pY1990 and TRIO pY2681
Phospho-peptides VRDLG(pY)VVEG (residues 1985-1994) and NPNYI(pY)DVPPE (residues 2676-2686) were synthesized and injected into rabbits or chickens at SCRUM (Japan).Each antibody preparation was affi nity-purifi ed using the immobilized antigen and the corresponding unphosphorylated peptide.Titers of the affi nitypurifi ed fractions were confi rmed by ELISA.Antibodies from either species produced essentially the same results.

RHOGEF GTP Exchange Assay In Vitro
We transfected HEK293T cells with pCX-T7-TRIO (WT or Y2681F) and pulled down the TRIO protein from cell lysates using anti-T7-conjugated agarose (MBL).Then, we mixed the TRIO fraction with recombinant RHOA or RAC1 in the presence of mant-GTP (Cytoskeleton) that emitted stronger fl uorescence when bound to RHO small GTPases.

Figure 2 .
Figure 2. Reduced expression of DAB1 blocks invasion and metastasis of mouse intestinal tumors, whereas its increase stimulates metastasis.

Figure 4 .Figure 5 .
Figure 4. ABL kinase inhibitor imatinib can block invasion of mouse intestinal tumors.A and B, imatinib suppresses invasion of endogenous colorectal cancer in Apc/Aes compound mutant mice ( n = 5 mice for each group).The same methods and keys were employed as in Fig. 1A and B .C, the intestinal tumor number or size was not affected by imatinib in Apc/Aes mice.Scale bars, 100 μm.Error bars, SD. *, P < 0.01.

Figure 5 .
Figure 5. (Continued) F, TRIO is Tyr-phosphorylated in the presence of ABL and DAB1 in colorectal cancer cells.HCT116 cells were transduced with or without an expression construct for T7-tagged TRIO (T7-TRIO), cotransduced with or without HA-tagged ABL1b (ABL1b-HA), and/or FLAG-tagged DAB1 (DAB1-FLAG).Phospho-Tyr (pY) contents were determined by Western blotting (WB) after immunoprecipitation (IP).For ABL1b, either WT or KD mutant was used.G, some tyrosine residues in the C-terminal half of TRIO are phosphorylated by ABL.WT or YF mutants of T7-TRIO were expressed simultaneously with ABL1b-HA and DAB1-FLAG, followed by Western blot analysis of pYs.H, Y1990 and Y2681 in TRIO are phosphorylated in the presence of ABL and DAB1.I, stimulation of colorectal cancer cell invasion in Matrigel by expressing TRIO WT, but not by TRIO Y2681F.Clonal RKO TetON derivatives were constructed to express ABL1b-HA/DAB1-FLAG simultaneously with T7-TRIO WT or T7-TRIO Y2681F in a doxycycline (Dox)-inducible manner, and assayed for protein expression and Matrigel invasion.J, cell-free RHOGEF GTP exchange assay.A representative result of three experiments is shown.K, TRIO pY2681 is critical for the RHOGEF activity.T7-tagged TRIO (WT or Y2681F) was purifi ed, and its RHOGEF activity for 10 minutes was determined as shown in J. WT[ima], WT TRIO protein purifi ed from imatinib (10 μmol/L)-treated HEK293T cells.All quantitative data are presented as the mean with SD. *, P < 0.01 in three independent experiments unless otherwise stated.

Figure 6 .
Figure 6.Phosphorylation of TRIO Y2681 in human colorectal cancer (CRC) is correlated with poorer prognosis.A, (left) patients of all stages were combined; 150 cases were TRIO pY2681-low, whereas 187 cases were TRIO pY2681-high ( P < 0.001 in log-rank test).Center, patients of stages II; the TRIO pY2681-low group ( n = 45) and the TRIO pY2681-high group ( n = 70, P = 0.015).Right, stage IV patients; the TRIO pY2681-low patients ( n = 11) and TRIO pY2681-high patients ( n = 46, P = 0.006).B, cytoplasmic staining for TRIO pY2681 in colorectal cancer cells (arrowheads).Compare with the adjacent normal mucosa (N) or lymphoid follicle (L).Boxed area is magnifi ed in the inset.C-E, staining for TRIO pY2681 is found in colorectal cancer cells invading into the stroma (arrowheads).F-I, immunohistochemical analysis of colorectal cancer sections for stem cell and EMT markers in relation to TRIO pY2681 staining.Stem cell markers LGR5 (F) and SOX9 (G), as well as cytokeratin 18 (CK18; H), were stained in the same series of sections as that stained for TRIO pY2681 (I).Similar lesions were selected and photographed at the same magnifi cation.Note that the staining for CK18 at the invading front is weakened (arrows).J, correlation between NICD and TRIO pY2681 immunostaining in human colorectal cancer.Shown is a representative set of adjoining serial sections (left and right).Arrows indicate positive staining whereas arrowheads indicate negative staining.Insets show higher magnifi cation of boxed areas.Hematoxylin counterstaining.Scale bars, 100 μm for B and C, 20 μm for F to I, and 50 μm for J.

Figure 7 .
Figure 7.A schematic representation of NOTCH signaling through DAB1-ABL-TRIO-RHO that stimulates colorectal cancer invasion and metastasis.Stromal cells, including endothelial cells (EC), can activate NOTCH signaling in colorectal cancer cells and upregulate DAB1 transcription, activating ABL when the AES level is low.Activated by DAB1, ABL enhances the RHOGEF activity of TRIO by phosphorylating Y2681.Thus, TRIO pY2681 is critical for the RHO activation that contributes to invasion, intra-and extravasation, and metastasis of colorectal cancer cells.In addition, a possibility remains that other products of RBPJ transcription as well as other tyrosine kinases may also be involved in colorectal cancer invasion and metastasis.These results provide a possible prognostic marker in TRIO pY2681 and a therapeutic strategy targeting ABL (e.g., using imatinib and related drugs) to prevent colorectal cancer metastasis.
. Colorectal cancer cells were treated with C3T (ref.19 ; Cytoskeleton), Y-27632 (ref.20 ; Wako), NSC23766, Rac Inhibitor II ( 21 ), or PP2 (ref.46 ; Calbiochem), imatinib ( 47 ), or dasatinib (ref.48 ; LC Laboratories) for 17 hours during the assays, after preincubation (for 6 hours for PP2, dasatinib, and imatinib; or 2 hours for RHO and ROCK inhibitors) and passage with drug concentrations used in the original reports above.Although up to approximately 15% survival inhibition was found in the MTS assay by dasatinib and imatinib, and up to 15% and 20% by C3T and Y-27632, respectively, they did not affect the interpretation of the invasion assays substantially.