B-Raf Inhibitors Induce Epithelial Differentiation in BRAF-Mutant Colorectal Cancer Cells

Colorectal carcinomas are characterized by a series of (epi)genetic alterations, including aberrant Wnt pathway activity and gain-of-function mutations in the KRAS or BRAF protooncogenes. Compared with advances in deciphering the events initiating colorectal carcinoma, little is known about the mechanisms underlying metastasis (1, 2). However, there is a pressing need to better understand and target metastatic colorectal carcinoma.

BRAF encodes a Ser/Thr kinase of the Ras/MEK/ERK pathway and is mutated in various tumor entities. The most common mutation, V600E, renders the kinase constitutively active and occurs in 11% of colorectal carcinoma (3). BRAF mutations are implicated as early “signature events” in the so-called serrated pathway describing the progression from serrated adenoma to colorectal carcinomas (2, 4). Importantly, BRAF-mutant colorectal carcinomas can be subdivided in two subgroups that differ in terms of microsatellite stability (MSS) and clinical outcome (5). The majority of BRAF-mutant colorectal carcinomas belong to the microsatellite instable (MSI) class, which poses the lowest risk for relapse of all colorectal carcinoma subtypes (6, 7). However, BRAF-mutant colorectal carcinomas with MSS represents a very aggressive entity and forecast a poor prognosis (8–10). Interestingly, BRAF-mutant colorectal carcinomas present with a distinct pattern of metastatic spread, in particular peritoneal carcinomatosis, and a shorter overall survival even after metastasectomy (11, 12).

The discovery of B-Raf^V600E has been translated into clinically applied kinase inhibitors, such as vemurafenib or dabrafenib. As single agents, they yield unprecedented responses in BRAF-mutant melanoma (13, 14). However, such an efficacy was not observed in the small groups of 21 patients with colorectal carcinoma treated with these compounds. Nevertheless, these inhibitors still elicited partial responses and stabilized disease in some patients (14, 15). Depending on their mutational landscape, BRAF-mutant colorectal carcinoma cell lines display differential sensitivity toward B-Raf or MEK inhibitors in tissue culture and xenograft models (4, 13, 16–18). Thus, these compounds still hold clinical promise, in particular if their combination with other substances is considered and once the complex pathology and genetic heterogeneity of colorectal carcinoma are better understood. Indeed, combination strategies show efficacy in preclinical colorectal carcinoma models as they target more than one critical node for proliferation and survival. These approaches prevent compensatory feedback loops and crosstalk, which currently limit the use of B-Raf inhibitors as single agents (4, 19–21). These preclinical drug combinations initiated several clinical trials that are currently ongoing (www.clinicaltrials.gov) and a first case report further demonstrates the utility of such combinations (22). However, to interpret the effect of such combinatorial approaches, it is necessary to obtain a more comprehensive understanding of their effects as monosubstances. Herein, we characterize three isogenic colorectal carcinoma cell line models in which B-Raf^V600E can be dynamically modulated. We demonstrate that B-Raf^V600E controls promigratory and proinvasive effectors and suppresses hallmarks of intestinal differentiation. Importantly, we report that clinically relevant B-Raf and MEK inhibitors mimic the effects of B-Raf^V600E depletion, suggesting that they promote a more differentiated and adhesive state, which in turn might improve prognosis and drug response.

Short tandem repeat–authenticated Colo-205 cells were purchased from CLS Cell Lines Service GmbH (Heidelberg, Germany) and cultured in RPMI-1640 with 10% FCS. Total passage time was under 6 months. Caco-2 and HT29 sublines and associated recombinant DNAs were described previously (23). All other lines were kindly provided by Stephan Feller (Martin-Luther-University Halle-Wittenberg, Germany) and cultivated like HT29 cells. Except for Colo-205, the identity of all lines was confirmed by SNP analysis (Multiplexion) in June 2014. Three-dimensional (3D) cultures were set-up in their culture medium as described previously (24). Details on cellular assays (clonogenic, wound healing, migration, and invasion as well as cell-cycle analyses) are described in Supplementary Data. Organotypic collagen I assays were set-up as described previously (25).

The following inhibitors were dissolved in DMSO: PLX4720 (Santa Cruz Biotechnology), dabrafenib, trametinib, vemurafenib (PLX4032), and gefitinib were purchased from Selleck Chemicals. Doxycycline was used at 2 μg/mL (24).

Immunofluorescence microscopy was conducted as described previously (24) using the antibody manufacturers' instructions. Topro-3 iodide was used as counterstain and bound primary antibodies were detected using the secondary antibodies Alexa Fluor 488 goat-anti-mouse IgG and Alexa Fluor 488 goat-anti-rabbit IgG (Invitrogen).

Western blotting was performed as described previously (23, 24). Detailed information on antibodies is provided in Supplementary Data. Blotted proteins were visualized with a Fusion Solo chemiluminescence reader and quantified using FusionCapt software (Vilber Lourmat).

Pulldown assays were conducted as described previously (26). PLX4720 or DMSO was added daily.

All procedures were carried out in accordance with local animal ethics committees (G-12/34 and G-13/116). Subcutaneous xenografts were generated by injecting 1 × 10⁶ HT29 cells (in 100 μL PBS) into severely immunocompromised Rag2^−/−;γc^−/− mice. Once tumors became measurable at d7, mice were randomly distributed to treatment groups and received doxycycline (2 mg/mL) via sucrose-supplemented (35 g/L) drinking water or not. Inhibitors were applied by oral gavage as indicated. Tumors were measured with calipers and subjected to histologic processing. Formalin-fixed and paraffin-embedded xenograft tumors were cut (3 μm), deparaffinized, subjected to standard operating procedures-driven antigen retrieval and antibody staining (primary antibodies for CK20, Cdx-2, PAS, Ki67, and secondary reagents according to routine diagnostic procedures) in a DAKO autostainer.

RNA was extracted using the Universal RNA Purification Kit (GeneMatrix) from Roboklon. RNA quality and integrity was verified using the Agilent 2100 Bioanalyzer system (Agilent Technologies). Transcript profiling is described in detail in Supplementary Data. Gene-expression data are available at GEO using the following link http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=edcfcmuszbudrst&acc=GSE50791.

Quantitative data are presented as mean ± SE. Multiple group and pair-wise comparisons were performed by ANOVA or the Student t test (two-tailed), respectively. A P value of ≤0.05 was considered statistically significant (***, P < 0.001; **, P < 0.01; and *, P < 0.05). Unless noted otherwise, bar diagrams report the mean of three independent experiments. Statistics of transcriptome data are described in Supplementary Data.

To evaluate endogenous B-Raf^V600E for its contribution to the phenotype of human colorectal carcinoma cell lines within an isogenic background, we generated conditional cell line models. For this approach, we chose two model cell lines, HT29 and Colo-205. Both lines display MSS (Supplementary Table S1), and are therefore likely derived from tumors belonging to the aforementioned high relapse risk subgroup. HT29 was generated from a primary colonic adenocarcinoma (Dukes stage B) and represents a “moderately differentiated” cell line characterized by growth in epithelial sheets. HT29 cells express E-cadherin but lack tight junctions (TJ; ref. 27). In contrast, the poorly differentiated Colo-205 cells represent metastatic cells derived from ascites of a patient with Duke's stage D colon carcinoma. Both lines were transduced with a nonsilencing control or two pTRIPZ constructs (3_11 and 1_14) allowing the doxycycline-regulated knockdown of B-Raf^V600E but not of B-Raf^wt (23). All experiments in the main figures were performed with the construct 3_11, key experiments are also shown with 1_14 (Supplementary Data).

First, we evaluated the characteristics of HT29 cells in conventional two-dimensional (2D) culture following B-Raf^V600E knockdown. Commensurate with our finding that B-Raf^V600E knockdown strongly reduces MEK/ERK phosphorylation in HT29 cells (23), colony growth was impaired, and cells accumulated in the G₁-phase (Fig. 1A and B). However, neither gross morphologic changes nor a significant increase in cell death were observed upon B-Raf^V600E knockdown (Supplementary Fig. S1A and S1B). Nevertheless, B-Raf^V600E knockdown specifically impaired migration (Fig. 1C). Conversely, ectopically expressed B-Raf^V600E induced an elongated morphology, faster spreading, and enhanced cell migration of Caco-2 cells, another line derived from a MSS adenocarcinoma (Supplementary Fig. S1C–S1F). This prompted us to analyze the expression of β-integrin subunits in both cell lines. Indeed, B-Raf induced β1- and β4-integrin in Caco-2 cells (Supplementary Fig. S1G), whereas B-Raf^V600E knockdown and the vemurafenib tool compound PLX4720 slightly reduced β1-integrin levels in HT29 cells (Fig. 1D). Given the link between Rac-GTPases and integrins, we next evaluated the influence of oncogenic B-Raf on the activation of Rac-1 and its splice variant Rac1b, which is linked to aggressive colorectal carcinoma behavior (28). Knockdown of B-Raf^V600E did not affect Rac1-GTP levels, but was accompanied by a significant reduction of Rac1b-GTP (Fig. 1D and E and Supplementary Fig. S2A).

The impact of cancer-associated mutations is increasingly studied using 3D tissue culture (24, 29). For example, Caco-2 cells are widely used to study epithelial effector functions and form cysts in 3D culture. Their single lumen can be enlarged by cholera toxin (CTX)–mediated increase in cyst fluid volume, an event that requires intact TJ and therefore reflects their functionality (29). To study the relevance of B-Raf^V600E signaling for the behavior of HT29 cells in 3D culture, single cells were seeded into Matrigel and assayed similarly to Caco-2 cysts (Fig. 2A; refs. 23, 29). In sharp contrast with Caco-2 cysts, CTX did not induce large lumina in HT29 spheroids in untreated or doxycycline-treated nonsilencing cells. Strikingly, spheroids with B-Raf^V600E knockdown reacted toward CTX treatment with the formation of large lumina (Fig. 2B and Supplementary Fig. S2B). Importantly, this phenotype was reproduced by administration of PLX4720, vemurafenib, dabrafenib, and the MEK inhibitor trametinib (Fig. 2C and Supplementary Fig. S2C). In contrast, the EGFR inhibitor gefitinib showed no effect on 3D morphology and lumen formation and did not further enhance the effects in combination with PLX4720. This indicates that inhibition of the B-Raf/MEK axis is necessary and sufficient for lumen formation and is not influenced by the crosstalk between the EGFR and oncogenic B-Raf (19, 20). Thus, depletion or pharmacologic inhibition of B-Raf^V600E enables HT29 spheroids to respond to CTX with the formation of large lumina.

Next, we analyzed the effect of B-Raf^V600E knockdown or inhibition on signaling in 3D cultures. As confirmed by Fig. 2D, induction of the V600E-specific shRNA strongly reduced B-Raf^V600E expression as well as MEK/ERK phosphorylation, both of which were not affected by CTX (Supplementary Fig. S2D). Likewise, PLX4720 markedly reduced MEK phosphorylation demonstrating potent B-Raf inhibition. However, ERK phosphorylation was mildly affected in PLX4720-treated cells compared with those with B-Raf^V600E knockdown. Nevertheless, the comparable reduction of the well-established ERK target gene product DUSP6, which integrates ERK activity over longer periods of time (30), confirms that PLX4720 had a long-lasting effect on ERK activity. In fact, loss of DUSP6 could increase the half-life of pERK in cells with B-Raf^V600E knockdown or inhibition explaining the differential effects on pMEK and pERK levels. In any case, the strong reduction in pMEK levels argues against the paradoxical ERK activation described for B-Raf inhibitors (23). We exclude waning PLX4720 concentrations in the long-term cultures as an explanation for pERK persistence since readdition of PLX4720 1 hour before lysis did not further reduce pERK levels (Supplementary Fig. S2E).

In agreement with the unchanged E-cadherin staining, no considerable alteration in E-cadherin expression was observed. However, β1-integrin was again reduced in B-Raf^V600E–depleted or –inhibited spheroids (Fig. 2D), which reflects our results of Caco-2tet and HT29 2D cultures. Collectively, we show that B-Raf^V600E increases the motility of HT29 and Caco-2 cells in 2D culture, controls promigratory molecules like Rac1b and β1-integrin, and enables HT29 spheroids to respond to CTX with lumina formation.

Next, we analyzed the effects of B-Raf^V600E knockdown or inhibition in Colo-205 cells. In 2D culture, loss of B-Raf^V600E led to a 3-fold increase in cell death (Supplementary Fig. S2F and S2G). Within the time frame of this experiment, the cells maintained their typical “lymphocytic” morphology (Supplementary Fig. S3A).

Subsequently, we analyzed the importance of B-Raf^V600E for the 3D phenotype of Colo-205 cells. Without doxycycline, Colo-205 cells containing either the nonsilencing or the V600E-specific shRNA grew as loose clusters showing a weak and diffuse staining of E-cadherin (Fig. 3A) and β-catenin (Supplementary Fig. S3B). Importantly, B-Raf^V600E knockdown caused the formation of compact spheroids with defined membranous E-cadherin or β-catenin localization. Analysis of 3D lysates revealed that E-cadherin was also increased on the protein level (Fig. 3B). This suggests that B-Raf^V600E confers the dispersive phenotype of Colo-205 cells in 3D culture and counteracts the function of adherens junctions (AJ). Indeed, PLX4720 and the clinically relevant inhibitors vemurafenib, dabrafenib, or trametinib induced a highly comparable “compaction” phenotype (Fig. 3C and D). However, gefitinib did not provoke such an effect. Thus, the “compaction” phenotype is specifically caused by loss of B-Raf^V600E expression, activity or downstream signaling.

As p120ctn stabilizes AJ by blocking E-cadherin endocytosis (31), we addressed the subcellular localization of this catenin relative in Colo-205 cells. In full agreement with our hypothesis that B-Raf^V600E impairs AJs, membranous localization of p120ctn was markedly increased upon oncoprotein knockdown or inhibition (Fig. 4A). This increase in membrane-associated p120ctn was correlated with its enhanced phosphorylation at Y228, a marker for its incorporation into AJs (31), in lysates from doxycycline- or PLX4720-treated Colo-205 spheroids (Fig. 4B).

Next, we monitored additional signaling events in the lysates from 3D cultures with B-Raf^V600E knockdown (Fig. 3B) or inhibition (Supplementary Fig. S3C). As seen for HT29 cells, doxycycline treatment markedly reduced B-Raf^V600E expression and MEK/ERK phosphorylation. We also noted that Rac1b-GTP, which is implicated as a negative regulator of E-cadherin–mediated cell–cell adhesion and as a promoter of cell motility (28), was markedly reduced by B-Raf^V600E knockdown in 2D culture (Fig. 4C and D and Supplementary Fig. S3D). This prompted us to analyze the effect of B-Raf^V600E on migration and invasion. This showed that vemurafenib significantly reduced both Transwell migration and invasion (Fig. 4E and F). In a 3D organotypic collagen I assay (25), Colo-205 cells with B-Raf^V600E knockdown failed to invade the matrix, whereas their untreated isogenic counterparts or cells expressing the nonsilencing shRNA were detectable in deep layers of the plug (Fig. 4G).

Next, we performed a time-resolved microarray analysis to investigate the transcriptome changes induced by B-Raf inhibition (see Supplementary Fig. S4A and S4B and Supplementary Table S2 for principal component/correlation analysis and a list of all regulated genes, respectively). Although this analysis revealed cell line–specific differences, about 130 PLX4720-responsive genes were commonly up- or downregulated following drug treatment in both cell lines (Fig. 5A). Looking at the 50 most up- or downregulated genes, PLX4720 reduced ERK target gene transcripts such as FOS or IER3 within the first 24 hours (Fig. 5B), again demonstrating long-lasting ERK inhibition. Interestingly, the ERK target gene ERBB receptor feedback inhibitor 1 (ERRFI1, Mig-6; ref. 32) was significantly downregulated in both cell lines (Fig. 5B–D). Thus, attenuation of oncogenic B-Raf signaling causes not only the loss of a rapid phosphorylation mediated (19), but also of a delayed negative feedback loop silencing the EGFR via the de novo synthesis of an EGFR inhibitor. Furthermore, we observed a 2.2-fold downregulation of CDC25C (Supplementary Table S2). This phosphatase is also phosphorylated in a B-Raf–dependent manner, and thereby responsible for the inefficacy of vemurafenib in many colorectal carcinoma lines (19).

Given these observations in our 3D transcriptomes and the well-established interplay between B-Raf^V600E signaling and EGFR activity in conventional tissue culture (19, 20), we next assessed EGFR phosphorylation in spheroid lysates. Interestingly, HT29 cells, which express prominent levels of this receptor tyrosine kinase (RTK; refs. 19, 20), reduce total EGFR levels upon B-Raf^V600E knockdown or inhibition (Supplementary Fig. S4C and S4D). However, if this lowered EGFR expression is considered, the phosphorylation of this RTK at its Grb2 recruitment site Y1068 was not significantly increased. This is in full agreement with findings in 2D culture (20), which suggested that vemurafenib allows a better coupling of the EGFR to downstream pathways rather than augmenting EGFR activity. In Colo-205 spheroids, however, total EGFR levels were unaffected by B-Raf^V600E knockdown or inhibition and, in agreement with other cell lines in 2D culture (19), both treatments augmented Y1068 phosphorylation (Supplementary Fig. S4E). As vemurafenib-mediated reactivation of EGFR signaling augments AKT phosphorylation in 2D culture (19, 20), we next analyzed the phosphorylation status of this kinase at its two key regulatory sites. Similar to 2D tissue culture experiments (19), phosphorylation of S473, an important residue for maximum AKT activation, was increased in HT29 and Colo-205 spheroid lysates. Interestingly, however, phosphorylation of T308, the even more relevant phosphorylation site for AKT activity, was not upregulated, suggesting that B-Raf^V600E rather suppresses the mTORC2–AKT than the PDK1–AKT axis (Supplementary Fig. S4C and S4E). These findings represent important details for combination therapies involving antibodies or inhibitors targeting the EGFR.

Importantly, both lines upregulate transcripts explaining the drastic phenotypic changes observed in 3D culture upon loss of B-Raf^V600E signaling (Supplementary Table S3). Indeed, many of the top 50 upregulated transcripts (Fig. 5B), for example, AKR1B10, CES, HPGD, RARRES1, TFF3, or TXNIP, are linked to epithelial differentiation and effector functions (Supplementary Table S4). Moreover, the master transcription factor of intestinal development and epithelial differentiation, Cdx-2, was significantly upregulated in Colo-205 and HT29 cells by PLX4720 treatment at the transcript (Supplementary Table S2) and protein level (Fig. 5C–E). As shown in Fig. 6A, this is also reflected by the upregulation of various previously identified Cdx-2 targets (33). For example, AMACR, a potential Cdx-2 target and marker for differentiated colorectal carcinomas, was upregulated in HT29 and Colo-205 cells (Fig. 5B and E) and also in three additional BRAF^V600E-mutant colorectal carcinoma lines (Colo-201, LS411, and HDC135), which we used for additional follow-up (Supplementary Fig. S4F). Furthermore, the TJ components claudin-15 (CLDN15) and claudin-18 (CLDN18) were upregulated in Colo-205 cells following B-Raf^V600E inhibition or knockdown. Interestingly, these claudins were not prominently upregulated in HT29 cells, in which we observed a significant induction of claudin-1 (CLDN1; Fig. 5C and D and Supplementary Fig. S5A–S5C). The reciprocal relationship between B-Raf^V600E and Cdx-2, claudin-1 and -18 was further confirmed in Caco-2tet cells, which spontaneously differentiate with increasing confluency. Here, these proteins were upregulated in cultures transfected with the empty and B-Raf^wt expression vectors, whereas B-Raf^V600E expression counteracted their upregulation (Fig. 5F).

Conversely, PLX4720 suppressed transcripts promoting migration and invasion. For example, β4-integrin (ITGB4), which was induced by B-Raf^V600E in Caco-2tet cells and contributes to colorectal tumor progression (34), was reduced in Colo-205 and to a lesser extent in HT29 cells (Fig. 5C and D and Supplementary Table S2). Furthermore, Fascin1 (FSCN1), a gene increasingly implicated in metastasis (35), was downregulated at the transcript level in Colo-205 and at the protein level in Colo-201, Colo-205, and LS411 cells (Fig. 5B and E and Supplementary Fig. S4F). In full agreement with a previous study linking Fascin1 to loss of differentiation (35), HT29 cells lacked Fascin1 expression (Fig. 5E). Likewise, myoferlin (MYOF), a gene linked to breast cancer progression (36) and metastasis of pancreatic carcinoma (37), was downregulated in Colo-205 cells (Fig. 5B and Supplementary Fig. S6A). B-Raf inhibition also altered the expression levels of transcripts controlling proteolytic processes. For example, ADAM19 and KLK3 were reduced in PLX4720-treated Colo-205 cells. Likewise, SERPINE2, which represents a promoter of tumor progression and metastasis in many tumor entities, was downregulated in both cell types. Conversely, the protease inhibitor SPINK1, an inhibitor of invasive processes and marker for a favorable prognosis in patients with colorectal carcinoma (38), was upregulated in both cell lines (Fig. 5B and Supplementary Fig. S6A). Importantly, our analyses also identified PLX4720-suppressed transcripts such as FGF19 and KIAA1199 (Fig. 5B–D and Supplementary Figs. S4F and S6A). Both are discussed as markers for poor prognosis in several entities (39, 40).

Furthermore, we identified transcripts in Colo-205 cells associated with stemness: PIWIL1, a critical epigenetic regulator implicated in self-renewal of normal and cancer stem cells (41), was strongly downregulated by PLX4720 (Fig. 5B and Supplementary Fig. S6A). Interestingly, transcripts of two genes controversially discussed as stemness markers, SMOC2 and OLFM4 (42), were upregulated by PLX4720 (Fig. 5B and Supplementary Fig. S6A). However, the well-established markers of intestinal crypts such as LGR5 and BMI (42) were not affected by PLX4720 (Supplementary Table S2) arguing against the induction of stemness. In fact, both cell lines loose stemness-associated transcripts following B-Raf inhibition (Supplementary Fig. S6B). The diverse cancer-related processes affected by PLX4720 treatment are summarized in Fig. 6B.

Next, we compared the PLX4720 response of both cell lines with clinical data, in which a 64–gene-expression signature for BRAF-mutant colorectal carcinomas was established (9). We mapped 57 of the 64 genes onto our transcriptome data, illustrating the similarities of our 3D cell culture systems with the in vivo transcriptomes of clinical colorectal carcinoma samples. By plotting the mean fold change after PLX4720 treatment individually for both classifier gene groups, a significant differential gene regulation over time was observed (Fig. 6C). Importantly, this classifier comprises several of the most-regulated transcripts of PLX4720-treated spheroids such as SPINK1, AMACR, CDX2, FSCN1, and PIWIL1. This demonstrates how the in vitro transcriptome of both lines switches from a transcriptome reflecting BRAF-mutant colorectal carcinomas toward one based on tumors lacking this mutation. Thus, many transcriptional alterations typically observed in clinical BRAF-mutant colorectal carcinoma samples might be reversible by pharmacologic intervention.

To analyze the role of oncogenic B-Raf signaling in vivo, xenograft experiments with HT29 cells containing nonsilencing or V600E shRNA constructs were performed. Doxycycline administration (starting at d7) stalled the growth of xenografts containing the V600E-specific shRNA construct, whereas tumors in the doxycycline-treated nonsilencing group or in both untreated groups grew exponentially (Fig. 7A and Supplementary Fig. S6C). Strikingly, B-Raf^V600E knockdown induced the formation of glandular structures (Fig. 7B), which was further confirmed by periodic acid-Schiff (PAS) and Cdx-2 staining (2, 43, 44). Importantly, glandular differentiation was not observed in the three control groups maintaining B-Raf^V600E expression. Doxycycline administration to HT29/V600E shRNA carrying mice reduced the number of cells expressing the proliferation marker Ki-67. In agreement with earlier studies (16, 17), tumor regression was also observed in mice treated with vemurafenib or trametinib (Fig. 7C). Furthermore, differentiated areas were observed in most vemurafenib-treated xenografts, albeit to various degrees (Fig. 7D). In agreement with the high efficacy of MEK inhibitors in colorectal carcinoma lines (16), glandular differentiation was more advanced and visible in all trametinib-treated xenografts. In both drug-treated xenograft groups, glandular differentiation was confirmed by PAS and Cdx-2 staining. Importantly, glandular differentiation was not observed in any of the DMSO-treated xenografts. Thus, the differentiation-suppressing effect of B-Raf^V600E/MEK/ERK signaling observed in vitro can be reproduced in xenografts in vivo.

BRAF^V600E confers a poor prognosis to patients with MSS colorectal carcinomas (8–10) and increases peritoneal metastasis (11, 12). However, the molecular and cellular mechanisms underlying these correlations remain ill-defined as most studies addressed oncogenic B-Raf in terms of its impact on colorectal carcinoma initiation by focusing on proliferation and survival (4, 19). Furthermore, studies relied on the phenotypic comparison of cell lines or tumors with either wild-type or mutant BRAF status, which, although insightful, only allows drawing conclusions of correlative nature. Here, we generated three conditional human colorectal carcinoma cell line models, which allow us to directly analyze the acute influence of B-Raf^V600E within an isogenic setting. We demonstrate that B-Raf^V600E enhances migration and invasion, impairs cellular junctions and controls epithelial differentiation markers of prognostic relevance.

For example, B-Raf^V600E suppresses the function of TJ, a hallmark of intestinal epithelia. We demonstrate that reduction of chronic B-Raf/MEK signaling induces large lumina upon CTX exposure, suggesting an improved sealing of cell–cell contacts. However, the expression and/or localization of critical TJ components or regulators, such as occludin, ZO-1 or scribble were not altered (Supplementary Fig. S6D and S6E and data not shown). This suggests that, in the presence of B-Raf^V600E, HT29 cells are able to form rudimentary but “leaky” TJs in which not all critical components are properly localized or expressed. Indeed, we demonstrate for the first time that B-Raf^V600E suppresses the expression of three distinct claudins. In HT29 cells, knockdown or inhibition of B-Raf^V600E upregulated the barrier forming and sealing claudin-1. Conversely, B-Raf^V600E strongly reduces claudin-1 in Caco-2tet cells, which provides an additional mechanism as to how this oncoprotein interferes with the development and maintenance of epithelial cysts (23, 45).

Unlike HT29 cells, Colo-205 spheroids with B-Raf^V600E knockdown or inhibition lack CTX-induced lumina, despite their prominent compaction phenotype (data not shown). Interestingly, Colo-205 cells with B-Raf^V600E knockdown or inhibition upregulated the pore-forming claudin-15 and -18, which contribute to cell–cell adhesion, and thereby fulfill a critical function in gastrointestinal development (46). Thus, these claudins, together with the sequestration of p120ctn into the reformed AJs (Fig. 4A and B), an event known to counteract the proinvasive character of free p120ctn (31), might contribute to the compaction of Colo-205 spheroids.

But how does B-Raf^V600E suppress claudins? To our knowledge, nothing is known about the mechanism regulating claudin-15 and -18 expression. Thus, our analyses connect these lesser-characterized claudins to B-Raf^V600E signaling. In contrast, there is a controversy about the mechanisms underlying claudin-1 expression. For example, the NF-κB, Wnt-, and MEK/ERK pathways as well as Snail1, Cdx-1/2, and Gata-4 are implicated in CLDN1 transcription (46, 47). Thus, our finding that claudin-1 is strongly suppressed by endogenous B-Raf^V600E links its expression to a clinically relevant oncogenic driver.

A major finding of our study is that the presence of B-Raf^V600E suppresses CDX2 expression at the transcript and protein levels in our three isogenic colorectal carcinoma cell line models. B-Raf or MEK inhibition also induced Cdx-2 expression in MSS (HT29 and Colo-205) and MSI (LS411 and HDC135) cells. Importantly, this phenomenon was also observed in HT29 xenografts, which presented with Cdx-2–positive glandular epithelial structures upon B-Raf^V600E knockdown or treatment with trametinib and, albeit to a lesser degree, with vemurafenib. Cdx-2 represents a master transcription factor regulating the differentiation, cell–cell adhesion, and polarity of intestinal epithelia (48), and indeed we observe several Cdx-2 targets being upregulated by B-Raf/MEK inhibition (Figs. 5B and 6A). The mechanisms controlling Cdx-2 expression are controversially discussed and implicate the Wnt, Notch, JNK, and ERK pathways as well as epithelial–mesenchymal transition regulators, epigenetic mechanisms, and the tumor microenvironment (7, 43, 48, 49). Interestingly, knockdown of Cdx-2 in Caco-2 cells disrupts cyst development (50), whereas its ectopic expression in Colo-205 cells improved E-cadherin function and cell–cell adhesion in 2D culture (51). Thus, both studies report phenocopies of experiments involving the ectopic expression or knockdown of B-Raf^V600E in Caco-2 cysts (23, 45) and Colo-205 cells (Fig. 3), respectively. These striking parallels imply that many aspects of the transformed phenotype evoked by B-Raf^V600E are mediated by Cdx-2 loss. Our identification of endogenous B-Raf^V600E as a Cdx-2 suppressor has clinical implications, because its loss is observed at the invasive front and is implicated in metastasis (43). Indeed, loss of Cdx-2 forecasts a poor prognosis and is frequently observed in right-sided colorectal carcinomas, which display a particularly high BRAF mutation rate (2, 44, 52). Furthermore, loss of Cdx-2 expression in MSI and MSS colorectal carcinoma tissue specimens is strongly correlated with BRAF^V600E and higher tumor grade (7, 53). Likewise, loss of expression of the Cdx-2 target gene CLDN1, which we identified as a B-Raf^V600E repressed gene, strongly predicts disease recurrence and poor patient survival (54). Thus, our isogenic model systems now provide a mechanism for the correlation between BRAF mutation status, Cdx-2, and claudin-1 expression by showing a dynamic and inverse relationship between B-Raf^V600E and the expression of Cdx-2 and its targets such as CLDN1 and AMACR (Figs. 5A and 6A). As Cdx-2 positively regulates a plethora of differentiation markers (Supplementary Table S4), our study suggests a mechanism for the aforementioned poor differentiation status (1, 2, 11). The differentiation-suppressing effect of oncogenic B-Raf is of particular interest in relation to recent studies reporting a novel classification system for colorectal carcinomas based on gene-expression profiling (1, 2). These studies show that BRAF-mutant colorectal carcinomas tend to be less differentiated and often display a stem-like signature.

To our knowledge, we are the first to describe the transcriptional alterations induced by a B-Raf inhibitor in 3D cultures of two colorectal carcinoma cell lines with an endogenous BRAF^V600E mutation. This identified transcripts associated with cellular adhesion, invasion, and metastasis, which might represent potential prognostic colorectal carcinoma markers. Several of these alterations at the transcript level were reflected at the protein level, not only in the profiled HT29 and Colo-205 cells but also in other colorectal carcinoma lines (Supplementary Fig. S4F). Some of these transcripts are known from other tumor entities, but have neither been implicated in colorectal carcinoma nor connected to B-Raf signaling. For example, although Fascin1 has been implicated as a Wnt target gene in colorectal carcinoma (35), our data show that B-Raf inhibition is sufficient to block the expression of this metastasis gene.

Furthermore, we identified B-Raf^V600E targets of ill-defined or unknown function. For example, KIAA1199 was significantly downregulated by B-Raf or MEK inhibition. KIAA1199 emerges as a marker of poor prognosis in several tumor entities, including colorectal carcinoma, but has never been linked to BRAF mutations before (39, 40, 55). Its activity as a promoter of cell migration in noncolorectal carcinoma cell lines (55) fits well to the strong reduction of invasive behavior of Colo-205 cells with inhibition or knockdown of B-Raf^V600E.

Thus, the identification of the mechanisms and cooperating signaling pathways by which oncogenic B-Raf suppresses differentiation and cell–cell adhesion represents a future task. Collectively, we show that B-Raf^V600E inhibition reduces the expression of transcripts associated with negative diagnostic value and increases those forecasting a more favorable outcome. Moreover, the datasets derived from our dynamic isogenic experimental systems complement gene-expression studies based on the comparison of clinical and therefore heterogeneous tumor samples (1, 2, 9).

Our discovery that B-Raf^V600E inhibition improves cell–cell adhesion by restoring AJs and TJs, impairs migration and invasion, and increases hallmarks of epithelial differentiation identifies a novel facet of inhibitors targeting the B-Raf/MEK/ERK pathway (Fig. 6B). Importantly, trametinib, which is part of combination therapies trialed in colorectal carcinoma, yields the same effects as B-Raf inhibitors in our 3D cultures and xenograft experiments. This implicates MEK as a key mediator of the differentiation suppression exerted by B-Raf^V600E. Although clearly beyond the scope of the present study, it would be interesting to assess the sensitivity of our isogenic cell line models toward conventional chemotherapy compounds, because differentiation status is often correlated with drug responses. Taken together, our study highlights the possibility for a novel therapeutic rationale for using B-Raf/MEK/ERK pathway inhibitors in colorectal carcinoma and potentially other carcinoma entities. Thus, these drugs could potentially limit metastasis, most likely, in combination with conventional chemotherapy or other targeted therapy compounds such as EGFR or dual PI3K/mTOR inhibitors (4, 17–20). Such combinations, which target various hallmarks of cancer, are currently in clinical trials for colorectal carcinoma. It will be interesting to see whether a differentiation-promoting effect can be observed in these clinical settings.

No potential conflicts of interest were disclosed.

Conception and design: R. Herr, M.A. Olayioye, H. Busch, M. Boerries, T. Brummer

Development of methodology: R. Herr, H. Andrlová, Y. Möller, N. Bittermann, S. Lassmann, T. Brummer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Herr, M. Köhler, H. Andrlová, F. Weinberg, Y. Möller, S. Halbach, M. Klose, N. Bittermann, S. Kowar, R. Zeiser, S. Lassmann, T. Brummer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Herr, M. Köhler, H. Andrlová, F. Weinberg, Y. Möller, S. Halbach, N. Bittermann, R. Zeiser, M.A. Olayioye, S. Lassmann, H. Busch, M. Boerries, T. Brummer

Writing, review, and/or revision of the manuscript: R. Herr, M. Köhler, F. Weinberg, S. Halbach, R. Zeiser, M.A. Olayioye, S. Lassmann, H. Busch, M. Boerries, T. Brummer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Mastroianni, S. Kowar, H. Busch, M. Boerries, T. Brummer

Study supervision: T. Brummer

Other (evaluation and interpretation of histology): L. Lutz

The authors thank Martin Werner for supporting histologic and immunohistochemical xenograft analyses, and Paul Timpson for discussion.

This work was funded by the DFG via the Collaborative Research Centre 850 (projects B4, C5, C6, Z1, and Z2), the Emmy-Noether-Program (T. Brummer) and SPP1395/InKomBio (M. Boerries, H. Busch, and M. Köhler). The Excellence Initiative of the German Federal and State Governments supports T. Brummer (EXC 294 BIOSS), and M. Köhler and S. Halbach (GSC-4; SGBM). This study was also supported by the BMBF through e:Bio 0316184D.

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.