Noncanonical TGFβ Pathway Relieves the Blockade of IL1β/TGFβ-Mediated Crosstalk between Tumor and Stroma: TGFBR1 and TAK1 Inhibition in Colorectal Cancer

It is well known that the communication between tumor cells and CAFs has far-reaching consequences for the progression and spread of cancer. However, during early stages of tumorigenesis, the crosstalk between the incipient neoplasia and the surrounding normal environment is controlled by the tissue-specific homeostatic balance (1). A finely tuned repertoire of cytokines and their receptors maintains this balance, which is particularly important in the colonic and rectal epithelia, where newly transformed tumor cells are in close contact with pericryptal colorectal fibroblasts. Once the balance is lost due to selective pressures, pericryptal or normal colorectal fibroblasts (NCF) are modified to become CAFs (2) in a process in which the microenvironment evolves to support tumorigenesis (3). Once altered, CAFs can help hiding tumor cells from the immune system or protect them from chemotherapeutic drugs. In these processes, a dialogue between cytokines/growth factors and their corresponding receptors is established, in which TGFβ plays a fundamental and paradoxical role, because it is one of the most important factors for boosting the prometastatic capacity of tumors (4). Other molecular dialogues are established between tumor and stroma sometimes in a context-dependent manner (5–7).

Targeting of such bidirectional crosstalk could form the basis of a strategy for fighting cancer. In fact, a lot of information and experience has been obtained from inflammatory diseases like arthritis, Crohn disease and asthma, where hindering the binding of ligand-receptor is a well-established approach (8). Pharmacologic targeting of TGFβ1 enhances the intratumoral spread of chemotherapy and reverts the stroma's activated status (9). Similarly, inhibition of Sonic Hedgehog signaling facilitates drug delivery in PDAC (10). Another archetypal example is stromal CXCL12 and its receptor CXCR4 expressed in tumor cells, where inhibitor AMD3465 blocks the action of the chemokine and consequently decreases AKT activation (11). Dozens of similar examples have identified, coinciding with the explosion of development of targeted therapies.

However, the main problem is that tumors still adapt quickly and develop resistant phenotypes because redundant signaling mechanisms overcome such blocking strategies. While most tumor types are kinase-addicted, the blocking of such ligand-dependent kinases quite often results in the activation of the pathway, although not by the reactivation of the kinase (12). This arises from the redundancy in signaling cascades in tumor cells (13), because different kinases operate in the same pathway or have a similar function, and different ligands act on the same receptor. Furthermore, soluble stromal or tumoral receptors can rescue tumor cells from therapies that hinder the crosstalk between stromal and malignant cells (14). For this reason, the design of therapeutic strategies must take into consideration relevant “hubs” of intracellular signaling. This is a difficult task, because these strategies are often accompanied by very severe side effects.

The TGFβ-Activated Kinase 1 (TAK1) protein, encoded by MAP3K7 gene, is a serine/threonine kinase that mediates signaling by IL1β (15), TNFα (16), LPS (17), TLR (18), although it was discovered because of its role in the noncanonical TGFβ pathway (19). In fact, TAK1 is a proinflammatory effector that contributes to the IKK complex activation and consequently activates NFκβ pathway (20). However, it can also activate other downstream MAPKs like JNK (21), P38, and ERK (22). TAK1 is also involved in tumorigenesis, and its inhibition induces apoptosis (23) and chemosensitivity (24). All these observations suggest that TAK1 is a point of convergence for signaling pathways that are activated by a broad spectrum of stimuli. In our particular context, TAK1 plays a central role in the proinflammatory response of the microenvironment.

Here we report that cotargeting of TAK1 and TGFBR1 in fibroblasts and CAFs is an interesting approach for avoiding the secretion of protective soluble factors from activated fibroblasts through crosstalk with tumor cells in colorectal cancer. In turn, malignant cells are rendered more sensitive to chemotherapy due to the absence of the protective environment normally provided by fibroblasts. Interestingly, in vivo experiments of orthotopic patient-derived xenografts and splenic cell line injection show that the combined inhibition had a deep impact in the dissemination and settlement of distant metastases, inhibiting the recruitment of a supportive stroma.

Primary NCFs, normal hepatic fibroblasts (NHF), and CAFs used in the study were isolated in our laboratory from surgical specimens after explicit patient's written consent had been obtained. NCFs and NHFs were isolated from at least 5 cm from the surgical margin of excised primary or metastatic tumors. Colorectal cell lines used are DLD1 (MSI-CMS1), HCT116 (MSI-CMS4), and HT29 (MSS-CMS3). Oxaliplatin-resistant derivatives (HTOXAR3 and DLDOXAR) were kindly provided by Dr. Martínez-Balibrea (Institut Germans Tries i Pujol, Badalona, Spain). All cells were cultured in DMEMF12 medium supplemented with 10% FBS, glutamine, and penicillin/streptomycin (Gibco) at 37°C in 5% CO₂. Cells were periodically tested for Mycoplasma contamination and were authenticated by short tandem repeat profiling.

Eight different isolated NCFs (clinical details in Supplementary Table online) were cocultured with DLD1 cells (coming from a stage III colorectal adenocarcinoma) in 3-μm pore size Transwell inserts for 96 hours in 2% FBS DMEMF12. Monocultures of same NCFs and DLD1 cells in the same culture conditions were used as corresponding controls.

IL1β and TGFβ1 were purchased from PeproTech. Neutralizing IL1β antibody was purchased from R&D Systems (reference AB-201-NA), diluted in water; TGFBR1 inhibitor III-CAS 356559-13-2 was purchased from Calbiochem, (5Z)-7-oxozeaenol (OXO) was purchased from Tocris, both diluted in DMSO. Oxaliplatin (L-OHP) was diluted in water and 5-fluorouracyl in saline buffer. For the in vivo studies, TGFBR1 inhibitor used was galunisertib, kindly provided by Eli Lilly & Co and was freshly suspended in 1% carboxymethylcellulose sodium salt, 0.5% SDS, 0.085% povidone, and 0.05% antifoam Y-30 emulsion.

RNA from eight paired primary NCFs, tumor cell–cocultured NCFs, paired CAFs, DLD1 monocultures, and cocultures of DLD1 cells with the corresponding eight NCFs, was extracted as described elsewhere (25).

Total mRNA was hybridized in an Affymetrix GeneChip Human Gene 1.0 ST Array. All computations and statistical analyses were performed using the R language. Microarray data were processed using the justRMA function in the simpleaffy package (26). The resulting data were used to look for genes that were differentially expressed between groups (monocultures vs. cocultures) using the Significance Analysis of Microarrays (SAM) test, available in the samr package (27). To obtain a reduced list of genes, we considered those with a false discovery rate (FDR) < 0.05 and a >2-fold change.

Gene-set enrichment analysis (GSEA; ref. 28) was carried out on a preranked list of genes determined by the SAM d-statistic to be differentially expressed in cocultured and monocultured NCFs and tumor cells (blue and red phenotypes, respectively, in GSEA plots). We used the H.HALLMARKS; C2.CP.ALL and C5.BP gene sets.

The WST-1 assay was performed following the manufacturer's instructions. Previously, six replicates of 2,000 cells/well (tumor cell lines) had been plated in 96-well plates and allowed to attach, then grown in DMEM/F12 at 37°C overnight. Next day, CM from treated cocultures were added accordingly. Cells were incubated for 5 days, treated with a range of doses against oxaliplatin (L-OHP), and the drug effect on tumor cells was calculated by normalizing the number of cells after 5 days of continuous treatment relative to the maximum number of cells in each CM. WST-1 assays were also tested directly on cells cocultured in 24-well Transwell inserts, seeding tumor cells in the bottom well and fibroblasts (either NCFs, NHFs, or CAFs) in the insert. Treatments were added when cells were attached to the culture surfaces and viability was assayed for 5 days.

Acquisition of CAF markers was assessed in 5-day cocultures by flow cytometry using anti-human CD90-APC (Miltenyi Biotec) to discriminate between fibroblasts (positive cells) from tumor cells (negative), and mouse anti-human CD49b-FITC (integrin A2) and mouse anti-human CD127-BV421 (IL7R). NCF and CAF monocultures were used as controls.

The patient gave written consent to donate a piece of tumor to generate the orthoxenograft model. The Institutional Ethics Committees approved the study protocol. Experimental design was approved by the IDIBELL animal facility committee (AAALAC–Unit 1155). All experiments were performed in accordance with the guidelines for Ethical Conduct in the Care and Use of Animals as stated in the International Guiding Principles for Biomedical Research Involving Animals, developed by the Council for International Organizations of Medical Sciences.

A patient-derived orthotopic colorectal cancer xenograft model (PDOX), generated from a locoregional relapse 5 years after completing adjuvant treatment, was orthotopically implanted into the cecum of Crl:NU-Foxn1nu mice. PDOXs were inspected every other day and monitored for the presence of breathing and behavioral problems. Tumor volume was monitored every 2 days by palpation. After reaching approximately 200 mm³, animals were randomized into control arm (n = 9, FUOX treatment, 5-fluorouracyl plus oxaliplatin) and test arm (n = 9, FUOX plus galunisertib plus OXO) and treated for 30 days. Metastatic dissemination was determined in macroscopic and microscopic examination (hematoxylin–eosin and Masson trichrome) in peritoneum, liver, lung, adrenal glands, kidneys, diaphragm, and brain. Survival was recorded following strict animal welfare criteria, supervised by a veterinarian, and normally did not coincide with the death of the animal but with criteria of euthanasia as a result of veterinary supervision.

The capacity for blocking the supportive stroma to promote distant metastases was assessed by injecting 8 × 10⁵ cells (HT29 and its L-OHP–resistant derivative HTOXAR3, and DLD1 and its L-OHP–resistant derivative DLDOXAR) into the spleen of Crl:NU-Foxn1nu mice. Two days after injection, the spleen was removed and a period of one week was allowed to pass for micrometastases to form. Next, animals were randomized into L-OHP (control) or L-OHP plus galunisertib plus OXO groups. The treatment lasted for 15 days, and after 30 days the animals were sacrificed to determine the presence of visceral and peritoneal metastases as described above.

To elucidate the contribution of tumor–stroma interactions for tumor growth and chemoresistance, we used a normal colonic fibroblast (NCF) coculture model from various patients affected by colorectal cancer and tumor cells (CRC DLD1 cell line). After 5 days of indirect contact through the mesh of a 3-μm Transwell insert, mRNA was obtained from each cell type. mRNA from monocultures of both types was also obtained for use as a reference for transcriptomic profiling. We hybridized mRNA in Affymetrix GeneChip Human Gene 1.0 ST expression arrays. Using a >2 (or −2) fold change (FC) and a false discovery rate (FDR) of q < 0.01, we obtained 886 differentially expressed genes (DEG) in the cocultivated NCFs with respect to monocultures (Supplementary Fig. S1; Supplementary Table S1). When we performed the same analysis on cocultured tumor cells under the same conditions of stringency, we obtained 1,066 DEGs in cocultured tumor cells compared with monocultured controls (Supplementary Table S2). In both cases, the large number of DEGs with low FDR values gives an idea of the profound transcriptomic changes induced by the crosstalk between these cell types.

To evaluate the biological importance of the coordinated variation in the change of expression observed in the two cocultured cell types, we performed GSEA on the complete list of ranked genes. The most notable pathways and biological processes enriched in cocultured fibroblasts are depicted in Fig. 1A and Supplementary Table S3 (Tumor cell GSEA in Supplementary Fig. S2). In addition, GSEA results revealed that the cocultured stroma and tumor cells acquired a transcriptomic profile enriched in pathways and biological processes, with considerable overlap with those of rectal cancer patients classified as nonresponders to neoadjuvant treatment according to our group's previous results obtained from microdissected endoscopic pretreatment biopsies (29).

We set out to determine whether the transcriptional changes suffered by the NCF fibroblasts when in paracrine communication with tumor cells were associated with the acquisition of a transcriptional pattern similar to that of their corresponding CAFs obtained from the same patient. In this way, this interaction could be used as a model for studying the participation and recruitment of fibroblasts during the early stages of tumor expansion and growth. To check this, we selected genes with a > 2-fold (or <−2-fold) change between NCFs and cocultured NCFs, with a simultaneous >2-fold (or <−2-fold) between NCF and paired CAFs. As observed in Fig. 1B, NCFs cocultured with tumor cells acquired a transcriptional profile largely similar to that of their corresponding CAFs. Two distinct clusters, CAFs and co-NCF in one, and monocultured NCF in the other, could be clearly identified. We also performed a principal component analysis (PCA) analysis in which genes of the first component generated two clusters, NCF in one, cocultured NCF and CAFs in the other, with a single misclassified sample (Supplementary Fig. S3).

We experimentally corroborated the overexpression of some of these markers in NCFs cocultured with tumor cells, in particular cell surface molecules such as CD49B (ITGA2) and CD127 (IL7R; Fig. 1C).

These results suggest that cocultured NCFs acquired a CAF-like phenotype.

Using single-sample GSEA (ssGSEA), we determined that DEGs in cocultured tumor cells conferred a bad prognosis on two different datasets of not treated patients with colorectal cancer [GSE33113, 90 patients stage II, HR = 2.78 (95% CI 1–7.76), Cox P = 0.05; GSE14333, 139 patients stages I–III, HR = 3.05 (95% CI 1.13–8.7), Cox P = 0.028, corrected for stage; Fig. 1D).

In summary, the coculture of tumor cells and NCFs produces radical transcriptional changes in both cell types. Consequently, the acquisition of these changes has an important repercussion on the prognosis of patients affected by colorectal cancer.

Our aim was to define the main mediating factors of the paracrine communication between NCF and epithelial tumor cells. For this purpose, an analysis of protein–protein interaction (PPI) was carried out, selecting from the transcriptomic data of cocultured cells (> 2-fold change and FDR q < 0.1 relative to corresponding monocultured cells) those mRNAs that encode proteins secreted in fibroblasts with a transcript that is in turn overexpressed and encodes a membrane receptor in epithelial cells, and vice versa. Detailed information is presented in the Supplementary Data and Supplementary Results (Supplementary Figs. S4–S6; Supplementary Table S4).

After integrating our transcriptomic/PPI data, we established our working hypothesis that cytokines released by NCFs stimulate tumor cells to secrete IL1β and TGFβ1. These soluble factors are responsible for the activation of NCF to convert them into CAFs. In turn, activated NCFs produce large amounts of IL1β, which potentiates the pathway, in an autocrine manner, as well as a large number of IL1β/TGFβ1 target proteins, mainly chemokines, cytokines, EGFR ligands and PAI-1 (SERPINE1), soluble factors with high affinity for receptors IL6R, IL11RA, (via JAK-STAT), and CCR6, CXCR3, EGFR, CCR2, FGFR, and cMET (via PI3K-AKT). These are all pathways involved in chemoresistance in colorectal cancer.

Transwell coculture experiments of NCFs with colorectal cancer cells were performed to evaluate the effects of soluble factors produced by the heterotypic crosstalk (Fig. 2A, left). Cocultures were intervened with agents blocking the action of soluble triggers according to our hypothesis (neutralizing IL1β antibody and a TGFBR1 inhibitor, or in combination). CM of these cocultures were used to test proliferation (Fig. 2A, second panel), migration (Fig. 2A, third panel), and sensitivity to drugs (Fig. 2A, fourth panel) in cells not previously exposed under these conditions. In the cells treated with CM in the presence of the two blocking agents, we observed an increase in proliferation, a decrease in migration and greater sensitivity to oxaliplatin (L-OHP), although the latter finding seems to be due mainly to the action of the neutralizing IL1β antibody.

DLD1 proliferation and survival experiments were also performed directly on the Transwell inserts, coculturing tumor cells with NCFs or CAFs. However, as shown in Fig. 2B, we did not find any benefit in terms of proliferation from any of the treatments (Fig. 2B, left panel solid bars), or from the addition of oxaliplatin at different doses (5, 8, 10, and 40 μmol/L; Fig. 2B, left, dashed bars). We only observed a trend toward increased survival when using the neutralizing IL1β antibody in cocultures with NCFs. Likewise, the cocultures with CAFs did not indicate any benefit of the various treatments in terms of oxaliplatin sensitivity (Fig. 2B, right).

Then, we wanted to check the signaling pathways altered by the blockade of IL1β and TGFBR1. As illustrated in Fig. 2C, tumor cells displayed a progressive decrease in levels of STAT3 activation. However, the levels of AKT phosphorylation changed in the opposite direction, increasing as we added the blocking agents to alter the crosstalk between cell types. We observed a similar compensatory feedback loop in fibroblasts, switching from the activation of P38 and SMAD2/3 pathways in control cocultures to the activation of NFκβ pathway as we added IL1β and TGFBR1 blocking agents (Fig. 2C, right). Next, we attempted to characterize the soluble factors present in the CM of 5-day cocultures of cells under the various treatment conditions to understand the results obtained from the Western blot analyses. To this end, we used a cytokine array, interrogating 174 soluble factors. Eighty of the 174 soluble factors appeared to be modulated by the combination of treatments, 31 decreasing in concentration with the addition of the blocking agents, particularly when the combination of the neutralizing IL1β antibody plus the TGFBR1 inhibitor was used (Fig. 3). Conversely, the concentration of 49 cytokines increased with the change in crosstalk. Such cytokine and growth factor switches might be responsible for the differential signaling observed in fibroblasts and the sustained AKT activation in tumor cells.

These results suggest that as we block the crosstalk between both cell types, compensatory mechanisms are produced where NFκβ pathway activation arises in fibroblasts through the noncanonical TGFβ pathway and TAK1, producing a cytokine turnover, which would exert redundant functions to those observed in the control cocultures.

The results obtained to this point suggested that an alternative mechanism was activating NFκβ. The TGFBR1 inhibitor was inhibiting the canonical TGFβ1 pathway and the neutralizing IL1β antibody that controls the IL1β-mediated inflammatory processes. In any case, NFκβ pathway was activated when the two main ligands responsible for fibroblast activation in our model were blocked. Thus, we reasoned that the inhibition of a central signaling node converging from the IL1β pathway and the noncanonical TGFβ1 pathway, the TAK1 kinase (MAP3K7) might be an interesting approach. Therefore, the simultaneous blockade of TAK1 and TGFBR1 would allow us to inhibit signaling mediated by IL1β and TGFβ1 (via a canonical pathway involving a TGFBR1 inhibitor, and a noncanonical pathway involving a TAK1 inhibitor) and counteract the activation of NFĸβ, maintaining the inhibition of the MAPK- and SMAD-dependent pathways.

We first examined the effect of TAK1 inhibitor OXO on the proliferation of cells in monoculture (Fig. 4A). OXO significantly reduced the proliferation of CAFs, while no statistically significant effect was observed in tumor cells. We assayed the effect of coculture CM, either control, OXO-treated or (OXO + TGFBR1 inhibitor)-treated, on the sensitivity of tumor cells to L-OHP. The CM from OXO + TGFBR1 inhibitor–treated cocultures clearly conferred less protection from L-OHP than cocultures with control-CM (Fig. 4B). In addition, the combined treatment of OXO + TGFBR1 inhibitor sensitized tumor cells to L-OHP in all cell lines tested in cocultures with either NCFs, NHFs, or CAFs (Fig. 4C; dashed colored bars). The combination also diminished the degree of proliferation of tumor cells in the absence of L-OHP (solid bars in Fig. 4C). Moreover, OXO inhibited the activation of P38 and P65 in CAFs after short (Fig. 4D left) or sustained (Fig. 4D middle) IL1β stimulation. The combination of OXO + TGFBR1 inhibitor was the most effective at inhibiting P38 and P65 phosphorylation in (IL1β + TGFβ1)-treated NCFs (Fig. 4D, right), and rendered those treated NCFs without acquired CAF markers, like FAP and IL7R (Fig. 4E). Nevertheless, when we assayed signaling directly in cells in coculture, we observed sustained AKT activation in tumor cells (Fig. 4F, middle), although we did succeed in inhibiting the activation of P65 and P38 in cocultured fibroblasts (Fig. 4E, right). Likewise, the concentration of some of the cytokines that we had previously found to be increased after 5 days in culture, such as IL11 and IL6, were significantly reduced with the new treatment strategy (Fig. 4G). Even so, the sustained activation of AKT seemed to indicate a new cytokine turnover. A second cytokine array revealed that some cytokines and growth factors were secreted into the culture medium with an opposite tendency to that previously noted with the treatments with neutralizing IL1β antibody and the TGFBR1 inhibitor (Fig. 5).

Thus, the combination treatment appeared to counteract the NFκβ activation in fibroblasts, reducing in turn the secretion of IL6, IL11 (ELISA), sensitizing tumor cells to oxaliplatin.

Once we had defined best in vitro treatment combination of those tested, we checked the in vivo effect of TAK1 and TGFBR1 inhibition in xenograft models on hindering the crosstalk between fibroblasts and tumor cells. First, we checked the efficacy of TAK1 inhibitor plus TGFBR1 inhibitor (galunisertib for all the in vivo experiments) in an orthotopic patient-derived xenograft model from a FOLFOX-resistant colorectal tumor with papillary histology (Fig. 6A). At the end of the experiment, three of the 9 animals were still alive with the FUOX + OXO + galunisertib treatment while only one of the animals was alive in the FUOX-treated arm. However, no statistically significant differences in survival were determined (Fig. 6B). However, there were differences in relation to the number of animals that had distant dissemination in the liver and peritoneum, significantly less in the group treated with OXO + galunisertib (Fig. 6C), as well as a trend (P = 0.07) toward a lower peritoneal tumor burden (Fig. 6D).

We also checked the inhibition of the metastatic potential in intrasplenic injection models of colorectal cancer cell lines (HT29 and its oxaliplatin-resistant derivative HTOXAR3, and DLD1 and its oxaliplatin-resistant derivative DLDOXAR). Following the design illustrated in Fig. 6E, animals were euthanized 30 days after treatment withdrawal. Macroscopic and microscopic examination revealed that animals receiving oxaliplatin plus OXO and galunisertib exhibited lower metastatic spreading than animals receiving oxaliplatin alone, the differences being statistically significant in HT29, DLD1, and DLDOXAR cells (Fig. 6F). Moreover, the peritoneal tumor burden was also lower in OXO plus galunisertib-treated animals in DLD1 and DLDOXAR cells (Fig. 6G).

Interestingly, when we characterized the xenografts grown in mice (particularly the PDX model and the HT29 and HTOXAR models), we noticed that mice treated with the combination of OXO plus galunisertib presented tumors with less infiltration by CAFs and less collagen and extracellular matrix deposition, as shown in Fig. 6H–J. Tumors grown in DLD1 and DLDOXAR experiments recruited few fibroblasts.

In summary, the combination of TAK1 and TGFBR1 inhibition diminishes the metastatic settlement of tumor cells and the recruitment of resident fibroblasts in the TME.

The crosstalk between malignant cells and their environment, particularly CAFs in cancers with a high tumor–stroma ratio, is a fundamental event in the development of the tumorigenic process, not only for molecular signaling (30), but also for providing the necessary fuel for a demanding cancer (31, 32). Here we report the combined use of a TAK1 inhibitor plus a TGFBR1 inhibitor to avoid the IL1β and TGFβ1-mediated conversion of resident fibroblasts into CAFs, either colonic or hepatic, by inhibiting their protumorigenic protective soluble factor secretion, decreasing extracellular matrix deposition and, consequently, rendering tumor cells more sensitive to chemotherapy and inhibiting the metastatic potential and CAFs recruitment.

Tumor cells use different soluble factors and cell adhesion molecules to recruit a favorable stroma to grow and evade chemotherapy (33, 34). Our protein–protein interaction analysis identifies tumoral IL1β and TGFβ1 as the most significant soluble factors responsible for activating resident fibroblasts and their conversion into CAFs. These, in turn, trigger a cascade of molecular signaling initially mediated by P38 and SMADs that induces a massive secretion of activating ligands of JAK/STAT and AKT in tumor cells. The blocking of IL1β with a neutralizing antibody and the TGFBR1 by means of a specific inhibitor, the first hypothesis we tested, is relieved by a SMAD-independent and TAK1-mediated transient JNK activation and a sustained NFκβ-mediated switch in the cytokine repertoire, while maintaining the silencing of the SMAD-dependent pathway (according to the nonphosphorylation of SMAD2 and 3, and the high levels of BMP-4 and BMP-7, which are markers of inactivation of the canonical TGFβ pathway). In fact, the crosstalk between effector kinases has been described in various cell models (35, 36), particularly in the case of P38 and JNK (37). In addition, in the absence of TGFBR1, either by genetic deletion of pharmacologic inhibition, TGFBR2 can activate the noncanonical TGFβ pathway (38). This alternative mechanism seems to be cell-dependent, being observed in cells expressing high levels of TGFBR2, such as fibroblasts. Thus, TAK1-mediated activation of the noncanonical TGFβ pathway induces a change in the soluble factors, causing, in turn, a decrease in the activation of the JAK/STAT pathway in tumor cells. However, malignant cells sustain the AKT activation, compensating for the blocking of molecular signaling in other connected pathways, like STAT3. In fact, STAT3 inhibition has been reported to be compensated by AKT as a stress response in peripartum cardiomyopathy (39). In addition, it has been previously proposed that this AKT feedback loop might be the consequence of the capacity of different ligands to activate the same receptors tyrosine kinase (RTK). RTK redundancy has already been described (12, 40, 41), because tumor cells are addicted to cytokine/chemokine–mediated RTK signaling. In fact, some of the cytokines/chemokines that appear in the CM after blocking cocultures with the neutralizing antibody plus TGFBR1 inhibitor are molecules that activate the PI3KCA/AKT pathway. Thus, such redundancy raises the problem of acting therapeutically on convergence hubs from different combinations of RTK and RTK/cytokine pairs, rather than acting directly on ligands and membrane receptors. In this way, we may circumvent the addiction of tumor cells to RTK signaling and the plasticity of stromal cells to produce a wide repertoire of soluble factors, depending on the microenvironmental conditions. Nevertheless, toxicity arises in many cases adopting such an approach. In this regard, TAK1 is a MAPK activated by proinflammatory cytokines, chemokines, and the growth factors IL1β and TGFβ1, among others. Although TGFβ1 operates in the SMAD-dependent canonical pathway, under several circumstances it might also signal through TAK1 activating ERK (38) NFκβ transcriptional program (42), as mentioned above. However, in recent years, the role of TAK1 in cancer has been circumscribed in malignant cells (23, 43). In the context of the crosstalk between fibroblasts and tumor cells, either at the primary site or in metastasis, TAK1 is a convergence hub for IL1β and noncanonical TGFβ pathways. Thus, the second hypothesis we tested was the blocking of both IL1β-mediated and TGFβ SMAD–independent responses with a TAK1 inhibitor, and the TGFβ SMAD-dependent response with the TGFBR1 inhibitor. Very recently, it was reported that the inhibition of stromal TAK1 diminishes the senescence-associated secretory phenotype (SASP) in vitro and in vivo (44). In contrast, from our coculture experiments between fibroblasts and tumor cells, inhibiting only TAK1 was not sufficient to sensitize as many tumor cells to chemotherapy as possible or to decrease the levels of protumoral cytokines to inhibit the relevant survival pathways in tumor cells. The main reason is that the TGFβ−canonical pathway can maintain the crosstalk between fibroblasts and tumor cells. The simultaneous blockade of TAK1 and TGFBR1 manages to inhibit proinflammatory signals mediated by IL1β- and TGFβ-dependent processes, of the canonical SMAD-dependent and noncanonical SMAD-independent types. Consequently, the levels of proinflammatory cytokines were reduced along with the in vitro viability of colorectal cell lines, irrespective of using NCFs or NHFs (as coculture models of the initial steps of tumorigenesis in a primary site or in a liver metastasis) or CAFs, either with chemotherapy (L-OHP) or without. Furthermore, this combination of inhibitors seems to alter the acquisition of the proinflammatory traits of the CAFs in coculture, as determined by the inhibition of the expression of FAP and IL7R, and the maintenance of a myofibroblast marker like αSMA. Therefore, the combination of TAK1 inhibitor plus TGFBR1 inhibitor was tested in vivo, where we observed a significant decrease in metastatic growth in a patient-derived xenograft (PDX) model and in various models of oxaliplatin-sensitive, resistant derivative pairs. Moreover, when we analyzed the generated tumors histologically and IHC, we found significant alteration of the stroma in animals receiving the combination of both inhibitors. The low level of Masson staining observed in these tumors might be an indication of TGFβ pathway impairment, while the decrease in fibronectin staining is a surrogate marker for NFκβ inhibition. In some cases, the tumor size was not reduced but the stroma recruitment. This could be an opportunity to treat tumors with chemotherapy without the protection conferred by the stroma at that precise point in the tumorigenic process, rather than only at the beginning of the treatment, when they have not yet been deprived of this protective stroma.

TAK1 inhibition induces apoptosis in Kras-dependent colorectal cancer cell lines (23). In our experiments, all the cell lines used were Kras-independent. Although some are MSI, we consider that this fact does not interfere with the results obtained because the absence of an immune system, both in the in vitro and in vivo experiments, does not recapitulate the supposed differential immunosuppressive response of MSI tumors. It has also been shown that tumors of subtypes with a better prognosis, such as CMS1 (MSI) if they present high expression of CAFs genes, are clustered as tumors with a very poor prognosis (45). Also, HCT116 (MSI) is cataloged as CMS4 (46). Thus, our results are attributable to the hampering of the crosstalk rather than to the direct effect of the TAK1 inhibitor on the tumor cells.

In conclusion, the blockade of the crosstalk between fibroblasts and colorectal tumor cells by means of a TAK1 inhibitor and a TGFBR1 inhibitor hinders the acquisition of protumoral and proinflammatory traits of fibroblasts and consequently renders tumor cells more sensitive to chemotherapy. In addition, the in vivo combination of both inhibitors impaired metastatic growth and the recruitment of stroma by tumor cells.

G. Capella has ownership interests (including patents) at, is a consultant/advisory board member for, and reports receiving commercial research support from VCN Biosciences. V. Moreno is a consultant/advisory board member for Ferrer, and reports receiving commercial research support from Universal DX and Bioiberica. A. Villanueva has ownership interests (including patents) at Xenopat SL. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R. Sanz-Pamplona, D.G. Molleví

Development of methodology: N.G. Díaz-Maroto, R. Sanz-Pamplona, F.J. Cimas, E. García, S. Gonçalves-Ribeiro, N. Albert, A. Villanueva, D.G. Molleví

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.G. Díaz-Maroto, M. Berdiel-Acer, G. Garcia-Vicién, V. Moreno, R. Salazar, A. Villanueva, D.G. Molleví

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Sanz-Pamplona, E. García, V. Moreno, D.G. Molleví

Writing, review, and/or revision of the manuscript: N.G. Díaz-Maroto, R. Sanz-Pamplona, F.J. Cimas, V. Moreno, R. Salazar, D.G. Molleví

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. García

Study supervision: D.G. Molleví

Other (partial funding): G. Capella

This work has been supported by grant PI10/1604 from the Fondo de Investigaciones Sanitarias of the Spanish Government, Fondo Europeo de Desarrollo Regional (FEDER) “Una manera de hacer Europa”/“A way of shaping Europe,” and AGAUR grant number SGR771. This work has been supported by grants PI10/1604 and PI15/0232. In addition, D.G. Molleví and E. García are recipients of the SLT002/16/00399 grant, funded by the Department of Health of the Generalitat de Catalunya by the call “Acció Instrumental d'incorporació de Científics i Tecnòlegs.”

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.