ERK1/2 Signaling Induces Upregulation of ANGPT2 and CXCR4 to Mediate Liver Metastasis in Colon Cancer

Progression from normal mucosa to carcinoma in colorectal cancer is driven by a sequential order of well-defined genetic alterations that affect the Wnt, MAPK, PI3K, and TGFβ signaling pathways (1). Alterations in MAPK signaling occur early during primary colorectal cancer development (1). Activating mutations in KRAS, NRAS, and BRAF, which are part of the RAS-ERK1/2 MAPK signaling cascade, are detected in nearly 50% of colorectal cancer cases (2–4). In addition, genetic variations in several members of MAPK signaling pathways are associated with the risk of developing colorectal cancer, as well as with overall survival (OS) after diagnosis with colorectal cancer (5).

However, recent clinical data emphasize the importance of MAPK signaling not only in primary colorectal cancer development but also in distant tissue colonization (6). OS after diagnosis of metastatic disease is shortest for patients with tumors that present alterations in the RAS pathway (7). In addition, having mutations in KRAS is associated with a higher risk of recurrence in patients after surgical resection of liver metastases from colorectal cancer (8–10). The presence of mutations in both KRAS and BRAF genes can influence the metastatic pattern of colorectal cancer, as patients with colorectal cancer who have a KRAS-mutant tumors have as well an increased risk of lung recurrence after primary tumor resection (11), whereas those with a BRAF mutations tend to develop metastasis to peritoneum and distant lymph nodes (12). Extensive genomic profiling has detected a high level of concordance in mutational status between colorectal cancer primary tumors and matched metastases (7, 13, 14), thus suggesting that metastatic development is not driven by the acquisition of additional mutational events. How RAS–ERK signaling regulates colorectal cancer metastasis, and how it determines metastatic patterns is still unknown.

The colorectal cancer cell lines were maintained in 5% CO₂ at 37°C in DMEM (Gibco) supplemented with glutamine (0.29 mg/mL), penicillin (100 U/mL), streptomycin (0.1 mg/mL), and either 5% FBS (for the cell lines SW620-P and SW620-LiM2 derivatives from SW620, SW480, and Colo26) or 10% FBS (for the HCT116 cell line); all supplements were purchased from Biological Industries. All cell lines were purchased from the ATCC but Colo26 was gift from the Batlle lab. All cell lines were authenticated for KRAS/BRAF mutations and tested routinely (biweekly) for Mycoplasma by PCR. Cell lines were not passaged more than 50 times.

Cells (2.5 × 10⁶) were seeded and treated for 24 hours with the MEK1/2 inhibitors U0126 (10 μmol/L, Cell Signaling Technology) or PD0325901 (100 nmol/L, Tocris) in DMEM supplemented with 0.1% of FBS.

293T cells were used for lentiviral production. Lentiviral vectors expressing short hairpin (shRNA) against human CXCR4, ANGPT2, ETV4, or ETV5 from Mission shRNA Library were purchased from Sigma-Aldrich (see sequences in Supplementary Material and Methods).

Retroviruses were produced using 293T cells as described previously (15).

The Ethical Committee of Animal Experimentation of the Government of Catalonia approved all animal work (protocol number 9317). Intrasplenic injections were done as previously reported, and liver metastasis development was followed twice a week by bioluminescence imaging using the IVIS-200 imaging system from Xenogen (Living Image 2.60.1 software; ref. 15). Treatment with IL10 antibody or IgG was initiated 7 days postimplantation of the cells and mice were treated three times per week with 1.5 μg of antibody. For the experiment using mouse colorectal cancer organoids, treatment with IL10 antibody or IgG was initiated 3 days after implantation of the cells and mice were treated three times per week with 5 μg of antibody.

Protein extracts obtained from whole cell lysates (40 μg) were fractionated in SDS-PAGE gels, transferred onto Immobilion-P (Millipore) membranes, and subjected to immunoblot analysis (antibody list in Supplementary Data; ref. 15).

Real-time qPCR was performed using TaqMan Gene Expression Assay (list in Supplementary Data; ref. 15).

Tissues were dissected, fixed in 10% buffered formalin (Sigma) and embedded in paraffin. Sections (2–3 μm thick) were stained with hematoxylin and eosin. Antibodies listed in Supplementary Materials.

Five million cells were seeded in 6-cm diameter plates. The medium was changed after 24 hours, and plates were incubated with 3 mL of media for a further 24 hours. Subsequently, supernatants were collected and analyzed by ELISA to detect IL10 or CXCL1 following manufacturer's instructions.

Four publicly-available Affymetrix microarray datasets where downloaded from the NCBI Gene Expression Omnibus repository or The Cancer Genome Atlas (TCGA) repository: GSE33113, GSE14333, GSE39582, and the TCGA colon datasets. The SW620 cell populations' dataset can be found at GSE142219.

Microarray samples from SW620 cell lines were processed using packages oligo from R and Bioconductor. Raw cell files were normalized using RMA background correction and summarization at the core transcript level. Chip probesets were annotated using the information provided by Affymetrix (details and references in the Supplementary Materials and Methods).

ERK1/2-controlled metastatic gene set (EMGS) genes were evaluated for pathway enrichment using a hypergeometric test. Gene sets derived from the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database as collected in R packages KEGG.db and org.Hs.eg.db. were used for these analyses. P values obtained from the hypergeometric test were corrected by multiple comparisons using the Benjamini–Hochberg FDR method (details and references in the Supplementary Materials and Methods).

Transcriptome analyses in human colorectal cancer tumors were carried out on 1,485 samples that were available in two public repositories listed in the Supplementary Materials. Microarray samples were processed separately for each dataset using packages affy and affyPLM from Bioconductor. Raw cel files were normalized using RMA background correction and summarization. Standard quality controls were performed in order to identify abnormal samples. Microarray intensities were corrected separately by metrics PM.IQR, RMA.IQR, and RNA.DEG as described previously. TCGA RNA-seq expression data were downloaded and processed as detailed in the Supplementary Material (details and references in the Supplementary Materials and Methods).

When not available in the clinical info, microsatellite instability (MSI) status was imputed in each dataset separately based on the expression of genes included in a published transcriptomic signature. Assignation to MSI or microsatellite stable (MSS) was performed according to results of a cluster analysis based on nonparametric density estimation on these correlation coefficients. Human colorectal cancer samples were annotated according to the consensus molecular classification, which was publicly available in the Synapse repository for most samples (n = 1,300). KRAS and BRAF mutations were annotated using data provided by the PanCanAtlas project (details and references in the Supplementary Materials and Methods).

Association with relapse was evaluated using a frailty Cox proportional hazards model. For association between gene expression levels and consensus molecular subtypes (CMS) or KRAS/BRAF mutation status, a mixed-effects linear model was used using R packages lme4 and lmerTest. Statistical significance was assessed by means of a log-likelihood ratio test, while Wald tests were used for pairwise comparisons. Sample groups of low, medium, and high expression levels were defined using the tertiles of the corresponding intensity distribution. Accordingly, HRs adjusted group means and their corresponding 95% confidence intervals were computed as measures of association. Only samples from patients diagnosed to be at stage I, II, or III were considered for analyses of time to relapse, for a total of 955 samples. All analyses were carried out using R (details and references in the Supplementary Materials and Methods).

We have recently shown that KRAS-mutated colorectal cancer cells with increased liver metastatic potential (SW620_LiM2) have increased ERK1/2 activity as compared with the poorly metastatic parental cells (SW620_P) or with lung potential (SW620_Lu) with the same mutational burden (15). To investigate the role of ERK1/2 in metastatic colorectal cancer cells, we first treated SW620_LiM2 cells with the MEK1/2 inhibitor U0126 and analyzed the resulting transcriptome for genes whose expression was reduced. These genes were then compared with those that were upregulated in metastatic SW620_LiM2 cells as compared with parental (SW620_P) cells (15). Using this approach, we identified genes that were controlled by ERK1/2 activity, and whose expression was upregulated in SW620_LiM2 cells compared with SW620_P cells (Fig. 1A). This comparison provided us with a list of 15 genes (1.75 fold-change and 5% FDR), which we named the EMGS(Supplementary Table S1). Next, we interrogated whether the EMGS was associated with disease progression in colorectal cancer. To this end, we used a pooled cohort of 955 stage I–III primary colorectal cancer tumors (GSE14333, GSE33113, GSE39582, TCGA-COAD) and found that high expression of genes within the EMGS was associated with an increased risk of recurrence (Fig. 1B) and mainly occurred at stage II of tumor progression (Supplementary Fig. S1). This association was independent of tumor MSI (Fig. 1C). Notably, tumors with annotated mutations in the KRAS and BRAF genes also had higher expression levels of the EMGS (Fig. 1D), thus confirming the association of the EMGS with RAS–ERK1/2 signaling.

Recent large-scale gene expression analyses of six independent classification systems of colorectal cancer tumors resulted in four CMS, whereby each subtype is characterized by specific biological phenotypes and different clinical and prognostic associations (16). We found increased expression levels of genes in the EMGS in the CMS1 and CMS4 subtypes as compared with the CMS2 and CMS3 subtypes (Fig. 1E). CMS1(MSI/immune) patients do not differ significantly in either OS or relapse-free survival (RFS) compared with those with CMS2 (epithelial/canonical) or CMS3 (metabolic) tumors, but they do have a significantly shorter survival after relapse than patients with other CMS subtypes (16). However, of all the subtypes, patients with CMS4 (mesenchymal/stroma) tumors have the worst OS and RFS (16). Notably, the highest EMGS expression levels were detected in CMS4 tumors, which are further characterized by the upregulation of genes involved in the epithelial-to-mesenchymal transition, as well as in tumor–stroma crosstalk. Therefore, the gene set regulated by RAS–ERK1/2 signaling appears to be associated with both recurrence and poorer prognosis in colorectal cancer. Of note, the same results were consistently observed when different fold-change thresholds were used for selected genes in the EMGS. Next, we studied whether changes in the expression of the EMGS were associated with specific biological functions. Using a hypergeometric test, we searched the KEGG database for pathways significantly enriched in the EMGS. Our results show the highest enrichment with the cytokine–cytokine receptor interaction pathway (Supplementary Table S2). The same results were obtained for various fold-change thresholds in the differential expression analysis. Focusing on this gene group, we identified five genes (AREG, CXCR4, KITLG, NGFR, and ANGPT2) that code for cytokines or cytokine receptors in the EMGS. Next, we asked whether the expression of any of these genes was associated with recurrence in the pooled cohort of 955 stage I–III primary colorectal cancer tumors. We found that only increased expression of ANGPT2 (angiopoietin-2) and CXCR4 (CXC motif chemokine receptor 4) associated with an increased risk of recurrence (Fig. 2A). ANGPT2 is a cytokine crucial for blood vessel remodeling and maturation (17), while CXCR4 is a G protein–coupled receptors, which is known to function as coreceptor for HIV entry (18, 19). These two proteins are upregulated in numerous human tumors (20–28).

The highest levels of ANGPT2 and CXCR4 expression were found in CMS1 and CMS4 colorectal cancer tumors (Fig. 2B). Subsequently, we validated that the expression levels of both ANGPT2 and CXCR4 increased in metastatic SW620_LiM2 cells as compared with the SW620_P cells (Fig. 2C; Supplementary Fig. S2A and S2B).

As both ANGPT2 and CXCR4 mediate extracellular signaling, we asked whether other genes involved in autocrine signaling were also differentially expressed in metastatic colorectal cancer cells. However, the CXCR4 agonist CXCL12 (CXCL12), the ANGPT2 receptor TIE2 gene (TEK), and the VEGFA were expressed at similar levels in between SW620_LiM2 and SW620_P cells (Supplementary Fig. S2C–S2E). This finding implies that autocrine signaling is not generally increased in the highly metastatic cells compared with the poorly metastatic cells.

Finally, treatment of SW620_LiM2 cells with the MEK inhibitors PD0325901 or U0126, or treatment of a panel of colorectal cancer cell lines (SW620_P, SW480, HCT116, and Colo26) with U0126, revealed that ANGPT2 and CXCR4 expression was dependent on RAS–ERK signaling (Fig. 2D–F). Hence, we conclude that expression of ANGPT2 and CXCR4 in colorectal cancer cells is regulated by ERK1/2 activity.

To functionally validate the role(s) of ANGPT2 and CXCR4 in colorectal cancer metastasis, we performed in vivo assays using xenograft mouse models. First, we used shRNAs to downregulate ANGPT2 and CXCR4 expression in SW620_LiM2 cells (Supplementary Fig. S3A), and then injected the cells intrasplenically into immunodeficient BALB/c nude mice to follow the kinetics of liver metastasis by quantitative bioluminescence imaging. Notably, independent or combined downregulation of ANGPT2 and CXCR4 delayed the onset of liver metastasis (Fig. 3A–C; Supplementary Fig. S3C and S3D). In addition, fewer mice injected with ANGPT2/CXCR4-downregulated SW620_LiM2 cells formed liver metastatic lesions (4/10) than in the control group (9/10; Fig. 3C), and the lesions that formed were significantly smaller (Fig. 3D). We then used retroviruses to overexpress both ANGPT2 and CXCR4 (or the empty vector, as a control) in the poorly metastatic SW620_P cell line (Supplementary Fig. S3B) and injected these cells intrasplenically into immunodeficient mice. Metastasis formation was followed by in vivo imaging and confirmed after the mice were sacrificed. Increased expression of ANGPT2 and CXCR4 conferred SW620_P cells with an increased ability to form liver lesions (observed in 8/10 mice) as compared with the control group (1/9; Fig. 3E and F). However, decreased levels of ANGPT2 and CXCR4 had no significant impact on the lung metastatic potential of SW620_LiM2 cells (Fig. 3G), in agreement with previous reports describing a role for p38 MAPK driving lung metastasis in this model (15). Similarly, lung metastatic potential from the liver of SW620_P cells was not increased upon expression of either CXCR4 or ANGPT2 (Fig. 3G). Finally, we confirmed that increased expression of ANGPT2 and CXCR4 did not confer SW620_P cells with the capacity to colonize the lungs when injected via tail vein (Fig. 3H). Altogether, these data indicate that ANGPT2 and CXCR4 regulate the ability of colorectal cancer cells to colonize the liver.

We next interrogated the molecular mechanisms that underlie liver metastasis formation driven by CXCR4 and ANGPT2. These experiments were performed in nude mice that lack T cells but retain the innate immune system. As both ANGPT2 and CXCR4 may contribute to tumor angiogenesis (29–32), we first scored the number of CD31⁺ endothelial cells in liver lesions developed by SW620_P or SW620_LiM2 cells. A comparison of size-matched lesions revealed an increased number of CD31⁺ cells in liver metastasis formed by SW620_LiM2 cells (Fig. 4A). Downregulation of CXCR4 alone in SW620_LiM2 cells did not affect the number of CD31⁺ cells in liver metastatic lesions (Fig. 4B). However, ANGPT2 depletion alone or in combination with CXCR4 significantly reduced the number of CD31⁺ cells in liver metastasis formed by SW620_LiM2 cells (Fig. 4C and D), suggesting that ANGPT2 plays a major role in the angiogenesis of liver metastasis formed by colorectal cancer cells.

Colon cancer metastasis follows a two-step hierarchical process to the liver and lung, although EMGS and the combine actions of ANGPT2 and CXCR4 seem to promote growth in the liver but not in the lung. The function of ANGPT2 and vascularization in general has a broad impact on metastasis. We therefore asked whether CXCR4 drives liver metastasis specifically. Indeed, SW620_LiM2 cells in which only CXCR4 had been depleted (with two independent shRNAs) either progressed slowly to colonize the livers or fail to generate metastasis in mice (Fig. 5A); in contrast, CXCR4 depletion did not significantly affect lung colonization (Fig. 5B). As CXCR4 knockdown in SW620_LiM2 cells did not affect the angiogenesis of liver metastases (Fig. 4B), we next focused on the role of CXCR4 in liver metastasis formation. We tested differences in various stromal populations and found a significant reduction in the infiltrating F4/80⁺ cells in liver metastatic lesions formed by CXCR4-depleted SW620_LiM2 cells as compared with those formed by control SW620_LiM2 cells (Fig. 5C). F4/80 is a well-established marker of murine macrophage populations, including Kupffer cells in the liver (33). Collectively, these results suggest that CXCR4 and ANGPT2 are downstream effectors of RAS–ERK1/2 signaling in colorectal cancer cells, which support liver colonization by triggering a supportive stromal response, involving at least vasculogenesis and macrophage recruitment.

To address the mechanism by which RAS–ERK1/2 signaling controls liver metastasis, we focused on CXCR4, as ANGPT2 has a well-known role in angiogenesis regulation. We first identified putative transcriptional factors whose expression levels correlated with that of CXCR4 in the pooled cohort of 955 stage I–III primary colorectal cancer tumors (34, 35). We found that high expression of CXCR4 correlated positively with the expression of the ETV5 transcription factor, with a partial correlation coefficient of 0.353 (P <2.22e-16; Fig. 6A). Indeed, high expression of ETV5 significantly associated with a higher risk of recurrence in patients (P = 2.5 × 10^–5; Fig. 6B). ETV5 belongs to the ETS transcription factor family, which also includes ERG, ETV1, and ETV4. ETS transcription factors promote cell proliferation and are involved in critical physiologic processes, including early development, organogenesis, and morphogenesis. The Etv4 and Etv5 double knockout mice do not develop kidneys, but this phenotype is not observed in the single knockout mice, suggesting that ETV4 and ETV5 may be functionally redundant in normal physiology. We found a small increase in ETV5 expression in SW620_LiM2 cells compared with parental cells (Supplementary Fig. S4A), and treating SW620_LiM2 cells with MEK inhibitors strongly downregulated the expression of both ETV4 and ETV5, confirming their regulation by RAS–ERK1/2 signaling (Fig. 6C). We also observed that downregulation of either ETV4 or ETV5 in SW620_LiM2 cells decreased CXCR4 expression (Fig. 6D), whereas ETV4 overexpression increased CXCR4 expression (Fig. 6E). Collectively, these results support that ETV4 and ETV5 regulate CXCR4 expression downstream of RAS–ERK1/2 signaling.

Previous data reported that CXCR4 expression is modulated by various cytokines, including TGFβ, IL10, IL4, TNFα, and IFNγ (36–39). As cytokines facilitate tumor cell–microenvironment crosstalk, which can strengthen the metastatic behavior of tumor cells, we used a cytokine array to study differences in cytokine production by SW620_P and SW620_LiM2 cells. We found that IL10 and CXCL1 were consistently expressed at higher levels in SW620_LiM2 than in SW620_P cells, at both mRNA and protein levels (Fig. 7A and B), although they were not included in the EMGS due to the use of a high stringent cut-off. Expression of IL10 and CXCL1 was also regulated by ERK1/2 signaling, as cells treated with either of two different MEK inhibitors showed reduced mRNA and protein levels of both cytokines (Fig. 7C and D). Moreover, MEK inhibition reduced the levels of CXCL1 and IL10 in the colorectal cancer cell lines SW480, HCT116, and Colo26 (Supplementary Fig S4B). Furthermore, shRNA-mediated CXCR4 downregulation led to decreased mRNA and protein expression of both CXCL1 and IL10 in SW620_LiM2 cells (Fig. 7D and E) as well as in SW620_P, SW480, and Colo26 cells (Supplementary Fig. S4C). Consistently, IL10 expression in stage II primary tumors associated with risk of recurrence in patients with colorectal cancer as previously observed for CXCR4 (Supplementary Fig. S4D). Next, we aimed to confirm CXCR4 and IL10 cancer epithelial cell dependency in clinical samples. To this end, we used the BRAF mutant–like signature (40) and the colorectal cancer intrinsic subtypes (CRIS) subtypes (41) to draw cancer cell–dependent associations instead of the CMS subtypes that reflect stromal population infiltration. Overall, we observed a significant correlation between the ANGPT2, CXCR4, ETV5, and IL10 genes and the BRAF mutant–like cancer cell signature (Supplementary Fig. S5A–S5D). Similarly, CXCR4 expression was associated with CRIS-A, -B, and -D subtypes (Supplementary Fig. S5E–S5G). Consistently, the CRIS-A subtype captures KRAS mutations; the CRIS-B subtype, the TGFβ pathway activity, the epithelial–mesenchymal transition, and poor prognosis; and the CRIS-D subtype, the WNT pathway activation. Collectively, these observations suggest that CXCR4 mediates the induction of IL10 and CXCL1 expression by ERK1/2 signaling.

We next tested whether blocking IL10 would prevent liver colonization by the highly metastatic SW620_LiM2 cell line after intrasplenic injection into nude mice. One week after injection of the cells, mice were treated with an antibody against IL10 or control IgGs, and cells homing to the liver was followed (Fig. 7F). Notably, we detected a reduction in liver colonization after IL10 antibody treatment as compared with the IgG-treated control group (Fig. 7F). This was confirmed by a reduction in metastatic lesions observed ex vivo in livers from mice injected with IL10 antibody (Fig. 7F). To extend this finding, we tested the effect of IL10 inhibition in a colorectal cancer immunocompetent mouse model. We used mouse KRAS^mut colorectal cancer organoids derived from Apcfl/fl, KrasLSL-G12D, Tgfbr2fl/fl, Trp53fl/fl, and Lgr5eGFP-creERT2 mice (34) and a neutralizing antibody against mouse IL10. As observed using nude mice and human colorectal cancer cells, we confirmed that anti-IL10 injection impaired liver colonization by colorectal cancer cells in immunocompetent mice (Fig. 7G). These results suggest that liver colonization by colorectal cancer cells relies on IL10 production, which is controlled by RAS–ERK1/2 signaling via CXCR4.

Accumulating evidence shows that the activity of signaling pathways present in the primary tumor cells is modified in metastatic cells. Consequently, this endows metastatic cells with an increased ability to colonize organs (35). Here we provide evidence that an increased RAS–ERK signaling in a mutant KRAS background could be a driver of liver metastasis formation in colorectal cancer. We identified a set of genes controlled by this signaling pathway that is associated with increased recurrence in patients with colorectal cancer (termed EMGS, for ERK1/2-controlled metastatic gene set). High levels of expression of the EMGS is associated with reduced survival for patients with both MSS and microsatellite-instable colorectal cancer tumors, thus implying that therapeutic targeting of ERK1/2 signaling pathway may benefit both groups of patients.

Interestingly, higher levels of EMGS expression were detected in the colorectal cancer subsets CMS1 and CMS4, which are characterized by an enrichment in genes that modulate the tumor microenvironment (16). More precisely, CMS1 tumors have high expression levels of genes associated with immune infiltration, whereas CMS4 tumors have high expression levels of genes associated with the epithelial-to-mesenchymal transition, the TGFβ signaling pathway, matrix remodeling, and angiogenesis (16). These findings are in line with the evidence supporting an effect of oncogenic KRAS signaling beyond cancer cells and affecting the tumor microenvironment (42). In particular, this pathway may affect the immune cell landscape in various kinds of tumor, thereby affecting their growth and progression (43–45). Furthermore, oncogenic KRAS may modulate tumor-associated angiogenesis by controlling the expression of VEGF or of cytokines such as IL8, CXCL1, and CXCL5 (46).

We now provide the evidence that the RAS–ERK1/2 axis controls the expression of the cytokine ANGPT2 and the cytokine receptor CXCR4 in colorectal cancer cells, which facilitates the development of liver metastases (Fig. 7H). ETV4 and ETV5 play a role in this metastasis process by controlling CXCR4 expression. Colorectal cancer cells with high levels of CXCR4 tend to migrate toward organs that have high levels of CXCL12, such as the liver or lung. However, our results show that an increased ANGPT2 and CXCR4 expression facilitates metastasis development in the liver but not in the lung. In addition, downregulation of ANGPT2 and CXCR4 also reduces the formation of liver metastases without affecting lung metastases. These differences in tissue-specific colonization could be attributed to the differences in the structure of liver and lung blood vessels. In this regard, liver sinusoidal blood vessels are formed by a discontinuous endothelium, which is more permissive for tumor cell extravasation than the tight junctions between the endothelial cells of lung capillaries. Therefore, ANGPT2 and CXCR4 might not be essential for tumor cell extravasation but control metastatic outgrowth.

Both ANGPT2 and CXCR4 can control tumor angiogenesis (29–32). However, in our mouse model, the number of endothelial cells in liver metastatic lesions was dependent exclusively on the expression levels of ANGPT2 but not CXCR4. Of note, the expression of VEGF, which has been reported to cooperate with ANGPT2 to promote tumor growth (47) and to regulate the expression of CXCR4 (48), was not significantly different between the poorly and highly metastatic cell lines SW620_P and SW620_LiM2, respectively. The cytokines ANGPT2 and ANGPT1 bind to the TIE2 receptor with similar affinity, but their binding elicits distinct responses in endothelial cells (49). Namely, ANGPT1 maintains endothelial cells in quiescence, while ANGPT2 promotes vascular remodeling and angiogenesis (49). We detected expression of ANGPT1 neither in SW620_P nor in SW620_LiM2 cells, which is in line with reports that ANGPT2 but not ANGPT1 can be highly expressed in human tumors, and that an imbalance in the ANGPT2/ANGPT1 ratio can promote tumor growth (50–54).

Increased levels of CXCR4 expression are found in more than 20 human tumor types, including ovarian, prostate, melanoma, neuroblastoma, and colorectal cancer (25–27, 55–58). In addition, CXCR4 has been connected with metastatic spread in breast, prostate, hepatocellular, and colorectal cancers, mostly by regulating cell migration toward surrounding tissues or toward CXCL12-enriched organs (26, 57, 59–61). Notably, CXCR4 downregulation in our model was associated with a reduced recruitment of F4/80⁺ cells to the metastatic lesions. Our data demonstrate, for the first time, that CXCR4 controls the expression of the cytokines IL10 and CXCL1, and we also provide evidence for a causal role of IL10 in supporting liver colonization. In patients with colorectal cancer, IL10 levels increase as the disease progresses, and high serum levels of IL10 correlate with poor survival of these patients (62, 63). In breast cancer, CXCL1 promotes lung metastasis; in contrast, in colon cancer, it may influence the formation of the liver premetastatic niche (64, 65). The mechanisms through which IL10 and CXCL1 promote liver metastasis from colorectal cancer tumors will be the subject of further investigation. In summary, we demonstrate that amplification of ERK1/2 signaling in KRAS-mutated colorectal cancer cells affects the cytokine milieu of these tumors, thus probably affecting tumor–stroma interactions and favoring liver metastasis formation.

R.R. Gomis reports other from Inbiomotion SL (member of the board of directors) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

J. Urosevic: Conceptualization, resources, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. M.T. Blasco: Conceptualization, data curation, formal analysis, investigation, methodology, writing-review and editing. A. Llorente: Resources, validation, investigation, methodology, writing-original draft. A. Bellmunt: Resources, data curation, formal analysis, investigation, writing-original draft. A. Berenguer-Llergo: Data curation, formal analysis, validation, methodology, writing-review and editing. M. Guiu: Investigation, methodology, writing-original draft. A. Cañellas: Investigation. E. Fernandez: Investigation. I. Burkov: Formal analysis, investigation. M. Clapés: Validation, investigation. M. Cartanà: Investigation. C. Figueras-Puig: Investigation. E. Batlle: Resources. A.R. Nebreda: Resources, writing-original draft, writing-review and editing. R.R. Gomis: Conceptualization, formal analysis, supervision, funding acquisition, methodology, writing-original draft, writing-review and editing.

We thank V. Raker for manuscript editing and IRB Barcelona Functional Genomics (J.I. Pons and D. Fernández), Histopathology (N. Prats), Advanced Digital Microscopy (J. Colombelli), and Flow Cytometry (J. Comas) Core Facilities for assistance. J. Urosevic was an AECC (Asociación Española Contra el Cáncer) Fellow. A. Llorente, A. Bellmunt., and C. Figueras-Puig were funded by the Spanish Government (MINECO-Formación de personal Investigador). I. Burkov was cofunded by FP7 Marie Curie Actions (COFUND program; grant agreement no. IRBPostPro2.0 600404). M.T. Blasco was funded by ISCIII/FEDER-CIBERONC and by the Spanish Government (Juan de la Cierva Formación-Postdoctoral Fellowship). R.R. Gomis, E. Batlle., and A.R. Nebreda are supported by the Institució Catalana de Recerca i Estudis Avançats. Support and structural funds were provided by the Generalitat de Catalunya (2014 SGR 535) to R.R. Gomis and A.R. Nebreda, and by the BBVA Foundation, the ISCIII/FEDER-CIBERONC, the “la Caixa” Foundation (ID 100010434), under the agreement <HR17–00092>, the Spanish Ministerio de Economia y Competitividad (MINECO) and FEDER funds (CIBEREONC and PID2019–104948RB-I00) to R.R. Gomis.

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.