Liver Metastasis Is Facilitated by the Adherence of Circulating Tumor Cells to Vascular Fibronectin Deposits Free

Metastasis comprises a sequence of events by which cancer cells disseminate from a primary location to a secondary growth site. Tumor cells first invade the surrounding stroma to reach the bloodstream, where, after anchorage-independent survival, leave the circulation and establish a new tumor mass. Although the estimated percentage of cancer cells that are able to complete the whole circuit is low (<0.02%; ref. 1), metastatic dissemination still represents the main life-threatening process in cancer (2–4).

Circulating tumor cells (CTC) represent the fluid phase of solid tumors. They have a key role in tumor dissemination, as at least a subpopulation of CTCs has the ability to initiate clonal metastatic lesions (5–7). Indeed, their presence in the bloodstream is used as a prognostic and a therapy response marker (8, 9). Once CTCs get trapped and/or attach to capillary beds at distant organs, they migrate through the endothelial layer in a process termed extravasation (2). The efficacy with which CTCs escape from the circulation depends on cell–cell interactions between cancer cells and endothelial cells and on the capacity of cancer cells to breach the endothelial layer and the underlying basement membrane. VCAM, ICAM, or L1 molecules, expressed by the vascular endothelium, interact with α4β1 and αvβ3 integrins, CD44 or MUC1 receptors present on cancer cells, facilitating cancer cell extravasation (2, 10).

We previously reported a global gene expression profile for CTCs derived from metastatic colorectal cancer (mCRC) patients (11). Interestingly, a high proportion of CTC-related genes are involved in integrin signaling and focal adhesion formation, such as talin1, vinculin, and kindlin (Supplementary Fig. S1). Focal adhesions are multi-protein complexes that assemble at integrin-clustered sites at the plasma membrane and regulate cell attachment to extracellular matrix (ECM) proteins (mainly fibronectin, laminin, and different types of collagen). However, the role of focal adhesions in anchorage-deprived cells, such as CTCs, remains unclear. Moreover, metastatic colonization at distant sites has been associated with matrix remodeling for niche formation, including the liver, the most frequent metastatic site in colon cancer (12). Here, we investigated the presence of vascular substrates to which CTC talin1-mediated focal adhesions might bind by assessing the localization of different ECM proteins in liver blood vessels. We propose a talin1-dependent mechanism for liver tissue colonization by CTCs.

EGFP-tagged paxillin and vinculin plasmids were kindly provided by C. Ballestrem (University of Manchester, Manchester, United Kingdom). LifeAct-GFP was obtained from O. Pertz (University of Basel, Basel, Switzerland). Luciferase reporter pLenti CMV V5-LUC Blast (w567-1), lentiviral packaging, and envelope vectors (psPAX2 and pmD2.G, respectively) were purchased from Addgene. GFP-talin1 was a gift from Anna Huttenlocher (Addgene plasmid #26724; ref. 13).

Mouse anti-α-tubulin (Clone DM 1A), rabbit anti-fibronectin and rabbit anti-laminin were from Sigma-Aldrich. Goat anti-talin (C20) and goat anti-fibronectin were from Santa Cruz Biotechnology. Rabbit anti-collagen I was from Abcam. Rabbit anti-collagen IV was from Millipore. Rabbit anti-VE Cadherin was from Cell Signaling Technology. Mouse anti-vinculin was obtained from M. Glukhova (Institut Curie, Paris, France). Mouse anti-β-Catenin, CD29 (clone 9EG7) against activated β1 integrins and IgG isotype control were purchased from BD Biosciences. Labeled-phalloidin, DAPI, and all secondary antibodies were purchased from Molecular Probes. Fibronectin (Life Technologies) was used at 20 μg/mL. SiR-Actin was from Spirochrome. Functional upstream domain (FUD) peptide was kindly provided by C. Albiges-Rizo (IAB, Grenoble, France).

HCT116, CT26, and HEK293 cell lines were obtained from ATCC in 2013–2014 and cultured in DMEM, 10% FBS, and 1% penicillin/streptomycin. Human umbilical vein endothelial cells (HUVEC) were obtained from Thermo Fisher Scientific in 2014 and cultured in endothelial basal medium EBM-2 (Lonza). For HUVECs, culture dishes were pretreated with 0.2% gelatin diluted in PBS for 30 minutes at 37°C. All cells were routinely tested for Mycoplasma using the Plasmotest kit (Invivogen).

CT26 LifeAct-GFP and HCT116-Luc cells were generated by lentiviral infection as described previously (14). HCT116-Luc cells were selected using blasticidine HCl (10 μg/mL). Talin1 knockdown HCT116 and CT26 cells were generated using lentiviral particles (Sigma Aldrich) following manufacturer's recommendations and selected with puromycin (1 μg/mL for HCT116 and 6 μg/mL for CT26). CT26 shtalin1 were transfected to stably express GFP-talin1 using Lipofectamine LTX and selected using G418 (0.5 mg/mL).

Around 3-mm³ pieces of mouse livers were fixed in 4% paraformaldehyde for 30 minutes, embedded in 4% agarose/DMEM solution using low melting point agarose (Thermo Fisher Scientific), and sectioned in 300-μm slices using a vibratome (Leica). Tissue slices were blocked with 1% BSA/3% FBS in PBS for 1 hour at room temperature stained for fibronectin, laminin, collagen I, and collagen IV (1:100 dilution) overnight at room temperature. Samples were incubated with secondary antibodies overnight at room temperature (1:200 dilution), washed, and mounted on microscopy slides. Imaging setup is described in Supplementary Methods.

HUVECs (10⁵) were seeded on a gelatin-coated 12-mm glass coverslip and incubated overnight to form a monolayer. A total of 4 × 10⁵ CT26 EGFP-Paxillin cells were seeded on top of the HUVEC monolayer in 10% FBS DMEM and incubated for 2 hours at 37°C to allow tumor cell–HUVEC contact establishment. Cells were then PFA fixed and stained for fibronectin and DAPI. Cell interactions were visualized with a 100× objective of a wide-field microscope (DM6000 B/M; Leica) equipped with a CCD camera (CoolSNAP HQ; Photometrics).

For transfilter migration assays, 10⁵ CT26 cells (ShControl and Shtalin1) were seeded on the top chamber of 24-well cell culture inserts (8-μm pore size; Millicell, Millipore) in 100 μL of DMEM containing 5% FBS. The bottom chamber was filled with DMEM supplemented with 20% FBS. After 24 hours, migrating cells were collected, stained with Calcein AM (5 μmol/L), and fluorescence was quantified on a FluoStar optima plate reader (emission wavelength: 530 nm, excitation wavelength 485 nm).

For transendothelial migration assays, 10⁵ HUVECs were seeded on the top chamber of a gelatin-coated transwell in EBM-2 complete medium and incubated for 4–6 hours at 37°C to form a monolayer, in which integrity was checked by addition of high molecular weight FITC-dextran (MW 40.000; Sigma Aldrich) to the top chamber (1 mg/mL) and assessing its passage to the bottom chamber. Then, 5 × 10⁵ CT26 cells (control and shtalin1; Calcein stained) were added to the top chamber in 10% FBS DMEM. Twenty-percent FBS DMEM was added to the bottom chamber. After 24 hours, transmigrated cells were collected and fluorescence emission was measured as above.

Results were represented as the relative number of shtalin1-migrating cells compared with control cells, defined as 100%, for each cell line.

A total of 3 × 10⁴ endothelial cells were seeded in gelatin-coated 96-well plates and incubated overnight for cell monolayer formation. For fibronectin blocking assays, 20 μg of anti-fibronectin antibodies (Sigma) were added in EBM-2 medium without serum for 4 hours. FUD was added at 500 nmol/L in EBM-2 medium every 24 hours for 48 hours. Cancer cells were stained with Calcein AM and plated at a concentration of 2.5 × 10⁵ cells/well over HUVEC monolayers. Cells were then incubated at 37°C for the specified times, washed three times to remove nonadherent cells, and fluorescence intensity was measured. Adhesion was represented as the percentage of adherent cells per condition.

Cell monolayer integrity was evaluated by two independent methodologies described in detail in Supplementary Methods.

The bottom glass of a culture chamber (PocMini-2, PeCon) was coated with 5 × 10⁵ HUVECs in EBM-2 medium at 37°C overnight for monolayer formation. The culture chamber was then connected to a syringe pump (NE-1000, New Era Pump Systems Inc.) and placed on a thermostated platform. After, Calcein AM–stained tumor cells (10⁵ cells/mL) were pumped through the culture chamber at a shear stress of 2.6 dynes/cm² for 1 hour, followed by a 10-minute wash with complete medium. After this time, the bottom glass from the culture chamber was removed and tumor cells attaching to the endothelial monolayer were indirectly quantified by fluorescence emission.

Under anesthesia, the mouse rib cage was opened and the heart exteriorized. An incision was made in the right heart atrium to create a blood exit. Approximately 5 mL of PBS was then injected in the left ventricle under a low and constant flow. After washing the blood out, evidenced by liver clearing, 5–7 mL of freshly prepared 4% PFA was injected using a syringe pump at 2 mL/minute rate. Whole-body stiffening and tail movements were positive controls of successful fixation. Livers were then excised, cut in approximately 3-mm³ pieces, and processed as described above.

To visualize cancer cell dynamics inside the liver vasculature, 10⁶ LifeAct-GFP cells diluted in 100 μL of PBS 1% FBS were intrasplenically injected in tomato mice. Mice were sacrificed 30 minutes after injection; livers were excised and kept at 4°C in Williams E medium supplemented with 1% antimycotic/antibiotic (Thermo Fisher Scientific) as described previously (15). Fresh liver tissue was embedded and vibratome sliced. Liver slices were mounted on top of a millicell cell culture insert (0.4 μm, 30 mm diameter; Millipore), which was then placed on a culture dish. Williams E medium was added only on the bottom surface of the filter to allow effective gas exchange and at the same time preventing tissue death. The imaging setup is described in Supplementary Methods.

Orthotopic, intracardiac, and intrasplenic mouse models were generated as described previously (16–18). Experimental details are included in the Supplementary Methods.

Primary colorectal carcinomas (n = 8) and metastasis (liver metastasis, n = 7; lung metastases, n = 7) were processed. The superficial noninvasive zone and the deep invasive area of the primary tumors were macroscopically dissected, ensuring similar tumor percentages. RNA was purified (TRIzol reagent, Invitrogen; RNeasy kit, Qiagen), cDNA was synthesized (MuLV reverse transcriptase, Life Technologies), and gene expression was evaluated using hydrolysis probes (Life Technologies). Data was represented as fold change relative to the expression in the superficial noninvasive area. GAPDH, ACTB, and RLPLO housekeeping genes were used as loading controls.

Blood from 20 healthy volunteers was extracted and analyzed as control samples. Talin1 expression levels in immunoisolated CTCs was performed as described previously (19). Protein tyrosine phosphatase receptor type C (PTPRC or CD45) was used as a reference gene as it detects hematopoietic cells unspecifically isolated, and we have previously reported its validity as a reference gene (19). C_q values (defined as the cycle number at which the fluorescence reached a fixed threshold value) for each transcript were normalized to 40 (maximum number of cycles), and this value to the 40-C_q value for CD45 [(40-C_q target)-(C_q CD45)].

Progression-free survival (PFS) and overall survival (OS) were defined as the time elapsed (in months) between treatment day 1 and the day reporting disease progression or death, respectively. Disease progression was defined following RECIST 1.1 criteria (20), as an increase in the number of metastatic lesions, growth of preexisting distant tumors in more than 20% of the initial size, or both. Prognostic groups for each analysis point (baseline and follow-ups) were set based on talin1 expression levels in CTCs as described previously (21). Patients were included into the high-CTC group when talin1 expression levels were above the cutoff, defined as the 75% expression percentile taking into account the whole patient set. Patients with talin1 expression below the 75% cutoff were included in the low-CTC group. Kaplan–Meier and univariate and multivariate Cox regression survival analysis were used to study associations between marker levels and PFS/OS. HR is defined as the relative risk of progression or survival for a specific group of patients compared with the control group. Survival tests were performed with SPSSv20.0 software, and considered significant when P < 0.05.

We immunostained 300-μm-thick liver sections from Tomato reporter mice (22) using specific antibodies for laminin, collagen I, collagen IV and fibronectin, to determine the putative ECM substrates for CTCs. This mouse model has ubiquitous expression of membrane-targeted Tomato, which enables a clear visualization of cell membranes in all cell types and provides a spatial reference for ECM protein localization.

All of the ECM proteins tested were highly expressed in hepatic blood vessels, with laminin expression restricted to higher diameter vessels and poorly expressed at liver sinusoids (Fig. 1B; Supplementary Fig. S2A and S2B). At the cellular level, all of the ECM proteins were localized to the basal side of endothelial cells as constituents of the basement membrane and stroma. A discontinuous layer of fibronectin was also found on the apical surface of endothelial cells, facing the lumen of the hepatic vasculature (Fig. 1A, inset; Supplementary Movie S1). These luminal fibronectin deposits were present in approximately 20% of the analyzed vessels in wild-type mice (Fig. 1D). Interestingly, we also found luminal fibronectin in mice that developed spontaneous primary intestinal tumors (by Notch overexpression and P53 deletion, NICD/p53^−/−; Fig. 1C; ref. 15). The number of hepatic vessels containing luminal fibronectin was increased compared with wild-type mice (Fig. 1D). Mutant mice that did not develop tumors had the same number of vessels positive for luminal fibronectin as wild-type mice, arguing against a mouse genotype effect. These results correlated with a general hepatic overexpression of fibronectin in the presence of primary tumors (Fig. 1F; Supplementary Fig. S3). Two of the analyzed mice also developed liver metastasis in addition to the primary intestinal tumors, but there was no difference in hepatic fibronectin expression or on its accumulation in the vessels compared with mice bearing only primary tumors (data not shown). To test whether fibronectin deposits are firmly attached to the endothelium, we cleared out the circulation of tumor-bearing mice by whole-body perfusion and fixation to eliminate soluble fibronectin. Reduction of fibrinogen, a protein that normally circulates in the blood, indicated successful liver clearing (Supplementary Fig. S4A). Although the frequency of fibronectin deposits was reduced, they were still detected after perfusion (Supplementary Fig. S4B). Fibronectin and fibrinogen were often found in the same structures localized at the luminal side, whereas laminin was found only at the adluminal surface (Supplementary Fig. S4C).

To rule out the possibility that the increased presence of luminal hepatic fibronectin deposits is specific for NICD/p53^−/− mice, we generated intestinal orthotopic tumors in SCID mice using HCT116 human colon cancer cells. Mice bearing primary tumors showed an increased frequency of hepatic luminal vascular fibronectin compared with control mice (P = 0.1273; Fig. 1E). These results suggest that the accumulation of hepatic luminal fibronectin deposits could be a general feature of mice during the development of intestinal cancer. Interestingly, we also found the same pattern of luminal fibronectin distribution in human livers containing colon cancer metastases (Supplementary Fig. S5).

Overall, these data show that liver blood vessels contain luminal fibronectin deposits and show that the presence of primary colon intestinal tumors increase the frequency of these accumulations.

We next investigated whether cancer cells could use endothelial fibronectin to promote attachment. For this, we set up an in vitro assay using a monolayer of HUVECs, on top of which we seeded CT26 cancer cells transiently expressing Paxillin-GFP. Endothelial cells produced and deposited fibronectin at their surface, organized in fibrillar structures (Fig. 2A). Cancer cells attached to endothelial cells, forming numerous focal adhesions that colocalized with fibronectin fibers (Fig. 2B). In vivo, after injection of paxillin-GFP CT26 cells in the spleen, we could detect cancer cells in the liver vasculature showing paxillin-positive aggregates at the cell edges, which colocalized with vascular fibronectin (Supplementary Fig. S6; Supplementary Movie S2).

To directly test the role of focal adhesion–mediated contacts between cancer and endothelial cells, we impaired focal adhesion formation in cancer cells by depleting one of the major focal adhesion constituents, talin1 (Supplementary Fig. S7A). Talin1 is not required for initial cell attachment and spreading, but it induces focal adhesion–dependent pathways by providing the mechanical link between integrins and the actin cytoskeleton (23). It is also one of the genes found to be highly expressed in CTCs from colorectal cancer patients (Supplementary Fig. S1; ref. 11).

As expected, talin1-depleted cells had fewer focal adhesions (Supplementary Fig. S7B–S7D) and decreased β1-integrin activation (Supplementary Fig. S7E). Talin1-depleted cells had reduced adhesion (Fig. 2C), also when plated on different extracellular matrices including collagen (I, II, and IV), laminin, fibronectin, tenascin, or vitronectin (Supplementary Fig. S8A and S8B). Reexpression of a WT GFP–tagged-talin1 rescued the adhesion capacity of cancer cells, demonstrating the specificity of the talin1 knockdown phenotype (Fig. 2C).

Talin1-depleted cells also displayed a significantly reduced ability to adhere to HUVEC monolayers compared with control cells both under static and flow conditions (2.6 dynes/cm²). Under flow, 7%–10% of control and 3%–4% of talin1-depleted cancer cells attached to the HUVEC monolayers. (Fig. 2D and E; Supplementary Fig. S8C and S8D).

Next, we investigated whether the binding of cancer cells to the endothelial layer is dependent on fibronectin. Fibronectin-blocking antibodies, as well as inhibition of fibronectin fibrillogenesis using FUD peptide (24), reduced cancer cell adhesion to the endothelial monolayer (Fig. 2F).

Altogether, these results show that cancer cells engage talin1-dependent focal adhesions that could mediate the interaction with fibronectin deposits on endothelial cells, possibly favoring docking of CTC during liver colonization in metastatic colorectal cancer.

Endothelial retraction induction by cancer cells is considered one of the first steps during transendothelial migration (25–27). To test the role of talin1 in this process, we performed an endothelial monolayer integrity assay that has been previously described (28). When no cancer cells were added to the HUVEC monolayer, the cell index (a direct indicator of monolayer integrity) was maintained over time. In contrast, when cancer cells were added on top of the endothelial monolayer, the cell index decreased as a result of endothelial cell retraction. Retraction of the endothelial monolayer was reduced in the presence of talin-depleted cells (Fig. 3A; Supplementary Fig. S9A). These results were confirmed using HUVECs cocultured with cancer cells. In the presence of ShControl cells, endothelial monolayers were disrupted, losing VE-Cadherin–mediated junctions. However, talin1-depleted cells were not able to induce endothelial retraction, with monolayers preserving adherens junctions (Supplementary Fig. S9B–S9D). In live imaging experiments, we did not observe endothelial cell death in the presence of cancer cells (data not shown), as reported previously (29). Our results indicate that cancer cell attachment and spreading over the endothelium are a prerequisite for endothelial retraction.

After interaction with endothelial cells, most of the control cancer cells spread out and extended numerous membrane protrusions. In contrast, the few focal adhesion–impaired cells that managed to attach remained round and rarely formed protrusions (Fig. 3B). Concordantly, when we examined focal adhesion dynamics in paxillin-GFP cells, we found that control cells formed and disassembled adhesions within 20 minutes (Fig. 3C; Supplementary Movie S3), which was sufficient for long-lived protrusions to form (Fig. 3D; Supplementary Movie S3). Adhesions in talin1-depleted cells were significantly shorter-lived, persisting for 10 minutes on average, resulting in the formation of rather small and unstable protrusions. We hypothesized that the stabilization of these cancer cell protrusions could be a hallmark during the first steps of extravasation, including endothelial retraction. Talin1 depletion also had a functional impact in the migratory ability of cancer cells, both in a transwell assay and across an endothelial barrier in a transendothelial migration assay (Fig. 3E and F).

Finally, we optimized a physiologically relevant ex vivo model to analyze the role of talin1 during liver colonization. For this, we used CT26 cells stably expressing Life-Act-GFP, which allows the real-time visualization of actin dynamics and membrane protrusion formation. Cells were injected intrasplenically into tomato mice, and 30 minutes postinjection, livers were excised, cut into 300-μm fresh sections, and imaged for 24 to 48 hours using a two-photon confocal microscope. In spite of obvious limitations due to the absence of blood flow, this model allows the analysis of cancer cell behavior in a complex three-dimensional environment. Control cells (70.9%, n = 34) formed actin-rich protrusions at their leading edge, and showed a highly dynamic behavior moving through smaller capillaries or into the liver parenchyma (mean cell displacement 22.1 ± 34.1 μm). Talin-depleted cells were mostly round-shaped (36.7%, n = 49 cells with protrusions), poorly migratory, and remained trapped at capillary straits (mean cell displacement 6.9 ± 14.1 μm; Fig. 3G; Supplementary Movie S4).

Altogether, these results indicate that talin1 in cancer cells is required during the initial steps of metastatic colonization and transendothelial migration, possibly by stabilizing cell membrane protrusions.

We next assessed the impact of CTC–endothelium interaction through talin1 on the metastatic potential of cancer cells in vivo, in two different metastasis scenarios.

First, we injected control and talin1-depleted cancer cells into the left heart ventricles of nude mice, mimicking the conditions experienced by a CTC while travelling in the bloodstream. Control cells survived in the circulation and rapidly formed tumors, mainly in bone tissue, adrenal glands, and ovaries. In contrast, mice injected with talin1-depleted cells displayed a significant reduction in the number of formed metastases (Fig. 4A).

Second, we analyzed how talin1 could influence liver tissue colonization in vivo by performing intrasplenic injections and evaluating metastasis formation over time. Nude mice were injected with Life-Act-GFP CT26 control and talin1-depleted cells and sacrificed at different time points (after 30 minutes, 8 hours, and 1, 3, 7, and 14 days). The number of control cells progressively decreased over time, and no cells were detectable after 3 days. Talin1-depleted cells followed the same tendency, but the number of cells was lower at each time point, especially after 24 hours (Fig. 4B). Moreover, the number of cells with membrane protrusions was lower in talin1-depleted cells at each time point, with a maximal difference 24 hours after injection (Fig. 4C). Although cells were barely detected after 3 days, control mice developed large liver tumors 14 days after injection, whereas mice injected with talin1-depleted cells showed a significant reduction in liver metastases (Fig. 4D). For both metastasis mouse models, the observed differences in metastasis formation were not due to proliferation/viability differences between control and talin1 depleted-cells (Supplementary Fig. S10).

Together, these results show that talin1 is essential for the effective completion of the metastatic colonization process from the bloodstream in two independent mouse models.

Finally, we investigated whether talin1 expression correlates with metastasis formation in colorectal cancer patients and whether it has a prognostic value. To test this, we analyzed the expression levels of talin1 in CTCs from a set of 50 mCRC patients at baseline, before the start of chemotherapy. Kaplan–Meier analysis revealed that patients with high-CTC-talin1 had shorter PFS (6.9 months vs. 12.3) and OS (10.5 vs. 24.6 months), as well as an increased HR, compared with low-CTC-talin1 patients (Fig. 5A and B).

To further investigate the role of talin1 expression levels in CTCs as a predictive marker to monitor therapy efficacy, we evaluated the expression levels of talin1 in CTCs after the administration of a first-line chemotherapy. Patients were classified as responders (R) or non-responders (NR) as previously reported (21; scheme in Fig. 5C). Kaplan–Meier analysis revealed that patients classified as nonresponders had shorter PFS (6.6 vs. 12.6 months) and OS (11.5 vs. 24.6 months), as well as an increased HR compared with responders (Fig. 5D and E).

Finally, to analyze the potential relationship between talin1 expression and tumor progression, we evaluated talin1 expression levels at the invasive and noninvasive areas of a series of 7 primary CRC tumors, and in 14 liver and lung metastases of an independent patient set. We found that talin1 was upregulated 1.7-fold (P < 0.01) in the invasive area of primary colorectal carcinomas, compared with the noninvasive area of the same tumor (Fig. 5F). A 4-fold talin1 upregulation (P < 0.001) was found in liver and lung metastatic tissue when compared with the noninvasive area of the primary tumor (Fig. 5F). This suggests a correlation between tumor progression and the expression of talin1 in tumor cells.

Taken together, these results show that talin1 expression levels in CTCs could be used as a prognostic and therapy response marker in mCRC.

The endothelial layer is the first barrier that CTCs encounter in their attempt to escape from blood vessels. Cancer cells must attach to endothelial cells to migrate through or between them to reach the underlying stroma and colonize distant organs (2). Although extravasation has been widely studied in normal leukocytes, much less is known about this process in cancer cells (2–4). Here we report a novel mechanism by which CTCs adhere to endothelial cells of liver blood vessels using fibronectin deposits as a substrate. This allows them to extravasate and form metastasis. Intravascular fibronectin accumulations were not present exclusively at the liver vasculature, as we could also find vessels positive for luminal fibronectin in mesenteric blood vessels of mice with peritoneal carcinomatosis (Supplementary Fig. S11). ECM on the luminal side of the vasculature could serve as a general docking site for CTCs. It has been reported that in the lungs' vasculature, laminin from the basement membrane could be exposed and thus become accessible for cancer cells, serving as an attachment point for extravasation (30). However, the type of luminal ECM most likely differs between organs. In all mice that we analyzed, we found only fibronectin and fibrinogen deposits in the luminal side of the hepatic vasculature, while we did not observe any laminin signal (Fig. 1; Supplementary Fig. S4).

Multiple sources could contribute to the formation of fibronectin deposits (31, 32). First, endothelial cells could either produce fibronectin or retain soluble plasma fibronectin produced by hepatocytes. Second, as the frequency of luminal liver fibronectin deposits increases in mouse bearing primary intestinal tumors, it is also possible that the primary tumor itself contribute to increased levels of fibronectin in the liver. The tumor (cancer or stromal cells) could either directly produce fibronectin or stimulate fibronectin production in hepatocytes or liver endothelial cells by secreting cytokines. Indeed, recent evidence has shown that fibronectin upregulation is important for liver premetastatic niche formation induced by pancreatic cancer–derived exosomes (12). Furthermore, studies in lung cancer suggest that plasma fibronectin could confer a prometastatic ability to cancer cells by activating α_Vβ3 integrins (33). These observations suggest that fibronectin could play a role in premetastatic organ conditioning. Indeed, there are multiple pieces of evidence suggesting that metastatic colonization often depends on substrate preparation prior cancer cell extravasation, usually by the release of soluble factors from primary tumors (34, 35).

On the basis of in vitro observations, it has been hypothesized that CTCs first roll over the endothelial layer, in a process termed “docking,” (mediated by cancer cell glycoproteins and corresponding endothelial selectins), then adhere to endothelial cells (“locking”) (usually mediated by CAMs and integrins) and finally migrate through the endothelial layer (2, 10, 36). Here we show that talin1 is involved in adhesion to endothelial fibronectin and for the formation of membrane protrusions that could help cancer cells to squeeze and migrate between endothelial cells. Indeed, our results are consistent with previously reported data in ovarian cancer, where talin1 has been shown to regulate mesothelial clearance by cancer spheroids in a fibronectin-dependent manner (37).

After successfully squeezing between endothelial cells, CTCs breach the endothelial basement membrane. Cancer cells use proteolytic F-actin–based structures called invadosomes to invade ECM such as the basement membrane (38). It has been recently shown in a breast cancer model that talin1 is required for matrix degradation. Talin1 regulates moesin-NHE1 recruitment to invadopodia, promoting further invasion and intravasation (39). Using a mesenteric basement membrane, we have obtained similar results suggesting that talin1 is essential for basement membrane invasion (unpublished observations). Thus, talin1 could play a role in the whole process of CTC extravasation from the circulation.

Collectively, the evidence we present here supports the existence of a complementary liver colonization mechanism for CTCs (talin1/fibronectin dependent), which has clear implications in metastasis formation, as demonstrated both in in vivo models of liver and distant dissemination, and in mCRC patients.

Our data also suggest fibronectin blocking as a possible therapeutic strategy to prevent metastasis formation. Talin1, besides playing a role in extravasation and metastatic spreading, is likely also needed for survival and growth in the secondary organs. Thus, it could be valuable marker not only to predict patient prognosis, but also as a tool to monitor treatment efficacy in mCRC. Further studies will pursue the evaluation of other focal adhesion–related proteins in the same process, to elucidate whether this is a mechanism that relies on talin1 expression, or it could be extended to focal adhesions in general.

No potential conflicts of interest were disclosed.

Conception and design: J. Barbazán, N. Elkhatib, R. López-López, M. Abal, D.M. Vignjevic

Development of methodology: J. Barbazán, L. Alonso-Alconada, V. Gurchenkov, G. van Niel, D.M. Vignjevic

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Alonso-Alconada, S. Geraldo, G. van Niel, R. Palmulli, B. Fernandez, T. Garcia-Caballero, R. López-López

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Barbazán, L. Alonso-Alconada, T. Garcia-Caballero, M. Abal, D.M. Vignjevic

Writing, review, and/or revision of the manuscript: J. Barbazán, L. Alonso-Alconada, S. Geraldo, T. Garcia-Caballero, M. Abal, D.M. Vignjevic

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

Study supervision: R. López-López, M. Abal, D.M. Vignjevic

Other (technical support): P. Viaño

We gratefully thank all the patients for their willingness to participate in the study. We thank C. Albiges-Rizo (IAB, Grenoble) for sharing FUD with us. We acknowledge the Cell and Tissue Imaging (PICT-IBiSA) and Nikon Imaging Center, Curie Institute, member of the French National Research Infrastructure France-BioImaging (ANR10-INBS-04), and all Vignjevic and Abal teams for support and scientific discussion.

This work was supported by ERC (starting grant STARLIN 311263 to D.M. Vignjevic). J. Barbazán is recipient of a Marie Curie Individual Fellowship (FiBRO).

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.