KLF4-dependent perivascular cell plasticity mediates pre-metastatic niche formation and metastasis

Lineage-traced perivascular cells demonstrate that phenotypic switching occurs in pre-metastatic sites

Perivascular cell phenotypic switching is characterized by loss of marker gene expression such as Myh11 and Acta2, and an increase in proliferation, migration, and ECM production⁷. Due to this reduced perivascular marker gene expression and the potential misidentification of these cells in their activated state, in vivo studies that carefully trace and investigate the function of phenotypically switched perivascular cells are required. To determine whether perivascular cells undergo phenotypic switching in pre-metastatic tissue, we used the recently described Myh11-ERT-creT2 ROSA-STOP-flox-eYFP lineage-tracing mice, wherein the perivascular-specific gene Myh11 promoter drives an inducible cre-recombinase (designated as “Myh11 lineage-tracing mice”) (Supplementary Fig. 1a)^17,19. In adult Myh11 lineage-tracing mice, tamoxifen induces stable expression of eYFP in vSMCs and pericytes, and enables the detection of cells expressing the Myh11 gene only at the time of tamoxifen administration, including pre-existing SMCs/pericytes and their progeny, even when this perivascular marker expression is subsequently lost^17,19. Importantly, we found that nearly all MYH11+ cells in the lungs of healthy Myh11 lineage-tracing mice treated with tamoxifen were eYFP+ and co-expressed MYH11 (Supplementary Fig. 1b). YFP+ cells were also ACTA2+, a known marker of perivascular cells and myofibroblasts (Supplementary Fig. 1c).

To interrogate the role of perivascular cells during metastatic development, we orthotopically injected metastatic melanoma B16-F10 or metastatic rhabdomyosarcoma M3-9M tumors into syngeneic Myh11 lineage-tracing mice and analyzed pre-metastatic lung at multiple time points for evidence of perivascular phenotypic switching in eYFP-expressing cells that have lost expression of perivascular markers MYH11 and ACTA2 (Supplementary Fig. 1d-e). We found that there is an increase in the total number of eYFP+ cells in the lungs of B16-F10 (Fig 1a-c, e, Supplementary Fig 2a) and M3-9M (Fig 1 h-j) tumor-bearing mice from day 10 to day 18 post-tumor injection. By careful confocal imaging across multiple pre-metastatic time-points, we demonstrated the loss of perivascular markers MYH11 and ACTA2, the displacement of perivascular cells away from blood vessels by day 3, the appearance of YFP+ cells away from blood vessels by day 5, and increase in eYFP+ MYH11- ACTA2- cells found in clusters within the parenchyma separated from the vasculature of pre-metastatic lung by day 10 post-tumor injection (Fig 1e, Supplementary Fig 2a). The total number of eYFP+ cells and the proportion of phenotypically modulated eYFP+ cells that were MYH11- ACTA2- was increased in pre-metastatic lung (Fig 1c, f, j, k). This expanded population of eYFP+ cells may be due to vSMC recruitment to these pre-metastatic sites, proliferation in situ, or a combination of both. Consistent with the increase in eYFP+ cell number identified in stained tissue sections, we determined that many eYFP+ MYH11- ACTA2- cells expressed the proliferation marker Ki67 (Supplementary Fig 2b). More eYFP+ cells were found at farther distances from blood vessels in the lungs of both B16-F10 (Fig 1d) and M3-9M (Fig 1l) tumor-bearing mice compared to control mice. These findings suggest that perivascular cells in pre-metastatic lung exhibited enhanced proliferation and migration either from existing lung vasculature or other unappreciated sites, which are hallmarks of perivascular phenotypic switching. This increase in phenotypically modulated perivascular cells was specific to pre-metastatic lung tissue in that a similar expansion of perivascular cells was not seen in other tissues of tumor-bearing mice such as the liver (Supplementary Fig. 3), to which the B16-F10 and M3-9M models rarely metastasize, implying that this process is associated with site-specificity of metastasis.

Using flow cytometric analysis, we confirmed and quantified the expansion of eYFP+ cells in pre-metastatic lungs of B16-F10 tumor-bearing mice over a range of time-points (Fig 1g, Supplementary Fig. 4a, b). Pathology review, IVIS imaging, and flow cytometry were utilized to assay for dual luciferase+ mCherry+ tumor cells and confirm that days 5 through 16 post-tumor injection were below the level of detection for metastasis (Supplementary Fig 1f, g). A significant 4.39-fold increase in total eYFP+ cell number was observed in the pre-metastatic lungs of B16-F10 tumor-bearing Myh11 lineage-tracing mice at day 10 post-tumor injection and persisting to later time-points, compared to HBSS-injected non-tumor-bearing Myh11 lineage-tracing littermate control mice (Fig. 1g). A concurrent increase in eYFP+ Ki67+ cells was observed at all pre-metastatic time-points, including at day 3 post-tumor injection prior to the increase in eYFP+ cells at day 10 post-tumor injection. Ki67 marks cells in all stages of the cell cycle leading to cell division, thus this early time-point may represent the early transition from a quiescent to cycling eYFP+ cell population²⁰. Together these data indicate that eYFP+ cells proliferate in pre-metastatic lung (Fig 1g, Supplementary Fig 4d). Similarly, a 3.2-fold increase in eYFP+ cells was observed in the pre-metastatic lungs of M3-9M tumor-bearing Myh11 lineage-tracing mice compared to HBSS-injected non-tumor-bearing Myh11 lineage-tracing littermate control mice (Supplementary Fig. 4c). A decrease in total eYFP+ ACTA2+ cells over time was observed in pre-metastatic lung of B16-F10 tumor-bearing Myh11 lineage-tracing mice (Fig 1g). These data were consistent with our analysis of immunofluorescently stained tissue sections of the similar but more specific eYFP+ ACTA2+ Myh11+ cell population, highlighting the complimentary nature of analysis by confocal imaging and flow cytometry to examine perivascular cells in pre-metastatic lung (Fig 1e, f, j, k). Together these data are indicative of marker down-regulation that is characteristic of perivascular phenotypic switching.

To further characterize perivascular lineage-tracing eYFP+ cells in our pre-metastatic models, we evaluated pericyte markers in eYFP+ cells in the Myh11 lineage-tracing model. We discovered many eYFP+ pericytes in the microcirculation of the heart, skeletal muscle and lung (Supplementary Fig. 5). A subset of eYFP+ cells within the microvasculature of non-tumor-bearing lungs of Myh11 lineage-tracing mice expressed the pericyte markers PDGFRβ and NG2 (Supplementary Fig. 5c). We identified eYFP+ cells dissociated from the microvasculature within pre-metastatic lungs of B16-F10 tumor-bearing Myh11 lineage-tracing mice (Supplementary Fig. 5c). Of note, Myh11 lineage traced eYFP+ cells in the pre-metastatic lungs of mice with B16-F10 tumors did not express macrophage markers F4/80 or LGALS3 as reported in atheromas¹⁷ (Supplementary Figure 4e). While PDGFRα+ cells are found in normal healthy lung, this total population was increased only at later pre-metastatic time-points (day 18) (Supplementary Fig 6a, c). By contrast, eYFP+ PDGFRα+ cells dramatically increased as early as day 10 post-tumor injection (Supplementary Fig. 6a), and we found that eYFP+ PDGFRα+ CD51+ cells increased throughout the time course in pre-metastatic lungs (Fig 1g). This was consistent with previous findings that identified the expansion of a PDGFRα+ fibroblast-like population in the pre-metastatic niche¹. Together, these data demonstrate that pre-metastatic lungs contain an expanded population of eYFP+ perivascular cells that are not in close association with blood vessels and may contribute to PDGFRα+ activated stromal cell population previously described in the pre-metastatic niche.

NG2+ pericytes contribute to the activated perivascular population in the pre-metastatic niche

In addition to the stable eYFP labeling of vSMCs in the Myh11 lineage-tracing model, eYFP+ pericytes were observed in many tissues, including the microvasculature of the heart and spinotrapezius muscle (Supplementary Fig 5a-b). Further, eYFP+ cells in the lungs of non-tumor-bearing Myh11 lineage-tracing mice expressed the pericyte markers NG2 and PDGFRβ (Fig 2a, Supplementary Fig 5c), and the total number of eYFP+ NG2+ cells increases in pre-metastatic lungs over time (Fig 2b). Together, these data suggested that pericytes could substantially contribute to the expansion of Myh11 lineage-tracing eYFP+ cells observed in pre-metastatic lung. Given that NG2+ perivascular cells are a known key component of stem cell niches, particularly in the bone marrow where they support hematopoietic stem cell maintenance¹¹ and fetal liver where they induce hematopoietic stem cell proliferation¹⁰, we sought to determine the contribution of the NG2-expressing pericyte population to the perivascular Myh11 eYFP+ population in pre-metastatic tissue.

To elucidate whether pericytes in particular participate in pre-metastatic niche formation, we generated conditional NG2-ERT-creT2 ROSA-STOP-flox-eYFP lineage-tracing mice (designated as “NG2 lineage-tracing mice”) (Supplementary Fig. 7a)²¹. Upon tamoxifen injection in adult mice, NG2-expressing pericytes are stably labeled by eYFP in many tissues (Supplementary Fig 7b, c). Using this NG2 lineage-tracing model, we observed an expansion of NG2 lineage-tracing eYFP+ cells in the lungs of B16-F10, M3-9M, and E0771 tumor-bearing mice that was comparable to the expanded perivascular population identified in the tumor bearing Myh11 lineage-tracing mice, suggesting that NG2+ pericytes represent an expanded perivascular population in the pre-metastatic setting (Fig. 2c-j, Supplementary Fig 7b). We confirmed and quantified this expansion by immunofluorescent image analysis and flow cytometric analysis (Fig. 2j-k, Supplementary Fig. 7d). eYFP+ cells were only rarely identified in the lungs of NG2 lineage-tracing mice that received non-metastatic melanoma B16-F0 tumors, suggesting that the increase in NG2 lineage-tracing eYFP+ cells correlated with metastatic potential (Fig. 2d). Together, the data reveal that perivascular cells, including NG2-expressing pericytes, exhibit expansion in the parenchyma of pre-metastatic lungs in a number of murine models of metastasis.

Tumor-derived factors induce KLF4 expression in perivascular cells of the pre-metastatic niche

Perivascular cell phenotypic switching is dependent on the expression of KLF4, Krüppel-like factor 4, in several disease models^{7,8,14,16,17,22} (Fig. 3a). Direct KLF4 binding onto the promoters of vSMC marker genes results in down-regulation of marker gene expression and a transition from a contractile SMC phenotype to a plastic, reactive state^17,22,23. We observed an increase in the percentage of eYFP+ cells that were KLF4+ in the pre-metastatic lungs of B16-F10 tumor-bearing Myh11 lineage-tracing mice at day 10 post tumor injection compared to HBSS-injected Myh11 lineage-tracing littermate control mice (Fig 3b-e). As such, we sought to determine whether tumor-derived factors might induce KLF4 expression in perivascular cells that, in turn, mediates perivascular cell plasticity and phenotypic switching.

Previous investigations have demonstrated that in vivo administration of tumor-derived factors such as tumor-conditioned media (TCM) or tumor-derived exosomes potentiates pre-metastatic niche formation^1,24–26. To determine whether perivascular cell activation in the pre-metastatic niche is initiated by tumor-derived factors, we intraperitoneally injected non-tumor-bearing Myh11-lineage-tracing mice to tumor-derived factors such as TCM or exosomes and assayed for phenotypic switching by analysis of Myh11 lineage-tracing eYFP+ cells that no longer express MYH11. eYFP+ MYH11- ACTA2- cells were increased in the lungs of Myh11 lineage-tracing mice exposed to tumor-derived factors that were no longer in close association with blood vessels, nor could be identified by perivascular marker expression, as was seen in tumor-bearing mice at pre-metastatic time points and indicative of their activated state (Fig 3f-j). Pre-conditioning of mice with B16-F10 exosomes or TCM followed by tail-vein injection of B16-F10 melanoma cells resulted in increased metastasis, indicating that pre-treatment with tumor-derived factors is sufficient for inducing perivascular cell plasticity and establishing a pre-metastatic niche that enhances tumor cell colonization and/or proliferation (Supplementary Fig 8a). These studies indicate that tumor-derived factors contribute to perivascular cell phenotypic switching and pre-metastatic niche formation.

The activated state of perivascular cells is characterized by a decrease in marker gene expression and an increase in proliferation, migration, and matrix production (Fig. 3a). To test whether tumor-derived factors induce KLF4-dependent perivascular phenotypic switching, we sought to establish an in vitro model by culturing vSMCs with either B16-F10 metastatic melanoma TCM or exosomes purified from B16-F10 TCM. qRT-PCR profiling of vSMCs exposed to TCM over the course of 72 hours revealed an increase in Klf4 gene expression as early as 2 hours post-treatment with TCM to levels similar to those seen by vSMCs treated with the cytokine PDGF-BB (Supplementary Fig 8b), a known inducer of KLF4-dependent phenotypic switching^23,27. We observed a subsequent decrease in expression of vSMC marker genes, including Acta2, Myh11, and Tagln (Supplementary Fig. 8b). Notably, TCM derived from the non-metastatic melanoma B16-F0 cell line do not induce these gene expression changes, indicating that tumor factor-induced vSMCs phenotypic switching may be specific to tumor metastatic potential (Supplementary Fig 8b).

We extended these gene expression data by performing RNA-seq on cultured vSMCs exposed to TCM over the course of 72 hours. We confirmed that Klf4 gene expression is elevated in vSMCs treated with B16-F10 TCM, and that the expression of vSMC marker genes Acta2 and Tagln were decreased compared to vSMCs treated with serum-free media (Fig 3k). Pathway analysis demonstrated that gene sets related to proliferation, migration, and invasion, all hallmarks of perivascular phenotypic switching, were activated over the course of 72 hours (Fig 3l). We confirmed an increase in proliferation by vSMCs cultured in B16-F10 TCM (Supplementary Fig. 8d), which was indicative of their activated, phenotypically-modulated state. Consistent with our finding that eYFP+ PDGFRα+ cells were increased in the pre-metastatic lungs of B16-F10 tumor-bearing mice (Fig 1g, Supplementary Fig 6b-c), vSMCs that were cultured in vitro with TCM upregulated this activated stromal marker (Fig. 3k, Supplementary Fig 8b-c), suggesting that phenotypically-modulated vSMCs may represent a PDGFRα+ cell population that was previously reported in pre-metastatic tissue in response to tumor-derived factors¹.

TCM from the metastatic B16-F10 cell line, but not from the non-metastatic B16-F0, is sufficient to induce vSMC phenotypic switching in vitro (Supplementary Fig 8b-c), indicating that factors secreted specifically by tumor cells with metastatic potential contribute to the activated vSMC phenotype. As such, we sought to determine which cytokines were elevated in the conditioned media from the metastatic B16-F10 cell line when compared to the non-metastatic B16-F0 cell line using a cytokine array (Supplemental Table 1). Of the 8 factors found to be elevated in this assay, WISP1, ANG2 and IL-1ß all have known roles in vascular biology and inflammation, and in some reports can induce KLF4 expression^28–33. Using a Klf4 promoter reporter that results in the secretion of luciferase upon Klf4 promoter activity, combined with qRT-PCR, we determined that all three cytokines were capable of inducing Klf4 expression in vSMCs to levels comparable to that induced by B16-F10 TCM, and siKlf4 knock-down significantly inhibited this activation (Fig 3m). WISP1 and IL1ß, but not ANG2, induced KLF4-dependent suppression of vSMC marker gene expression (Fig 3n). However, ANG2 neutralizing antibody pre-treatment of vSMCs exposed to B16-F10 conditioned media blocked TCM-induced vSMC proliferation (Extended Fig 8e), indicating that ANG2-induced effects on vSMC activation may work coordinately through KLF4 and other effector proteins.

In addition to B16-F10 TCM, B16-F10 tumor cell-derived exosomes were also capable of inducing perivascular phenotypic switching in vivo (Fig 3h-j), and vSMC phenotypic switching in vitro (Supplementary Fig 8c). Isolated melanocyte-, B16-F0- and B16-F10-derived exosomes were found to be of similar size distribution, at a peak of 133 nm, 85 nm and 99 nm, respectively (Supplementary Fig. 8g) and contained exosomal and extracellular vesicle markers (Supplementary Fig 8h). Fluorescently labelled B16-F0 and B16-F10 exosomes were found capable of entering vSMCs (Supplementary Fig 8i). We thus determined whether any of the three candidate cytokines were contained in metastatic tumor-derived exosomes. We found that only WISP-1, and not IL1ß nor Ang2, was found in metastatic tumor cell derived exosomes including in B16-F10 melanoma, MNNG HOS and HOS osteosarcoma, and Panc02 pancreatic cell lines. WISP-1 was not found in exosomes derived from the non-metastatic B16-F0 cell line nor a non-tumor melanocyte cell line, suggesting that the presence of exosomal WISP1 production may be specific to metastatic tumor cell lines that in combination with non-exosomal cytokines such as IL1ß and Ang2 may enhance tumor metastatic behavior (Fig 3o, Supplementary Fig 8f, Supplementary Fig 9a, d). Circulating WISP-1 was also found to be elevated in B16-F10 tumor-bearing mice at pre-metastatic time-points (Fig 3p) to levels equivalent to that found in B16-F10 exosomes generated in vitro (Supplementary Fig 8f) and administered in vivo to promote perivascular phenotypic switching (Fig 3h-j), whereby this cytokine may contribute to pre-metastatic niche formation (Supplementary Fig 8a). Neutralizing anti-β1 integrin antibodies that are known to interfere with WISP1 activity inhibited both B16-F10 exosome- and rWISP1-induced vSMC proliferation in vitro (Fig 3q). Together, the data suggest that multiple factors secreted by the metastatic melanoma B16-F10 cell line, including but not limited to WISP-1 containing-tumor-derived exosomes, may work in concert through the induction of KLF4 expression to contribute to vSMC phenotypic switching characterized by downregulation of vSMC marker gene expression and increased vSMC proliferation.

Tumor-derived factors induce perivascular phenotypic switching and enhance ECM that supports tumor metastatic behavior

Phenotypically-modulated perivascular cells increase production of ECM proteins, such as fibronectin and various collagens, and ECM-remodeling enzymes, such as matrix metalloproteinases in many disease settings^14,34. ECM modulation can dictate tumor cell fate and metastasis in a number of models^35–38. As has been demonstrated previously¹, we observed an increase in fibronectin staining in the lungs of B16-F10 tumor-bearing Myh11 lineage-tracing mice at day 10 post-tumor injection as compared to HBSS-treated littermate control mice (Fig 4a-b). Elevated fibronectin deposition in pre-metastatic lung was found more closely associated with eYFP+ cells than with eYFP- cells within the same immunofluorescently stained tissue sections (Fig 4c). eYFP+ aortic vSMCs from Myh11 lineage-tracing mice that were isolated and cultured ex vivo in either B16-F10 TCM or tumor-derived exosomes also exhibited an increase in fibronectin production when compared to those cultured in defined serum-free media (Fig 4d-e). We next used an in vitro model to test whether the matrix produced by phenotypically-modulated vSMCs provides a pro-metastatic environment and enhances metastatic behavior in tumor cells. vSMC phenotypic switching was induced in vitro by treatment with either B16-F10 TCM or tumor-derived exosomes for 72 hours, after which vSMCs were lysed from culture plates to leave behind vSMC-derived ECM (Fig 4f-g). B16-F10 tumor cells seeded onto plates containing ECM produced by phenotypically-modulated vSMCs with enhanced fibronectin (Fig 4g) resulted in increased B16-F10 tumor cell adhesion and proliferation compared to those B16-F10 tumor cells which were seeded on plates containing ECM produced by non-phenotypically modulated vSMCs (Fig. 4h, j). B16-F10 tumor cells increased migration across a transwell membrane towards ECM components derived from phenotypically-switched vSMCs educated by B16-F10 exosomes, B16-F10 TCM, or rWISP1, compared to ECM components derived from vSMCs conditioned with B16-F0 exosomes or melanocyte exosomes (Fig 4i). Fibronectin production by vSMCs cultured in B16-F10 TCM was KLF4-dependent in that siRNA-mediated knock-down of Klf4 resulted in a reduction of B16-F10 TCM-induced fibronectin production (Fig 4k, Supplementary Fig 9b), suggesting that activated vSMCs contribute to a pro-metastatic environment, at least in part, through KLF4-dependent fibronectin production. Indeed, B16-F10 tumor cells that are cultured on the ECM derived from phenotypically-switched vSMCs demonstrate an enhanced ability to form tumorspheres (Supplementary Fig 8j), suggesting that phenotypically-modulated vSMC-derived ECM production in response to tumor-derived factors promotes a stem-like environment that promotes features of metastasis-initiating tumor cells^39,40.

Klf4 knockout in perivascular cells limits phenotypic switching and pre-metastatic niche formation

To evaluate whether KLF4-dependent perivascular phenotypic switching is essential for pre-metastatic niche formation and metastasis, we crossed the perivascular cell-specific Myh11 lineage-tracing mice to Klf4 floxed (Klf4-flox) mice, effectively knocking out the Klf4 gene in all perivascular cells as previously described¹⁷. This Klf4-flox Myh11 lineage-tracing model enables simultaneous tamoxifen-inducible perivascular cell-specific Klf4 deletion (Klf4 Δ) and eYFP labeling prior to orthotopic tumor injection (Fig 4l). Despite the potential impact that perivascular phenotypic switching may have on primary tumor angiogenesis and tumor growth, we did not observe a significant difference in B16-F10 primary tumor diameters between Klf4 Δ and wild-type mice (Supplementary Fig.10a), nor involvement of eYFP+ perivascular cells in primary tumor architecture, with eYFP+ cells located only in normal adjacent tissue in full association with well-formed blood vessels (Supplementary Fig 10c). However, perivascular cell-specific Klf4 deletion resulted in a marked decrease in phenotypically-modulated eYFP+ ACTA2- cells in pre-metastatic lungs of B16-F10 tumor-bearing mice at day 10 post-tumor-injection when compared to the lungs of tumor-bearing Klf4 wild-type littermate control mice, as demonstrated by decreased numbers of phenotypically switched eYFP+ cells in the lung parenchyma by immunofluorescent staining and decreased numbers of eYFP+ cells and eYFP+ KLF4+ cells by flow cytometric analysis of whole lung (Fig. 4m-n). Consistent with a decrease in perivascular phenotypic switching with perivascular cell-specific Klf4 deletion, we observed decreased fibronectin in the lungs of B16-F10 tumor-bearing Klf4 Δ mice at day 10 post-tumor injection, compared to tumor-bearing Klf4 wild-type littermate control mice (Fig 4o-p), indicating that Klf4 expression in perivascular cells regulates fibronectin production during the process of pre-metastatic niche formation.

Perivascular cell-specific Klf4 deletion decreases early tumor cell colonization and metastatic burden

To determine the mechanistic role of KLF4-dependent perivascular mediated metastasis in early tumor cell seeding, we performed histological analysis of KLF4 wild-type and KLF4Δ lung sections at early metastatic time points and found decreased metastatic colonization (disseminated tumor cells and micrometastasis) in the lungs of KLF4Δ compared to KLF4 wild-type mice (Supplementary Fig 11a, b). To assess whether these changes in early tumor cell seeding impact metastatic burden, we stably transduced our B16-F10 metastatic melanoma, M3-9M metastatic rhabdomyosarcoma, and E0771 metastatic breast carcinoma cell lines with a mCherry-firefly luciferase fusion protein to detect the presence of tumor by immunofluorescence or bioluminescent signal. Analysis of late-stage metastatic lungs revealed a dramatic decrease in metastasis to the lungs of Myh11 lineage cell-specific Klf4 Δ mice in both the B16-F10 metastatic tumor model at day 23 post-tumor injection and M3-9M metastatic rhabdomyosarcoma model at day 26 post-tumor injection compared to Klf4 wild-type tumor-bearing mice (Figure 5a-d, Supplementary Fig. 11a, Supplementary Fig. 12a-b). No difference was seen in primary tumor growth in the B16-F10 or M3-9M models between Klf4 wild-type and Klf4 Δ mice by tumor volume measurements or bioluminescent signal (Supplementary Fig. 10a). While phenotypically modulated eYFP+ ACTA2- MYH11- cells were found in the late-stage metastatic lungs of both Klf4 Δ and wild-type mice, KLF4+ eYFP+ cells were minimally detected in Klf4 Δ lung sections in contrast to the abundant KLF4+ eYFP+ cells found in KLF4 wild-type lung sections, confirming perivascular cell-specific deletion of Klf4 in Klf4 Δ mice (Fig 5e). Further, there was a significant decrease in the average frequency of eYFP+ cells found at long distances from the closest PECAM+ blood vessels in the lungs of Klf4 Δ mice (Figure 5f).

Consistent with our findings in Myh11 lineage cell-specific KLF4 Δ mice, we found no significant difference in primary tumor growth (Supplementary Fig. 10a), but observed a dramatic decrease in metastasis to the lungs of NG2 lineage cell-specific Klf4 Δ mice in the M3-9M metastatic rhabdomyosarcoma model at day 29 (Fig 5 g-h, Supplementary Fig. 12c), and the E0771 metastatic breast carcinoma model at day 25 (Fig 5 i-j, Supplementary Fig. 12d). Together these data indicate that KLF4-dependent perivascular phenotypic switching is critical for metastatic progression in two genetic murine models of perivascular lineage-tracing and in three different orthotopic cancer models of spontaneous metastasis.

Inhibition of tumor cell binding to fibronectin recapitulates perivascular specific KLF4-dependent metastasis

To assess the specific role of phenotypically-switched perivascular cell fibronectin deposition in promoting tumor metastasis, we utilized siRNA-mediated knock-down of Klf4, Fn1, or both genes together, in vSMCs cultured with B16-F10 TCM. Following vSMC lysis, we seeded B16-F10 tumor cells on the resultant vSMC-deposited ECM (Fig 4f) and performed RNA-seq to determine vSMC specific KLF4- and fibronectin-dependent changes in tumor cell metastatic behavior. Pathways related to tumor cell proliferation, viability, and survival were most activated in B16-F10 cells that were cultured on the ECM of vSMCs treated with TCM compared to those treated with SFM. These pathways were not activated when either Klf4 or Fn1 were knocked down (Fig 6a). Hierarchical clustering analysis demonstrated that siKlf4, siFn1, or siKlf4-siFn1 double knock-down resulted in B16-F10 cell gene expression profiles that clustered together, and clustered with B16-F10 cells that were cultured on the ECM of vSMCs treated with non-target siRNAs (siNT) and serum-free media (SFM) (Fig 6b). 83 genes in the siKlf4-TCM condition overlapped with the siFn-TCM and siKlf4-siFn-TCM conditions, and only 8 genes in the siKlf4-TCM condition did not overlap with either siFn-TCM or siKlf4-siFn-TCM conditions (Fig. 6c). These data suggest that the alteration of B16-F10 gene expression when cultured on vSMC-generated ECM is regulated by exposure to KLF4-dependent fibronectin deposition, which in turn results in activation of pathways associated with enhanced tumor cell survival, viability, and proliferation.

To confirm the role of activated vSMC-dependent fibronectin deposition on tumor cell proliferation, we used neutralizing ß1 integrin antibodies to disrupt B16-F10 tumor cell binding to fibronectin. We found that B16-F10 proliferation is significantly reduced with neutralizing β1-integrin antibody treatment whether the vSMCs were first activated using B16-F10 TCM, recombinant WISP1, or B16-F10 exosomes (Fig 6d). B16-F10 cells did not exhibit elevated proliferation when cultured on the ECM generated by vSMCs cultured with SFM, melanocyte exosomes, nor non-metastatic B16-F0 exosomes, and neutralizing ß1 integrin antibody treatment did not further reduce proliferation in these conditions (Fig 6d). Together, these data suggest that tumor cell secreted factors, including WISP1, induce KLF4 expression in perivascular cells that, in turn, result in enhanced deposition of fibronectin-containing ECM which promotes ß1 integrin-mediated tumor cell proliferation and survival (Fig 6e). To test this hypothesis further, we used our M3-9M metastatic rhabdomyosarcoma model in KLF4 wild-type and KLF4 Δ mice where, upon pre-metastatic niche formation, we treated mice with either neutralizing ß1 integrin antibodies or RGDS peptides to disrupt tumor cell-fibronectin binding (Fig 6f). Analysis of metastatic lungs revealed a dramatic decrease in metastasis with either anti-ß1 integrin (Fig 6g-h) or RGDS peptide treatment (Fig 6i-j) in KLF4 wild-type mice to levels comparable to KLF4 Δ mice treated with either anti-ß1 integrin antibodies, RGDS peptides, or their respective controls. Neither anti-ß1 integrin nor RGDS peptide treatment decreased primary tumor volume in either KLF4 wild-type or KLF4 Δ mice as compared to their respective controls (Supplementary Fig. 10b). Together these findings indicate that tumor cell ß1 integrin-mediated fibronectin binding contributes to the pro-metastatic functions of KLF4-dependent perivascular phenotypic switching.

These findings suggest that KLF4-dependent perivascular cell phenotypic transition is important for pre-metastatic niche formation, and that the lasting effects of KLF4 targeting can be observed in the late-stage metastatic microenvironment that ultimately dictates metastatic outcome. These results demonstrate the potential utility of perivascular plasticity as a novel target to inhibit metastasis. KLF4-targeting to prevent metastasis may be of greatest use in the adjuvant setting following primary tumor removal, or in combination with chemotherapy or other microenvironment modulating agents to decrease tumor metastatic-initiating potential.

Mice

ROSA-STOP-flox-eYFP mice, MyH11-ERT-cre mice and NG2-ERT-cre mice were obtained from Jackson Laboratories. Only male mice inherit the MyH11-ERT-cre allele, and thus any experiments performed using this mouse model were done so using male mice that were ROSA-STOP-flox-eYFP +/+ and MyH11-ERT-cre +/−. The Klf4-flox mice were obtained through the Mutant Mouse Resource and Research Center (MMRRC), to whom they were donated by Klaus Kaestner. MyH11- ERT- cre mice were bred with the Klf4 floxed mice to obtain the Myh11-ERT-Cre Klf4 Δ mice. Genotyping was performed via PCR using previously published protocols. Cre activity was induced in mice aged 6–8 weeks via intraperitoneal injection (1 injection per day, 10 total injections, 10 mg ml⁻1 of tamoxifen in peanut oil). For drug treatment studies, mice were randomized within genotype grouping and drug treatment was performed blinded post-assignment. All experiments were approved by the NCI Animal Care and Use Committee and conducted under specific pathogen-free conditions at the NIH Animal Facility.

Cell lines

The B16-F0, B16-F10, E0771, and MOVAS cell lines were obtained from the ATCC. The M3-9M cell line was derived from the transgenic HGF/p53 knockout mice that develop spontaneous embryonal rhabdomyosarcoma tumors generated by Glenn Merlino and backcrossed to the Bl6 background by Crystal Mackall⁵⁸. All cell lines were authenticated via microarray analysis and tested mycoplasma negative. All tumor cell lines were transduced with a lentivirus expressing a mCherry-firefly luciferase fusion protein and flow sorted for 98% mCherry expression to establish stable cell lines. eYFP+ SMCs were isolated from the aortas of MyH11 lineage-tracing mice post-tamoxifen injection, via flow sorting on eYFP+ expression after negative selection of CD45/CD31/TER119 by Milltenyi biotinylated beads and depletion columns. eYFP+ SMCs were kept in culture for up to 7 passages and then discarded.

B16-F0, B16-F10, MOVAS and eYFP+ SMC cell lines were grown in DMEM with 10% FBS and 1% penicillin/streptomycin/glutamine. The M3-9M cell line was grown in RPMI with 10% FBS and 1% penicillin/streptomycin/glutamine. Serum free media for eYFP+ SMCs is DMEM supplemented with 1% penicillin/streptomycin/glutamine, 0.2 mM ascorbic acid, 5 ug ml⁻¹ transferrin, 2.8 ug ml⁻¹ insulin and 6.25 ng ml⁻¹ sodium-selenate. Serum-free media for all other cell lines are their growth media without FBS.

Orthotopic and tail-vein tumor models

Orthotopic tumor injections were administered with 5×10⁵ tumor cells in single-cell suspension of 100 µL of Hank’s Balanced Salt Solution (HBSS) using a sterile 27.5-gauge needle. The B16-F0 and B16-F10 melanoma cell lines were administered intradermally in the flank. The M3-9M rhabdomyosarcoma cell line was administered in the gastrocnemius muscle of the right hind leg. The E0771 breast carcinoma cell line was administered into the mammary fat pad. Tail-vein tumor injections were administered with 1×10⁵ tumor cells in single-cell suspension of 100 µL HBSS using a sterile 27.5-gauge needle into the lateral tail vein. Metastatic end-point was determined per experiment by maximum primary tumor burden and/or the appearance of moribund mice within all treatment groups as defined within protocols established by the NCI Animal Care and Use Committee.

Bioluminescent tumor cell tracking

For in vivo whole-animal imaging, anesthetized mice received 3 mg of D-luciferin (Gold Biotechnology) via intraperitoneal injection (i.p.). Luminescence readings were collected on a Xenogen IVIS 200, starting at 5 minutes post i.p. injection for 1 minute. For ex vivo tissue imaging, tissues were perfused with PBS, harvested, and incubated in 1 µg ml⁻¹ D-luciferin in PBS. Luminescent readings were collected starting at 5 minutes later, for 1 minute. Analysis was conducted using the Xenogen Living Image software package.

Immunofluorescence

Tissues were first cardiac perfused with PBS followed by 4% PFA. Tissues were then harvested, fixed overnight in 4% PFA at 4°C, then transferred to 30% glucose and incubated at 4°C until tissues sink, or up to 3 days. Tissues were then embedded in OCT compound (Tissue-Tek) and cryosectioned for staining. Sections were air-dried for 2 hours, washed in PBS, permeabilized in 0.5% Triton-X, blocked in 10% fish skin gelatin/10% donkey serum, and then stained with the following antibodies: 4,6’-diamidino-2-phenylindole (DAPI) (Invitrogen) as a nuclear marker; goat anti-GFP (ab6673, Abcam) for lineage marker eYFP; mouse anti-ACTA2 (SMA) conjugated to FITC (1A4, Sigma); rat anti-Myh11 (KM3669, Kamiya); rabbit anti-PECAM1 (AB28364, Abcam); rabbit anti-KLF4 (HPA002926, Sigma); rabbit anti-fibronectin (ab23750, Abcam); rabbit anti-PDGFRa (APA5, Abcam); rabbit anti-PDGFRb (Y92, Abcam); rabbit anti-Ki67 (ab16667, Abcam); mouse anti-NG2 conjugated to biotin (ab5320b, Millipore); and chicken anti-mCherry (ab205402, Abcam). Donkey secondary antibodies were used where necessary (Jackson ImmunoResearch).

Cells were plated and treated on autoclaved glass coverslips placed in sterile 6-well plates. Upon conclusion of the experiment, cells were fixed in 4% PFA, permeabilized in 0.5% Triton-X, blocked in 10% donkey serum, and then stained as detailed above.

Images for phenotyping were acquired with a Zeiss Axio Observer Z1 inverted microscope fitted with an Apotome.2. Confocal fluorescence images were acquired with a Zeiss LSM880 laser scanning confocal microscope equipped with a 34-channel spectral detector and 20× plan-apochromat (air, N.A. 0.8) and 63× plan-apochromat (oil, N.A. 1.4) objective lens (Carl Zeiss Microscopy, LLC, Thornwood, NY). A five-channel imaging configuration was used to acquire images from multi-labeled fluorescent samples. The fluorophores used for imaging, with respective laser excitation and fluorescence emission detection windows, were DAPI (405 nm laser, 415–490 nm emission), FITC (488 nm laser, 490–535 nm emission), Cy3 (561 nm laser, 563–585 nm emission), AlexaFluor 594 (594 nm laser, 596–631 nm emission), and AlexaFluor 647 (633 nm laser, 655–742 nm emission). Confocal images were collected with a 1024×1024 pixel frame size, and with the 20× objective lens a 0.28 µm X-Y pixel size, 3.9 µm optical slice thickness, and 4× image frame averaging, and with the 63× objective lens 0.09 µm X-Y pixel size, 1.2 µm optical slice thickness, and 4× image frame averaging. Imaging parameters were kept constant for all images in an experiment. Isotype and single fluorophore labeled control samples were imaged using the same five channel imaging configuration so that the potential degree of fluorescence cross-talk between image channels could be compensated for during image analysis. For extended field of view tile imaging, image frames were collected with 5% overlap and stitched together into one large image file using Zeiss Zen blue (v. 2.3) image processing software. After background subtraction based on the isotype and single-label control samples, the brightness and contrast were adjusted for each fluorescence channel using linear histogram stretching and the adjustments were made using the same parameters for all images in a dataset. The resultant images were exported as Tiff files and composed into figures using Adobe Photoshop. Deconvolution and figure preparation was performed using the Zeiss Zen software package, and all particle analysis was conducted using the CellProfiler 2.1.1 software package (cellprofiler.org)⁵⁹.

Image Analysis

eYFP+ particles were identified using CellProfiler 2.1.1 (cellprofiler.org)⁵⁹, whereby Parent particles were identified based on DAPI positive nuclei, and eYFP+ Child particles were identified based on Parent particles surrounded by eYFP+ staining. The total number of eYFP+ Child particles was quantified in three 20× tiled images (4×4 average tile area) per mouse, and reported as average eYFP+ particles per tiled field. KLF4+ eYFP+ cells were identified in eYFP+ Child particles by KLF4 staining restricted to the nucleus of each eYFP+ particle. The average proximity of eYFP+ particles to blood vessels was computed by determining the radial distribution of PECAM to eYFP+ Child particles, reported in bins of 3 µm and 27 µm radial distances from the nucleus of each eYFP+ particle. Each eYFP+ particle was then analyzed to determine the closest bin containing PECAM signal. Each bin was summed per tiled image, and then reported as an average bin signal per mouse. Signal density of fibronectin staining was determined in ImageJ by recording the integrated density in at least 3 fields of each stained sample.

For image analysis, the original images were used for mean intensity measurements, and the background subtracted based on single-color and isotype control images. Briefly, individual regions of interest (ROIs) were drawn for different representative regions within the images to measure the mean intensity for each fluorescence channel. For the series of single label control samples, the mean intensity was also measured for each of non-labeled channels and the highest background value recorded was used to subtract from the measurements made from the corresponding image channel in the multi-label samples. 10–25 ROIs were drawn in each image centered around DAPI stained nuclei and 5 representative images per lung specimen were included for analysis per dataset. The background subtracted mean fluorescence intensities for YFP positive lineage traced cells and YFP negative cells were plotted using Prism. Each dataset was analyzed independently by two investigators.

Flow cytometry

Lungs were first perfused with PBS and then insufflated with 1 mg ml⁻¹ Collagenase I, 10 µg mL^-1 Dispase II and 20 µg mL⁻¹ DNase I. Lungs were then harvested and placed in 10%FBS/5%BSA/PBS on ice. Lung tissue was finely minced, digested in collagenase I/Dispase II/DNase I at 37°C for 30 minutes with gentle agitation, and passed through a 70 µm nylon filter. Total cell counts per lung were determined using Trypan cell exclusion from single cell suspensions. Live cells were determined via exclusion of viability dye (eBioscience). eYFP+ cells were identified by their negative staining by antibodies to CD45 (30-F11, eBioscience), TER119 (TER119, eBioscience), and CD31 (390, Biolegend) in 0.5% BSA/PBS with 0.05% NaN₃, and positive eYFP signal. Extracellular markers included CD51 (RMV-7, eBioscience), PDGFRα (APA5, eBioscience), NG2 (ab5320b, Millipore) Mac-2 (M3/38, eBioscience), CD11b (M1/70, eBioscience), and F4/80 (BM8, eBioscience). Where biotinylated antibodies were used, an APC-e780 streptavidin (eBioscience) staining step was added. For intracellular/nuclear staining, cell surface markers were first stained as stated above followed by fixation/permeabilization/staining for all other markers using a Foxp3/Transcription Factor Staining Buffer Set (eBioscience). Because fixation decreases eYFP emission, eYFP+ cells were identified using an anti-GFP antibody (5F12.4, eBioscience). Other intracellular/nuclear antibodies included Ki67 (SolA15, eBioscience), ACTA2 (1A4, eBioscience), and KLF4 (HPA002926, Sigma). Unconjugated KLF4 staining was followed by a PE-conjugated secondary (eBioscience). Flow cytometry data was collected on a BD LSRFortessa, and analyzed using the FlowJo software package (Tree Star). Cell populations were reported as the proportion of live single cells identified by flow, multiplied by the total number of cells isolated from each lung.

Tumor-conditioned media and exosome preparation

Tumor cells were plated at 1×10⁵ cells ml⁻¹ in T75 flasks and allowed to adhere overnight in growth medium. Growth medium was then removed, and the cells were washed in PBS and cultured in serum free media for 72 hours. This tumor-conditioned media was collected, filtered through a 0.22 µm filter, aliquoted and stored at −80°C for downstream experiments. Exosomes were isolated from tumor-conditioned media prior to storing at −80°C via ultracentrifugation at 25,000 rpm, 4°C, for 2 hours. Pelleted exosomes were re-suspended in 35 µL HBSS, quantified for protein content by BCA assay, and stored in 1 µg µL⁻¹ aliquots at −80°C for downstream experiments. Aliquots were characterized using Nanosight to confirm uniform size distribution. For exosome labeling experiments, aliquots were thawed, diluted in 1 mL DilC, then mixed with 1 mL DilC containing 4 µL of PKH26 dye. After 5 minutes incubation at room temperature, 2 mL of 1% BSA was added for 1 minute to inhibit staining. Exosomes were then pelleted by ultracentrifugation and used directly in experiments. For in vivo pre-treatment experiments, 6 intraperitoneal injections were given over the course of 14 days of either 100 ul of tumor-conditioned media, or 10 ug of exosomes per injection.

Cytokine Array Analysis

Tumor-conditioned media from B16-F0 and B16-F10 was prepared as described above and then applied to RayBiotech C1000 antibody array as per manufacturer protocol. Membranes were analyzed by densitometry using the BioRad ChemiDoc imaging system.

Quantitative real-time PCR

Total RNA was isolated and precipitated using TRIzol followed by chloroform/ethanol washes. RNA concentration and quality was assessed using the Agilent 2100 Bioanalyzer. cDNA synthesis was performed using 1 µg RNA per sample and the Biorad iScript cDNA Synthesis Kit. qRT-PCR was performed using the Biorad SsoAdvanced master mix on a Biorad iCycler.

KLF4 knock-down and reporter plasmid

A KLF4 promoter-reporter plasmid that expresses the secretable Gaussia Luciferase was purchased from Genecopoeia, and transfected into plated MOVAS cells. 24 hours post-transfection, siRNAs targeting either KLF4 or non-target control were added to MOVAS cells according to manufacturer’s protocol (Dharmacon Accel siRNAs). 72 hours post-siRNA treatment, media was replaced with serum-free media containing recombinant proteins or tumor conditioned media. 16 hours after protein-containing treatment, media was removed and assayed for luciferase activity using the BioLux Gaussia Luciferase Assay Kit (New England BioLabs). Reporter plasmid activity is reported in luminescence as compared to empty plasmid control. Western Blots were also performed after TCM treatment of KLF4 knocked down vSMCs to probe for KLF4 (HPA002926, Sigma) and fibronectin (ab23750, Abcam), and were imaged using the BioRad ChemiDoc imaging system.

Western blotting

Exosomes or cell pellets were lysed in Cell Lysis Buffer (Cell Signaling) with protease and phosphatase inhibitor cocktail (Millipore). All lysates were cleared by centrifugation, and quantified by bicinchoninic acid (BCA) protein assay (ThermoFisher). 5 µg of protein from each sample were electrophoresed on 4–12% Bis-Tris Protein gels, and transferred to nitrocellulose membranes (ThermoFisher). Blots were probed with primary antibodies: anti-Angiopoietin-2 (2948, Cell Signaling), anti-WISP1 (ab178547, Abcam), anti-PDGF BB (ab23914, Abcam), anti-IL-1beta (ab200478, Abcam), anti-FN (1574-1, Epitomics), anti-KLF4 (HPA002926, Sigma), anti-GFP (ab6673, Abcam), anti-β-actin (8457s, Cell Signaling), anti-CD81 (Eat2 Thermo Scientific), anti-CD63 (ab193349, Abcam), anti-mouse CD9 (ab82390, Abcam), anti-human CD9 (ab92726, Abcam) , anti-Flotillin (18634, Cell Signaling), anti-Alix (2171, Cell Signaling), anti-Annexin V (8555, Cell Signaling), LAMP2 (ab203224, Abcam), anti-HSP70 (4876, Cell Signaling), anti-EpCAM (2626, Cell Signaling), and anti-GM130 (12480, Cell Signaling), then probed with the secondary antibody anti-rabbit IgG-HRP (7074s, Cell Signaling), or anti-mouse IgG-HRP (7076, Cell Signaling), or anti-Armenian hamster (PA1-32045, Thermo Scientific). HRP-based detection was performed using SuperSignal West Substrate (Thermo Scientific).

RNA sequencing

RNA sequencing was performed using the standard operating procedure of the CCR Genomics core facility. mRNA samples were prepared using Illumina TruSeq Stranded mRNA kit (Illumina). Samples were pooled and sequenced using the NextSeq500 platform in high output mode. Demultiplex was done allowing 1 mismatch in the barcodes. All reads for each sample were trimmed adapters and low quality bases using Trimmomatic software and aligned to the reference using STAR software. The data normalization and statistical analysis used to identify differentially expressed genes was performed using the Partek Flow installed in the NIH Helix cluster (https://partekflow.cit.nih.gov). The fastq was aligned to mm10 genome using STAR-2.4.1d aligner, the transcript abundance was estimated using the Partek E/M algorithm based on mm10-Ensembl Transcipts release 82 transcriptome model. Gene counts were normalized to total read count per sample and the differential gene expression (GSA) algorithm was used to detect differentially expressed genes. Pathway analysis was conducted using Ingenuity Pathway Analysis (Qiagen), using a 2-fold expression cut-off for all comparisons.

Wisp-1 ELISA

Exosomes as prepared above were assayed using the R&D Mouse/Rat WISP1/CCN4 Quantikine ELISA kit. Mouse plasma samples were obtained from terminal cardiac punctures and processed according to manufacturer’s recommendations.

In vitro SMC-derived matrix experiments

SMCs were plated at a density of 1×10⁵ cells mL⁻¹ and allowed to adhere to culture plates overnight. Cells were washed with PBS and media was replaced with either serum-free media, tumor-conditioned media, exosomes, or recombinant cytokines. 72 hours post-media treatment, SMCs were lysed in a solution of 1% SDS, 1% Triton-X-100, and 20 mM NH₄OH for 5 minutes. The remaining matrix was washed in PBS 3 times and used immediately for downstream experiments.

Tumor cell adhesion was assessed by plating 1×10⁴ tumor cells mL⁻¹ onto SMC matrix, and then removing the tumor cells after 2 hours. Tumor cell confluence per well was determined using an IncuCyte Zoom (Essen BioScience) and accompanying software. Tumor proliferation was assessed using CellTiter Glo following the manufacturer’s provided protocol. Tumor migration was assessed by plating tumor cells on top of a 0.8 µm transwell membrane in serum-free media above an ECM-coated well. Percent migration was counted as the number of DAPI+ cells that were identified under the transwell membrane at 8 hours post-plating.

Tumorsphere formation was assessed by allowing tumor cells to proliferate on SMC-generated matrix for 72 hours, after which tumor cells were harvested, counted, and plated in ultra low-attachment plates (200 cells per well) in DMEM/F12 supplemented with B27 supplement, 20 ng mL⁻¹ epidermal growth factor, 10 ng mL⁻¹ basic fibroblast growth factor, 5 µg mL⁻¹ insulin, and 0.4% FBS. Cells were cultured for 1 week, after which spheres were imaged using a phase-contrast microscope at 40× magnification. Tumorsphere were distinguished from cellular aggregates by their solid spherical shape with a diameter of at least 50 µm, and reported as the average number of tumorspheres per 40× field.

Statistical analysis

All analysis was performed using GraphPad Prism version 6.04, www.graphpad.com, with the exception of the Jonckheere-Terpstra test which was performed in R. Sample size was determined by running an a priori power analysis using G*Power with data collected from pilot experiments⁶⁰. Data are expressed as mean +/− standard deviation. Unless otherwise stated, statistical differences were evaluated using an unpaired, nonparametric Student’s t-test. Data were evaluated for normality using the D’Agostino-Pearson omnibus normality test. Two-tailed Fisher’s exact test was used for categorical data. Jonckheere-Terpstra test was performed for data that represent independent values over time. Significant differences between experimental groups were represented as *P<0.05, **P<0.001, ***P<0.0001, and ****P<0.00001.