Endothelial-to-mesenchymal transition (EndMT) occurs during development and underlies the pathophysiology of multiple diseases. In tumors, unscheduled EndMT generates cancer-associated myofibroblasts that fuel inflammation and fibrosis, and may contribute to vascular dysfunction that promotes tumor progression. We report that freshly isolated subpopulations of tumor-specific endothelial cells (TEC) from a spontaneous mammary tumor model undergo distinct forms of EndMT in response to TGFβ stimulation. Although some TECs strikingly upregulate α smooth muscle actin (SMA), a principal marker of EndMT and activated myofibroblasts, counterpart normal mammary gland endothelial cells (NEC) showed little change in SMA expression after TGFβ treatment. Compared with NECs, SMA+ TECs were 40% less motile in wound-healing assays and formed more stable vascular-like networks in vitro when challenged with TGFβ. Lineage tracing using ZsGreenCdh5-Cre reporter mice confirmed that only a fraction of vessels in breast tumors contain SMA+ TECs, suggesting that not all endothelial cells (EC) respond identically to TGFβ in vivo. Indeed, examination of 84 TGFβ-regulated target genes revealed entirely different genetic signatures in TGFβ-stimulated NEC and TEC cultures. Finally, we found that basic FGF (bFGF) exerts potent inhibitory effects on many TGFβ-regulated genes but operates in tandem with TGFβ to upregulate others. ECs challenged with TGFβ secrete bFGF, which blocks SMA expression in secondary cultures, suggesting a cell-autonomous or lateral-inhibitory mechanism for impeding mesenchymal differentiation. Together, our results suggest that TGFβ-driven EndMT produces a spectrum of EC phenotypes with different functions that could underlie the plasticity and heterogeneity of the tumor vasculature. Cancer Res; 75(7); 1244–54. ©2015 AACR.
Endothelial-to-mesenchymal transition (EndMT) is defined as the loss of endothelial-specific factors and gain of mesenchymal features that accompany the morphogenesis of specific tissues, especially those of the heart (1). EndMT also occurs in various pathologic conditions, including cerebral cavernous malformations, cardiac and kidney fibrosis, vein stenosis, and cancer (2–7). In these diseased states, aberrantly regulated EndMT results in unscheduled conversion of endothelial cells (EC) into diverse mesenchymal-lineage cell types, especially myofibroblasts, that may dissociate from the vessel wall and can be found throughout the affected tissue (8–14). EndMT coincides with genome-wide reprogramming that allows ECs to exist in a variety of phenotypic states but may also cause vascular dysfunction that contributes to disease progression (9, 15, 16).
The concept that ECs may “drift” toward mesenchymal-like cell types was shown many years ago in TGFβ-treated EC cultures (17). TGFβ, and other members of the TGFβ superfamily, impart plasticity to ECs by activation of specific transcription factors (e.g., Snail, Slug, and Twist) that interact with chromatin modifiers to create genome-wide epigenetic reconfigurations (15, 18). Whether acquisition of the mesenchymal program in EC is a partial and transient or stable and transmittable change in cellular specification is not clear. The context dependency of TGFβ, combined with its cell-type specific activity, has made it challenging to understand how ECs react to sustained TGFβ signaling, particularly in complex tissues such as tumors (19, 20). In addition, EC cellular responses to TGFβ may proceed differently in different types of EC owing to heterogeneous forms of mesenchymal differentiation throughout the vasculature that is tumor-type and/or tumor-stage dependent.
Recently, we isolated normal mammary gland ECs (NEC) and mammary tumor ECs (TEC) from a transgenic tumor model (21). We were surprised to find a wide range of mesenchymal-like genetic signatures among these different types of ECs challenged with TGFβ in vitro. Furthermore, although some TGFβ-stimulated mesenchymal genes (e.g., smooth muscle actin α; SMA) were blocked by addition of basic FGF (bFGF), expression of other mesenchymal genes was augmented by the combination. These results require a refined assessment for how EndMT is defined across different vascular beds. In principle, EndMT may be characterized by a spectrum of intermediate and reversible mesenchymal-cell phenotypes, especially in pathologic tissues. Moreover, ECs resist specific conversion to SMA+ myofibroblast-like cells when challenged with TGFβ through secretion of bFGF; thus, ECs maintain their own fate and differentiation by a self-regulatory mechanism that counteracts TGFβ activity.
Materials and Methods
EC isolation, cell culture, and media
Mammary NECs and TECs were isolated from C3-TAg (FVB/N C31-TAg) mice and FVB wild-type mice, respectively (21). We previously isolated prostate TECs from TRAMP mice (8), and K-Ras lung TECs were isolated from K-RasG12D mouse lung tumors (22). Fluorescently labeled acetylated low-density lipoprotein (DiI-Ac-LDL; Biomedical Technologies) was incubated with lung TEC clones to examine cell purity under a fluorescent microscope (23). EC clones were maintained in 1 g/L d-glucose DMEM (LG-DMEM, Gibco) supplemented with 10% FBS, 10% Nu-Serum IV (BD), 5 ng/mL bFGF (Peprotech), 10 ng/mL VEGF A (Peprotech), and 20 USP U/mL heparin (Sigma). Mouse mesenchymal stem cells (MSC) were purchased from Gibco and were maintained in DMEM/F-12 (Gibco) with 10% MSC-qualified FBS (Gibco). All media were supplemented with antibiotic–antimycotic (Gibco).
Unless otherwise stated, 5 ng/mL bFGF and/or 10 ng/mL TGFβ2 in 20% FBS or 1% FBS LG-DMEM was used to treat cells for 48 hours before protein or RNA extraction. BMP-2 (Peprotech), BMP-6 (Peprotech), and BMP-7 (Peprotech) were used at 100 ng/mL. Small-molecule inhibitors were added to the media for one hour before TGFβ2 treatment. The following reagents used were: TGFβ2R inhibitor (SB431542, Sigma), PI3K inhibitor LY294002 (Sigma), Akt inhibitor VIII (EMD Millipore), Smad3 inhibitor SIS3 (Calbiochem), mTOR inhibitor rapamycin (Sigma), MEK (MAPK/ERK kinase) 1/2 inhibitor U0126 (Calbiochem), bFGF neutralizing antibody (Millipore, 05-117), and FGF receptor (FGFR) kinase inhibitor (Calbiochem).
Total RNA isolation was completed using an RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions, and cDNA synthesis was carried out using an iScript cDNA Synthesis Kit (Bio-Rad). qPCR was run in triplicate in 10 μL/reaction with 2× Maxima SYBR Green (ThermoFisher) on an Applied Biosystems Step One Plus analyzer. The threshold cycle (Ct) values were determined by Step One Software 2.2.2 by Applied Biosystems. Ct values of Gapdh gene expression were used as an endogenous control. The relative expression of each gene was quantified using the formula: 2e(Ct of Gapdh − Ct of gene X) = fold increase of reference gene expression. Primer sequences are available upon request. Heatmaps were generated using Gene-E (http://www.broadinstitute.org/cancer/software/GENE-E/.)
Western blot analysis
Cells were lysed in RIPA buffer complemented with phosphatase and protease inhibitor cocktails (Sigma) for protein extraction. Protein concentrations were determined by Bradford assays, and approximately 30 μg per sample was used for Western blotting. For non-phosphorylated protein detection, membranes were blocked and antibodies were added in 5% milk TBS with 0.1% Tween 20 (TBST), and for phosphorylated protein detection, 5% BSA TBST was used. Membranes were incubated with primary antibodies at 4°C overnight and then with secondary antibody at room temperature for one hour. Primary antibodies: 1:1,000 mouse anti-SMA (Sigma, A5228), 1:1000 rabbit anti-phospho-Ser 465/467 SMAD2 (pSMAD2; Cell Signaling Technology, 3108), 1:1,000 rabbit anti-phospho-Ser 423/425 SMAD3 (pSMAD3; Millipore, 0713-89), 1:1,000 rabbit anti-SMAD2 (Cell Signaling Technology, 5339), 1:1,000 rabbit anti-SMAD2/3 (Cell Signaling Technology, 8685), 1:1,000 rabbit anti-fibronectin (Abcam, ab2413), 1:1,000 rabbit anti-bFGF (Sigma), 1:1,000 rabbit anti-phospho-Thr 202/204 ERK (pERK) 1/2 (Cell Signaling Technology, 4370), 1:2,000 rabbit anti-ERK1/2 (Cell Signaling Technology, 9102), 1:1,000 rabbit anti-VEGF receptor 2 (VEGFR2; Cell Signaling Technology, 55B11), and 1:2,500 rabbit anti-GAPDH (Cell Signaling Technology, 5174). Secondary HRP-conjugated antibodies: 1:10,000 horse anti-mouse and 1:10,000 goat anti-rabbit antibodies (Vector Laboratories).
Wound closure scratch assay and live imaging
Cells were plated at 1.0 × 105 cells/well in 6-well plates. Twenty-four hours later, the monolayer was gently scratched with a 200 μL pipette tip across the center of the well. An Olympus IX70 Inverted Live Cell System was used for time-lapse imaging of the cells at a minimum of four locations/well at 20-minute intervals until the scratch wound was completely closed. The images were acquired with the Volocity 6.2 software package (PerkinElmer) and analyzed using TScratch software (available at: http://www.cse-lab.ethz.ch) according to the developers' instructions. (24). The open areas on the images were quantified with the software's automated image analysis and expressed as relative area closure with an arbitrary area unit assigned by the software. Phase contrast images were captured with a Hamamatsu ORCAR2 camera.
Matrigel tube formation assay
Growth factor-reduced Matrigel (Corning, 356230) was first plated into 96-well plates and allowed to set for 30 minutes at 37°C. Cells were preincubated in 10% FBS medium with or without TGFβ2 for 16 hours before being detached and plated in 10% FBS medium with or without TGFβ2 in Matrigel-containing wells in triplicate at a density of 1.0 × 104 cells/well. Phase contrast images (4×) were taken on an Evos XL Core Cell Imaging System (Life Technologies) at approximately 7 hours. Images were processed with ImageJ using the “find edges” feature to enhance the contrast. Quantification was done by counting vessel-like cords that were formed by at least two nonadjacent cells. A 3 × 3 grid was superimposed on each image, and at least four random squares were counted per image to obtain an average number of tubes per image.
Flow cytometry was performed using a BD Accuri C6 Flow Cytometer as previously described with data analyzed post hoc using FloJo (version X; ref. 21). Antibodies: PE rat anti-mouse CD31 (BD Pharmingen, 553373), PE rat anti-mouse CDH5 (VE-cadherin; BD Pharmingen, 562243), and rat anti-mouse IgG (BD Pharmingen, 55393) at a ratio of 1.5 μL antibody to 100 μL of cell suspension.
Tumor studies in mice
C3-TAg mice were provided by the Mouse Phase 1 Unit from the Lineberger Comprehensive Cancer Center at University of North Carolina (UNC) Chapel Hill (Chapel Hill, NC). Tumors were harvested when mice were approximately 5 months of age. ZsGreenCdh5-Cre mice were generated by crossing R26 ZsGreen mice (Jackson Laboratory, 007906) and Cdh5-Cre mice (Jackson Laboratory, 006137). E0771 murine mammary tumor cells (CH3 BioSystems, 940001) or PyVMT tumor cells isolated from the PyVMT mice were suspended in Matrigel (Corning, 356234) at a density of 1.0 × 107 cells/mL, and 100 μL of cell suspension was orthotopically injected into the mammary fat pats of 7-week-old female ZsGreenCdh5-Cre mice (25). Tumors were harvested when they reached 1 cm3 in size, and normal mammary glands from age-matched littermates were harvested at the same time. Lung tumors were obtained from K-RasG12D mice (22). Tissues were fixed in 4% paraformaldehyde and cryoprotected in 30% sucrose-PBS before OCT embedment and cryosection. All mouse experiments were carried out under approval of the Institutional Animal Care and Use Committee at UNC-Chapel Hill.
Immunofluorescence methods were described previously (8). Antibodies used were: 1:100 rat anti-mouse CD31 antibody (BD, 550274), 1:200 Alexa Fluor 488 goat anti-rat antibody (Invitrogen, A11006), and 1:500 monoclonal mouse anti–α-SMA Cy3 antibody (Sigma, C6198). Slides were mounted with Vectashield Hardset Mounting Medium with DAPI (Vector Laboratories) and imaged on a Zeiss CLSM 710 or 700 Spectral Confocal Laser Scanning Microscope.
Three-dimensional confocal microscopy
Sections were stained as described above, and imaged on a Zeiss CLSM 710 Spectral Confocal Laser Scanning Microscope using cubic voxels to capture the z-dimension. Three-dimensional projections, orthogonal slices, and the Supplementary Movie (Movie S1) were generated using ImageJ (http://imagej.nih.gov/ij/).
TGFβ pathway qPCR array
The array was carried out in duplicate using the Mouse TGFβ Signaling Targets RT2 Profiler PCR Array (SA Biosciences, PAMM-235Z) according to the manufacturer's instructions. Cells were plated and exposed to TGFβ2 and/or bFGF for 48 hours. Total RNA was purified with an RNeasy Mini Kit (Qiagen) and cDNA synthesis was performed with an iScript cDNA Synthesis Kit (Bio-Rad). ABI 7900HT Quantitative PCR System was used for the qPCR array and the data were analyzed with SA Biosciences' RT2 Profiler PCR Array Data Analysis Online Software.
Conditioned media treatment of TEC-H8
Conditioned media (CM) collected from TEC-H8 treated with or without TGFβ2 (10 ng/mL) for 48 hours were concentrated using Amicon Ultra-15 Centrifugal Filter Unit (Millipore) or Pierce Protein Concentrators (Thermo Scientific). For Western blot analysis of bFGF, 25 μL of the concentrated CM was used, and the remaining concentrated CM were then divided equally to treat a new batch of TEC-H8 for 48 hours in the presence or absence of freshly added TGFβ2. Cell lysates were collected for Western blot and qPCR analyses.
All values are expressed as ± SEM. Results were analyzed by a Student t test or ANOVA using GraphPad Prism 5 software. P values less than 0.05 were considered significant.
Results and Discussion
Isolation of TECs with an intermediate EndMT phenotype
We recently isolated multiple NEC and TEC clones from wild-type mice and spontaneous mammary tumors in C3-TAg mice using a high-fidelity cloning method producing pure EC populations uniformly positive for CD31 and CDH5 (Fig. 1A and B; ref. 21). Surprisingly, a number of TEC clones showed elevated basal mRNA expression of the EndMT and myofibroblast differentiation marker, SMA (Acta2; Fig. 1C). On the other hand, expression of endothelial markers Cd31 and Vegfr-2 were variable among all NEC and TEC clones and did not correlate with the levels of Acta2 mRNA (Fig. 1C). When challenged with TGFβ2, TEC-H8 (a clone with high basal SMA expression) upregulated SMA protein expression, whereas an NEC clone (B12) and TEC-A2 (a SMA-low TEC clone) were relatively unresponsive (Fig. 2A). TEC-H8 were responsive to all three TGFβ isoforms and similar results were obtained using an additional mammary TEC clone with high SMA mRNA (TEC-D8) and mesenchymal-like prostate TECs previously described by us (Fig. 2B and C; ref. 8). As non-EC (e.g., SMA+ fibroblast) contamination can occur during EC isolation, we performed staining with SMA and CD31 to exclude this possibility. All individual cells in the TEC clone examined expressed CD31, and confocal images clearly revealed colocalization of SMA+ stress fibers and CD31 in the same cells post-TGFβ2 treatment (Fig. 2D). SMA+ ECs have been observed in the luminal surface of healthy thoracic aorta, but their density is markedly increased in atheromatous aorta (26). The appearance of SMA+ ECs correlates with a proinflammatory state, such as in tumors or fibrosis, where TGFβ is also highly upregulated (27, 28). It is possible that the tumor microenvironment promotes the conversion of SMA− ECs into SMA+ ECs, or favors the enrichment of rare but preexisting SMA+ EC located throughout the mammary gland vasculature.
To test whether there were any functional differences between SMA-high TECs and NECs, we carried out a time-lapse wound-healing migration assay using TEC-H8 and NEC-B12 clones. NEC-B12 challenged with TGFβ2 showed little difference in migration compared with untreated controls, whereas TEC-H8 migration was inhibited by approximately 40% (Fig. 3A). TGFβ2 also markedly enhanced TEC-H8 contraction in collagen gel contraction assays, while exerting little effect on NEC-B12 (D.J. Kim; unpublished data). These results are consistent with previous studies showing impaired migration of SMA+ myofibroblasts and are perhaps related to higher contractility and increased focal adhesions associated with filamentous actin stress fibers (29–31). In contrast, formation of in vitro vascular structures was inhibited by 60% to 80% in NEC-B12 and SMA-low TECs (TEC-A2) challenged with TGFβ2 but were increased by 40% in TEC-H8 (Fig. 3B). Taken together, our data suggest that subpopulations of EC are differentially receptive to TGFβ stimulation: some are characterized by high SMA induction, decreased migration, and stabilized vascular structures, whereas others respond oppositely.
TGFβ induces diverse genetic signatures in different types of ECs
We next used qPCR to compare the expression pattern of 12 mesenchymal marker genes in TGFβ2-treated NEC-B12 versus TEC-H8 cultures. Strikingly, we found the pattern of gene expression to be entirely opposite when these genes were hierarchically clustered (Fig. 4A). For example, although NEC-B12 strongly upregulated mesenchymal markers, including calponin 1 (Cnn1), transgelin (Tagln), cadherin 11 (Cdh11), and endosialin (Cd248), TEC-H8 strongly increased the expression of Acta2 (SMA), fibronectin 1 (Fn1), platelet-derived growth factor (PDGF) receptor β (Pdgfrβ), and desmin (Des). Basal FN1 protein expression was also slightly higher in TEC-H8 and was further increased after TGFβ2 stimulation as shown by Western blotting (Supplementary Fig. S1A). Another cluster of mesenchymal genes, including tenascin c (Tnc), endoglin (Eng), and vimentin (Vim), were slightly elevated in TEC-H8 and their levels were variably altered by TGFβ2 treatment. The EC markers Cdh5, Cd31, and Vegfr1 were either moderately decreased or unchanged by TGFβ2 challenge. Longer periods of TGFβ2 treatment could result in sustained downregulation of these and other EC-specific genes. Notably, the pericyte marker Ng2 was not detected in NECs or TEC-H8 with or without TGFβ2 treatment, ruling out the presence of contaminating pericytes in these cultures (data not shown).
Because both SMAD-dependent and SMAD-independent pathways are implicated in mesenchymal gene regulation during EndMT, we used pharmacologic inhibitors to evaluate different mechanisms of TGFβ2-induced SMA expression in TECs (32). We found that TGFβ regulated SMA expression in TECs via PI3K, Akt, and SMAD3, but not through mTOR (Fig. 4B). To determine whether differential expression of SMADs could account for the divergent responses to TGFβ2 in NEC versus TEC cultures, we assessed levels of pSMAD2 and pSMAD3 using Western blotting. We found similar basal pSMAD2 and pSMAD3 levels in both NEC-B12 and TEC-H8 cultures (Supplementary Fig. S1B and S1C). TGFβ2 further induced SMAD2 and SMAD3 phosphorylation in the two cell types to a similar level, with comparable total SMAD2/3 expression, indicating that TGFβ2 signaling is not altered in TEC-H8 compared with NEC-B12. However, TGFβ2-activated SMADs regulate distinct groups of target genes in NECs versus TECs, suggesting that either additional cofactors are differentially recruited to mesenchymal gene promoters in the two cell types or that a different set of mesenchymal genes may be poised for transcription in different types of ECs.
Only a fraction of tumor vessels contain SMA+ ECs in vivo
We found that primary EC clones displayed a spectrum of SMA expression upon TGFβ2 stimulation, suggesting that not all ECs respond identically to TGFβ2 challenge. To test this possibility in vivo, we fate-mapped tumor endothelium using ZsGreenCdh5-Cre reporter mice to indelibly mark all ECs (Supplementary Fig. S2A and S2B). Three-dimensional confocal imaging of these tumors at high magnification clearly demonstrated that SMA and ZsGreen localized in the same cells (Supplementary Figs. S2C and S3 and Supplementary Movie S1). Consistent with our in vitro findings, we observed that in two different orthotopically implanted mammary tumors, only a minor fraction (∼1%–10%) of tumor blood vessels contained SMA+ ECs, whereas few if any SMA+ ECs were found in normal mammary glands (Fig. 4C). On the other hand, blood vessels in spontaneous C3-TAg mammary tumors and K-RasG12D lung tumors showed a broad range (0%–25%) of CD31 and SMA coexpression in ECs (Fig. 4D). The higher frequency of CD31+/SMA+ ECs in these genetically engineered mouse (GEM) models could be due to the longer tumor growth period, or in the case of lung, be related to underlying plasticity inherent within the vascular bed (i.e., different vascular beds show differential proclivity to undergo EndMT). However, it is challenging to discriminate between SMA+ pericytes and the closely underlying CD31+ ECs using immunostaining in these models and these results must be interpreted with caution; therefore, the frequency of CD31+/SMA+ ECs in these GEM models could be overestimated.
These results suggest that ECs variably acquire SMA expression in the tumor models we evaluated, but SMA+ ECs may be more common in other tumor types; for example, in pancreatic tumors where fibrosis is a prominent feature and the percentages reported are markedly higher (∼40%; ref. 5). SMA+ ECs might appear only transiently during different stages of tumor progression, or are generated continuously, but can revert to SMA− ECs, depending on the balance of EndMT promoting (or inhibiting) factors present in the tumor microenvironment. Though ZsGreenCdh5-Cre mice are a high-fidelity EC lineage-tracing model, it is also conceptually possible that SMA+ fibroblasts or other mesenchymal-lineage cells may acquire EC markers through mesenchymal-to-endothelial transition (MEndT), which is a reverse process of EndMT. Although EndMT has been observed in a wide range of conditions, the phenomenon of MEndT in tumors is still a matter of debate (33). In contrast with EndMT, which may arise spontaneously in vitro, generation of ECs from lineage-committed mesenchymal cells requires enforced induction of multiple EC-selective transcription factors in addition to TGFβ inhibition, indicating that MEndT demands restrictive conditions and precise temporal activation of specific regulatory factors (34). Therefore, SMA+ ECs in our model are more likely to be originated from endothelial-lineage cells via EndMT rather than mesenchymal-lineage cells through MEndT.
bFGF opposes the expression of some TGFβ target genes but augments the expression of others
Similar to epithelial-to-mesenchymal transition (EMT), we expected that conversion of ECs to mesenchymal-like cells was dynamic and could be reversed upon removal of TGFβ2 from the culture medium. Indeed, we found that when TGFβ2 was removed and cells were returned to their normal growth medium, which contains bFGF (EC media), SMA expression was rapidly lost and TECs regained their typical EC morphology (Supplementary Fig. S4A and S4B). The same cell population upregulated SMA again after being returned to TGFβ2-containing media, indicating that some TECs may readily morph between SMA− endothelial and SMA+ mesenchymal-like phenotypes. We further determined that bFGF, but not VEGF, suppressed TGFβ2-induced SMA expression (Fig. 5A). Similar to interstitial cells in the heart, bFGF neutralized TGFβ2-stimulated SMA expression in TECs at 500 pg/mL, a concentration that was 20-fold lower than the TGFβ2 concentration added to the culture medium (Fig. 5B; ref. 35). Remarkably, acidic FGF (aFGF), the closely related isoform of bFGF, did not neutralize SMA expression in TGFβ2-challenged cells at the same concentrations. Suppression of TGFβ2-induced SMA expression by bFGF was confirmed at the protein level by immunofluorescence (Fig. 5C). Simply removing bFGF from the culture medium promoted TECs to drift toward a mesenchymal-like phenotype indicated by a moderate increase in SMA expression, especially in cells maintained at subconfluent conditions (Supplementary Fig. S4C). Moreover, these subconfluent cells were highly receptive to TGFβ stimulation compared with confluent cultures and showed robust SMA induction and more pronounced VEGFR2 suppression (Supplementary Fig. S4C), indicating that loss of cell–cell contact enhances TGFβ responses. Thus, bFGF functions both as a potent EC mitogen and a “specification factor” that prevents ECs drift toward mesenchymal-like cell types, particularly in the presence of TGFβ.
Next, we carried out a TGFβ2 pathway qPCR array to comprehensively assess the expression of 84 TGFβ2 target genes in NEC versus TEC cultures, either with or without addition of bFGF. The results showed that, as predicted, TGFβ2 induced distinct expression patterns in NEC versus TEC cultures (Fig. 5D). Furthermore, although bFGF counteracted TGFβ2-induced expression of the mesenchymal genes Acta2 and thrombospondin 1 (Thbs1), TGFβ2 and bFGF synergistically activated other genes such as Notch1 and S100a8, which were shown previously to regulate tumor angiogenesis (Fig. 5D; refs. 36, 37). Although repressing SMA, bFGF rescued the expression of Vegfr-2, which was suppressed by TGFβ2 challenge; however, bFGF only marginally downregulated expression of the transcription factor Snail, which was previously identified as a master regulator of EndMT (Supplementary Fig. S4D; ref. 38). Interestingly, bFGF expression was increased by TGFβ2 stimulation, suggesting a possible autocrine or paracrine feedback loop in ECs to counteract TGFβ2-induced expression of specific target genes, namely SMA (Supplementary Fig. S4D).
To further confirm the protective role of bFGF in neutralizing TGFβ-induced SMA expression, we examined gene expression patterns in an additional C3-TAg TEC clone with relatively low basal SMA mRNA (TEC-G8). bFGF completely suppressed TGFβ-induced SMA expression in TEC-G8, even though the SMA levels were significantly lower than those in the SMA+ clone TEC-H8 (Supplementary Fig. S5A and S5B). To assess the interaction between bFGF and TGFβ in ECs derived from a different tumor model, we isolated lung TECs from K-RasG12D mouse lung tumors (22). All clones were virtually 100% positive for DiI-Ac-LDL uptake and CD31, and strongly expressed VEGFR2 (Supplementary Fig. S6A and S6B). Interestingly, unlike C3-TAg TEC clones, the K-Ras TEC clones showed elevated basal SMA protein that was detectable by Western blotting even in EC media that contained bFGF, suggestive of a “partial” EndMT phenotype in EC subpopulations derived from lung. The expression of SMA and other myofibroblast markers, including Col1a1, Fn1, and Tagln, in these K-Ras clones was further stimulated by TGFβ2 in the absence of bFGF, although to a lesser extent when compared with SMA+ mammary TECs (Supplementary Fig. S6B and S6C). Addition of bFGF variably antagonized the effect of TGFβ on these genes, indicating that bFGF may act through similar pathways in different types of ECs to maintain EC fate (Supplementary Fig. S6C). Although Twist1 and Twist2 have been implicated in the induction of mesenchymal genes during EndMT and EMT; surprisingly, we did not observe an upregulation of Twist1 or Twist2 by TGFβ (39). It is possible that induction of these genes in ECs requires prolonged stimulation with TGFβ or they play only a marginal role during EndMT.
Previous studies have demonstrated that bFGF can antagonize TGFβ through the MAPK/ERK kinase pathway to suppress fibrogenic effects in epithelial cells and revert EMT, a process closely related to EndMT (40). Providing a further link between bFGF and TGFβ signaling, basal SMAD2 phosphorylation and TGFβR1 levels are increased when FRS2, an FGFR coactivator, is depleted using shRNA (41). In lymphatic ECs, bFGF was reported to suppress TGFβ-stimulated SMAD2 activation via MAPK signaling (42). However, we found that although bFGF strongly induced EKR1/2 activation, it exerted no observable effects on TGFβ2-stimulated SMAD2 phosphorylation and only weakly reduced SMAD3 phosphorylation after a 60-minute TGFβ2 stimulation (Supplementary Fig. S7A). Furthermore, inhibition of MAPK/ERK only mildly disabled the opposing effect of bFGF on TGFβ-induced SMA (Supplementary Fig. S7B), indicating that bFGF may counteract TGFβ signaling through additional mechanisms.
Bone morphogenetic proteins (BMP), which are members of the TGFβ superfamily, can interact synergistically or antagonistically with TGFβ to fine-tune cellular differentiation. For example, BMP-2 controls cardiac valve formation through Snail1-mediated EMT during heart development, whereas BMP-7 was reported to attenuate TGFβ-induced EndMT in cardiac fibrosis (4, 43). In addition, BMP-6 acts as a major inhibitor of renal fibrosis as loss of BMP-6 increases SMA and FN1 expression in obstructed kidneys (44). However, we found that treatment with BMP-2, 6, or 7 did not overly affect SMA or FN1 expression stimulated by TGFβ2 in ECs (Supplementary Fig. S7C), suggesting that interaction between TGFβ and BMP signaling may be cell-type dependent. More specifically, different types of ECs may respond differently to combinations of TGFβ and BMP signals. Taken together, our results suggest that EndMT, while sharing many features with EMT, is a distinctly regulated process that is variably regulated in different vascular beds, or in different disease settings.
ECs challenged with TGFβ secrete their own bFGF, which suppresses mesenchymal-like differentiation in secondary cultures
Because bFGF potently blocks TGFβ-stimulated conversion to SMA+ mesenchymal cells, and ECs upregulate bFGF in response to TGFβ, we asked whether a cell autonomous mechanism in ECs could counteract TGFβ challenge. Consistent with our findings, a recent study using microarrays also showed that ECs challenged with TGFβ upregulate bFGF mRNA (2). First, we confirmed that TGFβ2 dose dependently increased bFGF mRNA expression, with a maximum approximately 10-fold increase after 10 ng/mL TGFβ2 treatment (Fig. 6A). TGFβ2 also time dependently increased bFGF expression, with maximum levels peaking at approximately 10-fold above untreated control cells after 48 hours. Next, we harvested the CM and cellular lysates of TGFβ2-challenged TECs. We found a striking time-dependent increase in bFGF secretion and a slight increase in bFGF expression in the cellular lysates after TGFβ2 stimulation (Fig. 6B). Notably, we observed that secreted bFGF migrated at three different molecular weights, which is consistent with a previous study reporting multiple splice variants of bFGF (45).
To determine whether EC-derived bFGF could oppose TGFβ2 activity, we challenged TEC cultures with TGFβ2 and then harvested the CM to rechallenge secondary cultures (Fig. 6C, a). We found that CM collected from TECs stimulated with TGFβ2 induced approximately 2-fold less SMA expression in secondary cultures treated with cell-free media containing equal amounts of TGFβ2 (Fig. 6C, b and c). The decrease in SMA expression in secondary cultures correlated with the presence of secreted bFGF protein in the CM. qPCR analysis of additional mesenchymal and EC genes, including Tagln, Pdgfa, Thbsp1, Cd31, and Cdh5, further showed that the CM of TGFβ2-challenged primary cultures could oppose the effects of TGFβ2 in secondary cultures. In contrast, S1008a, which can be induced synergistically by TGFβ2 and bFGF as revealed by the TGFβ pathway array (refer to Fig. 5D), showed a similar expression pattern after the CM treatment, whereas the expression of Snail was not affected. To test whether the TGFβ2-antagonizing effect of the CM was specifically due to an increased bFGF production by TECs, we used a bFGF-blocking antibody (BA) and an FGFR kinase inhibitor (KI) to neutralize bFGF activity. As expected, blocking bFGF signaling by either compound increased TGFβ2-stimulated SMA mRNA expression in TECs (Fig. 6C, d). However, neutralizing bFGF with a BA or KI only had a modest effect on SMA expression, possibly because SMA mRNA levels were already maximized by the addition of TGFβ2.
Our results have shown that ECs exhibit heterogeneity in their responses to TGFβ: some express high-basal SMA and react to TGFβ by generating SMA+ myofibroblast-like cells with distinct functions, whereas others have low-basal SMA and transition to SMA− fibroblast-like cells (Fig. 7). These results are remarkably similar to what has been reported for epithelial cells and suggest parallels between the processes of EMT and EndMT (46). However, ECs that have undergone a partial mesenchymal-like transition may be directed to a stable SMA+ mesenchymal-like phenotype, perhaps after prolonged TGFβ stimulation or disruption in bFGF signaling (41). In addition to regulation by growth factors and cytokines, EndMT may also be controlled by epigenetic barriers within the different subpopulations of ECs that guide them toward one lineage or the other. These epigenetic restrictions could explain heterogeneity in EC responses to TGFβ signaling. EC markedly increases bFGF production in response to stimulation by TGFβ2; thus, one EC that receives TGFβ could protect neighboring ECs from mesenchymal-like differentiation through a lateral inhibitory mechanism involving bFGF secretion (47). A similar mechanism of lateral inhibition involving VEGF/Notch signaling is well known in ECs, as it occurs during fate determination of tip cells during sprouting angiogenesis (48). In tumors, organ fibrosis, and other chronic inflammatory conditions, the extent of EndMT will likely be determined by local concentrations of several interacting cytokines and growth factors, including TGFβ, bFGF, and BMPs, and perhaps on organ-specific properties and heterogeneity of the ECs that form the microvasculature (49). Notably, in endoglin- (a TGFβ2 co-receptor) deficient mice, EndMT in tumor vessels was exacerbated and the frequency of metastases was increased (13). On the other hand, endoglin deficiency delayed resistance to an antiangiogenic therapy targeting VEGF. These findings bring to light the importance of further understanding the molecular mechanisms that promote and inhibit EndMT in tumors and in other pathologic conditions.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: L. Xiao, D.J. Kim, A.C. Dudley
Development of methodology: L. Xiao, D.J. Kim, N. Xu, S.G. Pattenden, S.V. Frye, A.C. Dudley
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Xiao, C.L. Davis, J.V. McCann, J.M. Dunleavey, A.K. Vanderlinden, N. Xu, M. Onaitis, E. Monaghan-Benson
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Xiao, D.J. Kim, J.M. Dunleavey, N. Xu, M. Onaitis, A.C. Dudley
Writing, review, and/or revision of the manuscript: L. Xiao, D.J. Kim, J.M. Dunleavey, K. Burridge, A.C. Dudley
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.J. Kim, C.L. Davis, X. Xu, K. Burridge
Study supervision: S.V. Frye, A.C. Dudley
The authors thank the UNC Microscopy Services Laboratory in the UNC Department of Pathology and Laboratory Medicine for technical assistance with confocal and live imaging, and K. McNaughton and J. A. Ezzell at the Histology Core of the UNC Department of Cell Biology and Physiology for processing and sectioning histology samples, and Dr. Shelly Earp from the Lineberger Cancer Center for sharing the PyVMT tumor cells.
A.C. Dudley was supported by a grant from the NIH (R01-CA177875). L. Xiao is a fellow in the HHMI-funded translational medicine program at UNC Chapel Hill.
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.