Caveolin-1 Enhances Tissue Factor Pathway Inhibitor Exposure and Function on the Cell Surface* □ S

Tissue factor pathway inhibitor (TFPI) blocks tissue factor-factor VIIa (TF-FVIIa) activation of factors X and IX through the formation of the TF-FVIIa-FXa-TFPI complex. Most TFPI in vivo associates with caveolae in endothelial cells (EC). The mechanism of this association and the anticoagulant role of caveolar TFPI are not yet known. Here we show that expression of caveolin-1 (Cav-1) in 293 cells keeps TFPI exposed on the plasmalemma surface, decreases the membrane lateral mobility of TFPI, and increases the TFPI-dependent inhibition of TF-FVIIa. Caveolae-associated TFPI supports the co-localization of the quaternary complex with caveolae. To investigate the significance of these observations for EC we used RNA interference to deplete the cells of Cav-1. Functional assays and fluorescence microscopy revealed that the inhibitory properties of TFPI were diminished in EC lacking Cav-1, apparently through deficient assembly of the quaternary complex. These findings demonstrate that caveolae regulate the inhibition by cell-bound TFPI of the active protease production by the extrinsic pathway of coagulation. blood vivo . of VIIa factors and thrombogenic responses

Tissue factor pathway inhibitor (TFPI) blocks tissue factor-factor VIIa (TF-FVIIa) activation of factors X and IX through the formation of the TF-FVIIa-FXa-TFPI complex. Most TFPI in vivo associates with caveolae in endothelial cells (EC). The mechanism of this association and the anticoagulant role of caveolar TFPI are not yet known. Here we show that expression of caveolin-1 (Cav-1) in 293 cells keeps TFPI exposed on the plasmalemma surface, decreases the membrane lateral mobility of TFPI, and increases the TFPI-dependent inhibition of TF-FVIIa. Caveolae-associated TFPI supports the co-localization of the quaternary complex with caveolae. To investigate the significance of these observations for EC we used RNA interference to deplete the cells of Cav-1. Functional assays and fluorescence microscopy revealed that the inhibitory properties of TFPI were diminished in EC lacking Cav-1, apparently through deficient assembly of the quaternary complex. These findings demonstrate that caveolae regulate the inhibition by cell-bound TFPI of the active protease production by the extrinsic pathway of coagulation.
Tissue factor pathway inhibitor (TFPI) is the endogenous regulator of TF-FVIIa activity. TFPI is a Kunitz-type protease inhibitor, which binds FVIIa via Kunitz-1 and FXa via Kunitz-2. Formation of the TF-FVIIa-FXa-TFPI complex provides sustained repression of the TF pathway (8). TFPI was first described as a soluble plasma protein that binds through its C terminus to yet uncharacterized anionic sites on the cell surface. Nevertheless, the majority of TFPI circulating in plasma is C-terminal truncated and bound to plasma lipoproteins, and as such, has significantly less inhibitory activity than the full-length TFPI (9).
TFPI in vivo is mainly produced by endothelial cells (EC), has an intact C terminus, and associates with the plasma membrane through mechanisms that are not fully identified (3, 10 -13). Heparin releases a portion of the full-length, functionally active cell-associated TFPI, either from cell surface-binding sites or from intracellular stores (14,15). However, the bulk of heparin-resistant cellular TFPI is released by phosphatidylinositol-phospholipase C in vitro, which indicates that most cellular TFPI associates with the cell surface via a glycosylphosphatidylinositol link (10 -12, 16, 17). Furthermore, endogenous TFPI in resting endothelium (10), monocytes (18), and the ECV304 cell line (11,16) partitions in low-density fractions insoluble in cold detergent (lipid rafts). In monocytes and ECV304 endogenous TFPI inhibits efficiently FVIIa-TF activity through translocation of the TF-FVIIa-FXa complex in lipid rafts (11). These findings suggest that cell-bound TFPI, particularly the lipid raft-associated pool, plays a critical role in regulating cell surface FVIIa-TF and FXa activity.
Lipid rafts are domains rich in cholesterol and sphingolipids that can exist by themselves or as caveolae. Caveolae are small (50 -80 nm) plasma membrane invaginations that have a protein "coat" composed of caveolin family members. Caveolin-1 (Cav-1) is an integral membrane protein and the principal component of caveolae (19). The role of Cav-1 in caveolae formation was confirmed in Cav-1-deficient cells, which lack morphologically identifiable caveolae (20). Expression of Cav-1 in these cells induces caveolae formation (21). Caveolae are multifunctional organelles in which Cav-1 plays a direct role in various events, such as membrane trafficking and cellular signal transduction (reviewed in Ref. 22).
We reported previously that TFPI is localized in caveolae in EC both in vitro and in vivo (3,10). The mechanism of TFPI association with caveolae and the anticoagulant role of caveolar TFPI are not yet known. We report here that Cav-1 regulates the distribution and function of TFPI in HEK293, a cell system where we controlled the expression of Cav-1 and TFPI by transfection. We show for the first time that caveolae keep TFPI associated with the cell surface and enhance the anticoagulant activity of the inhibitor. Furthermore, using RNA interference (RNAi) to deplete HUVEC and EA.hy926 cells of Cav-1, we show that EC which lacks Cav-1 displayed severalfold enhanced procoagulant activity. In conclusion, we identified Cav-1 as an active regulator of TFPI-dependent inhibition of TF-FVIIa activity, which adds the hemostatic function as a novel dimension to the biological significance of caveolae.
Freeze-fracture EM-Freeze-fracture EM was performed as described (28) on wild-type and Cav-1 expressing cells.
Proteolytic Activity of Cell Surface TF-FVIIa-Confluent monolayers in 24-well plates were tested with a two-stage chromogenic assay (12,16). 50 g/ml cycloheximide was added to prevent de novo protein synthesis. To de-encrypt TF and expose anionic phospholipids the cells were briefly incubated with 1 M ionomycin. Several concentrations of FVIIa (0.2, 10, or 50 nM) were added for 30 min at room temperature, and then excess FX was added (5,200, and 500 nM, respectively). Samples were taken during a 30-min incubation time at 37°C, and quenched in ice-cold 50 mM Tris-buffered saline (TBS), pH 8.8, containing 25 mM EDTA and 0.1% bovine serum albumin. FXa generated was determined from the hydrolysis of the chromogenic substrate S-2765. In antibody-blocking experiments, 50 g/ml anti-TFPI IgG or 100 g/ml of the anti-TF mAb 5B7 were included during the incubation with FVIIa.
Determination of TFPI Activity by Functional Assay-The activity of cell surface TFPI was measured with a modification of a two-stage chromogenic assay in which we estimated the residual capability of TF-FVIIa pre-formed mixtures to activate FX after incubation with TFPI-exposed cells. The original assay (10, 12) made use of 0.4 nM FVIIa and 1:320 dilution of thromboplastin made from one vial reconstituted with 1 ml of 50 mM TBS, pH 7.35. Here, TF-FVIIa complexes were pre-formed by mixing up several combinations of thromboplastin dilutions and FVIIa concentrations for 30 min at 37°C in the presence of 15 mM CaCl 2 . Each mixture was tested for the linearity and the rates of FXa generation, first in the absence of any TFPI, and then in the presence of serially diluted normal human plasma. Three mixtures containing (nM FVIIa ϩ thromboplastin dilution) 0.2 ϩ 1:700, 10 ϩ 1:15, and 50 ϩ 1:3 were chosen. TFPI-bearing monolayers were then incubated for 30 min at room temperature with each pre-formed TF-FVIIa complex, either alone or in the presence of 50 g/ml anti-TFPI IgG (to block the available TFPI). The supernatants were removed and any unbound TF-FVIIa present was tested for activation of FX (added as 10 times excess over FVIIa) by using S-2765. The activity of TFPI was extrapolated from standard curves constructed with serial dilutions of normal human plasma, which was assigned a TFPI functional potency of 1 unit/ml.

SDS-PAGE and Western
Blotting-Cell monolayers were scraped in ice-cold 0.1 M TBS, pH 7.8, containing 1 mM phenylmethylsulfonyl fluoride, 1% aprotinin, 1 mM sodium orthovanadate, 5 mM EDTA, and 0.02% NaN 3 . The cells were lysed with 1% Triton X-100 and 60 mM n-octyl-␤-Dglucopyrannoside in TBS plus inhibitors for 30 min at 37°C. After brief centrifugation, aliquots of lysates containing 50 g of total protein were precipitated with 40% trichloroacetic acid on ice and resolved by nonreducing SDS-PAGE (NuPAGE® MES 4 -12% polyacrylamide gradient gel, Invitrogen). After electrotransfer, the polyvinylidene difluoride membranes were blocked for 1 h with 10% FCS and 0.5% Tween 20 in phosphate-buffered saline (PBS), incubated for 1 h with primary antibody diluted in PBS containing 1% bovine serum albumin and 0.1% Tween 20, washed with 0.3% bovine serum albumin and 3% Tween 20 in PBS, and then incubated for 1 h with appropriate secondary antibodies conjugated with alkaline phosphatase or horseradish peroxidase. The blots were developed with the DAB substrate kit for horseradish peroxidase or Vector Red AP substrate kit I (Vector Labs).
Triton X-114 Cell Extraction-Cell monolayers on ice were scraped in ice-cold 0.1 M TBS, pH 8.1, plus inhibitors, and lysed in pre-condensed 1% Triton X-114 in the same buffer by incubating for 1 h on ice with repeated mixing. Debris was removed by centrifugation and phase separation was induced for 5 min at 37°C, followed by brief centrifugation to separate detergent and aqueous phases. The same buffer lacking detergent was added and the extraction repeated once before the separated phases were assayed or frozen down. In some experiments cells were treated with 1 unit/ml phosphatidylinositol-phospholipase C (Glyko) for 1 h at room temperature before extraction with Triton X-114.
Immunofluorescence-Immunofluorescence was carried out as described (10). The effect produced by FVIIa/FX on the distribution of TF, TFPI, and Cav-1 was studied in cells fixed with 3% paraformaldehyde in PBS (1 h at room temperature) at the end of the FX activation assay. Intracellular immunostaining was achieved on cells permeabilized with 0.1% Triton X-100.
Images were collected using a Nikon Eclipse inverted microscope equipped with a confocal Nikon C1 system, using a computer-controlled 488-nm argon laser to excite EGFP, and a helium-neon laser whose green line (543-nm) excites Cy3 and red line (633-nm) excites Cy5 or TO-PRO (nuclear stain). Samples mounted with Vectashield (Vector) were observed with a PlanApochromat oil-immersion objective (ϫ60, NA: 1.4). Images were processed using MetaMorph (Universal Imaging). The overlap was quantified using Adobe Photoshop (Adobe Systems), by determining the % of protein A that co-localizes with protein B (29). A-channel minus B-channel gave the non-overlapping A elements. Their intensity was subtracted from the total A-channel, and the resulting overlap signal was expressed as mean % of the total signal for each channel. Triple co-localization was analyzed similarly. Fluorescence intensity measurements were performed on a minimum of 50 cells/group, randomly chosen from at least five pictures for each experimental condition. Experiments were repeated three to four times.
Statistical significance of the differences between groups was determined by t test (Microsoft Excel), and the differences were considered significant when p Ͻ 0.05. Descriptive statistics (median values and range) and correlation analysis were performed with InStat (Macintosh).
Fluorescence Recovery after Photobleaching (FRAP)-For live cell recordings, 293 cells stably expressing EGFP-TFPI, Cav1-HPC4ep, or both, were grown on coverslips and mounted in a perfusion chamber at 37°C (Bioptech FCS2). The chamber was installed on the stage of the inverted microscope equipped with a temperature controlled ϫ63 oil immersion objective. The chamber was perfused with phenol red-free culture medium at 37°C. Bleaching in 3-5 spots in regions of interest was performed at 37°C with the 488-nm laser line at full power and full transmission for 2 s. Observation of fluorescence recovery over time was done at full laser power and 1% transmission, to avoid significant photobleaching. Image J software (public domain) was used to measure pixel intensities in regions of interest, and to correct for overall bleaching by comparison with areas not exposed to the full power of the laser. The % recovery (recovery values Ϭ mean of pre-bleach values ϫ 100) shows how much fluorescence returns to the area out of the amount of fluorescence before photobleaching.
Depletion of Cav-1 in EA.hy926 and HUVEC by RNAi-Post-transcriptional silencing of the Cav-1 expression was achieved with two duplex RNA oligonucleotides (Dharmacon Research), each composed of 21 bases of sequence that is conserved in Cav-1 from several mammalian species (30). The sense strands of the duplexes were 5Ј-CCUgAUU-gAgAUUCAgUgCdTdT-3Ј and 5Ј-gAgAAgCAAgUgUACgAcgdTdT-3Ј (31). EC were seeded on coverslips to reach ϳ60% confluence on the following day. Silencing (si) RNA duplexes (25 nM final concentration) mixed with TransIT-TKO (Mirus) in Opti-MEM I were added to the cells in complete medium without antibiotics. Cells were incubated for 2-3 days and analyzed by confocal microscopy after immunostaining. Controls were as described under supplemental materials.
Supplemental Data-Details of the following protocols are available: construction and expression of pEGFP-C 2 /TFPI(C) and pSVZeo-Cav1-HPC4ep, primers, and conditions for reverse transcriptase-PCR, characterization, and selection of 293 cell clones, and RNAi experimental controls. Fig. S1 demonstrates that EGFP-TFPI has distribution and function similar to natural TFPI, therefore is a reliable marker for endogenous TFPI.

EGFP-TFPI Is a Reliable Marker for Endogenous TFPI-
Using EA.hy926, we compared the expression and activity of EGFP-TFPI (a chimera with EGFP fused at the N terminus of TFPI) with the properties of native TFPI. Western blotting showed a protein with molecular mass ϳ65,000 Da immunoreactive with both anti-GFP and anti-TFPI IgGs, which were absent from native (N) and cells expressing EGFP only (GFP) (Fig. S1, A). The distribution and properties of native and chimerical TFPI were similar. Fluorescence imaging showed that 98% of EGFP-TFPI overlapped the anti-TFPI IgG staining ( Fig. S1, B, a, d, and e), indicating that the EGFP signal originated from the chimera and not from EGFP alone. Both EGFP-TFPI and native TFPI co-localized with endogenous Cav-1 (b, c, and f).
Enzyme-linked immunosorbent assay (ELISA) for TFPI antigen confirmed that EC overexpressing EGFP-TFPI synthesized and secreted two times more TFPI than native cells ( Fig.  S1, C, a). TFPI-mediated inhibition of exogenously added TF-FVIIa was similarly increased (Fig. S1, C, b). EGFP-TFPI matched the Triton X-114 partition properties of native TFPI, a finding that suggests direct or indirect glycosylphosphatidylinositol anchoring. Enzymatic cleavage of the glycosylphosphatidylinositol anchor with phosphatidylinositol-phospholipase C (10, 11, 32) substantially reduced TFPI partitioning into the detergent phase of Triton X-114 ( Fig. S1, C, c).
Taken together, these data establish that EGFP-TFPI is a valid fluorescent indicator of native TFPI, and can be used to study the distribution and function of TFPI in live as well as in fixed cells.
Analysis of the Expression of EGFP-TFPI and Cav1-HPC4ep in 293 Cells-In HEK293 transiently expressing Cav1-HPC4ep (a chimera where the small HPC4 epitope was fused to the C terminus of Cav-1), immunostaining with mAb HPC4 (24) revealed the typical distribution of native Cav-1 (Fig. S2, a-c). In cells expressing EGFP-TFPI 99% of the chimera overlapped the immunolabeled TFPI, showing that no cleaved EGFP was present ( Fig. S2, d-f). Co-localization of each protein with its tag exceeded 90% (Fig. S2).
Five stable cell clones were selected for each experimental condition. The cell clones were re-named: "TFPI" (express EGFP-TFPI only); "Cav" (cells express Cav1-HPC4ep alone); and "TFPI ϩ Cav" (express both proteins). As detailed in Fig.  S3, the total level of TFPI was not significantly different between TFPI cells and TFPI ϩ Cav cells (panel A). Although the constitutive secretion of TFPI was significantly decreased in the presence of Cav-1. The capability of TFPI cells to prevent endogenous TF-dependent FX activation was significantly diminished as compared with TFPI ϩ Cav cells (Fig. S3, B).
Native (wild type, wt) cells were also tested, either quiescent or after transfection with pEGFP-C 2 . There was no significant variation in TFPI distribution or function between these two conditions; therefore we used wt cells in all subsequent experiments.
Wt and Cav cells expressed very low levels of TFPI mRNA, whereas TFPI cells and TFPI ϩ Cav cells expressed mRNA for both TFPI and EGFP-TFPI (Fig. 1A, semi-quantitative reverse transcriptase-PCR). Lysates of TFPI cells and TFPI ϩ Cav cells displayed a protein band of ϳ65 kDa that was immunoreactive with both anti-GFP and anti-TFPI IgGs, but was absent from wt cells (Fig. 1B). Cav cells and TFPI ϩ Cav cell lysates contain a protein that was immunoreactive with anti-Cav-1 IgG (band at ϳ20 kDa), present only as trace in lanes "wt" and "T." The equivalent band appeared in EA.hy926 lysates used as positive control (lane EA). Immunofluorescence confirmed that wt cells contained very low levels of TFPI and Cav-1 (Fig. 1C, a and b), and that EGFP-TFPI largely co-localized with Cav-1 (j-l).
Similar to TFPI in EC, EGFP-TFPI in 293 cells partitioned predominantly in the detergent phase of Triton X-114 (Fig. 1D). The equivalent of the 1:3 ratio between the water-soluble and detergent-soluble TFPI normally found in EC was achieved in 293 cells only in the presence of Cav-1 (Fig. 1D). The total amount of TFPI was not significantly different among the TFPI cells and TFPI ϩ Cav cells.
Cav-1 Expression in 293 Cells Induces Formation of Caveolae-We used freeze-fracture EM to analyze the formation of invaginations/caveolae in wt and Cav cells. In wt cells (Fig. 2a) the external leaflet of the membrane displayed a smooth aspect, hence lack of caveolae. Cav cells exhibited pits with dimensions and aspect typical of caveolae (Fig. 2b), thus confirming that expression of Cav-1 in deficient cells induced invagination of caveolae.
Kinetics of Fluorescent TFPI Were Different in the Presence and Absence of Cav-1-We used FRAP to quantify the kinetic properties of EGFP-TFPI in the plasma membrane of living cells. The fluorophore was bleached in "regions of interest" at the peripheral EGFP-TFPI-labeled rim of each cell and the fluorescence recovery was followed for 10 min. Non-bleached molecules diffused into the bleached areas according to their membrane mobility.
In the presence of Cav-1, clusters next to regions of interest maintained their fluorescence after bleaching and did not move into the bleach area, suggesting that only a little lateral diffusion of fluorescent TFPI took place (not shown). Fig. 3 shows the FRAP curves after curve fitting of the means of 10 independent recordings for each cell clone. The recovery of EGFP-TFPI in TFPI ϩ Cav cells was much slower than in TFPI cells. The mobile fraction of EGFP-TFPI was ϳ35% in the presence of Cav-1 and ϳ85% in its absence (Fig. 3, p Ͻ 0.01).
Expression of Cav-1 in 293 Cells Increases Both Cell Surface TFPI and the Inhibition of FVIIa-TF by TFPI-From the measurement of TFPI and Cav-1 fluorescence intensity after immunostaining (Fig. 4A), it results that expression of Cav-1 in TFPI ϩ Cav cells doubled or even tripled the amount of TFPI exposed on the cell surface (Fig. 4A, compare TFPI ϩ Cav cell clones (TC) 1, 2, 4, and 5, with TFPI cell clones). Only TFPI ϩ Cav cells displayed a significantly positive correlation (r 2 ϭ 0.92) between Cav-1 expression and the level of cell surface TFPI. To determine the functional activity of the surface exposed TFPI, we assessed its inhibitory potency, both against exogenously added TF, and toward the endogenous TF, which is normally expressed by HEK293.
The functional activity of cell surface TFPI tested against exogenous TF-FVIIa was significantly different between the clones. We measured the capability of the TFPI exposed on the cell surface to prevent the activation of FX by pre-formed complexes consisting of 10 nM FVIIa and 1:15 diluted thromboplastin added in the overlying cells medium. The activity of TFPI, expressed as arbitrary units, was extrapolated from standard curves made with serial dilutions of normal human plasma, run in parallel in identical conditions. In Fig. 4B we illustrated for each cell clone the activity of cell surface TFPI expressed as milliunits, together with the corresponding TFPI antigen (ng) measured by enzyme-linked immunosorbent assay on cell monolayers (all values were normalized to 10 6 cells).
TFPI cells displayed a significant increase of both antigen and activity of cell surface TFPI as compared with wt cells (p Ͻ 0.01 for all the clones). Nevertheless, there was no positive correlation between the activity and either TFPI or Cav-1 antigen levels (Fig. 4C, r 2 ϭ 0.31 and 0.35, respectively). In TFPI ϩ Cav cells, increases of up to five times of cell surface antigen led to an enhancement of TFPI activity of as much as 40 times over the wt cell levels (Fig. 4B, right panel; note the different scale of the y axis). In comparison with TFPI cells, the average 2.5-fold increase of TFPI antigen on the cell surface of TFPI ϩ Cav cells could be translated into as much as 10 times higher inhibitory potency (Fig. 4B). After normalization for TFPI antigen levels it results that TFPI on TFPI ϩ Cav cells were, on average, 3.5 times more potent than TFPI on TFPI cells in inhibiting exogenous TF-FVIIa activity. Specificity controls included parallel assays run in the presence of 50 g/ml anti-TFPI IgG, which abrogated almost entirely the activity of TFPI (not shown).
Only in Cav-1-expressing cells did the activity of cell surface TFPI correlate positively with the TFPI antigen, and implicitly with Cav-1 levels (Fig. 4C, r 2 ϭ 0.95 and 0.93, respectively). The inhibitory activity of cell surface TFPI was severely impaired in the absence of Cav-1, even for cells that displayed equal or higher amounts of TFPI antigen (Fig. 4C, compare TFPI cell clones 4, 3, 5, and 1, with TFPI ϩ Cav cell clone 3). Conversely, the presence of Cav-1 alone was enough to bring the activity of even very small amounts of TFPI antigen, such as the one present in Cav cells, up to levels that were equivalent with the activities measured on TFPI cells (Fig. 4B, compare Cav cell clones 3, 4, and 5 with TFPI cell clones 2, 4, and 5).
Next, we tested the inhibitory potency of TFPI against endogenous TF on TFPI cell clone 1 and TFPI ϩ Cav cell clone 4. TF mRNA determined by reverse transcriptase-PCR and total TF antigen in cell lysates were comparable (not shown), indicating that the expression of TF was similar among the 293 cell clones. TFPI cells and TFPI ϩ Cav cells displayed similar amounts of cell surface TF antigen (11 and 12.2 pM, respectively). FX was efficiently activated after the addition of FVIIa to 293 cell monolayers, following a brief incubation with ionomycin to expose the cryptic cell surface TF and/or phosphatidylserine. Inclusion of cycloheximide precluded changes because of de novo protein synthesis. Preincubation of cells with inhibitory anti-TF mAb 9-5B7 blocked the generation of FXa, confirming that FX activation was dependent on TF-FVIIa.
The overall capability of the cells to activate FX, assessed after inhibiting the available TFPI with anti-TFPI IgG, was defined as "total FXa." FXa generation curves for both TFPI cells and TFPI ϩ Cav cells were linear over time for each concentration of FVIIa tested (Fig. 5A). The specific potential of the cells to generate FXa in the presence of TFPI, assessed using non-inhibited assays, was significantly different between the two cell clones (Fig. 5A). Whereas FXa generation increased continuously on TFPI cells, the activation of FX on TFPI ϩ Cav cells stayed at very low levels, and, if it increased slightly, it was with a significant delay and slower rate. The differences of FXa generation were larger than what would be accounted for only by the differences in cell surface TFPI antigen: 45 pM in TFPI cells and 100 pM in TFPI ϩ Cav cells.
The differences between total FXa (open symbols in Fig. 5) and the FXa generated in 15 min in the absence of anti-TFPI IgG (closed squares and circles) represent the amount of FX whose activation was prevented by TFPI, therefore giving the measure of the functional activity of cell surface TFPI (closed triangles). As shown in Fig. 5B, TFPI ϩ Cav cells displayed significantly higher capabilities to prevent FXa generation than TFPI cells for all three concentrations of FVIIa tested (p Ͻ 0.01). Even after normalization of the values to account for the differences in the cell surface antigen, TFPI on TFPI ϩ Cav

FIG. 4. Cav-1 supports TFPI-dependent inhibition of the procoagulant activity of exogenous TF-FVIIa.
A, measurement of TFPI and Cav-1 fluorescence intensity on each of the five clones was performed after immunostaining for cell surface TFPI (non-permeabilized cells, rabbit anti-TFPI IgG/donkey anti-rabbit IgG-Cy3) and intracellular Cav-1 (permeabilized cells, mouse anti-Cav-1 IgG/donkey anti-mouse IgG-Cy5) . The positive correlation between fluorescence intensities of Cav-1 and cell surface TFPI was highly significant (r 2 ϭ 0.92) only for TFPI ϩ Cav cells. B, cell surface TFPI inhibits FX activation by pre-formed TF-FVIIa complex added in the medium. The activity values of cell surface TFPI (black bars) for each cell clone were extrapolated from the inhibitory potency measured for TFPI in serially diluted normal human plasma, which is arbitrarily assigned 1 unit/ml. The amount of TFPI antigen on the cell surface determined by enzymelinked immunosorbent assay (ng, gray bars) is given for comparison. Values are mean Ϯ S.D. of triplicate assays. Both TFPI activity and antigen were normalized for cell numbers (10 6 cells). C, correlation analysis was performed for each cell clone to analyze the relationship between cell surface TFPI activity and either TFPI antigen or Cav-1 levels. Only Cav-1-expressing cells displayed significant positive correlation for both conditions (r 2 ϭ 0.95 and 0.93, respectively). n ϭ 100 for all the conditions. For simplicity, we represented median values of fluorescence intensities for each clone. cells was still ϳ2 times more active in inhibiting FX activation than on TFPI cells.

Cav-1 Co-localizes with TF-FVIIa-FXa-TFPI in 293
Cells-Using fluorescence microscopy we studied the spatial relationship between Cav-1, TF, and TFPI in the presence of FVIIa/FX. Cells kept in assay buffer (control) or incubated with FVIIa and FX were fixed with 3% paraformaldehyde in PBS. Cell surface TF was immunostained on non-permeabilized cells with mAb 10H10 followed by donkey anti-mouse IgG-Cy3. EGFP-TFPI was visualized through its intrinsic fluorescence. To detect Cav-1, and thus to observe the relationship between the cell surface TF and TFPI, and the submembrane Cav-1, the cells were permeabilized with 0.1% Triton X-100 and immunostained with rabbit anti-Cav-1 IgG followed by donkey antirabbit IgG-Cy5.
In the absence of FVIIa/FX, TFPI cells displayed little to no co-localization between TF and TFPI, and the triple co-localization was scarce (Fig. 6A, a, yellow, and c, white). Quantitatively, 20% of the total TF co-localized with TFPI, and ϳ15% of the total TFPI co-localized with TF.
After adding FVIIa/FX to TFPI cells the overlap between TFPI and TF increased to 30% of both total TFPI and total TF. TFPI patched with TF over the traces of Cav-1 present (Fig. 6A, d-f) resulting in the odd "rosette" formation (f, arrow).
Such co-localization was no longer observed if TFPI ϩ Cav cells were incubated with inhibitory anti-TFPI IgG before addition of FVIIa. The activity assays indicated that this treatment blocked the capability of TFPI to inhibit FX activation. Fluorescence microscopy now showed that TF largely failed to co-localize with TFPI and/or Cav-1 (not shown). The percentage of TF overlapping TFPI, Cav-1, or both was brought down to the control levels for all the conditions.
Silencing of Cav-1 Expression in EC Decreases the Surface Exposure and Activity of TFPI-The role of Cav-1 in the distribution and function of TFPI in EC was studied by siRNA depression of Cav-1. Incubation of EA.hy926 and HUVEC with two siRNA duplexes homologous to highly conserved sequences in cav1, from several species, led to a significant decrease of Cav-1 expression, as assessed by immunofluorescence (Fig. 7,  EA.hy926, compare A, a with B, a). For quantification, images of mock and siRNA cells were collected keeping the confocal parameters unchanged. The intensity of fluorescence was Caveolin-1 Modulates TFPI Function measured on three groups of images taken during three different experiments (total evaluated: 170 -200 cells for each condition). Results were similar between HUVEC and EA.hy926. Median and range values of the fluorescence intensity measured on HUVEC are represented in Fig. 7C. Cav-1 fluorescence in siRNA-treated HUVEC was ϳ6-fold diminished as compared with mock cells (p Ͻ 0.001). The fluorescence intensity of Cav-1 was ϳ10% of mock cells levels in 50% of the siRNA cells. 18% of the siRNA-treated cells exhibited fluorescence intensity equivalent to mock cells. The mean fluorescence intensity of the remaining cells was 20% (Ϯ5) of the mock cell levels. TFPI exposed on the surface of Cav-1 siRNA cells was also decreased, for both EA.hy926 (Fig. 7, compare A, b with B, b, and Fig. S4,  compare A with C, b), and for HUVEC. TFPI fluorescence was ϳ3-fold diminished versus mock cells (Fig. 7C, p Ͻ 0.01). TFPI fluorescence intensity was ϳ20% of the mock cells level in 40% of the siRNA cells. 16% of the cells displayed fluorescence intensity for TFPI similar with mock cells. The remainder displayed fluorescence intensity of 30% (Ϯ7) of the mock cells levels.
The surface exposure of TFPI correlated positively and highly significantly with the expression of Cav-1, for both EA.hy926 (not shown) and HUVEC, as indicated by the correlation analysis of the fluorescence intensity performed on 201 cells (mock) and 174 cells (Cav-1 siRNA) following double immunostaining for TFPI and Cav-1 (Fig. 7D, r 2 ϭ 0.86 for mock cells, and r 2 ϭ 0.94 for siRNA cells).
The RNAi experimental controls (see supplemental materials) verified that the above described effects were due indeed to specific Cav-1 gene silencing. Neither TFPI nor Cav-1 were affected by transfection of cells with siGLO RISC-Free siRNA, a non-functional non-targeting siRNA (Fig. S4, A and B), but decreased drastically when the cells were co-transfected with Cav-1 siRNA (Fig. S4, C). Positive controls consisted of EC transfected with siCONTROL lamin A/C siRNA (Fig. S5). Whereas the expression of lamin A/C was strongly decreased (compare A, b and c, with B, a and d), neither cell surface TFPI (B, b and c) nor Cav-1 (B, e and f) were affected by depression of lamin A/C.
Silencing of Cav-1 Decreases the Activity of TFPI on the EC Surface-The activity of TFPI was assessed through the inhibition of TF-FVIIa-mediated activation of FX. To study endogenous TF-dependent FXa generation, we induced TF expression in EA.hy926 and HUVEC through incubation with 10 ng/ml r-tumor necrosis factor-␣ for 5 h (16). Fig. 8A displays FXa generation curves on mock-transfected and Cav-1 siRNAtreated HUVEC. The overall capability of the cells to activate FX (total FXa, open symbols), assessed in the presence of anti-TFPI IgG, was linear over time for all three concentrations of FVIIa tested, as well as similar among mock-and Cav-1 siRNA-transfected HUVEC, thus suggesting equivalent expression of TF.
FXa generation measured in non-inhibited assays (Fig. 8A, closed squares and circles) varied with the FVIIa concentration, as well as between the mock-and siRNA-transfected cells. Whereas the cells in both groups proved efficient in inhibiting FX activation at 0.2 nM FVIIa (Fig. 8A, a), only TFPI on mock cells was able to keep FX activation at low levels for 10 and 50 nM of added FVIIa (Fig. 8A, b and c). Meanwhile, the generation of FXa increased constantly on Cav-1 siRNA-transfected HUVEC.
The amount of FX whose activation was prevented by TFPI on HUVEC was calculated for each condition as described for 293 cells. As shown in Fig. 8B (closed triangles), the inhibitory capability of TFPI on Cav-1 depressed cells was significantly impaired (ϳ6 times diminished, p Ͻ 0.005). After normalizing for differences of the surface TFPI antigen, 150 pM for normal HUVEC and 63 pM for Cav-1 siRNA-transfected cells, the two cell groups had similar inhibitory potency against the low level of FXa generated by 0.2 nM FVIIa (ϳ80% of the total FXa). When challenged with larger amounts of FVIIa and faster rates of FX activation, TFPI on mock cells preserved its inhibitory potency almost intact, being still capable of preventing the generation of 60 -80% of the total FXa in the presence of 10 and 50 nM FVIIa. Conversely, equimolar levels of TFPI were ϳ3 times less potent inhibitors of FXa generation (p Ͻ 0.0001) in the absence of Cav-1, thus allowing the activation of FX to proceed almost uninhibited (Fig. 8B, b).
Similar differences were observed when the inhibitory efficiency of cell surface TFPI was tested against exogenous TF-FVIIa (Fig. 8C). Although TFPI in mock cells proved equally efficient in inhibiting FX activation for all the concentrations of TF-FVIIa tested, the inhibitory capabilities of TFPI could diminish by as much as 7-fold in Cav-1 siRNA-transfected HU-VEC in the presence of 50 nM FVIIa (Fig. 8C).
The inhibitory activity of TFPI in EA.hy926 was also significantly diminished following Cav-1 depression, albeit to a lower extent than in HUVEC. After normalization of the values for cell surface antigen, we found that the capability of TFPI to inhibit TF-FVIIa-dependent generation of FXa was ϳ1.5 times lower in Cav-1 silenced cells than in mock cells (not shown).
Effect of Cav-1 Silencing on the Formation of the TF-FVIIa-FXa-TFPI Complex-At the end of the activity assay, both HUVEC and EA.hy926 cells were fixed and double immunostained for cell surface TFPI (mAb TFPIK-9/horse anti-mouse IgG-FITC) and TF (goat anti-human TF IgG/donkey anti-goat IgG-Cy 5 ), then permeabilized and labeled with rabbit anti-Cav-1 IgG/donkey anti-rabbit IgG-Cy3. Fig. 9 illustrates the distribution and overlap between the three proteins. Measurement of fluorescence intensity and calculation of co-localization percentages were done as for 293 cells. Mock cells exhibited strong patching of cell surface TF and TFPI over areas rich in Cav-1 (Fig. 9, A, a-d, and B, a-d, ϳ80% mean overlap), indicating that the quaternary complex did co-localize with caveolae (panels d, white indicates triple co-localization). The colocalization reached almost 100% on the apical surface and on the cell edges (A and B, panels d, yellow arrows).
Such co-localization was no longer observed if EC were incubated with inhibitory anti-TFPI IgG before addition of FVIIa. Similar to the 293 cell clones (see above), the percentage of TF overlapping TFPI, Cav-1, or both, decreased to ϳ30% (not shown).
Although expressing lower intensity of fluorescence for cell surface TFPI than mock cells (Fig. 9, A and B, panels e and f), the cells where Cav-1 was silenced displayed normal levels of TF (panels g). Mean fluorescence intensity for TF (arbitrary units) was not significantly different among HUVEC and EA.hy926 cells. The values determined for mock cells and siRNA cells with normal levels of Cav-1 were 114 Ϯ 11 and 109 Ϯ 10 (HUVEC), respectively, 100 Ϯ 11 and 90 Ϯ 12 (EA.hy926). For Cav-1 siRNA-transfected cells with low levels of Cav-1 we measured 106 Ϯ 11 (HUVEC) and 95 Ϯ 7.2 (EA.hy926) arbitrary units of TF fluorescence intensity. The response of Cav-1 siRNA cells was heterogeneous, most likely because of the relatively significant proportion of cells in this population that still had normal levels of Cav-1 and TFPI (Fig. 9A, e-h; note the very low magnification images of Cav-1 siRNA-transfected HUVEC).
The triple co-localization TF/TFPI/Cav-1 was similar between mock cells and the population of siRNA cells that had normal levels of Cav-1 (white; Fig. 9B, d and h, yellow arrows). In contrast, we observed very little co-localization between TF and TFPI and/or Cav-1 in Cav-1-depressed cells (ϳ15% average overlap, Fig. 9, A and B, panels h). DISCUSSION We showed here that Cav-1 and/or caveolae actively regulate the anticoagulant activity of TFPI, apparently through a dual mechanism: by stabilizing the exposure of the inhibitor on the plasmalemma surface and by creating a microenvironment that enhances the down-regulation of TF-FVIIa activity by TFPI.
Despite reports trying to identify the factors that regulate the inhibition of TF-FVIIa activity by TFPI, the molecular mechanism of TF-FVIIa-FXa-TFPI complex formation is not yet fully understood. In ECV304 and monocytes, TFPI mediates the translocation of the quaternary complex to lipid rafts (11,16,18). In EC, where the majority of membrane TFPI resides in caveolae/lipid rafts, disturbing the rafts with cholesterol-extracting agents decreases the activity of TFPI (10). Because membrane TFPI is now believed to play the major role in down-regulation of TF-FVIIa activity on cell surfaces, we decided to investigate whether Cav-1/caveolae regulate the TFPIdependent inhibition of TF-FVIIa activity.
Our strategy involved first the expression of tagged fulllength TFPI and Cav-1 in HEK293, a cell line that is naturally deficient of TFPI and Cav-1. We found that the properties and functional activity of EGFP-TFPI matched those of the endogenous TFPI in EA.hy926 cells. We thus confirmed that EGFP-TFPI is a valid fluorescent indicator of native TFPI. Next we confirmed by immunofluorescence and freeze-fracture EM that the expression of Cav-1 in HEK293 induced the appearance of caveolae proper, as it was described for other cell types (21,33).
Analysis of HEK293 expressing TFPI, Cav-1, or both proteins, by live cell and (immuno)fluorescence microscopy, and by functional assays, revealed novel aspects of TFPI distribution and function. In the absence of Cav-1, TFPI had predominant intracellular localization and was largely secreted. Without affecting the overall levels of TFPI, Cav-1 expression instrumented a shift in the distribution of TFPI, with the inhibitor becoming predominantly associated with the plasma membrane surface and less secreted into the medium.
FRAP results showed that cell surface TFPI was significantly less laterally mobile in the presence of Cav-1. Under standard culture conditions Cav-1 becomes highly immobile once it reaches the plasma membrane, and shows very limited lateral diffusion and exchange with the intracellular pool (34). The fluorescence recovery profiles of EGFP-TFPI in TFPI ϩ Cav cells (present paper, Fig. 3) and caveolae-associated GFPtagged Cav-1 (35) are strikingly similar. We suggest that Cav-1/caveolae retain and stabilize TFPI on the membrane surface, hence the increased surface exposure and decreased lateral mobility of EGFP-TFPI in Cav-1 expressing cells.
The functional impact of TFPI localization in caveolae was analyzed through the inhibition of TF-FVIIa activity, as reflected by inhibition of FXa generation. Activation of FX by a pre-formed mixture of 10 nM FVIIa, which is the equivalent of the plasma level of FVII (7), and thromboplastin (crude TF) was almost completely inhibited by the TFPI on TFPI ϩ Cav cells, but not on TFPI cells. As observed on the 293 cell clones that express different levels of TFPI and Cav-1, the inhibitory activity of TFPI was dependent on the level of antigen only when Cav-1 was also present. In the absence of Cav-1, varia- tions of cell surface TFPI antigen had no noticeable impact upon the functional capabilities of the inhibitor. Accordingly, equal or even larger amounts of TFPI were always far less active against exogenous TF in TFPI cells.
When assessing the inhibitory potency of TFPI against endogenous TF, we used concentrations of FVIIa meant to span both sides of 10 nM (the equivalent of the plasma level of FVII). The higher concentration of FVIIa (50 nM) was intended to mimic locally increased levels of FVIIa. These may occur in normal circumstances, because of FVIIa interactions with phospholipids and/or proteoglycans (7), or in pathological conditions where a hypercoagulable state arises. Regardless of the concentration of FVIIa added, equimolar amounts of TFPI prevented twice as much FXa generation in the presence of Cav-1. The increase in TFPI activity was independent of the level of TF, which was also not affected by Cav-1 expression. TFPI ϩ Cav cells were able to keep FX activation at very low levels for both 10 and 50 nM added FVIIa, preventing, in both cases, ϳ85% of total FX activation. The fact that TFPI cells were capable, in similar conditions, to inhibit only ϳ20% of FX activation suggests that the presence of Cav-1 was essential for the preservation of the TFPI inhibitory potential, especially when high levels of FVIIa, and consequently FXa, were present. Altogether, these findings suggest that, when expressed in naturally deficient cells such as HEK293, Cav-1 actively enhanced the anticoagulant activity of TFPI, probably through a combination of increased TFPI antigen retention on the cell surface and microenvironment modifications brought about by the formation of caveolae.
Next we verified that Cav-1 played a similar role in EC. To the best of our knowledge, this is the first direct proof that caveolae (or Cav-1) regulate the activity of TFPI against TF-FVIIa in EC. Using two well characterized duplex RNA oligonucleotides (30,31), we achieved both Cav-1 depletion and considerable reduction of cell surface TFPI antigen and activity in both HUVEC and EA.hy926. This indicates that Cav-1 on the cytosolic face of the membrane might control the exposure of TFPI on the plasmalemma surface.
Cav-1 depletion decreased the capability of TFPI to downregulate the activity of endogenous TF-FVIIa, an effect that was more visible for high FVIIa concentrations. Similar with 293 cells, equimolar levels of TFPI were ϳ3 times less potent in preventing TF-FVIIa-dependent FX activation in the absence of Cav-1. As a result of combined lower cell surface TFPI antigen and diminished functionality, the procoagulant activity of Cav-1 depressed EC could increase by as much as ϳ5 times.
TF-bearing circulating microparticles derived from leukocytes and/or other blood cells probably represent a very significant source of procoagulant species in several diseased states. We sought to find out whether cell surface TFPI on EC could inhibit the procoagulant activity of preformed TF-FVIIa complexes, used as an in vitro equivalent of microparticle-associated TF-FVIIa. Accordingly, we assayed in HUVEC the inhibitory activity of TFPI against several mixtures of FVIIa (including the 10 nM equivalent of plasma FVII) and crude TF. Depletion of Cav-1 decreased by ϳ6-fold the potency of cell surface TFPI to block FX activation. Furthermore, high levels of FVIIa-TF and, consequently large amounts of FXa, could overwhelm the inhibitory capability of TFPI in the absence of Cav-1.
The nature of the interaction(s) that keeps TFPI in caveolae is under investigation in our group. TFPI associates through a direct or indirect glycosylphosphatidylinositol anchor with lipid rafts (10,17), and both cholesterol and sphingolipids influence the association of TFPI with lipid rafts in EC. 2 Because caveolae are highly enriched in cholesterol and sphingolipids, it is conceivable that these lipids mediate the caveolar distribution and function of TFPI. Whether such a mechanism is instrumental only in cells that express caveolins, as suggested by the lack of effect of cholesterol depletion on the inhibition of TF-FVIIa by TFPI in HEK293 (36), remains to be determined.
As indicated by the fluorescence microscopy, a high percentage of the cell surface TF and TFPI becomes co-localized with the submembrane Cav-1 during and/or after the formation of the quaternary complex. Blocking TFPI-FXa and TFPI-FVIIa interactions with anti-TFPI IgG prevents the redistribution of TF and the overlap of the complex with Cav-1. We suggest that the complex either forms within caveolae or reaches them afterward. Regardless of the mechanism, the association of TFPI with caveolae plays the determinant role in targeting the TF-FVIIa-FXa-TFPI complex to caveolae.
How does Cav-1, a protein located on the inner surface of the membrane, influence localization and function of outer surface proteins such as TFPI and TF? In normal conditions, lipidassociated proteins, TFPI included, reside very briefly within transient rafts, which are extremely small in size and very dynamic (37). Clustering of the small rafts and their associated proteins leads to the formation of larger, less mobile rafts that can be further stabilized by Cav-1. When cells are exposed to FVIIa/FX, TFPI forms complexes with TF-FVIIa/FXa regardless of the presence of Cav-1. In Cav-1-deficient cells, which likely contain less cholesterol in the membrane (35), the strength of the interaction between the complex and rafts may be too low, or the residence time too short, to "lock" the quaternary complex in the condensed cholesterol-sphingolipid domains. If the complex dissociates, a dynamic equilibrium between the formation and the dissociation of the complex would be expected, with a significant fraction of TF-FVIIa being left free to generate FXa at any given moment. This would explain the continuous generation of FXa in Cav-1-deficient cells. The presence of Cav-1, which poses a barrier into the lateral mobility of TFPI, would produce a longer lasting inhibitory effect once the complex is stabilized in the cholesterol-sphingolipidrich environment of caveolae/rafts.
Our results suggest that caveolae could concentrate the interacting molecules in particular regions of the cell surface and facilitate the formation of the quaternary complex. The process may have high biological relevance because it identifies Cav-1 and/or caveolae as a key factor in the regulation of TFPI-dependent inhibition of TF-driven coagulation pathway in EC. As such, our findings convey a novel function to caveolae, namely active regulation of hemostasis.