Tissue factor (TF) is a transmembrane cofactor that binds and promotes the catalytic activity of factor (F) VIIa. The TF/VIIa complex activates FX by limited proteolysis to initiate blood coagulation and helps provide the thrombin burst that is important for a stable thrombus. TF is present both in the extravascular compartment, where it functions as a hemostatic envelope, and the intravascular compartment, where it contributes to thrombus formation, particularly when endothelial disruption is minimal. The regulation of its cofactor function appears to differ in the two compartments. Intravascular TF derives predominately from leucocytes, with either monocytes or neutrophils implicated in different models of thrombosis. This TF exists mostly in a non-coagulant or cryptic form and acute events lead to local decryption of TF and FX activation. A variety of experimental observations imply that decryption of leucocyte surface TF involves both a dithiol/disulfide switch and exposure of phosphatidylserine. The dithiol/disulfide switch appears to involve the Cys186-Cys209 disulfide bond in the membrane-proximal domain of TF, although this has not been demonstrated in vivo. Activation of a purinergic receptor or complement has recently been observed to decrypt TF on myeloid cells and a dithiol/disulfide switch and the oxidoreductase, protein disulfide isomerase, have been implicated in both systems. The molecular mechanism of action of protein disulfide isomerase in TF encryption/decryption, though, remains to be determined.
Early studies detailed the selective cellular expression of tissue factor (TF) in astrocytes, smooth muscle cells, epidermus, renal glomeruli, vascular adventitia and placenta, but not in cells that come into contact with flowing blood . This led to the traditional concept of the hemostatic envelope by which TF is sequestered from the circulation by an intact endothelium. It is still widely taught that the initiation of thrombosis and hemostasis in vivo occurs when a breach in the vessel wall lining exposes TF to circulating coagulation factors. However, this concept is increasingly being challenged as evidence supports the presence of TF in the intravascular space. TF microparticles from the blood of healthy volunteers have been found to shorten the lag time in calibrated thrombin generation assays , and several groups have demonstrated the presence of TF in the circulation at detectable levels by antigenic and functional assays [2-6]. These results remain controversial, with some groups suggesting that the measurements of circulating TF are spurious due to collection artefacts .
Others have failed to detect measureable TF activity in plasma or TF associated with microparticles isolated from unstimulated whole blood and have suggested that the discrepant results are partly due to non-specific anti-TF antibodies [8, 9]. In addition to full-length TF, an alternatively spliced TF (asTF) has been described, which lacks the transmembrane domain and has an altered C-terminus due to loss of exon 5 . Although initial studies suggested that asTF was procoagulant and delivered to the developing thrombus , subsequent studies have shown that asTF is not secreted, non-coagulant and may have a greater role in angiogenesis than coagulation [11, 12]. Most immunological studies do not differentiate between full-length TF and asTF and there is controversy with regards to the level of TF in the circulation. Initial studies suggested that 10–30% of circulating TF could be asTF, although more recent studies suggest that asTF is not normally secreted, but has a role in tumour cells.
Regardless, circulating TF has been found within the platelet and/or fibrin-rich thrombus in several studies. TF is found by immunostaining of clots formed by ex vivo perfusion of blood from healthy human volunteers over collagen slides or through coronary stents [6, 13]. In these studies, the TF in the platelet thrombus was associated with leucocytes or microparticles and derived from whole blood. Murine models of thrombosis confirm these findings. Saphenous vein clots in a murine ferric chloride thrombosis model were TF positive by immunostaining . Using mice in which TF expression was ablated in the smooth muscle cells, TF antigen from the circulation, not from the vessel wall, was incorporated throughout the thrombus . Real-time imaging of thrombus formation using high speed intravital fluorescence microscopy shows that circulating TF is delivered to the leading edge of the developing platelet thrombus after laser stimulation [16, 17].
Mouse models of thrombosis have been useful in determining the functional role of circulating TF in arterial thrombosis. The laser injury model of thrombosis  has been developed as a minimal injury model. Thrombus formation in this model is dependent on endothelial activation  and on thrombin  but not on the platelet-collagen interaction . This contrasts with other injury models of thrombosis in which the stimulus for thrombus initiation includes endothelial disruption. These include the ferric chloride and rose Bengal models, and the inferior vena cava ligation model of deep venous thrombosis (DVT) where subendothelial vascular wall TF is thought to be exposed to the circulation. It is worth noting that the laser injury model, like the vena cava stenosis model of DVT described below, is a non-occlusive model of thrombosis. These models allow continuous delivery of microparticles and other circulating forms of TF to the thrombus. This contrasts with occlusive models of thrombosis in which delivery of circulating components is curtailed at the time of vessel occlusion. This may impact on the influence of circulating TF on thrombus formation.
Comparing and contrasting these mouse models of thrombosis has allowed some clarification of the role of circulating TF in arterial thrombosis. Reciprocal bone marrow transplant studies between wild-type mice and mice expressing only 1% human TF show that hematopoietic-derived TF contributes to platelet thrombus and fibrin formation, particularly the propagation phase of thrombus formation . Of interest, the distribution of fibrin deposition in the thrombus was different in the mice lacking hematopoietic TF, with fibrin deposition predominantly adjacent to the vessel wall rather than throughout the thrombus. This suggests that hematopoietic-derived TF is required to provide the substrates for fibrin stabilization in the thrombus as it develops away from the vessel wall.
These results directly contrast with results using endothelial disruption models of thrombosis and the same bone marrow transplant strategy. The rose Bengal and tail bleeding models both showed that TF from the hematopoietic compartment was not required, while non-hematopoietic TF was essential . Deletion of TF from the smooth muscle cell resulted in significant attenuation of large vessel thrombus formation in response to ferric chloride injury . In this model, TF from the circulation was detected within the fibrin-rich clot; however, this circulating TF was insufficient to compensate for the deletion of TF from the smooth muscle compartment. The mechanism of ferric chloride injury has previously been characterized as being due to endothelial denudation resulting in exposure of subendothelial matrix components, including collagen [21, 23-25]. In light of this, one interpretation of the studies summarized above is that the degree of injury to the endothelium determines the relative contribution of circulating TF to the size and stability of thrombus formation: in endothelial disruption models, vascular wall TF may be sufficient to drive fibrin formation; however, in situations with minimal injury to the endothelium, circulating TF is required for the propagation and possibly the initiation of arterial thrombosis. However, the extent of denudation of endothelium in the ferric chloride injury has recently been called into question [26, 27], so the interpretation of the relationship between subendothelial TF and thrombosis in this injury model is less clear. Clarification may be gained from ferric chloride injury studies of mice where TF is deleted from both hematopoietic and endothelial cells. If subendothelial TF in this model contributes to thrombus formation then it presumably is exposed to the circulation.
Early studies of TF in animal models of venous thrombosis indicated that there was a role for circulating TF in DVT. A study in which a collagen-coated thread was inserted into the jugular vein of rabbits found that TF-positive leucocytes accumulated in the thrombus and that circulating TF was required for fibrin formation . The most common murine model used for venous thrombosis involves inferior vena cava ligation. This is not the best model for studying circulating TF because ligation restricts the delivery of circulating cells and microparticles to the thrombus. In addition, ligation injures the endothelium and exposes the circulation to subendothelial TF and upregulates endothelial TF . Bearing in mind these limitations, results suggest a role for TF in initiation as well as propagation of DVT in this model. Injection of MPs into mice after inferior vena cava ligation increases the thrombus weight in the early phase but not in the plateau phase of the thrombosis . In a rat ligation model, there was a rapid accumulation of TF-positive leucocytes within the thrombus in conjunction with P-selectin expression in endothelial granules . Similar to the laser injury model, mice deficient in either PSGL-1 or P-selectin have smaller thrombi in the inferior vena cava ligation model ; however, deletion of circulating TF does not have an effect on thrombus weight in this model. In a non-occlusive inferior vena cava DVT model that uses flow restriction but not occlusive ligation, the role of circulating TF in DVT was able to be elucidated with greater clarity . Deletion of TF from the myeloid cells in this flow restriction DVT model resulted in a decrease in TF delivered to the clot as detected by immunostaining. This decrease was correlated with a marked functional defect in thrombus initiation, indicating that myeloid TF significantly contributes to thrombus initiation in DVT. Thus it appears that circulating TF has a role both in venous and arterial thrombosis.
The source of circulating TF remains a controversial area. One basic question is whether circulating cells or microparticles deliver functional TF to the thrombus. Exogenous monocyte-derived microparticles were delivered to the thrombus in the laser injury model and disruption of the P-selectin/P-selectin glycoprotein ligand 1 (PSGL-1) interaction resulted in decreased TF delivery to the thrombus . Leucocytes were not involved in this model of thrombosis in the first 3 min, while PSGL-1-mediated TF delivery occurred within seconds of endothelial injury . Together, this strongly suggests that the TF in the thrombus was derived from monocytes and in the form of microparticles. Similarly, in a non-occlusive murine model of venous thrombosis, TF was more strongly associated with monocytes than neutrophils. In this DVT model, neutrophils, monocytes and platelets cooperated to propagate the venous thrombosis and initiation was reliant on myeloid-derived TF in particular . More recently, TF-positive neutrophils have been implicated in the initiation of thrombus formation in the laser injury model of thrombosis . Exogenous and endogenous neutrophils were seen to accumulate at the vessel wall 3–5 s after laser stimulation, and blocking neutrophil accumulation at the thrombus attenuated TF accumulation, implying that TF delivery was via neutrophils. The differences observed in the recruitment of leucocytes to the thrombus in the different studies using the same laser injury model are yet to be elucidated. However, the fact that different cells dominate in models that use different thrombosis stimuli is likely to reflect the heterogeneity in mechanisms of thrombosis.
In summary, evidence from conditional TF deletion mice and bone marrow transplant studies shows that TF is present in the circulation and potentially functional. Furthermore, blood-borne TF can be found within the platelet thrombus using multiple techniques: visualization of TF by real-time fluorescence in intravital models of murine thrombosis, or immunohistochemistry staining or electron microscopy after excision of the clot. The source of the blood-borne TF appears to be predominately leucocyte derived, with either monocytes or neutrophils implicated in different models of thrombosis. The functional contribution of the blood-borne TF is dependent on the type of stimulation used to induce thrombus formation. Rather than being contradictory, these differences are likely to indicate that the various models use different processes and that ‘thrombosis’ is a heterogenous disease.
Encrypted and decrypted tissue factor
The finding that TF is in contact with the circulation but does not trigger obvious coagulation implies that the majority of the cofactor circulates in a coagulation inactive or cryptic form and that acute events lead to local decryption of TF and blood coagulation. The features of cryptic and decrypted TF have been defined using cell culture models. Cryptic TF slowly binds FVIIa (Kd of 5–20 nm) and cleaves a peptidyl substrate but not FX, while decrypted TF rapidly binds FVIIa (Kd of < 1 nm) and cleaves both a peptidyl substrate and FX [33-35]. Binding of VII/VIIa to cryptic TF takes 1–2 h to reach equilibrium, while binding to decrypted TF is established within a few minutes. The cryptic form of TF identified on cultured cells appears to reflect the procoagulant inactive TF observed in the circulation, in that circulating TF does not activate FX until it is decrypted.
The mechanism(s) by which encryption and decryption of TF occurs remains a hotly debated topic. The different proposals are discussed below. The mechanism that has received a lot of attention in recent years involves a thiol/disulfide switch in TF (Fig. 1).
Tissue factor encryption
Four different mechanisms have been proposed to keep TF in a cryptic configuration until it is required for coagulation. It is probable that all mechanisms contribute to TF encryption to some extent and that the relative importance of each process may well be different on different cells and in different contexts.
Sequestering of TF in cholesterol-rich microdomains or lipid rafts inhibits activation of FX in some cells . In human embryonic kidney cells, for example, disruption of rafts with methyl-β-cyclodextrin resulted in a 3-fold stimulation of TF procoagulant activity . Dimerisation and oligomerisation of TF on the cell surface has also been proposed to encrypt TF by preventing interaction with FX [38, 39]; however, TF dimers have not been consistently observed in other studies. Exposure of phosphatidylserine on the surface of blood and vessel wall cells is essential for productive assembly of coagulation complexes  and a negatively-charged phospholipid surface is required for maximal TF activity [41, 42]. It has been proposed that a neutral or anticoagulant phospholipid environment is what encrypts TF . The extent of the contribution of the phospholipid environment to TF encryption has been questioned in recent years by evidence that the cofactor is subject to post-translational modification involving one of its two disulfide bonds .
Cryptic TF on the surface of different cells has been found to contain unpaired cysteine thiols [45-48], which implies that one or both of the cofactor disulfide bonds can exist in the reduced dithiol form. The extracellular region of TF consists of two fibronectin type III domains that are made up of two antiparallel β-sheets. The N-terminal domain contains a typical structural disulfide bond (Cys49-Cys57) that links across the two β-sheets. The disulfide bond in the C-terminal (Cys186-Cys209) or membrane-proximal domain straddles the F and G strands of the same antiparallel β-sheet and is exposed to solvent [47, 49, 50]. It is this bond that appears to be reduced in cryptic TF .
The TF Cys186-Cys209 disulfide has a –RHStaple configuration, which is the most common configuration of the functional disulfide bonds known as allosteric bonds [52, 53]. Once proteins are made by cells they are chemically modified in a variety of ways to control where, when and for how long they function. There are three basic types of post-translational modifications . Type 1 is covalent modifications of amino acid side chains and type 2 is hydrolytic cleavage or isomerization of peptide bonds. Type 3 post-translational modification is cleavage of the disulfide bonds that link pairs of cysteine residues in the mature protein. Some of these bonds are dynamic and can be reductively cleaved by oxidoreductases or by thiol-disulfide exchange in order to control protein function; such bonds are termed allosteric disulfides.
The cystine residue is defined by five angles that are calculated from eight atoms. Twenty possible cystine configurations are possible based on the sign of each of the five angles and all of the different types have been identified in protein crystal structures [53, 54]. Of the 20 disulfide bond configurations, three are emerging as allosteric configurations. The most common allosteric configuration is the –RHStaple. The incidence of this configuration amongst the allosteric bonds is six times the incidence of this bond type in all disulfides in the PDB [52, 54]. A defining feature of –RHStaple bonds is the close proximity of the α-carbon atoms of the two cysteine residues, which can impart a significant strain on the bond [53, 55].
Cryptic TF binds FVIIa slowly and with weaker affinity than the decrypted cofactor and the complex cannot activate FX [33-35]. While TF in a procoagulant lipid environment substantially increases activation of FX , TF in a neutral lipid environment or the lipid-free extracellular domain of TF still functions as a cofactor for FX activation . In keeping with the nature of cryptic TF, though, ablation of the Cys186-Cys209 disulfide bond in full-length or soluble TF destroys procoagulant activity [46, 48, 57, 58]. In the initial studies, mutation of the Cys186-Cys209 disulfide bond by replacing either or both cysteines with alanine or serine resulted in poor expression of the mutants at the cell surface. A valid argument was made that extrapolating functional differences between wild-type and disulfide mutant TF when the expression levels of the proteins are very different could lead to incorrect conclusions . To address this question, cells expressing comparable levels of wild-type and disulfide mutant TF were created and assessed for procoagulant activity . The disulfide mutant was completely devoid of coagulant activity, which validates the initial conclusion. Additional support for a functionally inactive reduced TF comes from recent studies of murine TF. The membrane proximal disulfide bond in TF is conserved in all species from the pufferfish, Takifugu rubripes, to Homo sapiens [61, 62]. In contrast to expression of the human disulfide mutant protein, expression of the murine mutant was comparable to wild-type protein in baby hamster kidney cells . As for the human protein, ablation of the membrane-proximal disulfide bond in murine TF (Cys190-Cys213) abolished procoagulant activity.
A standard redox potential of −278 mV was determined for the Cys186-Cys209 disulfide of recombinant soluble TF and the Cys186 and Cys209 sulfur atoms of reduced TF were found to be within 3–6 Å of each other . This redox potential and spacing of the unbonded sulfur atoms implies that the cysteines are primed for disulfide bond formation. A key question for the thiol/disulfide switch mechanism of TF encryption/decryption is how the bond is maintained in the reduced state?
There is evidence for reaction of the unpaired TF cysteines, particularly Cys209, with nitric oxide and/or glutathione [45, 46, 62]. Alkylation of one or both of the cysteines by these compounds may restrict disulfide formation. In addition, the thioredoxin/thioredoxin reductase/NADPH system  has recently been implicated in control of cellular TF activity . This redox system reduced the Cys186-Cys209 disulfide bond in recombinant soluble TF and ablated procoagulant activity, while inhibition of thioredoxin expression in human breast cancer cells increased surface TF procoagulant activity. The level of cellular thioredoxin protein was reduced 5-fold, which resulted in a 3-fold increase in TF activity. Thioredoxin knockdown did not affect TF or thioredoxin reductase protein levels in the cells. Notably, the thioredoxin catalytic and TF Cys186-Cys209 disulfide bonds have very similar standard redox potentials (−270 and −278 mV, respectively), so the energetics of the system are not inconsistent with thioredoxin and its cofactors maintaining cysteines 186 and 209 in a reduced state on the cell surface. The link between these two systems requires confirmation but is a topic that deserves further exploration.
Tissue factor decryption
There is no doubt that a negatively charged phospholipid surface is required for maximal TF activity [41, 42]. Exposure of phosphatidylserine on TF bearing cells, though, does not account for full TF decryption. Saturating concentrations of annexin V, which quenches phosphatidylserine on the cell surface, does not completely inhibit TF procoagulant activity. Inhibition in the range 50–80% has been observed in different studies [35, 47, 65, 66]. The other event required for full TF activity appears to be oxidation of TF Cys186 and Cys209 cysteine residues to the cystine disulfide bond. Much of the evidence for this chemical event has been gathered using non-physiological activators of TF and so the relevance of these studies should be and has been questioned . Recent studies using physiological activators of TF, though, have supported the observations with the artificial activators.
The calcium ionophore, ionomycin, cell disruption by freeze/thaw and a number of different thiol oxidizing agents decrypt TF on cultured cells. Cryptic TF contains unpaired cysteine thiols that are diminished upon activation of the cofactor with ionomycin, which correlates with a conformational change in the vicinity of the Cys186-Cys209 disulfide bond [45-47]. HgCl2 is a very efficient activator of cellular TF that has been proposed to function by oxidizing cysteines 186 and 209 to a cystine [47, 63, 68]. This proposal is supported by the finding that HgCl2 oxidizes the Cys186/Cys209 dithiol in soluble TF at concentrations that decrypt TF on cells . In addition, the dithiol cross-linkers, bis-maleimides and trivalent arsenic, and methyl methanethiolsulfonate, at concentrations where it oxidizes closely spaced cysteines to a cystine, are also efficient activators of cell surface TF [47, 63]. Moreover, thiol alkylators that block disulfide bond formation inhibit TF decryption, both in vitro [46, 47] and in vivo [45, 69]. For example, the increase in fibrin formation seen with the infusion of TF-positive monocyte microparticles in mice is abrogated when the microparticles are pretreated with the thiol alkylator, 3,3'-dithio-bis(6- nitrobenzoic acid) (DTNB) . Some have argued that binding of FVIIa to TF masks the membrane proximal disulfide, thus preventing access to the bond by modifiers . This argument only applies if the interaction is covalent. As FVIIa binding to TF is a reversible interaction, the protease will be constantly associating and dissociating from its cofactor in the circulation. Even if the dissociation rate is very slow, FVIIa will not preclude access to the TF disulfide by redox modifiers.
Antibodies that neutralize protein disulfide isomerase (PDI) inhibit TF decryption on cultured cells  and TF-mediated thrombosis in murine models [45, 69]. PDI is an oxidoreductase that resides mostly in the endoplasmic reticulum in mammalian cells, but is also secreted and influences protein dithiol/disulfide exchange events on the cell surface . Inactivation of PDI with specific antibodies and by alkylation of the active site thiol groups results in inhibition of platelet thrombus size and fibrin formation in both micro- and macrovascular models [45, 69]. There is clearly a platelet-independent role for PDI in coagulation. Elimination of platelet accumulation via platelet depletion, αIIbβ3 integrin blockade or deletion of the PAR4 thrombin receptor on platelets all result in normal fibrin levels in the laser injury model of thrombosis; however, infusion of anti-PDI antibodies or small molecule inhibitors results in abrogation of fibrin deposition [71, 72]. PDI can influence S-nitrosylation  and S-glutathionylation  of proteins and there is evidence for both of these modifications of the Cys209 thiol of cryptic TF [45, 46, 62]. It is also possible that the chaperone function of PDI, in addition to its redox activity, is involved in TF encryption/decryption.
Two mechanisms of TF decryption by physiologically relevant mediators have recently been described and both are consistent with a thiol/disulfide switch in TF. First, stimulation of purinergic receptor 7, a ligand-gated ion channel, by ATP decrypts TF on mouse myeloid cells primed with lipopolysaccharide or interferon γ . The decryption is inhibited by the thiol alkylators, DTNB and nitric oxide. TF-bearing microparticles are released with purinoceptor receptor 7 stimulation, which is attenuated by an antithrombotic anti-PDI antibody, and the procoagulant potential of the microparticles is inhibited by DTNB. Second, antithymocyte globulin is a potent decrypter of TF on monocytic cells, which is dependent on lipid raft integrity and complement activation . Decryption of TF by antithymocyte globulin is prevented by thiol alkylators, an anti-PDI antibody and a small molecule PDI inhibitor, and does not require maximal phosphatidylserine membrane exposure. These rapid decrypters of TF link coagulation with myeloid cell ion channel activity and the complement system.
Conclusions and perspectives
Thrombosis is initiated by TF in the circulation. The cofactor is present on myeloid cells and microparticles in mostly a cryptic configuration that requires decryption before it can productively bind FVIIa and activate FX. A number and variety of experimental observations imply that decryption involves both formation of a disulfide bond between unpaired cysteine residues 186 and 209 in cryptic TF and exposure of phosphatidylserine on the cell surface. Both events appear to be required for full TF procoagulant activity. Proof of the thiol/disulfide switch in TF, however, is reliant on a direct method of measuring the redox state of Cys186 and Cys209 in vivo, which is currently not available. The possibility that the redox change occurs in or between other proteins that participate in TF encryption/decryption cannot be excluded at this time. Two physiologically relevant decrypters of TF have been identified, involving activation of a purinergic receptor or complement on myeloid cells, and PDI has been implicated in both systems. The molecular mechanism of action of PDI in TF encryption/decryption remains to be determined.
Disclosure of Conflicts of Interest
The authors state that they have no conflicts of interest.