José A. López, Baylor College of Medicine, Thrombosis Research Section, BCM 286, N1319, One Baylor Plaza, Houston, TX 77030, USA. Tel.: 713-798-3480; fax: 713-798-3415; e-mail: email@example.com
Summary. Hemostasis at sites of blood vessel injury and its pathologic counterpart, thrombosis, involve a complex interplay between several blood elements: soluble proteins of the blood coagulation system, blood cells (most prominently platelets) and cell fragments, and elements of the vessel wall (endothelial cells and, at sites of injury, the exposed matrix and deeper cellular components). In this review, we focus on ways in which specialized membrane microdomains known as lipid rafts are involved in various phases of hemostasis and thrombosis.
In 1972, Singer and Nicolson  published a seminal review that coalesced the then rapidly evolving concepts of the structures of biologic membranes. They proposed that biologic membranes exist as fluid structures with a lipid membrane within which proteins are interspersed randomly and are free to diffuse laterally. They termed this the ‘fluid mosaic’, the mosaic being composed of the randomly interspersed proteins. This model precluded the presence of long-range order within the fluid mosaic, as this would prevent random distribution of the membrane components. The authors cautioned, however, that ‘this absence of long-range order should not be taken to imply absence of short-range order in the membrane’. They speculated that such short range order could be mediated by specific protein–protein or protein–lipid interactions leading to local clustering of components within the membrane, but that the ‘long-range distribution of such aggregates would be expected to be random over the entire surface of the membrane’. Over the past several years regions of the membrane with short-range ordered structure, which are vital to the function of the cell, have indeed been described, but their order is largely a consequence of lipid–lipid interactions. These regions of ordered lipids have been variously designated, but a useful descriptive term is ‘rafts’, meant to depict distinct structures floating within a sea of fluid, randomly moving lipid.
Lipid rafts are dynamic assemblies of sphingolipids and cholesterol, their formation driven by the tight self-association of the long, saturated acyl chains of the sphingolipids, and the planar hydrophobic structure of cholesterol (Fig. 1) . The length and saturated nature of the sphingolipid acyl chains are characteristics that separate them from the more abundant phospholipids of cell membranes, allowing them to pack more closely [3–5] and promoting a phase separation between sphingolipid-rich regions and phospholipid-rich regions. Alone, the sphingolipids would form a solid-like gel phase with a high melting temperature, in stark contrast to the disordered, low melting temperature of the phospholipids. Cholesterol has important effects on the structure of the sphingolipid-rich domains, inducing a ‘liquid-ordered’ phase (lo) in which the acyl chains are tightly packed as in the gel phase, but have a much greater degree of lateral mobility . Cholesterol also promotes the phase separation between sphingolipid-rich and phospholipid-rich regions, a property that explains how membranes with relatively low contents of sphingolipids (such as biologic membranes) are able to form liquid-ordered domains. This property of cholesterol explains why depletion of cholesterol from cell membranes is an effective means of disrupting rafts (see below).
Ordered microdomains comprise approximately 10% of the total cell surface area, a fraction that varies with cell type, the cholesterol and sphingolipid content of its membrane, and the state of cell activation. Each raft in an unactivated cell has been estimated to contain approximately 3500 sphingolipid molecules and 10–20 protein molecules . It is not known what limits the size of individual rafts within plasma membranes, especially in the cell's unactivated state, but it is clear that the rafts are usually free to diffuse laterally and may coalesce to form larger structures which facilitate transmembrane signal transduction.
Rafts are believed to enhance signaling functions by providing platforms that physically concentrate receptors, downstream kinases, and adaptor proteins involved in signaling pathways (Fig. 1). Clustering of raft components upon receptor ligation and cross-linking increases the local concentration of signaling molecules, facilitating the generation of signals. Key roles for rafts have been described in signaling from a large number of membrane receptors, prominent among these being the T- and B-cell antigen receptors and Fc receptors [8–10].
Rafts also provide the portals of entry for numerous intacellular pathogens, including viruses (e.g. HIV-1, influenza virus) and pathogenic bacteria (Mycobacterium bovis, Escherichia coli, Campylobacter jejuni), and parasites. In addition, intracellular parasites such as Toxoplasma gondii and Plasmodium falcipurum reside in host cells within parasitophorous vacuoles rich in raft proteins and lipids that resist fusion with lysosomes [11–13].
Rafts play important roles in almost all phases of hemostasis and thrombosis, including: (i) the initial adhesion of platelets to sites of vessel injury and transmission of intracellular signals produced by this adhesive interaction; (ii) capacitative calcium entry in platelets; (iii) the enzymatic reactions that occur on the platelet surface that initiate and maintain coagulation; and (iv) the generation of procoagulant microvesicles that increase the likelihood of thrombosis at sites distant to the vessel injury (Fig. 2).
Platelets recognize sites of vessel injury or atherosclerotic plaque rupture through a highly specialized complex on their surfaces, the glycoprotein (GP) Ib-IX-V complex, which attaches the platelets to the injured site by binding matrix-bound von Willebrand factor (VWF) . GP Ib-IX-V complex binding to VWF (which itself is bound to vessel wall collagen) decelerates the platelets and allows engagement of other receptors for matrix proteins, prominently two collagen receptors, GP VI and integrin α2β1, and integrin αIIbβ3, a receptor for fibrinogen. Signals from the GP Ib-IX-V complex, facilitated by signals through GP VI, appear capable of activating the two integrins to bind their ligands with higher affinity . The functions of two of the receptors involved in this complex adhesive process, the GP Ib-IX-V and GP VI/Fcγ complexes, depend on their localization to lipid rafts.
Role of rafts in GP Ib-IX-V complex functions
The GP Ib-IX-V complex contains four polypeptides: GP Ibα and GP-Ibβ, bound by a juxtamembrane disulfide bond, and GP IX and GP V, which associate with the complex non-covalently. Glycoprotein Ibα, the largest of the four polypeptides, is responsible for almost all the ligand-binding functions of the complex, containing binding sites for VWF, thrombin, coagulation factors XI and XII, P-selectin, and the leukocyte integrin Mac-1 (integrin αMβ2), all within a sequence of about 300 amino acids at its N-terminus. The ligand-binding domain is separated from the plasma membrane by a mucin-like stalk known as the macroglycopeptide, which facilitates the interaction with VWF as platelets attach to the vessel wall from rapidly flowing blood . Glycoprotein Ibβ and GP IX are much smaller polypeptides that are required for the surface expression of the complex , whereas GP V has been reported as having an ancillary role in binding collagen , with both GP V knock-out mouse platelets and human and rat platelets blocked with an anti-GP V antibody demonstrating diminished collagen-induced aggregation. GP V also appears to negatively modulate the signals produced when thrombin binds the complex . GP V-null mice demonstrate neither a bleeding nor a thrombotic phenotype; on balance, the increased sensitivity of their platelets to thrombin  appears to be balanced by their diminished responsiveness to collagen .
The GP Ib-IX-V complex was first reported to localize to membrane microdomains by Dorahy et al. in 1996 , but these researchers did not investigate or speculate on the functional consequences of this association. We developed interest in the possible localization of the complex to lipid rafts when it became clear that cysteine acylation, particularly by palmitate, is a feature of many raft-localized proteins . Glycoprotein Ibβ and GP IX each contain one palymitoylated cysteine located immediately following a single transmembrane domain . We examined the possibility that the GP Ib-IX-V complex was located in lipid rafts, using the conventional method of sucrose density centrifugation of cold detergent platelet lysates. In resting platelets, approximately 10%–15% of the complex localized to fractions 3, 4, and 5 of 12 equal fractions from the sucrose gradient, a region to which the raft marker GM1 ganglioside also localized . Of interest, the percentage of the total complex that localized to this fraction increased about threefold when platelets were activated by VWF, but not when they were activated by ADP. Activation and increased raft localization of the complex was accompanied by an increase in complex palmitoylation, suggesting that acylation of the complex was driving it into rafts. This presumption must be viewed skeptically, however, given that the cysteines that are most likely palmitoylated are not uniformly conserved in the GP Ibβ and GP IX orthologs from other species.
Raft association of the complex has important functional consequences. Disruption of rafts by depletion of membrane cholesterol markedly blunted platelet aggregation induced by porcine VWF or human VWF/ristocetin . More importantly, cholesterol-depleted platelets perfused over a surface coated with the VWF A1 domain (containing the GP Ibα-binding site) failed to adhere to the surface. Another GP Ib/VWF-dependent function, shear-induced platelet aggregation, was also markedly inhibited by cholesterol depletion.
At least one other important function of the GP Ib-IX-V complex requires that it be localized to rafts: its binding of coagulation FXI. Binding of FXI to platelets facilitates its activation by thrombin and is required for the efficient secondary burst of thrombin generation after the extrinsic pathway is inactivated by tissue factor pathway inhibitor (TFPI). Deficiency of this reaction in Bernard-Soulier syndrome platelets (congenitally deficient in the GP Ib-IX-V complex ) may account for their observed deficiency in prothrombin consumption , and likely contributes to defective hemostasis in these patients. The role of rafts in coagulation is discussed in detail below.
Finally, localization of the GP Ib-IX-V complex to lipid rafts also facilitates its association with key signaling molecules, including the platelet Fc receptor, FcγRIIA. It has long been appreciated that the GP Ib-IX-V complex associates physically and functionally with FcγRIIA , although the reason for the unbalanced and inconstant stoichiometry of the association (the GP Ibα:FcγRIIA ratio varies between about 10:1 and 50:1) suggested that the association was not one typical for membrane receptors. The fact that FcγRIIA only associates with the raft fraction of the complex  may help to explain the peculiar stoichiometry. The true stoichiometry of association has not been determined. Functional association of the two receptors is demonstrated by the fact that antibodies against GP Ib can block platelet aggregation induced by immune complexes (an FcγRIIA-mediated event)  and antibodies against FcγRIIA inhibit shear-induced platelet aggregation, a phenomenon mediated by GP Ib-IX-V .
Co-localization within lipid rafts may also explain the observed association of the GP Ib-IX-V complex and another immunoreceptor tyrosine-based activation motif (ITAM)-containing protein, the FcRγ chain, an association also likely to be involved in signaling subsequent to complex ligation [28,29].
Role of rafts in GP VI functions
Lipid rafts have also been shown to participate in platelet activation after GP VI binds collagen. GP VI is constitutively associated with FcRγ, the signaling subunit of the complex [30–32]. The GP VI/FcRγ complex has been shown to be functionally associated with lipid rafts, although there are contradictory reports as to the nature of this association. Both Wonerow et al.  and Ezumi et al.  demonstrated that GP VI is highly enriched in the lipid raft fractions of resting platelets and that its distribution is largely unaffected by receptor cross-linking. In disagreement, Locke et al.  reported that GP VI was not constitutively raft-associated and that it was recruited to rafts only upon stimulation of platelets or GP VI expressing RBL-2H3 cells. The authors also showed that the raft localization of GP VI is dependent upon its association with FcRγ, as GP VI mutants unable to couple with FcRγ did not localize to lipid rafts following receptor cross-linking.
FcRγ is phosphorylated upon platelet stimulation with collagen; one study found that phosphoylated FcRγ was exclusively located in rafts . In contrast, however, Wonerow et al.  found phosphorylated FcRγ to be equally distributed between the raft and non-raft fractions of stimulated platelets. Despite these discrepancies, however, all reports agree that GP VI-mediated signaling events, including protein tyrosine phosphorylation and platelet activation and secretion, are sensitive to depletion of membrane cholesterol [33–35].
Rafts and other platelet proteins
Other platelet proteins have been shown to localize to rafts, including the scavenger receptor CD36  and the tetraspanin CD63 . In addition, other classes of proteins abundantly represented in platelets have been reported to localize to rafts in other cells including G-protein-coupled receptors .
Not all platelet receptors with adhesive or signaling functions localize to rafts, however. A small fraction of the most abundant platelet receptor, αIIbβ3, is found within the raft fractions on sucrose density centrifugation [21,24] but the relative quantities in raft and non-raft fractions do not change on platelet activation .
Rafts and chilled platelets
Rafts also play a key role in the activation of platelets exposed to cold. Platelet lipids undergo at least two phase transitions when platelets are cooled, one at 30 °C during which sphingolipid-rich regions are entering the lo phase, and the second at approximately 15 °C, when phospholipid-rich regions enter the gel phase [39,40]. Platelet chilling is also accompanied by raft clustering, which is prevented by prior cholesterol depletion . Raft coalescence could be the mechanism through which GP Ib-IX-V complexes cluster when platelets are chilled , a phenomenon that correlates with enhanced clearance of chilled, rewarmed platelets. Cold-induced clustering of the GP Ib-IX-V complex allows the platelets to bind liver Kupffer cells through the phagocyte integrin Mac-1, which apparently is the first step in the phagocytosis of the chilled platelets. The binding of GP Ibα from chilled platelets to Mac-1 is unlike the association of the two receptors that has been previously described. Mac-1 on activated leukocytes uses its I domain to bind platelet GP Ibα ; the binding sites within both the I domain  and GP Ibα (Han and J. A. López, unpublished observation) having been identified. By contrast, the interaction of chilled platelets with Kupffer cells (or their in vitro surrogates, THP-1 cells) involves carbohydrate moieties within the GP Ibα N-terminus that bind the Mac-1 lectin domain . Chilled platelets were reported to bind VWF normally ; however, in our laboratory we recently found that chilled and rewarmed platelets are also hyperreactive with respect to VWF binding, attaching more readily under flow to immobilized VWF and being sensitive to lower concentrations of ristocetin than control platelets (Han and J. A. López, unpublished observation). Clustering and functional alteration of the GP Ib-IX-V complex may thus be a consequence of raft microdomain clustering.
Rafts and the platelet membrane skeleton
The fact that a significant portion of the GP Ib-IX-V complex localizes to rafts may alter interpretation of many earlier experiments describing translocation of signaling molecules to the cytoskeleton. Platelets possess two distinct cytoskeletal fractions, one representing the bulk of the actin filaments and forming an extensively cross-linked structure at the core of the platelet (the actin cytoskeleton), and the other, known as the membrane skeleton, which forms a submembrane lattice of short actin filaments connected to the plasma membrane through cytoskeletal adaptors that bind the cytoplasmic tails of membrane proteins. The GP Ib-IX-V complex functions as one of those membrane anchors, attaching the membrane skeleton to the membrane through the binding of the GP Ibα cytoplasmic tail to filamin A, an actin-binding protein.
In unstimulated platelets, the two cytoskeletal components are loosely connected, and can be separated by differential centrifugation after the platelets are lysed in cold 1% Triton X-100. Both components are insoluble in the detergent, but the actin cytoskeleton can be isolated from the detergent lysate by low speed centrifugation (15 600 g for 5 min), whereas sedimentation of the membrane skeleton requires much higher g forces (100 000 g for 1 h) . When platelets are activated by agonists, the membrane and core cytoskeletons are extensively cross-linked, such that both now sediment at low g forces. Numerous papers have reported the translocation of signaling molecules to the actin cytoskeleton upon agonist stimulation [45–48], based on their association with the low speed pellet in the activated platelets but not in unstimulated platelets. We have found that the conditions used to isolate the membrane skeleton remove much of the rafts from the lysate – not surprisingly, because the conditions to isolate the membrane skeleton resemble those used to isolate detergent-resistant raft fractions, with the exception that the rafts are isolated over sucrose gradients (to obtain the ‘floating’, lipid-rich fraction) and generally for longer centrifugation times. If, after activation, the platelets are treated with N-ethylmaleamide to dissociate GP Ibα from filamin A, many of the proteins reported to ‘translocate’ to the cytoskeleton upon platelet activation are no longer found in the low-speed pellet (A. D. Munday and López, unpublished observation). Thus, it appears that when Triton X-100 lysates of activated platelets are centrifuged at low speeds, not only are proteins directly linked to the cytoskeleton removed from the lysate, but so are proteins associated with GP Ib-IX-V complex-containing rafts. The one caveat that must be added to this interpretation is that N-ethylmaleamide may have unintended effects that may complicate interpretation of these experiments.
Rafts and blood coagulation
Increasingly, evidence is emerging that rafts also play a role in blood coagulation, both in its initiation and maintenance. The major co-factor required for the initiation of coagulation, TF, has been reported to localize to caveolae (considered a specialized form of detergent-resistant microdomain) in endothelial cells where it is inactivated by the primary inhibitor of the TF pathway, TFPI, the principal isoform of which localizes to rafts by virtue of a phosphatidylinositol glycan link . Other data now indicate that raft localization of TFPI may not be necessary for its ability to inhibit TF , although raft integrity is necessary for tonic inhibition of TF, indicating that another component of rafts may participate in ‘encrypting’ TF.
The TF produced by monocytes exposed to inflammatory stimuli such as tumor necrosis factor-α or bacterial lipopolysaccharide (LPS) also localizes to rafts, which facilitate its release into the blood in the form of membrane blebs called microvesicles . Substantial recent evidence indicates that TF-bearing microvesicles participate in the formation of platelet thrombi by binding P-selectin on the platelet surface and incorporating into the developing thrombi where they stimulate the formation of fibrin . That this might be physiologically or pathologically significant was supported by the findings of Hrachovinova et al., who demonstrated that increased concentrations of TF-bearing microvesicles decreased clotting times in mice with hemophilia . Localization to rafts also endows the TF-bearing microvesicles with a receptor that enables them to dock to platelets in developing thrombi, P-selectin glycoprotein ligand-1 (PSGL-1) . Studies of LPS-treated monocytes showed that both TF and PSGL-1 were enriched in raft fractions, and found in the microvesicles from these cells and in plasma microvesicles in concentrations that exceed their concentration on the cells of origin . Consistent with this being a consequence of raft localization, the microvesicles were virtually devoid of the transmembrane phosphatase CD45, which was abundant in the cells of origin but excluded from the raft fractions. The ability of microvesicle-associated TF to initiate coagulation in the developing platelet thrombus is facilitated by the fusion of the microvesicles with the activated platelet, a process that requires an interaction of microvesicle PSGL-1 with platelet P-selectin and the presence of phosphatidylserine on the microvesicle surface (fusion is blocked by preincubating the microvesicles with annexin V) . Incorporation of the microvesicle's TF cargo into the platelet membrane allows all of the reactions of the coagulation cascade to occur on one membrane. Such a mechanism solves the conundrum of how the protease FXa, once formed, can find the platelet membrane on which it participates in the prothrombinase complex. Any alternative mechanism for coagulation initiation in vivo must explain how FXa arrives on the platelet membrane through the sea of protease inhibitors present in the plasma (most notably antithrombin) or how it is able to penetrate the thrombus from a site of production in the blood vessel adventitia . It has also been proposed that fusion of TF-bearing microvesicles with activated platelets may decrypt TF, allowing it to participate in initiating coagulation . Indeed, we have found that the ability of the TF/VIIa complex to cleave FX increases when TF-bearing microvesicles fuse with platelets . Whether this involves movement of the newly acquired TF from rafts has not been determined.
The propagation phase of coagulation, which begins with the conversion of FXI to the active protease by thrombin, also involves reactions that occur on rafts. FXI activation on the platelet surface requires its binding to a small number of high affinity sites on the platelet membrane . It is now appreciated that these sites involve a subset of GP Ib-IX-V complexes on the platelet membrane, specifically GP Ibα within the complex, given that glycocalicin, the isolated ectodomain of GP Ibα, in the presence of high-molecular weight kininogen and zinc ion, can support thrombin activation of FXI as well as one of the most potent non-physiologic co-factors of the reaction, dextran sulfate . On the platelet surface, however, involvement of GP Ibα in FXI activation requires both platelet activation and intact rafts, as it is inhibited by depletion of membrane cholesterol [56,57]. A model has been proposed for FXI activation on the platelet surface whereby thrombin binds GP Ibα through exosite II and FXI through exosite I, with the bound FXI then binding an adjacent GP Ibα molecule through its A3 domain . This model requires that GP Ib-IX-V complexes be clustered sufficiently closely to allow formation of this quaternary complex, and may explain the requirements of rafts for FXI activation.
Rafts may also be involved in the other reactions of the coagulation cascade that eventuate in the production of thrombin. It is well known that both the tenase (FVIIIa and FIXa) and prothrombinase (FVa and FXa) complexes require the externalization of phosphatidylserine on activated platelets to dock and concentrate the reactants so as to efficiently catalyze substrate cleavage. In these homologous reactions, the co-factors (FVIIIa and FVa) bind phosphatidylserine surfaces through their C domains and the proteases (FIXa and FXa) bind through their γ-carboxy glutamic acid-containing GLA domains. Phosphatidylserine movement from the inner leaflet to the outer leaflet of the membrane bilayer requires an increase in the cytoplasmic concentration of calcium ion . The elevation in cytoplasmic calcium involves store-operated Ca2+ entry (SOCE) , which is activated when endoplasmic reticulum calcium stores are depleted . This phenomenon was shown to require raft integrity, as phosphatidylserine exposure on the surface of human erythroleukemia (HEL) cells was markedly blunted by cholesterol depletion. The process also involves the transient receptor potential channel 1, which localizes to rafts where it interacts with calcium signaling proteins .
Externalized phosphatidylserine itself may localize to rafts, as already suggested above by the observation that membrane microvesicles, which are rich in external leaflet phosphatidylserine, appear to arise from raft-rich regions of the plasma membrane. Indeed, in B lymphocytes undergoing apoptosis, the externalized phosphatidylserine has been found to be localized to raft fractions , which are presumably the membrane regions from which apoptotic blebs arise, a situation similar to the genesis of TF-bearing microvesicles.
Another phenomenon that argues for localization of outer leaflet phosphatidylserine to lipid rafts is the budding of enveloped viruses from cells. Such viruses, which include HIV-1  contain a membrane coat rich in outer leaflet phosphatidylserine. That they arise from rafts is strongly suggested by the relative paucity within the envelope of raft-excluded proteins from the cells of origin and by the efficacy of membrane cholesterol depletion in preventing virus budding .
This review should make it clear that specialized microdomains on the surfaces of cell in contact with blood have key roles in hemostasis: both in platelet adhesion and coagulation. It is likely that many more roles for rafts in hemostasis will be discovered. In each of the examples cited here, with the possible exception of tonic inhibition of TF function by raft-localized TFPI, rafts enhance the efficiency of the hemostatic process, and, by extension, the possibility of thrombosis. This raises the possibility that measures that reduce membrane cholesterol may provide relatively mild, but global, antithrombotic benefits. Could a decreased risk for thrombosis be one of the unintended benefits of statin therapy? Recent studies suggest that it is [66,67]. These and other questions about the roles of rafts in hemostasis and thrombosis await answers.