Polyphosphate suppresses complement via the terminal pathway

Inorganic polyphosphate (polyP) is a linear polymer of orthophosphate, linked by phosphoanhydride bonds.^1-3  It is found in all mammalian cells and lower organisms, localized in lysosomes, dense granules, mitochondria, and nuclei. The polymer varies in length from cell to cell and in different organisms, ranging from 60 to 100 units in human platelets to up to thousands of phosphate units in some bacteria.^4-6  In platelets, polyP is localized in dense granules⁷  and released on activation, whereupon it is found in platelet-rich thrombi at concentrations of 1 to 3 μM.⁸  At physiologic pH, each internal unit has a monovalent negative charge, and thus the polymers are highly anionic. This property led to the finding that polyP is a physiologic surface on which factor XII, prekallikrein, and high molecular weight kininogen assemble for contact activation of coagulation.⁹  Subsequent studies confirmed that polyP is prothrombotic and proinflammatory in in vivo mouse models^8,10  and that there are multiple steps in the coagulation cascade at which polyP acts to achieve this end.^8,11-17  The effects of polyP on coagulation are concentration- and size-dependent.¹⁴  Thus, platelet-sized polyP (P_60-100) primarily accelerates thrombin-mediated activation of factor XI and factor V, whereas larger-size polyP triggers coagulation via contact activation of factor XII and enhances fibrin polymerization.

The observation that polyP modulates coagulation raised the question as to whether polyP also regulates the evolutionarily related blood-borne proteolytic cascade complement. The complement system comprises more than 30 soluble and membrane-bound proteins, contributing to innate and adaptive immunity and aiding in disposal of danger-associated molecular patterns (for reviews, see Morgan¹⁸  and Ricklin and Lambris¹⁹ ). Complement activation is achieved via 3 pathways: the lectin pathway, classical pathway, and alternative pathway. These pathways converge with C3 convertase-mediated transformation of C3 into C3a and C3b. The C3a anaphylatoxin recruits leukocytes and activates platelets.²⁰  C3b deposition on bacteria promotes phagocytosis by leukocytes and is required for formation of the C5 convertase that cleaves C5 into C5a and C5b. C5b rapidly binds to C6, forming a tight C5b,6 complex, which then binds to C7, yielding the C5b-7 complex. This attaches to the outer leaflet of a target membrane. The subsequent addition of the heterotrimeric C8αβγ stabilizes and anchors the now C5b-8 complex to the cell by inducing a conformational change in C8 and burying a hydrophobic tail through the lipid bilayer. Multiple C9 subunits finally join for assembly of the C5b-9 pore-like, lytic membrane attack complex (MAC).²¹ 

Coagulation and complement are coordinately regulated to limit blood loss, eliminate pathogens and damaged cells, and promote healing. Although often viewed as distinct, they are highly integrated, with several recently identified cross-talk pathways.^22-30  For example, thrombin activates C5 and enhances formation of the MAC.³¹  Conversely, C5a promotes expression of tissue factor and plays a role in the antiphospholipid antibody-associated fetal loss syndome.³²  Several other biochemical connections have been documented.^29,33,34 

In spite of evidence of links between coagulation and complement, the possibility that polyP regulates complement in mammalian systems is unexplored. In a mutant Neisseria meningitidis, lack of the polyphosphatase that normally degrades polyP renders the bacteria resistant to complement-mediated death, indicating that polyP facilitates evasion from complement-mediated killing.³⁵  We tested the hypothesis that polyP provides a bridge between complement and coagulation in the human system.³⁶  We show that polyP significantly suppresses complement via the terminal pathway (TP) and that this occurs in a concentration-dependent, size-dependent, and ion chelation-independent manner. PolyP binds to and destabilizes C5b,6, interfering with optimal assembly and membrane binding of the MAC. The findings reveal a novel mechanism by which complement is regulated and underline the complex relationship between coagulation and complement.

Chemicals were purchased from Sigma Aldrich (St. Louis, MO) unless otherwise stated. PolyP preparations were synthesized, size-fractionated, and quantified by the malachite green assay,¹⁴  with concentrations expressed in phosphate monomers (monomer formula: NaPO₃). The number of phosphate units making up polyP is indicated in the subscript. Gelatin veronal buffer (pH 7.4), normal human serum (NHS), and human plasma-derived C5b,6, C7, C8, and C9 were from Complement Technology Inc. (Tyler, TX). Rabbit erythrocytes (rRBCs) were freshly obtained from citrated venous blood after venipuncture performed by Animal Care Services personnel at the University of British Columbia (Vancouver, BC, Canada), with approval by the University of British Columbia Animal Ethics Committee. Chicken erythrocytes (cRBCs) were from Colorado Serum Company (Denver, CO). Hemolytic assays were performed in 96-well nontreated microplates from Corning (Amsterdam, The Netherlands). Sypro Orange was from Invitrogen (Burlington, ON, Canada).

Erythrocyte lysis assays^18,37  were used to measure complement-mediated hemolytic activity (supplemental Data, available on the Blood Web site). Lysis was first determined relative to the control of 100% lysis with H₂O. For total hemolytic and TP assays, concentrations of NHS and C5b,6 to initiate complement activation were established from pilot studies to obtain 70% to 80% erythrocyte lysis at 30 minutes (supplemental Figures 1-3). Results were normalized to this amount of lysis and shown in figures as “relative RBC lysis.” Results are representative of experiments performed a minimum of 3 times each.

Native polyacrylamide gel electrophoresis (PAGE), analytic gel filtration, and differential scanning fluorimetry (DSF; also referred to as thermal shift assay)³⁸  were used to assess polyP-protein interactions. Methods for the first 2 are described in supplemental Data. For DSF, 25 µL reaction mixes consisting of 0.4 mg/mL protein (0.14 mg/mL for C5b,6), varying concentrations of polyP diluted in Hepes buffered saline (20 mM N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid and 150 mM NaCl at pH 7.4), and 5× Sypro Orange were loaded into MicroAmp Fast Optical 96-well Reaction Plates (Applied Biosystems, Carlsbad, CA). Samples were heat denatured in an Applied Biosystems StepOnePlus Real-Time PCR System, using a ramp configuration starting at 25°C and increasing at 1°C min⁻¹ to 95°C. Fluorescence (in relative fluorescence units) was measured every 30 seconds.³⁸  Data for each curve were normalized to the maximum and the minimum of the curve.

Analyses were performed with GraphPad Prism version 5.0 (San Diego, CA). Where indicated, 1-way analysis of variance with Bonferroni’s multiple comparison tests were performed. Results shown are means ± standard error of the mean. Statistical significance refers to P < .05.

We first assessed the effects of polyP on total complement-mediated hemolytic activity. NHS diluted to yield 70% to 80% lysis was preincubated with varying concentrations of polyP that include more than 1000 orthophosphate units (NaPO₃) per chain (polyP_>1000) and that promote coagulation.¹⁴  Orthophosphate (P₁) was used as a control. Reactions were initiated by the addition of rRBCs, and hemolysis was measured after 30 minutes (Figure 1). P₁ and polyP_>1000 suppressed hemolysis in a concentration-dependent manner. However, at equivalent molar concentrations of the monomeric form, polyP_>1000 was strikingly more effective at suppressing hemolysis. The concentration required to achieve half maximal inhibition of lysis (IC₅₀) was ∼3.5 mM for P₁ and ∼350 µM for polyP_>1000 (Figure 1). Because complement activation via the alternative and classical pathways is dependent in part on the presence of ionic calcium and magnesium, it is likely that the observed suppression of total hemolytic activity by monophosphate concentrations exceeding ∼5 mM was induced primarily by chelation of those ions. However, chelation would not account for the enhanced activity of polyP compared with P₁, as the ionic strengths of the polyP and P₁ at equivalent monomeric concentrations are similar. Thus, polyP suppresses total hemolytic activity.

We examined the effect of polyP on the TP of complement. The TP is initiated by rapid binding of C5b to C6 and the subsequent and sequential binding of C7, C8, and several C9 molecules to form the C5b-9 MAC. This pathway is ion-independent, and thus any effects of polyP would also be independent of its capacity to chelate cations.

We first studied the effects of polyP on the TP of complement in NHS. TP activity was initiated by adding a limiting amount of exogenous purified C5b,6 to achieve 70% to 80% lysis after 30 minutes. A molar excess of EDTA was used to prevent upstream activation of complement and generation of endogenous C5b,6. In the absence of C5b,6, there was no detectable complement-mediated cRBC lysis (data not shown). P₁ at concentrations ranging from 0.1 to 1000 µM had no effect on complement-mediated hemolysis in NHS via the TP (Figure 2A). In contrast, polyP_>1000 inhibited the TP in NHS in a concentration-dependent manner, with an IC₅₀ of ∼10 µM polyP_>1000 (Figure 2A). To verify that the effects observed were dependent on the integrity of the polymers, we pretreated polyP with CIAP, which cleaves polyP into monomeric units.³⁹  CIAP alone had no effect on the TP, whereas CIAP treatment of the polyP_>1000 completely abrogated its ability to dampen hemolysis via the TP (Figure 2B). In addition, coincubation of up to 2 mM P₁ in combination with polyP_>1000 at a concentration of 200 µM had no effect on the suppressive properties of polyP_>1000 (Figure 2C).

The effects of polyP on coagulation are dependent on chain length. We therefore tested the effect of different chain lengths of polyP on TP hemolytic activity. Medium chain-length polyP (polyP_40-160) had a similar concentration-dependent suppressive effect as polyP_>1000 on TP-mediated hemolysis (Figure 3A). However, shorter-length polyP (polyP_<30) was less potent, and an IC₅₀ could not be achieved even with 5 mM polyP_<30. A wider range of different polyP chain lengths was examined (Figure 3B). At equivalent 100 µM concentrations, P₁, diphosphate (Na₄P₂O₇) (P₂), and triphosphate (Na₅P₃O₁₀) (P₃) had no effect on TP hemolytic activity. However, polyP with a mean length of 22 orthophosphate units (polyP₂₂) significantly dampened TP hemolytic activity, and the extent of suppression by longer-chain polyP increased in a size-dependent manner. PolyP released from thrombin-stimulated platelets also suppressed TP complement activity (not shown).

We next examined whether polyP suppresses TP hemolytic activity through direct interactions with TP components (C5b,6, C7, C8, or C9) or whether other serum factors are required. Purified TP components C7, C8, and C9 were therefore used as the source of complement, rather than NHS. TP hemolytic activity was measured by sequentially adding cRBCs, varying concentrations of polyP, and then C5b,6, C7, C8, and C9. Similar to findings with NHS, polyP dose-dependently suppressed lysis, whereas P₁ had no effect (Figure 4A). The IC₅₀ of polyP_>1000 in this system was ∼1 µM. The effect of polyP was not specific to the species of the target erythrocytes, as TP hemolytic activity was also suppressed by polyP when human erythrocytes were used (data not shown). Overall, the findings indicate that polyP interferes with TP complement-mediated lysis of RBCs by binding directly to 1 or more of the TP complement components or to the target red cell membrane.

We delineated the steps in the TP at which polyP interferes with the function of the MAC to lyse erythrocytes. PolyP was added at different steps of the TP assay (ie, before C5b,6, after C5b,6, after C7, after C8, or with C9; Figure 4B). PolyP_>1000 at a concentration of 200 µM suppressed lysis to less than 5% of maximal lysis (ie, without polyP) when added before or immediately after C5b,6. In contrast, polyP had no effect on cRBC lysis in the TP assay when added after C7, after C8, or with C9. Equivalent results were obtained with polyP_60-100 (data not shown). Thus, polyP interferes with optimal MAC function/assembly by destabilizing C5b,6, limiting normal C5b,6 interaction with C7, or causing further downstream TP complexes (C5b-7, C5b-8, or C5b-9) to be unstable or incapable of attaching or inserting into the RBC membrane. Once C5b-7 forms, however, polyP can no longer modulate MAC function.

To determine whether polyP interferes with incorporation of the terminal complex onto or into the membrane of cRBCs, we added varying concentrations of polyP to the cells, followed by a 5-minute incubation with equimolar concentrations of purified C5b,6, C7, and C8. Under these conditions, the C5b-8 complex integrates into the membrane, causing minimal lysis. Cells were pelleted, and the amount of unbound C5b, reflecting unbound C5b-8, was assessed by Western immunoblot under reducing conditions (supplemental Figure 4A). More C5b was recovered in the supernatant when polyP was included, and this effect was dose-dependent. In a similar manner, by incubating C5b,6 and C7 (without C8 or C9), the presence of polyP caused reduced binding of the C5b-7 complex to the erythrocyte membrane (supplemental Figure 4B). The findings indicate that polyP interferes with binding/integration of the C5b-7 and C5b-8 complexes to/into the cell membrane.

We investigated whether polyP interacts directly with different components of the TP and whether it alters the stability and, thus, the function of these components. In the first approach, native PAGE was used to assess polyP–protein binding. Platelet-derived polyP binds to thrombin (distribution coefficient, ∼5 nM), but not to prothrombin¹³ ; findings that are evident on native gels (Figure 5). Similar to thrombin’s interaction with polyP, we showed that C5b,6, C6, and C7 all exhibited band shifts in the presence of polyP consistent with a physical interaction under these experimental conditions.

In a second approach, we performed gel filtration studies with either C5b,6 or C7 in the presence or absence of polyP_>1000. PolyP had no effect on the chromatogram for C7 (Figure 6A). This was somewhat surprising, given the gel shift data, but is likely attributable to different buffer conditions. Notably, coincubation of polyP caused a shift to a higher oligomerized state in the C5b,6 elution profile (Figure 6B), indicating a direct interaction of C5b,6 with polyP. The broader C5b,6-polyP peak is consistent with polyP destabilizing C5b,6. In the absence of C6, C5b is highly unstable and forms aggregates,⁴⁰  and thus C5b was not independently examined.

In the third approach, DSF was used to determine the thermal stability of the complement proteins in the presence and absence of polyP. This technique involves heat-denaturing proteins and exposing internal hydrophobic regions to the aqueous environment, detected by the fluorescent dye Sypro Orange.³⁸  Interactions between a protein and a binding partner often change the thermal stability of the protein, observed as a shift in the melting (denaturation) curve. In the absence of polyP, the melting curve of C5b,6 (Figure 7A) exhibits 3 transition phases (arrows), likely as a result of the differential thermal stabilities between C5b,6 domains. When C5b,6 was incubated with polyP_>1000, a dose-dependent leftward shift in the second transition phase at 59.5°C was observed (expanded in Figure 7B), evidence that polyP binds to a distinct domain on C5b,6. PolyP had no effect on the melting curve of C5 but caused a dose-dependent destabilization shift in the C6 melting curve (Figure 7C-D, respectively), suggesting that polyP interacts with C5b,6 via C6. In spite of the gel shift showing an interaction between polyP and C7, only a minimal polyP-induced shift in the DSF C7 curve was observed, and only in the presence of a high concentration (1 mM) of polyP (Figure 7E). In a similar manner, although thrombin binds to polyP,¹³  we also did not detect a polyP-induced change in thrombin’s thermal stability measured by DSF (supplemental Figure 5). PolyP_60-100 had the same effect as polyP_>1000 for all proteins tested.

We have shown that polyP dampens complement via the TP, that this occurs in a concentration- and polymer size-dependent manner, and that the mechanisms involve, at least in part, destabilization of C5b,6, thereby interfering with optimal lytic function of the C5b-9 MAC. This is the first account of a complement-inhibitory function for polyP in a eukaryotic system. Previously, Zhang et al³⁵  used mutant forms of N meningitidis, which lack the exopolyphosphatase that cleaves polyP, and showed that these bacteria are resistant to complement-mediated killing. Moreover, exogenous polyP protected wild-type bacteria from complement-mediated death. Thus, in these bacteria, polyP is a weapon of survival, used to overcome host innate immunity.⁴¹  Our findings are in line with theirs, supporting the notion that polyP in humans may dampen innate immune responses via suppression of complement.

Previous studies established that polyP initiates activation of coagulation and inflammation via the contact pathway, accelerates generation of factor Va by factor Xa and thrombin, and enhances the stability of the fibrin clot.^9,16  The biochemical studies were validated in vivo in mice, where injections of polyP caused a thrombotic and proinflammatory response, the latter mediated by increased generation of factor XIIa, kallikrein, and bradykinin.^8,10  That polyP can inhibit complement and limit the innate immune response appears at odds with its well-characterized procoagulant and proinflammatory properties. How can this be explained?

It is not unprecedented in the coagulation system for a molecule to have dual and functionally opposing activities that are delicately balanced to maintain homeostasis. For example, thrombin, through dynamic allosteric changes and interactions with cofactors, receptors, and substrates, elicits procoagulant and anticoagulant properties that are orchestrated to ensure rapid and localized responses to injury and bleeding while preventing unrestrained thrombosis (reviewed in DiCera⁴²  and Lechtenberg et al.⁴³  Thrombin is also promiscuous in its role in regulating inflammation and innate immunity. It induces proinflammatory cytokine release,^44-46  is chemotactic, and activates complement C5.^31,47  Conversely, thrombin enhances vascular endothelial barrier protection⁴⁸  and generates activated thrombin activatable fibrinolysis inhibitor, which, in turn, inactivates proinflammatory mediators bradykinin, osteopontin, C3a, and C5a.⁴⁹  It therefore is not surprising that polyP has diverse functions. PolyP in the plasma⁵⁰  or released from cells may provide local protection to host cells (eg, platelets, endothelial cells, leukocytes) while inducing proinflammatory/procoagulant effects at other sites. These properties may be differentially exhibited on the basis of the pathophysiologic situation, the cellular source of the polyP, the presence of ions in the target pathways, the extracellular concentration of the polyP, and/or the size of the polymer at the site of action.

Polymer size is indeed important in the function of polyP in the coagulation system, where it dictates the specific steps in the cascade at which it optimally performs.¹⁴  More extensive study of the role of polyP in the multiple complement pathways may similarly reveal size-dependent differential regulation. Our studies suggest that polymer size may distinguish the effects of polyP in coagulation and complement. PolyP preparations with lengths less than 30 orthophosphate units had no effect on plasma clotting time,¹⁴  whereas polyP_>1000 shortens the clotting time at concentrations of 5 to 20 μM (supplemental Figure 6). In vivo studies that revealed the prothrombotic and proinflammatory properties of polyP were performed, using polymers with chain lengths that exceeded 45 orthophosphate units,^8,10  and thus, it is not known whether shorter-length polymers would have the same effect. The shortest-length polymers used in our in vitro studies were polyP₂₂ and polyP₃₁, sizes that would be expected to have minimal effects on coagulation. Notably, these could still suppress the TP. Because we have not yet evaluated the role of polyP in vivo in animal models of human disease associated with excess complement activation, we can only speculate that using shorter forms of polyP might selectively reduce activation of complement through the TP. It is also possible that lower concentrations than what have been used in vivo^8,10  might spare the prothrombotic and proinflammatory effects of polyP and favor complement inhibition and dampening of inflammation. These studies, using a range of sizes and concentrations of polyP in different disease models, will be important to perform, as therapies targeting polyP might be tailored to independently modulate coagulation and complement to meet patients' needs.

Platelets play a central role in maintaining hemostasis and promoting thrombus growth and are also critical in immune surveillance (reviewed in Rondina et al⁵¹ ). In addition to being a source of polyP, they express several complement receptors, store complement regulatory and activating factors, and provide a surface for generation of C3a, C5a, and MAC assembly.^20,52,53  Although platelets participate in the innate immune response by promoting complement activation, this is tightly regulated by platelet-derived negative regulators of complement, such as factor H, clusterin, CD55, and CD59. Indeed, the platelet must protect itself from excess activation and complement-mediated destruction. This is accomplished in part by the release of the major fluid phase-negative regulator, factor H.⁵⁴  PolyP, released from activated platelets, is believed to participate in the initiation and activation of coagulation but might also modulate the local inflammatory response by suppressing complement activation. In humans, low platelet levels of polyP are found in patients with dense granule storage pool diseases,^55,56  such as Hermansky Pudlak Syndrome⁵⁷  and Chediak Higashi Disease.⁵⁸  Patients with these disorders exhibit a bleeding diathesis but, interestingly, often suffer from largely unexplained recurrent, life-threatening infections. These may be a result of leukocyte defects. Alternatively, the immune defects may be partly caused by alterations in complement that could be partially attributed to reduced polyP release. It will be interesting to evaluate complement activation in these patients.

We explored the mechanisms by which polyP suppresses complement activation. As an anionic polymer, polyP chelates divalent metal ions.⁵⁹  Initiation and activation of complement until formation of the C5 convertases is dependent on Mg²⁺ and Ca²⁺. In our studies using the total hemolytic activity assay, increasing concentrations of monophosphate (P₁) suppressed cell lysis, but only at molar concentrations equivalent to or greater than the concentrations of ionic calcium and magnesium in the assay system and required for complement activation. PolyP_>1000 at equivalent molar concentrations was much more potent in suppressing complement-mediated cell lysis, suggesting that chelation of calcium and magnesium ions was not the only mechanism. Although we subsequently focused on examining the effect of polyP on the TP, which is ion-independent, preliminary findings from our laboratory indicate that polyP binds to C1-esterase inhibitor, factor B, and factor H, but not to factor D, factor I, or C3b (supplemental Figure 7). Further studies will uncover the relevance of these findings with respect to regulating complement via 1 or more of the complement activation pathways.

The TP of complement is, in effect, a point of no return that, without intervention by negative regulators, results in formation of the MAC. Three major negative regulators of the TP that prevent cell lysis have been described, all of which variably bind to TP components to prevent formation and integration of the MAC. CD59 is a glycosylphosphatidylinositol-linked protein that binds to C8 and C9 and prevents C9 from polymerizing.⁶⁰  CD59 deficiencies are associated with paroxysmal nocturnal hemoglobinuria.⁶¹  Clusterin binds to C7, C8, and C9, inducing a conformational change that reduces the capacity of the C5b-9 complex to integrate into the membrane.^62,63  Vitronectin binds to C5b-7 in the fluid phase (but not to C5b,6), allowing formation of a soluble MAC that cannot bind to the membrane.⁶⁴  PolyP is a fourth negative regulator of the TP that also binds to TP components to reduce cell lysis.

Several lines of physicochemical evidence led to the conclusion that polyP binds to and destabilizes the earliest component of the TP, such that formation of a functional MAC is reduced. Electromobility shift assays in native gels showed that polyP directly interacts with several complement components. These findings, which do not necessarily indicate a functional relationship, prompted us to use other approaches to establish the mechanisms. In gel filtration studies, polyP binds directly to C5b,6, whereas DSF clearly shows that polyP destabilizes C5b,6 in a concentration-dependent manner. By sequentially adding the TP components to the cRBCs in the TP hemolytic assay, we determined that polyP significantly suppresses lysis when added before or after C5b,6, but not after C7, and that polyP interferes with C5b-7 and C5b-8 anchoring to the membrane. Further structure function studies will be necessary to identify the molecular structures involved and whether there are additional mechanisms.

In summary, we have uncovered a novel mechanism by which complement is negatively regulated. Future studies will focus on delineating the in vivo relevance, using mouse models of human disease and exploring how polyP might regulate other steps in the complement system. The interplay among polyP, complement, and coagulation is important to understand, as alterations in their cross-talk may manifest as disease and diagnostic and therapeutic insights may be gained.

J.M.W. is supported in part by a scholarship from the Canadian Institutes of Health Research. J.H.F. is supported by a Banting Fellowship. E.M.C. is supported by operating grants from the Canadian Institutes of Health Research, the Natural Sciences and Engineering Research Council of Canada, and the Canada Foundations for Innovation. He holds a CSL Behring Research Chair and a Tier 1 Canada Research Chair in Endothelial Cell Biology, is an adjunct scientist with the Canadian Blood Services, and is a member of the University of British Columbia Life Sciences Institute. N.C.S. is supported by operating grants from the Canadian Institutes of Health Research and holds a Tier 1 Canada Research Chair in Antibiotic Discovery and a Senior International Research Scholar award from the Howard Hughes Medical Institute. This work was supported in part by National Institutes of Health Grant R01 HL47014 (to J.H.M.).

Contribution: E.M.C. directed all research and wrote the paper; J.M.W. designed and performed the experiments and helped write the paper; J.H.F. and M.J.K. helped design the studies, established conditions for hemolytic assays, and helped write the paper; V.L. and L.M.O. performed hemolytic assays and gel filtration; G.A.W., E.L., and N.C.S. helped perform and analyze DSF and gel filtration; and S.A.S. and J.H.M. prepared polyP and helped in designing experiments and preparing the paper.

Conflict-of-interest disclosure: J.H.M. and S.A.S. are coinvestigators on patents pertaining to medical uses of polyphosphate. E.M.C., J.M.W., M.J.K., J.H.F., and L.M.O. are coinvestigators on a patent pertaining to medical uses of polyphosphate. The remaining authors declare no competing financial interests.

Correspondence: Edward M. Conway, Centre for Blood Research, 4306-2350 Health Sciences Mall, University of British Columbia, Vancouver, BC V6T 1Z3, Canada; e-mail: ed.conway@ubc.ca.