Complement activation by heme as a secondary hit for atypical hemolytic uremic syndrome

Atypical hemolytic uremic syndrome (aHUS) is a rare kidney-predominant thrombotic microangiopathy (TMA) associated with formation of fibrin-platelet clots in the glomerular microvasculature leading to mechanical hemolysis.¹  This disease is related to a dysregulation of the complement alternative pathway (AP), as shown by the identification of genetic or acquired abnormalities in AP regulators or activators in more than 60% of patients.²  aHUS has an incomplete penetrance among mutation carriers, and a triggering event is presumably required for the disease manifestation. This primary hit can be an infection, pregnancy, drug, or other cell-activating event. Sera from aHUS patients carrying mutations deposit complement on endothelial cells activated by inflammatory cytokines,³  suggesting continuous aggression on the endothelium. However, not all infections trigger aHUS in patients. It is frequently the second pregnancy postpartum period that is associated with the disease.⁴  Therefore, other factors appear to be necessary to exceed the threshold of tolerable endothelial stress leading to severe TMA lesions.

In acute phase TMA, mechanical hemolysis induces the release of hemoglobin into the bloodstream.⁵  This hemoglobin is readily oxidized to ferric hemoglobin (methemoglobin) which, in turn, liberates heme. In physiological conditions, cell-free hemoglobin and heme are promptly scavenged by haptoglobin and hemopexin, respectively, which prevent their accumulation and limit their toxicity.⁶  However, at high concentrations of circulating hemoglobin, as in hemolytic diseases, these detoxification systems are overwhelmed,⁷  causing undesirable cell, tissue, and organ injury. The effects of heme exposure on the vascular system have been widely reported and include, in particular, heme iron–induced oxidative stress, vascular inflammation with recruitment of leukocytes, hemolysis,^8-10  and thrombosis.^11,12  Nevertheless, the participation of hemolysis in the development of TMA, as in aHUS, is not well studied.

Heme has been shown to activate the complement AP and to promote the deposition of C3 activation products on erythrocytes, as in malaria.^13,14  Complement AP activation has also been reported during the acute phase of other hemolytic disorders, such as sickle-cell disease,¹⁵  β thalassemia major,¹⁶  and thrombotic thrombocytopenic purpura,¹⁷  but the participation of hemoglobin or its breakdown products in complement activation has not been investigated.

Here, we studied the effect of heme on complement activation and endothelial damage in the context of TMA in aHUS and potentially in other hemolysis-related diseases.

The oxidized form of heme (hemin [ferriprotoporphyrin IX], designated as heme), hematoporphyrin IX, and hemoglobin were from Sigma. Stock solutions of 10 mM hemin and hematoporphyrin in 50 mM NaOH and 145 mM NaCl, stored at 4°C in the dark, were diluted just before use in phosphate-buffered saline (PBS) or M199 medium. All purified complement proteins and complement immunodepleted sera were from Complement Technologies. C3 from Calbiochem and Factor B (FB) from Quidel were also used. Recombinant FB was expressed in HEK293T cells as described.¹⁸  Complement enzyme-linked immunosorbent assay (ELISA) kits were from Quidel and HyCult.

Serum samples were obtained from healthy blood donors or from aHUS patients from the French aHUS cohort.¹⁹  Approval was obtained from the Institut National de la Santé et de la Recherche Médicale (INSERM) institutional review board for these studies. Informed consent was provided according to the Declaration of Helsinki. Samples harboring six different mutations have been tested: three C3 mutations^3,20  (5 patients) and three Factor H (FH) mutations.¹⁹  An AB blood group serum obtained from the French blood transfusion center Etablissement Français du Sang was used as a common standard in all experiments (normal human serum [NHS]).

Primary human umbilical vein endothelial cells (HUVECs) and conditionally immortalized glomerular endothelial cells (GEnCs)^18,21  were cultured as described previously.³  They were incubated with heme and then serum, as described in the Figure 2, 3 and 6 legends. Cells in 24-well plates were detached, labeled, and analyzed by flow cytometry (FACSCalibur). Cells grown in 96-well plates were fixed with paraformaldehyde and analyzed by ELISA. Each condition was assayed in sixplicate wells (3 wells for the ELISA and 3 wells treated with crystal violet and analyzed by colorimetry to normalize the results with cell numbers). For fluorescence microscopy, HUVECs were grown in endothelial growth medium 2 (EGM2; Lonza) on gelatin-coated slides, stimulated, then fixed with paraformaldehyde and labeled as described in the Figure 2C legend. They were then analyzed with a Zeiss LSM700 confocal microscope, and photographs were processed with Image J64 software. Cell-free supernatants, centrifuged twice at 350g to remove residual cells, were assayed for von Willebrand factor (vWF) with a home-made sandwich ELISA.²² 

C3, C3b, C3a, C3d, FH, Factor D (FD), or FB were immobilized on the GLC sensor chips of ProteOn XPR36 equipment according to the manufacturer (BioRad). Six different concentrations of heme (starting with 300 nM) or heme-exposed proteins (C3, C3b, FB, or FH) were injected as analytes.

The C3/C5 convertase formation was followed by consecutive injections of native C3 (80 µg/mL), then FB (50 µg/mL) with FD (0.1 µg/mL), C3 again, and C5 (20 µg/mL) last on heme-exposed immobilized C3 and C3b. Alternatively, C3, C3b, FH, FB, or FB + FD were exposed to heme of different molar excess, diluted in cascade, and injected on a protein-coated chip.

Absorbance spectra of hemin were recorded by using a Unicam Helios b UV-vis spectrophotometer, as described previously.²³  Human C3, C3b, or C3a (Calbiochem; CompTech) was diluted to 500 nM in PBS. Aliquots of heme resulting in final concentrations of 0.01 to 20 μM were added to the optical cell–containing proteins and to a reference optical cell containing PBS only. The absorbance spectra in the wavelength range of 350 to 700 nm were recorded (300 nm/min).

The molecular composition of native and 10-fold molar excess heme-exposed C3 was analyzed by using the fast protein liquid chromatography Akta Purifier 10 system with a 30-mL Superose-6 column equilibrated with PBS. Detection was set at dual wavelengths: 280 nm for proteins and 400 nm for heme.

The atomic coordinates of C3 (PDB 2A73²⁴ ) were used for the molecular docking. HexServer (http://hexserver.loria.fr/) was used to accommodate the heme molecule to C3, as described previously for C1q.²³  Criteria for final C3-heme complex selection were based on the total energy of binding. Visualization was done by PyMol (www.pymol.org).

Heme induced fluid-phase complement activation in NHS, as shown by a dose-related increase of C3a, C5a, and sC5b9 levels, measured by ELISA (Figure 1). Similar levels of C3a, C5a, and C5b9 were measured in NHS, in NHS-EGTA-Mg²⁺, and in C2-depleted serum. The latter two conditions allow AP activation only. No activation was observed in FB-depleted serum, which is deprived of AP activation. Adding back purified FB restored the heme-induced complement activation capacity. No complement activation was observed with an iron-devoid derivative of heme, hematoporphyrin, under the same conditions (data not shown).

To mimic the situation of intravascular hemolysis, human endothelial cells, HUVECs, and GEnCs were pretreated with increasing heme concentrations, washed, and incubated with NHS. Cell-bound C3 fragments (markers of complement activation) were analyzed by flow cytometry after cell detachment or by ELISA and immunofluorescence on adherent cells. As shown in Figure 2A-C, the level of cell-bound C3 was minimal on resting cells when compared with the immunoglobulin G1 isotype control. Exposure to heme resulted in a concentration-dependent increase of C3 deposits on both HUVECs and GEnCs detected by flow cytometry (Figure 2A), ELISA (Figure 2B), and by immunofluorescence microscopy (Figure 2C). Complement deposits were still observed when heme-treated cells were submitted to up to 5 washes before the addition of complement (data not shown), revealing a cell-directed effect of heme. Complement activation by heme-treated cells occurred via the AP, since C3 deposits were persistent in the presence of EGTA-Mg²⁺ and in C2-depleted serum but strikingly diminished in FB-depleted serum (Figure 2D). The addition of purified FB to the FB-depleted serum restored C3 fragment deposition. These effects were observed with both HUVECs (Figure 2D) and GEnCs (data not shown). Importantly, C5b9 complexes were also detected by flow cytometry on heme-treated cells (Figure 2E), reflecting the formation of a C5 convertase. Exposure of endothelial cells to heme up to 200 µM for 1 hour did not induce apoptosis and necrosis as measured by Annexin V and propidium iodide labeling (data not shown).

To test whether complement dysregulation or overactivation in sera induces modification in C3 deposition on heme-exposed endothelial cells, sera from normal controls or from aHUS patients with previously described FH mutations (FH-aHUS)¹⁹  or C3 mutations (C3-aHUS)^3,20  were used. Exacerbation of the C3 deposition was observed by flow cytometry on GEnCs (Figure 3A) and on HUVECs (data not shown) in contact with patients’ sera compared with normal controls (n = 30). C3 deposition on resting cells was enhanced in sera from about half the patients, although on heme-exposed cells, it was significantly increased with sera from all of the patients when compared with 30 normal sera. Immunofluorescence microscopy observations of heme-exposed HUVECs incubated with sera from patients with C3 mutations showed increased C3 deposition, cell retraction, and/or detachment when compared with cells incubated with normal sera (Figure 3B). C3 deposition was similarly augmented on heme-treated HUVECs incubated in conditions mimicking aHUS sera, that is, with FB-depleted serum restored with a recombinant FB with aHUS gain-of-function mutation D254G¹⁸  compared with the wild-type FB (Figure 3C); in the presence of FI-depleted (Figure 3D) or FH-depleted (Figure 3E) sera, as a model for FI and FH deficiency; or in the presence of blocking anti-FH antibody directed against the N- and C-termini of FH, as found in patients with autoimmune aHUS^25,26  (Figure 3F).

To find out whether heme activation of AP was a result of its interaction with a particular complement protein, the binding of heme to C3, FB, FH, and FD was tested by SPR. Strong, specific, and dose-dependent binding to surface-immobilized C3 (hence C3b like C3(H₂O)²⁷ ) and 50% less to FH and FB was detected (Figure 4A). No binding to FD was found. Further, heme was shown to bind the C3a fragment of C3 in contrast to C3b and C3d (Figure 4B). C3(H₂O), C3b, and C3d were functionally active, since they bound FH (data not shown), excluding artifacts due to protein alterations.

To further examine the binding of heme to C3 and its fragments, C3, C3b, and C3a were exposed to heme, and alterations in the specific electronic configuration of heme were then studied by absorption spectroscopy. The absorbance spectrum of oxidized heme showed a ΔA increase in the molar extinction coefficient at the Soret band (around 400 nm) in the presence of C3 and C3a and a 4-nm and 8-nm red shift, respectively (Figure 4C). C3b showed only ΔA with no red shift. The titration of C3a and C3b with increasing concentrations of heme resulted in saturable ΔA (at about two- to sixfold heme excess), although titration of C3 did not reach saturation even at 20-fold molar excess of heme, suggesting C3 oligomer formation (Figure 4D).

To gain further insight into the possible location of the heme binding site(s) in the C3 molecule, molecular docking was performed. HexServer returned 14 top-scored models with total calculated energy of the system between −397.61 kJ/mol and −306.06 kJ/mol (Figure 4E). The first binding site (4 models) was located near the anaphylatoxin (ANA) domain (C3a after cleavage by the C3 convertase). The second site (8 models) was less than 10 Å from the thioester bond (Cys988-Gln991) in the thiol ester-containing domain (TED) at the interface with macroglobulin domain 2 (MG2) and macroglobulin domain 8 (MG8) (Figure 4F). The last site (2 models) is located in close proximity to the second one at the interface between the TED and MG2 domains.

The possibility that heme induces homophilic C3 interactions was assessed by SPR. Heme-exposed C3 injected in the fluid phase was able to bind in a dose-dependent manner to C3 (Figure 4G), but not to C3b (data not shown) immobilized on the chip. Heme-exposed C3b on the chip did not interact with either C3b or C3 (data not shown). C3 on the chip, exposed to increasing heme concentrations, bound native C3 injected in the fluid phase (first step of the SPR profile in Figure 5A) in a heme dose-dependent manner.

To assess the stability of heme-induced C3-C3 complexes, the molecular composition of C3 after exposure to heme was analyzed by size exclusion chromatography. A fraction of heme-exposed C3 was eluted earlier than native C3, corresponding to homophilic C3 interactions and formation of dimers and higher-order oligomers (Figure 4H). These oligomers showed stable heme binding, since some heme, detected by its specific absorbance at 400 nm, co-eluted with the oligomeric protein detected by the 280-nm absorbance. Free heme has a low molecular weight and is retained by the column.

To explain the observed heme-induced AP activation and in view of the heme-induced homophilic C3 interaction described earlier, the effect of heme on the C3 convertase formation was investigated by SPR (Figure 5). C3 (Figure 5A) and C3b (Figure 5B) were immobilized on the sensor chip and exposed to increasing concentrations of heme. As in Figure 4, a dose-dependent binding of heme to C3, but not to C3b, was detected. A fixed amount of C3 was then injected, which bound strongly to heme-exposed C3 but only minimally to C3b. Further fixed concentrations of FB and FD were flowed onto the chip, to form the C3 convertase. A dose-dependent increase of the signal was detected on heme-exposed C3-coated flow cells but there was little or no effect on C3b-coated flow cells. Subsequent injections of fixed concentrations of C3 followed by C5 resulted again in a dose-dependent increase of binding to the heme-exposed C3 surface. Since a slight heme binding to FB was detected (as shown in Figure 4A), the effect of heme on the ability of FB to interact with C3 and to form a C3 convertase was tested. No difference in the interaction of FB with C3 and C3b in the presence of FD was detected after FB exposure to up to 10-fold molar excess of heme (Figure 5C-D).

Since heme binding to FH was detected in Figure 4A, the influence of heme on the binding of FH to its C3 and C3b ligands was studied. Heme-exposed C3 bound FH more strongly compared with native C3, reaching more than a threefold increase at 10-fold molar excess of heme (Figure 6A). In contrast, heme-exposed C3b bound FH similarly to native C3b binding (data not shown). The increased C3-FH binding was a consequence of C3 exposure to heme, since FH exposure to up to 10-fold molar excess of heme did not alter its interaction with C3 and C3b (Figure 6B).

Heme treatment of HUVECs resulted in a dose-dependent increase of FH binding to the cell surface upon incubation with NHS (Figure 6C). Of note, no FH was detected on HUVECs in the absence of serum by using a mixture of three anti-FH monoclonal antibodies (mAbs) (data not shown). In order to test whether FH bound to heme-treated cells was functionally active and able to control C3 deposition, cells were incubated with PBS, with C3-depleted serum (as a source of FH without C3 deposition), or with FH-depleted serum (allowing C3 deposition with no FH), followed by a second step of NHS incubation (Figure 6D). Preincubation with FH contained in the C3-depleted serum resulted in a 40% lower C3 deposition than the control preincubated in PBS. By contrast, preincubation with FH-depleted serum resulted in 20% higher C3 deposition than with the NHS control. Moreover, inhibition of FH-mediated control by cell preincubation with an FH 19-20 construct able to interact with the cell membrane and to compete with FH²⁸  resulted in a dose-dependent decrease of FH binding and concomitant increase of C3 deposition on the cell surface (Figure 6E).

Heme-induced complement activation on endothelial cells could be the result of either defective control or local overactivation or both. To study the role of complement regulators, expression levels of CD59, decay-accelerating factor (DAF), and membrane cofactor protein (MCP) were measured on resting or heme-treated HUVECs by flow cytometry. As shown in Figure 7A, heme treatment decreased the membrane expression of DAF and MCP in a dose-dependent manner, with no modification of CD59 levels. Heme at 50 µM resulted in about 50% loss of MCP and DAF when compared with untreated cells.

P-selectin has been described as a C3b-binding protein, capable of initiating complement activation.²⁹  Thus, we analyzed the membrane expression of P-selectin on endothelial cells and observed a time-dependent increase of this expression during heme treatment (Figure 7B). The rapid expression, in less than 30 minutes, excluded a de novo synthesis but suggested a mobilization of Weibel-Palade bodies. Indeed, a secretion of vWF was observed, as measured by ELISA, in parallel with P-selectin membrane expression (Figure 7C).

We report that heme activates complement in the fluid phase and on endothelial cells and induces a prothrombotic phenotype by a set of five probably co-existing mechanisms (summarized in Figure 7D). Our data allow us to propose a novel hypothesis in which hemolysis would serve as a secondary hit for aHUS manifestation. A triggering event preceding the acute phase of aHUS may lead to formation of an initial thrombus responsible for erythrocyte rupture and heme release. We suggest that heme amplifies complement activation and creates a positive feedback loop that sustains and perpetuates the histologic TMA lesions. Upon exceeding a threshold, this heme-induced complement overactivation becomes a second hit for the manifestation of a full-blown aHUS. This may shed new light on aHUS pathogenesis and explain, at least in part, the incomplete penetrance of the disease, which would develop only after such an amplification of primary pathogenic events. First, we found that the presence of heme in NHS is sufficient to trigger fluid phase complement AP activation with release of anaphylotoxins C3a and C5a, as well as sC5b9, in agreement with previous reports.^13,30  Heme has a dual role in complement, activating AP but inhibiting the binding of C1q to its ligands.²³  We investigated the mechanism of AP complement activation and showed that it results from the interaction of heme with the C3 molecule, which favors C3 homophilic interactions and leads to the formation of an active AP C3/C5 convertase. Molecular docking revealed that one of the most likely heme-binding sites on C3 is adjacent to the thioester bond. The insertion of the redox active heme at this site may favor the thioester bond hydrolysis and the transition of native C3 to hydrolyzed C3*. Importantly, hydrolyzed C3* is able to bind FB and form a C3 convertase, contrary to native C3. Therefore, heme-induced generation of C3* could explain the fluid phase AP activation in heme-treated serum (Figure 7D, step 1). Heme is known to mediate protein covalent cross-linking and oligomer formation³¹  and indeed induces the formation of C3/C3 complexes. Those are particularly critical for the formation of a fluid phase C5 convertase, which requires C3b/C3b covalent linkage for an efficient affinity for C5.³²  Iron-devoid heme derivative hematoporphyrin did not activate complement, strongly suggesting that iron redox activity is involved.⁵  This is reminiscent of previous data showing that H₂O₂ and neutrophil-derived oxidants trigger the complement AP and may also activate C5 directly.³³ 

Circulating free heme concentrations are about 6 to 30 μM in sickle-cell disease and thalassemia⁷  but have not been reported in aHUS. Mechanical hemolysis in the renal microvasculature in aHUS probably results in local heme concentration, at the site of hemolysis and before dilution in the blood flow, much higher than concentrations measured in the peripheral blood. The concentration-dependent effects of heme on complement and endothelium could, at least in part, explain why vascular lesions are mainly restricted to microvessels in TMA. The 100 μM heme concentration used in these in vitro experiments, which corresponds to the destruction of less than 0.05% of bloodstream erythrocytes,¹³  thus seems physiologically reasonable.

This work is the first demonstration that free heme triggers complement AP activation on endothelial cells. Although little or no C3 deposits were observed on resting cells, they reached significant levels on heme-treated HUVECs and GEnCs. Moreover, cell-bound C5b9 was detected, reflecting the formation of a C5 convertase and the release of C5 fragments C5a and C5b9 known to activate endothelial cells and to promote endothelium permeability.³⁴ 

Imbalance between complement activation and regulation may explain the observed AP activation on endothelial cells (Figure 7D, step 2). Indeed, a decrease of DAF (CD55) and MCP (CD46) expression was detected after exposure of HUVECs to heme, similar to previous observations of downregulated complement regulators on retinal epithelial cells submitted to brief oxidative stress.³⁵  DAF and MCP are responsible for the protection of endothelium against autologous complement attack. Their downregulation has been reported on apoptotic and necrotic cells,³⁶  but we did not detect significant increase of apoptosis with heme under the experimental conditions used.

In contrast to the decreased expression of MCP and DAF, exposure of endothelial cells to heme resulted in enhancement of FH binding to the cell membrane. FH thus appears to compensate for the loss of membrane-bound complement inhibitors DAF and MCP, as has been described on apoptotic cells.³⁷  Heme-treated cells would then be particularly sensitive to complement attack in the presence of defective FH, as is found in more than 30% of aHUS patients presenting with FH mutations or anti-FH autoantibodies.^1,19,38  Indeed, levels of C3 deposits on heme-treated cells increased in FH-depleted serum or in the presence of the FH SCR 19-20 fragment known to compete with FH²⁸  or with anti-FH blocking antibodies.

FH enhanced binding to heme-treated cells could result from FH interaction with oxidized epitopes on the heme-exposed membrane³⁹  or with heme itself, bound to the lipid bilayer.^40,41  Indeed, a direct binding of heme to FH was detected by SPR in this study. Heme-exposed C3 also bound more strongly to FH compared with native C3. Heme-exposed FH interacted normally with C3 and C3b and, once bound to the cell membrane, it maintained its functional activity.

Complement activation on endothelial cells could also result from the expression of P-selectin (Figure 7D, step 3). Indeed, P-selectin has been identified as a C3b-binding protein, and on cells expressing P-selectin, this expression alone was sufficient to activate the complement system marked by C3b deposition, C3a generation, and C5b9 formation.²⁹  We observed that heme treatment of endothelial cells induced membrane expression of P-selectin concomitantly with the secretion of vWF. This probably results from iron-induced oxidative stress, since oxygen radicals were shown to induce exocytosis of Weibel-Palade bodies and membrane P-selectin.⁴²  The exposure of P-selectin could thus focus the AP activation on the surface of heme-challenged endothelial cells. Of note, the main function of P-selectin is the recruitment of leukocytes expressing its ligand PSGL-1, and heme treatment could thus favor inflammation at the surface of endothelial cells. vWF secretion could recruit platelets and contribute to a prothrombotic state of heme-exposed endothelial cells. The overall effects of heme reported here are reminiscent of those of Shiga toxin, a primary trigger of HUS, which induces P-selectin expression on endothelial cells, resulting in C3 binding and alternative pathway activation together with a vWF-mediated thrombus formation.⁴³ 

Heme also binds directly to the cell membrane phospholipids,^40,44  where it could activate complement as in the fluid phase (Figure 7D, step 4). In addition, the endothelial cell retraction, which occurs after treatment with heme⁴⁵  and exposes the complement activating subendothelial matrix⁴⁶  may lead to heme-dependent C3 deposition on the endothelium (Figure 7D, step 5). We indeed observed a preferential localization of C3 deposits along intercellular junctions.

In HUS, we hypothesize that multiple potential triggers such as viral infection, pregnancy, or drugs lead to initial microthrombi formation. Erythrocyte fractionation on these microthrombi would result locally in microvessels in concentrations of heme sufficient to promote the release of vWF and the activation of complement and the release of C5a and C5b9, thus amplifying the initial thrombosis and inflammation. This amplification would normally be limited by the homeostatic control of complement and by the cleavage of vWF by the 13th member of a disintegrin-like and metalloprotease with thrombospondin type 1 motif, 13 (ADAMTS-13). Genetic or antibody-induced alterations of complement proteins in aHUS or of ADAMTS-13 in HUS would synergize with hemolysis, thus decreasing the threshold of the initial triggers to overcome homeostatic control and trigger an explosive endothelial injury and thrombosis. Moreover, heme-induced complement activation and endothelial injury may not be restricted to HUS and could contribute to the pathogenesis of thrombotic events observed in hemolytic episodes of different etiologies.

The authors thank Prof P. Mathieson and Dr S. Satchel (Bristol, United Kingdom) for providing the GEnC cell line, Prof Paul Morgan (Cardiff, United Kingdom) for his gift of the anti-C5b9 mAb, Dr Sakari Jokiranta (Helsinki, Finland) for giving us the FH SCR 19-20 construct, Dr Julie Rayes for her help with the vWF analysis, and Prof P. Lesavre (Paris, France) for the fruitful discussions. Part of the cytometric analysis was done at the Centre d’Imagerie Cellulaire et de Cytomètrie, Centre de Recherche des Cordeliers UMRS 872 (Paris, France). Centre d'Imagerie Cellulaire et de Cytomètrie is a member of the Université Pierre et Marie Curie Flow Cytometry network.

This work was supported by grants from Agence Nationale de la Recherche (ANR Genopath 2009-2012 09geno03101I), Assistance Publique-Hôpitaux de Paris (Programme Hospitalier de Recherche Clinique (AOM08198), European EU FP7 grant 2012-305608 (EURenOmics Consortium), by Association pour l'Information et la Recherche sur les maladies Rénales Génétiques France, and by INSERM.

Contribution: L.T.R., L.H.-M., J.D.D., and V.F.-B. designed the project and initiated the research; M.F., L.T.R., L.H.-M., and J.D.D. designed and performed the experiments, did the analysis, and prepared the figures; F.T. and C.P. performed experiments; M.F., L.T.R., L.H.-M., J.D.D., and V.F.-B. wrote the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Lubka T. Roumenina, Cordeliers Research Center, INSERM UMRS 872, team 13; 15 rue de l’Ecole de Medecine, entrance E, floor 3; 75006 Paris, France; e-mail: lubka.roumenina@crc.jussieu.fr.