Murine systemic thrombophilia and hemolytic uremic syndrome from a factor H point mutation

Complement is an important innate immune system that protects the host from infection.¹  However, if not properly regulated, activated complement has the potential to cause significant tissue injury. In humans and mammals, a number of cell surface and plasma proteins function to inhibit complement attack of host cells, and dysfunction in these proteins has been linked to complement-mediated diseases.²  Among the major complement regulators, factor H (FH) is an abundant plasma glycoprotein that inhibits alternative pathway (AP) complement activation.^3,4  It regulates complement activation, both in plasma and on the cell surface, by accelerating decay of the AP C3 convertase C3bBb and by acting as a cofactor for factor I–mediated cleavage and inactivation of C3b.⁵  FH is composed of 20 short consensus repeat (SCR) domains. Its complement-regulating function resides in 4 N-terminal SCRs. The other SCRs, especially SCR 19 and 20 at the C terminus, are involved in binding host cells through interaction with surface-deposited C3b/C3d and host cell–specific constituents on the cell surface.^6-9 

Recent data from genetic studies in animals and humans have linked FH dysfunction to several complement-mediated pathologies. Mutations in the FH gene leading to absent or low-level FH protein in plasma impaired fluid phase and cell surface control of complement activation, and this caused secondary complement deficiency due to C3 and factor B consumption.^10,11  Rare heterozygous mutations in human FH, mostly in SCR 19 and 20, are associated with atypical hemolytic uremic syndrome (aHUS), a prototype of thrombotic microangiopathy (TMA) that causes renal failure.^12-14  A single-nucleotide polymorphism in the human FH gene (Y402H) in SCR 7 significantly increased the risk of age-related macular degeneration (AMD).^15-18 

In the present study, we introduced a single amino acid mutation (W1206R) to mouse FH by gene targeting in order to better understand the physiological role of FH as a cell surface complement regulator. The W1206R change in mouse FH corresponded to W1183R mutation in human FH found in multiple aHUS families.^19,20  We found that the W1206R mutation in the mouse FH impaired its interaction with host cells but did not affect its plasma complement-regulating activity. Homozygous mutant mice carrying this mutation developed renal TMA as expected. Surprisingly, the mutant mice also developed systemic thrombophilia involving large blood vessels in multiple organs, including liver, lung, spleen, and kidney. Approximately 30% of mutant mice displayed symptoms of stroke and ischemic retinopathy, and approximately half died by 30 weeks of age. Mutant mice showed endothelial and vascular smooth muscle cell dysfunction with significantly elevated plasma levels of von Willebrand factor (VWF) and a lower ratio of high- to low-molecular-weight VWF multimers. Genetic deficiency of complement C3 and factor D prevented both the systemic thrombophilia and renal TMA phenotypes. These results demonstrate a causal relationship between complement dysregulation and systemic angiopathy and suggest that complement activation may contribute to various human thrombotic disorders involving both the micro- and macrovasculature.

Recombinant mouse FH SCR 19-20 was produced in HEK cells as described previously.²¹  Production of full-length mouse FH protein followed a similar procedure, and experimental details are described in supplemental Methods (available at the Blood Web site).

All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Pennsylvania. Mutant mice were either generated using C57BL/6J embryonic stem cells or backcrossed onto C57BL/6J background. The source and generation of FD^−/−, C3^−/−, and DAF^−/−Crry^−/−C3^−/− mice have been reported previously.^21-23  The W1206R mutant mouse was generated by homologous recombination in embryonic stem cells following procedures previously described,¹¹  and more detailed are provided in supplemental Methods. Wild-type (WT), heterozygous, and mutant mice used for comparative studies were all littermates from heterozygous breeding.

The survival curve was analyzed using GraphPad Prism (GraphPad, La Jolla, CA) as described previously.¹¹ 

A lipopolysaccharide (LPS)–dependent enzyme-linked immunosorbent assay (ELISA) was used to measure total serum AP complement activity as published previously²⁴  and described in detail in supplemental Methods.

Hemolytic activity of WT or W1206R mutant mouse serum was assessed as reported previously¹¹  using red blood cells (RBCs) from DAF^−/−Crry^−/−C3^−/− mice and 50% test serum and is described in detail in supplemental Methods.

Serum creatinine (Cr) was measured using a CRE-III plus kit (Kainos, Tokyo, Japan). Blood urea nitrogen (BUN) was measured using serum and urea nitrogen reagents (Sigma-Aldrich, St Louis, MO) as described previously.²⁵  Complete blood count (CBC) was measured using an automated hematology analyzer (XT-2000iV, Sysmex) at the Translational Core Laboratory of the Children’s Hospital of Philadelphia.

Washed mouse blood platelets were prepared as described previously.²⁶  Platelets (0.5 mL, adjusted to 2 × 10⁸ cells/mL) were incubated at 37°C with buffer or AYPGYK (250 μM), 2 MeSADP (50 nM), or collagen (5 mg/mL). Light transmission was monitored with a PAP4 light scattering platelet aggregometer (BioData Corporation, Horsham, PA) for 3 minutes.

All data collection, including pathology scoring, was performed in a blinded manner. Mantel-Haenszel log-rank test, Student t test or 1-way analysis of variance (ANOVA) followed by Bonferroni’s post hoc test were used for statistical analysis as appropriate, and a P value of .05 was considered to be significant.

Additional experimental details are provided in supplemental Methods.

The W1183R mutation in SCR 20 of human FH is found in multiple aHUS patients.^19,27,28  Based on sequence alignment, we identified W1206 as the murine equivalent of W1183 in human FH (Figure 1A). Previous studies using a recombinant protein corresponding to human FH SCR 19-20 showed that W1183R mutation increased its binding to C3b and heparin and abolished its interaction with RBCs.^29,30  We used similar assays to test functional changes in the W1206R mutation in mouse FH. We expressed and purified recombinant mouse FH SCR 19-20 and W1206R variant in HEK cells (Figure 1B). We found that W1206R mutant exhibited an increased binding of mouse FH SCR 19-20 to plate-coated heparin (Figure 1C). At the lower concentration tested, it also showed increased binding to plate-coated C3b (Figure 1D). In a hemolytic assay, WT mouse FH SCR 19-20 caused lysis of complement-susceptible RBCs from DAF^−/−Crry^−/−C3^−/− mice (Figure 1E) because it lacked the N-terminal complement-regulating SCRs and acted as a competitive inhibitor of natural FH in RBC binding. In contrast, W1206R mutant of FH SCR 19-20 failed to induce RBC lysis in the same assay (Figure 1E), suggesting that it lost the ability to interact with RBCs. We also expressed and studied full-length mouse FH and W1206R mutant (Figure 1F-G). Compared with WT mouse FH, W1206R mutant had a larger retention volume on a heparin column, suggesting that full-length FH carrying this mutation also had increased binding to heparin. Finally, we assessed FH function and confirmed that W1206R mutation did not affect the cofactor activity of murine FH in factor I–mediated cleavage of C3b (Figure 1H).

To assess the in vivo consequence of W1206R mutation in murine FH, we generated by gene targeting a mutant mouse (supplemental Figure 1). Western blot and ELISA analysis of plasma from WT (FH^W/W), heterozygous (FH^W/R), and homozygous (FH^R/R) mice showed that mutant FH protein was stable and present at a higher level (115% by densitometry estimation of western blot) in FH^R/R mice than normal FH in FH^W/W mice (Figure 2A and supplemental Figure 2). In contrast, we detected lower levels of plasma intact C3, factor B (FB), and C5 in FH^R/R mice (50%, 52%, and 86% of the FH^W/W mouse level for C3 α-chain, FB, and C5, respectively, by densitometry or ELISA) (Figure 2B-D). Confirming the reduction in C3 and FB levels, plasma AP complement activity, as assessed by an LPS-based ELISA,²⁴  was significantly lower in FH^R/R mice (Figure 2E). Despite the reduction in C3, FB, and C5 levels, FH^R/R but not FH^W/W or FH^W/R mouse serum caused marked lysis of complement-sensitive RBCs from DAF^−/−Crry^−/−C3^−/− mice (Figure 2F). Thus, while W1206R mutation in murine FH caused a reduction in plasma complement levels, it led to increased complement activity on the cell surface due to impaired interaction of FH with host cells, a finding that is consistent with data from in vitro experiments using recombinant proteins (Figure 1).

At weaning (3-4 weeks), most FH^R/R mice were noticeably smaller and thinner compared with their FH^W/W and FH^W/R littermates. This deficiency in body weight persisted in a group of mice that we examined from 4 to 18 weeks of age (Figure 2G). We also followed a large cohort of FH^W/W, FH^W/R, and FH^R/R mice up to 30 weeks of age for survival. Of 147 FH^R/R mice genotyped, 48% died by 30 weeks, whereas none of the 275 FH^W/W and 310 FH^W/R mice died during this time period (Figure 2H). To investigate the cause of death, we studied FH^R/R mice for signs of aHUS, which is characterized by thrombocytopenia, anemia, and kidney failure. A CBC revealed severe thrombocytopenia and anemia in FH^R/R mice at 4 and 8 weeks of age (Figure 2I-J) but a normal white blood cell count (supplemental Figure 3A). Consistent with low hemoglobin (Hb) levels, we found significantly increased reticulocyte count in FH^R/R mice (supplemental Figure 3B), likely the result of compensatory hematopoiesis. These results suggested that thrombocytopenia and anemia were caused by peripheral blood destruction rather than bone marrow production failure. We measured BUN and serum Cr levels as markers of kidney function at 4, 12, and 20 weeks of age. BUN in FH^R/R mice was progressively increased with age (38 and 116 mg/dL at 4 and 20 weeks, respectively), and values were significantly higher than that of FH^W/W and FH^W/R mice within each age group (Figure 2K). Although Cr was also higher in FH^R/R mice, it varied considerably among different FH^R/R mice (Figure 2L).

Periodic acid–Schiff (PAS) staining of kidney sections showed mesangial expansion and double contours of glomerular basement membrane with capillary microthrombi and endothelial swelling with narrowing of the capillary lumen in FH^R/R mouse glomeruli (Figure 3A). Hyalinosis and thrombosis were also detected in interlobular arteries and arterioles of FH^R/R mice (Figure 3A). We analyzed 30 FH^R/R mice (1-30 weeks of age) and found that all contained aHUS-like pathological changes in the kidney, with 53%, 68%, 52%, and 78% of the glomeruli showing expanded matrix and endothelial swelling, microthrombi, large vein thrombi, and arteriolar hyalinosis, respectively (supplemental Tables 1 and 2). These pathological changes were not observed in age-matched FH^W/W and FH^W/R mice (Figure 3A; supplemental Table 1). Immunofluorescent staining of C3 and fibrin, markers of complement activation and vascular injury, respectively, showed positive glomerular deposition of both, with weak C3 staining showing a granular pattern in the mesangium and strong fibrin staining in capillary lesions (Figure 3B and supplemental Figure 4). Electron microscopy analysis revealed subendothelial expansion with fluffy granular material without immune-type deposits and mesangial interposition with basement membrane duplication (Figure 3C; supplemental Table 3A). These findings suggested that FH^R/R mice developed severe aHUS that might have contributed to premature death.

We found that ∼30% of FH^R/R mice exhibited severe neurological abnormalities, including hindlimb paralysis and circular movement (supplemental Video 1). Histological examination of these mice consistently showed ischemic brain injury and the presence of thrombi and/or intracerebral hemorrhage in the brain (Figure 3D). These findings suggested that FH^R/R mice were prone to developing thrombotic and hemorrhagic strokes, which promoted us to further examine signs of thrombosis and vascular injury in other organs. By histology, we found thrombi in medium and large blood vessels in all organs examined, including liver, lung, spleen, and kidney (Figure 3E; supplemental Table 2). Most of these thrombi were found in veins (eg, portal veins of the liver, splenic veins, and interlobular and renal veins of the kidney) and pulmonary arteries (Figure 3E). Portal vein thrombosis was particularly prevalent and was present in the liver of >80% of FH^R/R mice on histology examination. Using ultrasonography, we were also able to detect a large clot (2.9 × 1.3 mm, sagittal view) in the portal vein of a 20-week-old male FH^R/R mouse (supplemental Video 2).

We assessed the correlations among stroke and ischemic brain injury, renal pathology, and organ thrombosis in 14 FH^R/R mice with symptoms of stroke and 11 FH^R/R mice without. As expected, we detected ischemic brain injury only in mice with symptoms of stroke (supplemental Table 2). On the other hand, TMA-like renal pathology was detected in all mice regardless of the neurological phenotype. We found that mice with symptoms of stroke were more likely to have multiorgan thrombosis, but portal vein thrombosis appeared to be common and was less correlated with stroke symptoms (supplemental Table 2).

To understand the chronology of thrombus formation and renal pathology development, we examined by histology of embryonic day 15.0 fetuses and 1-week-old FH^R/R mice. Fetuses were collected from heterozygous breeders and sectioned sagittally as a whole, whereas liver and kidney were dissected from postnatal 1-week-old FH^R/R mice for histology. We found no thrombus formation in any of the fetal organs, nor were any pathological features detected in the fetal kidney (supplemental Figure 5A; supplemental Table 3C). On the other hand, we detected mesangial expansion in glomeruli in all 3 of the 1-week-old FH^R/R mice examined and kidney thrombi in 1 and liver thrombi in 2 of the 3 mice (supplemental Figure 5B; supplemental Table 3C).

In addition to thrombosis in internal organs, we observed blood vessel occlusion in the eyes of 70% of FH^R/R mice (n = 21, 5-10 weeks old) (supplemental Table 3B). Using in vivo retinal photography, we found signs of ischemic injury, including retinal edema, whitening, and cotton wool spots in most of the mutant mice examined (Figure 4A-D; supplemental Table 3B). Using fluorescein angiography, we observed dropout of retinal vasculature, central and branch retinal artery occlusions, and perivascular leakage (Figure 4A-D; supplemental Video 3). In a fluorescein angiography test, the time required for the dye to fill the retinal vasculature in FH^R/R mice was significantly increased (45 vs 300 seconds for FH^W/W and FH^R/R mice, respectively, Figure 4F). Approximately 9% of FH^R/R mice (2 out of 23 mice examined by funduscopy, 5-10 weeks old) also had retinal detachments (Figure 4E), most likely resulting from exudation secondary to infarctions in the choroidal vasculature. These were visible grossly in some mice as leukocoria (white pupil).

To test the hypothesis that vascular injury in FH^R/R mice contributed to thrombotic or organ dysfunctional phenotypes, we used myography to measure endothelial cell–dependent and independent relaxations of small mesenteric artery fragments from FH^W/W and FH^R/R mice. Administration of acetylcholine, an agonist of endothelium-dependent relaxation pathways mediated by nitric oxide, prostaglandin I₂, and endothelium-derived hyperpolarizing factor, demonstrated a clear difference between FH^W/W and FH^R/R mice (Figure 5A). The FH^R/R mouse blood vessel fragment had reduced dilation in response to acetylcholine stimulation. This difference disappeared with the addition of nitro-L-arginine methyl ester or combinations of nitro-L-arginine methyl ester, indomethacin, and tetraethylammonium, respective inhibitors for endothelial nitric oxide synthase in the nitric oxide pathway, cyclooxygenase in the prostaglandin I₂ pathway, and potassium channel in the endothelium-derived hyperpolarizing factor pathway (Figure 5A). Administration of sodium nitroprusside (SNP), an endothelium-independent vasodilator that can directly act on vascular smooth muscle cells, showed no significant difference in blood vessel relaxation at low or high doses, but a detectible difference was observed at medium doses. These data suggested that endothelium-dependent and independent relaxations of smooth muscle cells were impaired in FH^R/R mice. Although we have not characterized blood pressure changes in the mutant mice, many aHUS patients are known to develop hypertension,^31,32  which can be caused by thrombotic microangiopathy, including impairment in smooth muscle cell function.

Thrombotic thrombocytopenic purpura (TTP) is another prototype of TMA that shares many clinical features with aHUS but is primarily caused by severe deficiency of ADAMTS13, a plasma metalloprotease that cleaves von Willebrand factor (VWF).^33-36  Under normal conditions, ultralarge VWF multimers are secreted from Weibel-Palade bodies and cleaved into small multimers by ADAMTS13. Severe deficiency of plasma ADAMTS13 activity results in accumulation of ultralarge VWF multimers on endothelial cell surface or in circulation.³⁷  To confirm that the pathogenic mechanism in FH^R/R mice is distinct from that of TTP, we measured ADAMTS13 activity, VWF antigen concentrations, and VWF multimer distribution in FH^R/R mice and FH^W/W controls. Contrary to TTP conditions in humans and in mice, we detected increased plasma ADAMTS13 activity (Figure 5B), increased plasma VWF antigen levels, and a lower ratio of high- to low-molecular-weight VWF multimers in FH^R/R mice (Figure 5B-D).

In an attempt to understand the thrombotic phenotype, we used intravital microscopy to examine FeCl₃-induced thrombus formation in mesenteric arteries and veins of FH^W/W and FH^R/R mice.^38,39  This experiment was conducted under relatively gentle conditions in order to maximize the chance of picking up a prothrombotic phenotype in the mutant mice. We also tested platelet sensitivity to agonist stimulation in an aggregation assay and by fluorescence-activated cell sorting analysis of 2 activation markers, αIIbβ3 and P-selectin. Interestingly, we did not find FH^R/R mice with preexisting thrombophilia and aHUS phenotypes to be more sensitive than FH^W/W mice to FeCl₃-induced thrombus formation under the experimental setting used, nor did we observe platelet hypersensitivity to agonist stimulation in these mice (supplemental Figure 6).

Since intact C3 and FB levels were decreased and plasma AP complement activity was lower in FH^R/R mice, we tested the hypothesis that dysregulation of AP complement on the cell surface was responsible for disease development in FH^R/R mice. We crossed FH^R/R mice with C3-knockout (C3^−/−) and factor D–knockout (FD^−/−) mice and generated FH^R/RC3^−/− and FH^R/RFD^−/− mice. We found that both FH^R/RC3^−/− and FH^R/RFD^−/− mice lived normally, with none developing stroke symptoms, whereas their FH^R/R littermates showed the neurological phenotype and high mortality rate observed earlier (Figure 6A). Likewise, while FH^R/R littermates increased their BUN from 5 to 10 weeks of age, the levels of BUN in FH^R/RC3^−/− and FH^R/RFD^−/− mice were normal and remained unchanged with age (Figure 6B). Furthermore, unlike their FH^R/R littermates, FH^R/RC3^−/− and FH^R/RFD^−/− mice had normal platelet counts and Hb levels (Figure 6C), and displayed no evidence of renal pathology, nor was thrombus formation detected in liver and other organs (Figure 6D; supplemental Table 4). Finally, funduscopy showed completely normal-looking retina, and fluorescein angiography revealed normal retinal blood flow in FH^R/RC3^−/− and FH^R/RFD^−/− mice (Figure 6D; supplemental Table 4). Thus, stroke, thrombophilia, retinopathy, and aHUS in FH^R/R mice were all mediated by AP complement.

We have shown in this study that a single amino acid mutation in SCR 20 of mouse FH led to AP complement dysregulation on the cell surface and produced an unexpectedly severe phenotype of renal TMA as well as systemic thrombophilia involving large blood vessels. Mutant mice failed to thrive, had lower body weights, and developed multiorgan thrombi, stroke, retinopathy, and progressive renal disease. Our in vitro experiments demonstrated that the W1206R mutation altered mouse FH binding to heparin and C3b, and impaired its interaction with RBCs but did not affect the cofactor activity for factor I–mediated cleavage of C3b. Accordingly, the mutant FH is expected to have retained its fluid-phase AP complement-regulating activity. Thus, unlike mice with FH knockout or hypomorphic mutation^10,11,40  in which fluid-phase AP complement is uncontrollably activated and consumed, complement dysregulation is expected to have occurred only on the cell surface in FH^R/R mice, and cell surface complement activation and consumption likely accounted for the observed lower plasma C3, FB, and C5 levels. The mechanism for slightly elevated plasma FH in the mutant mice remains to be determined. The W1206R mutation may have increased FH messenger RNA and/or protein stability and half-lives. Additionally, lower tissue binding and retention of mutant FH due to its impaired interaction with host cells may have led to a higher plasma level.

The extent of thrombophilia in FH^R/R mice was notable, and a direct causal relationship was found between complement dysregulation and thrombotic angiopathy. Apart from renal failure, occlusion of vital organs such as the brain, lung, and heart likely contributed to the premature mortality of FH^R/R mice. Indeed, we observed sudden death of FH^R/R mice of different ages on numerous occasions. The onset of organ thrombosis and renal injury appeared to be fast, occurring within the first week of life, but not during fetal stage in a setting where the mother was heterozygous for the W1206R mutation and thus had functional FH in maternal blood which could have crossed the placenta.^41,42  An alternative explanation for the lack of phenotype in fetuses is that AP complement was not yet fully functional. The precise mechanisms for the observed macrovascular thrombophilia phenotype remain to be determined but several hypotheses may be tested in future studies. In the absence of FH protection on the cell surface, terminal complement activation could have damaged vascular endothelial and smooth muscle cells, creating a prothrombotic state of the vessel wall. Platelets and neutrophils or monocytes could also have been activated by membrane attack complex and C5a, respectively, leading to the release of tissue factor from the stimulated leukocytes,^43,44  platelet and leukocyte aggregation, and clot formation both in capillaries and in large blood vessels.^45,46  Notably, however, when tested in vitro and in vivo, platelet function in mutant mice with systemic thrombophilia and aHUS disease appeared to be impaired rather than increased. The thrombocytopenia phenotype suggested that platelets were being consumed, shortening their survival in the circulation. Given that, the in vivo results of FeCl₃-induced experimental clotting study likely reflected both the thrombocytopenia and an as-yet-unexplained defect in platelet function found in the in vitro studies.

TTP is a form of TMA that is caused by genetic or acquired ADAMTS13 deficiency leading to formation of ultralarge VWF multimers.³⁷  Many studies have explored a possible mechanistic overlap between TTP and aHUS.^47-49  Complement dysregulation may occur in TTP patients and exacerbate the TMA phenotype,^50-52  and several recent studies have also demonstrated an interaction between FH and VWF.^53,54  It is clear, however, that the pathogenic mechanism of FH^R/R mice is distinct from that of TTP. Instead of lower ADAMTS13 activity, we detected higher ADAMTS13 activity in FH^R/R mice, possibly as a feedback response to increased VWF release from damaged endothelium and activated platelets. Thus, despite higher plasma VWF levels, the relative ratio of ultralarge to smaller VWF multimers was lower in FH^R/R mice. Supporting a complement-dependent mechanism, we showed that AP complement deficiency by either C3 or FD deletion completely prevented disease manifestations in FH^R/R mice.

In addition to its implication in TMA, FH dysfunction has also been linked to the pathogenesis of dry AMD. A common single-nucleotide polymorphism encoding a Y402H change in SCR 7 of human FH significantly increases AMD risk.^15-18  A rare mutation, R1210C, in SCR 20 of human FH originally found in some aHUS patients,^55,56  has been reported to be a high penetrant mutation for dry AMD,^57-59  but there is no known connection between retinopathy and the W1183R mutation in human FH. By funduscopy, we found severe retinal injury in a high percentage of FH^R/R mice. We have not characterized the specific pathology of retinopathy in these mice, and whether retinal injury is secondary to vascular occlusion only or there is also direct complement-mediated attack on retinal cells remains to be investigated.

In summary, by creating a mouse with a FH point mutation that impaired FH activity on cell surface only, we have revealed a surprising role of FH in preventing macrovascular thrombosis as well as TMA and renal disease Our data suggest that dysfunction or variation of FH activity on the cell surface could be implicated in various human thrombotic disorders such as stroke and retinal arterial or venous occlusion, in addition to primary or secondary TMA. The FH^R/R mouse will be a useful model to study the interaction between complement and thrombotic pathways and to test novel anti-complement therapies for related human diseases.

The authors are grateful for the chimeric mouse production and electron microscopy services provided by the Transgenic and Chimeric Mouse Facility and the Electron Microscopy Core of the Perelman School of Medicine, University of Pennsylvania, respectively, and for the mouse CBC analysis provided by the Clinical and Translational Research Center of the Children’s Hospital of Philadelphia.

This work is supported by National Institutes of Health, National Institute of Allergy and Immunological Diseases grants AI1174190, AI44970, AI085596, and National Institutes of Health, National Eye Institute grant EY023709.

Contribution: W.-C.S., T.M., and Y.U. designed experiments and interpreted data; Y.U., I.M., D.S., D.G., L.Z., Z.C., S.G., J.B., Y.M., and T.M. performed experiments; M.P. performed pathological analysis; S.S. and Y.W. contributed reagent generation and mouse colony maintenance; W.-C.S. and Y.U. prepared figures and wrote the manuscript; and H.W., X.L.Z., L.B., M.P., and J.D. interpreted data and critically edited the manuscript.

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

Correspondence: Wen-Chao Song, Department of Systems Pharmacology and Translational Therapeutics, Perelman School of Medicine, University of Pennsylvania, Room 1254 BRBII/III, 421 Curie Blvd, Philadelphia, PA 19104; e-mail: songwe@mail.med.upenn.edu.