Glucagon-Like Peptide 1 Receptor Activation Attenuates Platelet Aggregation and Thrombosis

Type 2 diabetes (T2D) is associated with a number of risk factors that contribute to an increased risk of atherothrombotic events, including hypertension, dyslipidemia, obesity, and chronic inflammation, as well as endothelial and platelet dysfunction (1). Platelets are small, versatile, anucleate cells in the circulation that play critical roles in both early and late stages of atherothrombosis, contributing also to cell-based thrombin generation and blood coagulation (2). Subjects with T2D exhibit a prothrombotic state, including increased production of coagulation factors; decreased production of fibrinolytic factors; and a propensity to platelet activation, aggregation, and adhesion (1,3,4). Compounding the latter, subjects with T2D show reduced sensitivity to antiplatelet drugs, such as aspirin and clopidogrel (5,6), and manifest a higher incidence of cardiovascular events (1,6,7).

Although the currently available antidiabetic agents have been effective at lowering blood glucose levels and preventing microvascular disease, until the recent EMPA-REG study (8), it had been exceedingly difficult to demonstrate the beneficial effects of normalizing blood glucose levels in major adverse cardiovascular events (i.e., macrovascular events). Indeed, in some studies, aggressive glucose lowering has been associated with an increase in the incidence of cardiovascular events, including death (9). In this context, significant interest has been generated by incretin-targeted therapies, which include glucagon-like peptide 1 (GLP-1) receptor (GLP-1R) agonists and dipeptidyl peptidase-4 inhibitors (DPP-4is), both of which are under active investigation in long-term cardiovascular safety studies. The first of those studies examining DPP-4is, namely the SAVOR-TIMI 53, EXAMINE, and TECOS studies, have demonstrated no increase in cardiovascular events (10–12). By contrast, relatively short-term studies (13–16), in arguably lower-risk patients treated with GLP-1R agonists and DPP-4i, have suggested beneficial effects on cardiovascular event rates. Because the rather modest and brief improvements in glycemic control observed in these studies are unlikely to explain the observed benefits, we wondered whether GLP-1–targeted therapies may reduce macrovascular event rates in these particular patient populations by directly inhibiting platelet aggregation and thrombus formation. Indeed, this premise could be extrapolated from the one small uncontrolled study (17) that demonstrated the curious ability of DPP-4i sitagliptin to inhibit thrombin-induced platelet aggregation in vitro.

Here we show that a human megakaryocyte cell line expresses a functional GLP-1R, and that treatment with the GLP-1R agonist exenatide inhibits human and mouse platelet aggregation in vitro, thrombus formation ex vivo, and mouse arterial thrombosis in vivo. Given the importance of nitric oxide (NO) signaling, and more specifically type III endothelial NO synthase (eNOS)–derived NO, to platelet biology (18,19), and the known abilities of GLP-1 and exenatide to activate eNOS in human endothelial cells (20,21), we here also define a role for eNOS in exenatide-inhibited mouse arterial thrombosis in vivo. Together, these findings provide potential mechanisms for observed improvements in cardiovascular outcomes in patients with T2D who were treated with GLP-1R agonists, and a potential rationale for their selective use in subjects with diabetes who are at particular risk of atherothrombosis.

cDNA was synthesized from MEG-01 DNAse-treated total RNA. Primers designed to span the 5′ and 3′ untranslated regions (UTRs) of the human GLP-1R mRNA were used to amplify a 1.5-kb full-length GLP-1R cDNA using Phusion High-Fidelity PCR Master Mix (Thermo Scientific Finnzymes Molecular Biology Solutions; ThermoFisher Scientific, Ottawa, Ontario, Canada). The amplicon was gel eluted using a QIAquick Gel Extraction kit (catalog #28704; Qiagen, Toronto, Ontario, Canada) and cloned into the expression vector pEF-IRES-puro 6. The resulting clone was sequenced to confirm the orientation of the insert within the expression vector as well as to validate that the cloned insert contained full-length GLP-1R cDNA.

MIN6 cells and MEG-01 cells were washed with Hanks’ buffered salt solution and resuspended in binding buffer (as described previously) (22,23) with 100 nmol/L fluorescent probe (EP40-BF) or 100 nmol/L unlabeled exenatide for 1 h in the dark. Cells were washed 2× with Hanks’ buffered salt solution and resuspended in PBS with 2% BSA. Flow cytometry was performed using a MACSQuant Analyzer (Miltenyi Biotec, San Diego, CA), and analysis was performed using FCS Express 4 software.

Total RNA was extracted from MEG-01 cells, human skin fibroblast (catalog #CC-2511; Lonza, Mississauga, Ontario, Canada), and human lung tissue using TRIzol (Invitrogen). Total RNA was then reverse transcribed into cDNA. Commercially available pancreatic cDNA was used as a positive control for GLP-1R gene expression (catalog #HD-313; Zyagen, San Diego, CA). Previously published GLP-1R primers spanning the boundaries of exons 9–10 and 10–11 were used to quantify GLP-1R expression via quantitative RT-PCR (qRT-PCR) (24). The final primer concentrations used were 0.4 μmol/L with 60°C annealing temperature for all primers. All samples were run in triplicate on a Roche LightCycler 480 System, and GLP-1R gene expression was normalized to the housekeeping gene TATA binding protein.

A cAMP Enzyme Immunoassay Kit (catalog #581001; Cayman Chemicals, Ann Arbor, MI) was used to measure changes in intracellular cAMP induced by GLP-1. MEG-01 cells were cultured to a concentration of 2 × 10⁵/mL and transferred to a 24-well plate (BD Falcon, San Jose, CA) with 500 μL/well culture medium. 3-Isobutyl-1-methylxanthine was added to each well at a concentration of 0.5 μmol/L and was allowed to incubate for 10 min at 37°C to inhibit phosphodiesterase activity. GLP-1(7–36), in concentrations of 0.1, 1, 10, and 100 nmol/L, was added to the wells and incubated for 15 min at 37°C. As a positive control for cAMP generation in MEG-01 cells, prostacyclin (PGI₂) was added at a concentration of 0.02 ng/mL to control wells and was incubated for 10 min at 37°C. PBS was used as a negative control. After incubation, cells were lysed by adding EDTA to a concentration of 10 mmol/L. Lysed cells were transferred to 1.5 mL Eppendorf tubes and boiled for 5 min at 95°C. Samples were spun down at 10,000g for 15 min at 4°C. Supernatant was removed and frozen at −80°C. The cAMP Enzyme Immunoassay Kit assay was then performed according to manufacturer instructions.

Aggregation of isolated mouse and human platelets in platelet-rich plasma or gel-filtered platelet preparations was assessed at 37°C using a computerized Chrono-Log aggregometer (Chrono-Log Corporation, Havertown, PA), as previously described (25,26). Two hundred fifty microliters of a 2–3 × 10⁸/mL platelet solution was added to a cylindrical glass cuvette (Chrono-Log Corporation) with a siliconized magnetic stir bar (Chrono-Log Corporation). The aggregometer was set to a stir rate of 1,000 rpm. Both human and murine platelets were stimulated to aggregate with various agonists, such as thrombin (0.25–0.5 units), ADP (5 µmol/L), and collagen (5–10 µg/mL), and aggregation was assessed by aggregometer using Agro-Link software (Chrono-Log Corporation). Once stimulated, the assay was allowed to run until the percentage of light transmission reached a plateau, and the platelets had thus reached their maximum aggregation. Prior to stimulation with agonists, platelets were incubated with exenatide (H-8730; Bachem) at increasing concentrations (0.1–10 nmol/L) for 15 min at 37°C, or with PBS as a control.

Perfusion assays were performed as previously described (25–27). Perfusion chamber slides (μ-slide VI^0.1; ibidi, Martinsried, Germany) were coated with 100 μg/mL collagen type I (Kollagenreagens Horm; Nycomed) at 4°C for 18 h. Whole blood was labeled by adding 1 μmol/L DiOC6. Perfusion was performed at a flow rate of 1,500 s⁻¹ for human blood and 1,800 s⁻¹ for mouse blood using a syringe pump (Harvard Apparatus, Holliston, MA). Images of platelet adhesion aggregation and thrombus growth were recorded under a Zeiss Axiovert 200M Inverted Fluorescent Microscope (60× W objective; Zeiss) using a CoolSNAP Camera (Photometrics). Platelet aggregation and thrombus growth were quantitated by dynamics of platelet mean fluorescence intensity over the period of 3 min of perfusion using the Slidebook program.

All protocols were approved by the Animal Care Committees of the Toronto General Research Institute or Keenan Research Centre and conformed to the guidelines of the Canadian Council on Animal Care. C57BL/6 wild-type mice were purchased from Charles River (Montreal, Quebec, Canada), eNOS knockout (KO) (Nos3^tm1Unc) mice were purchased from The Jackson Laboratory (Bar Harbor, ME), and the origin, characterization, and genotyping of Glp1r^−/− (KO) mice have been described previously (28). Experiments were performed on 6- to 12-week-old mice housed for at least 2 weeks.

Male adult mice were anesthetized, and a tracheal tube was inserted to facilitate breathing. Antibodies, anesthetic reagent (pentobarbital, 0.05 mg/kg body wt; Abbott Laboratories, Toronto, Ontario, Canada), and exenatide (60 nmol/kg body wt i.v.) (29,30) were administered by a jugular vein cannula. The cremaster muscle was prepared under a dissecting microscope and superfused throughout the experiment with preheated bicarbonate buffer saline. Platelets were labeled by injecting a DyLight 649–conjugated rat anti–mouse GP1bβ antibody (0.1 μg/g; EMFRET Analytics). Multiple independent upstream injuries were performed on a cremaster arteriole with the use of an Olympus BX51WI Microscope with a pulsed nitrogen dye laser. The dynamic accumulation of fluorescently labeled platelets within the growing thrombus was captured and analyzed using Slidebook software (Intelligent Imaging Innovations). Blood glucose levels were monitored throughout the experiment and remained constant.

Six-week-old male C57BL/6 mice were irradiated with a single 10-Gy dose of radiation (γ Cell 40; Nordion International) 3 h prior to transplantation. Bone marrow cells were harvested from either Glp1r^−/− or wild-type mice femurs and resuspended in PBS. A total of 2.0 × 10⁶ bone marrow cells per mouse was injected intravenously with a 28-gauge needle. Mice were allowed to recover for 6 weeks prior to undergoing experiments.

To assess the effects of exenatide on a metric that depends on both platelet function and coagulation in vivo, murine tail bleeding assays were also performed (31,32). Mice were given a single injection of exenatide (60 nmol/L/kg i.v.), anesthetized, and placed on a heating pad for 10 min. The distal tail (5 mm from the tip) was amputated with surgical scissors, and the remaining tail was immediately immersed in 10 mL of 0.9% NaCl prewarmed to 37°C. The time required for spontaneous bleeding to cease was recorded.

The CBA085 Innozyme DPP-4 Immunocapture Activity Assay (Merck Millipore, Darmstadt, Germany) was used to assess DPP-4 activity in human platelets and MEG-01 cells. Protein from MEG-01 cells, human platelets, and lung tissue for use in the assay was extracted as described above. Blood was collected from a healthy donor into a BD Vacutainer tube (3.2% buffered sodium citrate) and spun down at 190g for 10 min. The plasma fraction was removed and stored in aliquots at −80°C for use in the assay. As an experimental control, half of the plasma samples were treated for 20 min with 20 µg/mL MK-626 (Merck Millipore) to block DPP-4 activity. The remainder of the assay was performed according to the manufacturer instructions.

Given the variety of concerns with currently available GLP-1R antibodies (33), we began our studies by exploring the expression of GLP-1R at the mRNA level. Using primers designed to span the entire GLP-1R coding sequence, including both 5′ and 3′ UTRs, we were able to clone and sequence full-length human GLP-1R mRNA from a human megakaryocyte cell line (MEG-01). Analysis of the transcript size and sequence confirmed identity with the known human GLP-1R (Fig. 1A) (GenBank Accession #KR138540).

Next, qRT-PCR was used to measure the levels of GLP-1R mRNA expressed in MEG-01 cells relative to other human tissues. This analysis revealed GLP-1R mRNA expression levels in MEG-01 cells to be <20% of the levels in the human pancreas, but >10-fold more than in the human lung. No GLP-1R signal was detected in negative control samples that included human skin fibroblasts and non–reverse transcribed MEG-01 mRNA (Fig. 1B). Furthermore, the binding of a fluorescently tagged exenatide peptide that was developed to probe GLP-1R expression (22) was evident in MEG-01 cells (Fig. 1C).

The GLP-1R is a Gαs-coupled receptor, and its activation in the pancreatic β-cell leads to adenylate cyclase–dependent increases in cAMP (34). Accordingly, intracellular cAMP responses to GLP-1 and the GLP-1R agonist exenatide were used to test the functional activity of the GLP-1R expressed in MEG-01 cells. These experiments showed dose-dependent increases in intracellular cAMP in response to GLP-1 (Fig. 2A) and exenatide (Fig. 2B), with maximal responses observed at the 10 nmol/L concentration of both agonists. Although the maximal cAMP response produced by these agents was considerably less than that evoked by PGI₂ (Fig. 2), the level of cAMP achieved with GLP-1 and exenatide is similar to that achieved in other cell types known to express the GLP-1R (35).

To assess the consequences of GLP-1R activation on platelet aggregation, both human and mouse platelets were incubated with increasing concentrations of exenatide (0.1–10 nmol/L) and were stimulated to aggregate with thrombin. On aggregometry traces, the first wave of thrombin-induced aggregation is known to correspond to platelet activation, with the second steeper wave corresponding to platelet α-granule and dense-granule release (36). At 6 min after stimulation with thrombin, during the second wave of aggregation, exenatide-treated human (Fig. 3A and B) and mouse (Fig. 3C and D) platelets manifest decreased aggregation in gel-filtered platelets compared with PBS-treated control cells, suggesting an inhibitory effect of GLP-1R activation on platelet granule release. Similar inhibitory effects of exenatide were also evident in human platelet aggregation stimulated with ADP and collagen in platelet-rich plasma and gel-filtered platelets, respectively (Supplementary Fig. 1).

Using collagen-coated glass perfusion chambers, fluorescently labeled platelet aggregation and thrombus formation in whole human and mouse blood were monitored under a high shear rate (human 1,500 s⁻¹, mouse 1,800 s⁻¹) in exenatide-treated (10 nmol/L) versus PBS-treated (equal volume) blood. Even greater than the effects of exenatide on platelet aggregation in vitro, thrombus formation in whole blood from both human (Fig. 4A and B) and mouse (Fig. 4C and D) was markedly inhibited by the GLP-1R agonist ex vivo.

Using our well-established quantitative laser injury cremaster arterial thrombosis model (26,31,32), we next determined that exenatide inhibited thrombus growth in adult male mice compared with PBS-injected controls. Moreover, the thrombi that did form in exenatide-treated animals were smaller, detached more easily, and embolized more rapidly compared with PBS-treated controls (Fig. 5A and B and Supplementary Video).

Similar experiments were performed in mice with streptozotocin-induced diabetes, because this model has previously been shown to have enhanced thrombosis (37). As expected, greater thrombus formation was observed in hyperglycemic mice compared with normoglycemic controls, with exenatide inhibiting thrombus growth to a similar degree in both models (Fig. 5C and D).

To dissect whether the antithrombotic effects of exenatide observed in vivo were the result of GLP-1R activation on circulating cells, such as platelets, or were the potential consequence of putative GLP-1R–mediated effects on endothelial or other cell types, lethally irradiated mice were transplanted with bone marrow from Glp1r^−/− versus wild-type mice. After successful bone marrow reconstitution (6 weeks), recipient mice were subjected to the cremaster arterial laser injury model. Importantly, mice reconstituted with bone marrow that lacked a functional GLP-1R had greater thrombus formation than those that received wild-type bone marrow (Fig. 6, green vs. gray). Furthermore, the antithrombotic effect of exenatide was markedly attenuated in mice lacking a functional GLP-1R in bone marrow–derived cells (Fig. 6, pink vs. red). Together, these results suggest a physiological role for GLP-1R in bone marrow–derived cells in thrombosis.

However, several studies (38,39) have also shown that exenatide is a vasodilator, and that exenatide treatment in human umbilical vein cells causes the release of NO, which, in addition to being a vasodilator, is a potent inhibitor of platelet function (20). As such, we also investigated whether exenatide would function in mice with a genetic absence of the endothelial NOS gene (eNOS^−/−). Compared with PBS-injected controls, the extent of in vivo thrombus formation after laser injury was not reduced by exenatide in eNOS^−/− mice (Fig. 7).

Finally, to exclude the possibility of an effect of exenatide on coagulation and hemostasis, mouse tail vein bleeding times were measured. No significant differences in bleeding times were observed between exenatide- and PBS-treated animals with (Supplementary Fig. 2A, wild-type) or without (Supplementary Fig. 2B, Glp1r^−/−) a functional GLP-1R.

Given that DPP-4is are widely used members of the incretin class of antidiabetic drugs, we also sought to explore the role of DPP-4 in platelets and the megakaryocyte cell line used in this study. DPP-4 is an aminopeptidase that is expressed in many tissue cell types as a membrane-spanning protein, and is also found in the plasma in a soluble form. An immunocapture activity assay (Merck Chemicals) revealed that neither MEG-01 cells nor platelets harbor any endogenous DPP-4 activity (Supplementary Fig. 3).

The current study shows that human megakaryocytes express a functional GLP-1R, and that GLP-1R activation with the GLP-1R agonist exenatide plays a role in inhibiting mouse and human platelet aggregation in vitro, mouse and human thrombus formation under flow conditions ex vivo, and mouse thrombus formation in vivo. This may have important implications for subjects with T2D, who have platelet dysfunction and are at an increased risk of adverse macrovascular events.

In a meta-analysis, Monami et al. (40) suggested that GLP-1–targeted therapies may reduce the risk of acute myocardial infarction and cardiovascular mortality in T2D patients. Moreover, they suggested that these beneficial effects are the result of mechanisms other than reductions in conventional risk factors such as obesity, blood pressure, and lipids. In this context, our report on the antiplatelet and antithrombotic effects of the GLP-1 analog exenatide represents a potential mechanism underlying these short-term clinical findings, in what were arguably smaller groups in lower-risk populations than those being studied in several large cardiovascular safety studies currently underway.

The investigators of ELIXA, a cardiovascular safety study of the GLP-1 analog lixisenatide in a high-risk post–acute coronary syndrome population of subjects with diabetes, have recently published results (41) stating that lixisenatide was noninferior to placebo for cardiovascular safety. Based on this early finding, we can only speculate as to whether the biology we report here will translate into reductions in thrombosis-dependent major adverse cardiovascular events in other studies.

To our knowledge, there exist no published data on the effects of GLP-1 or its analogs on platelet function and thrombosis. Regrettably, neither the small mechanistic clinical studies of GLP-1 analogs published to date nor the larger cardiovascular safety studies currently underway have included substudies aimed at platelet function. Having said that, our search of clinicaltrials.gov revealed one study (NCT01408862, yet to be published) of 20 healthy participants, whose updated information suggests that GLP-1R is expressed on human platelets. However, this online update also suggests that neither GLP-1 nor its metabolite GLP-1(9–36) had an effect on the aggregation of platelets isolated from these healthy volunteers. Given our results, we believe that human platelet function should be tested in clinical studies of GLP-1 analogs, particularly those including exenatide or its longer-acting formulations.

Interestingly, Gupta et al. (17) proposed that sitagliptin, a DPP-4i, inhibits platelet aggregation by interfering with tyrosine phosphorylation of the platelet plasma membrane Ca²⁺ ATPase channel, thereby limiting the accumulation of intracellular Ca²⁺. Given that there should have been no GLP-1 present in their platelet preparations, and that we have been unable to document any DPP-4 activity in platelets (or MEG-01 cells) (Supplementary Fig. 3), the results of Gupta et al. (17) are somewhat surprising. We can only speculate as to whether their results in vitro suggest a “direct” action of sitagliptin on platelet biology (i.e., independent of GLP-1 or DPP-4). With regard to their demonstration of an in vivo effect of sitagliptin on platelet aggregation, their use of a small group of subjects with diabetes (N = 50) treated with sitagliptin for 1 and 3 months was neither blinded nor controlled with placebo or an active comparator.

Our study also has several limitations. First, we have been unable to directly demonstrate the existence of GLP-1R protein on megakaryocytes or platelets. Recently, commercially available antibodies for GLP-1R have been shown to be nonspecific (42). Although we did obtain a new antibody believed to recognize human GLP-1R (33), this reagent did not show sufficient specificity for GLP-1R in our hands (data not shown). Having said that, our evidence implicating GLP-1R–dependent platelet biology is substantive, including the cloning and sequencing of full-length GLP-1R mRNA from MEG-01 cells, the binding of a fluorescent exenatide to MEG-01 cells, and the functional studies described above. Second, the results of our bone marrow transplant experiments suggest that a cell type other than bone marrow–derived cells may also contribute to the antithrombotic effects of GLP-1R activation. Although in vivo thrombus formation still occurred in mice receiving Glp1r^−/− bone marrow, the extent of thrombosis was considerably less than that observed in mice receiving wild-type bone marrow (Fig. 6). As such, we have not excluded the possibility that exenatide mediates antithrombotic effects, at least in part, via effects on the endothelium. Although our results in eNOS^−/− mice suggest that this putative downstream target of GLP-1R activation is central to the in vivo antithrombotic effects of exenatide (Fig. 7), it remains controversial as to whether platelets express eNOS (18,43). Because of this, the contribution of endothelial cells to the platelet-inhibiting effects of exenatide will require future studies with an endothelial cell–specific KO of GLP-1R.

The paradigm of endothelium-derived signaling molecules playing important roles in platelet function and thrombosis is well established. For example, endothelium-derived PGI₂ inhibits platelet aggregation by activating a G-protein–coupled receptor in platelets, increasing their intracellular cAMP levels through adenylate cyclase, with subsequent cAMP/cAMP-dependent protein kinase signaling inhibiting virtually all platelet-activating mechanisms (44,45). In this context, intracellular cAMP levels in platelets play a key role in maintaining hemostasis. Given our demonstration of both GLP-1 and exenatide causing increased cAMP levels in megakaryocytes, we believe that similar mechanisms mediate the ability of exenatide to directly inhibit platelet aggregation in vitro. In addition to this, the more potent antithrombotic effects of exenatide observed in vivo appear to depend on its activation of eNOS.

The expression of a functional GLP-1R on megakaryocytes and the inhibitory effects of GLP-1 and exenatide on platelet aggregation and thrombus formation have several implications. First, this could help to explain early clinical data suggesting improved cardiovascular outcomes in GLP-1 analog–treated patients (16). Second, T2D is associated with increased platelet reactivity and risk of thrombus formation, higher incidences of myocardial infarction and stroke (46), and diminished sensitivity to widely used antiplatelet drugs, such as aspirin and clopidogrel (5). In this context, our suggestion that exenatide might reverse the “prothrombotic” phenotype of T2D is of clinical significance. Indeed, if our findings translate to subjects with T2D who have high cardiovascular risk, treatments that include GLP-1R activation may help to reduce the burden of major adverse cardiovascular events in this population. Third, because platelets also play a role in atherosclerotic plaque formation and leukocyte activation (47–49), inhibiting platelet activation with a GLP-1 analog may also contribute to antiatherogenic and anti-inflammatory effects (30,50). Fourth, because intravenous exenatide did not prolong mouse tail-bleeding time, its antithrombotic actions are not accompanied by effects on coagulation or hemostasis. This important finding argues against the possibility of exenatide causing an excess of bleeding complications. Taken together, these implications may confer significant therapeutic advantage to GLP-1R agonists.

Acknowledgments. The authors thank Dr. Tianru Jin, Dr. Michael Wheeler, and Dr. Daniel Drucker of the University of Toronto for helpful discussion and access to specific reagents.

Funding. This work was funded in part by an Operating Grant to H.Ni from the Canadian Institutes of Health Research (MOP 119540), an Investigator-Initiated Research grant to H.Ni and M.H. from Merck Canada (IISP #40363), and a Grant-in-Aid to M.H. from the Heart and Stroke Foundation of Canada (T6757). M.H. was supported by a Career Investigator Award from the Heart and Stroke Foundation, Ontario Provincial Office.

Duality of Interest. The discoveries described in this report form part of U.S. provisional patent application US61/721,819. M.H. has received other investigator-initiated research support from Amylin/BMS/Astra-Zeneca and Novo Nordisk. M.H. has received consulting fees and/or speaking honoraria from Astra-Zeneca, Boehringer-Ingelheim, Eli Lilly, Janssen, Merck, Novo Nordisk, and Roche. No other potential conflicts of interest relevant to this article were reported.

Author Contributions.A.C.-V., A.R., and M.A.S. conceived, designed, and conducted the experiments; analyzed the data; and wrote and revised the manuscript. X.R.X., X.L., O.E.-M., and H.No. conducted the experiments and analyzed the data. Y.W. and T.A. conceived, designed, and conducted the experiments and analyzed the data. E.S. conducted the experiments, analyzed the data, and wrote and revised the manuscript. H.Ni and M.H. conceived and designed the experiments and wrote and revised the manuscript. R.W. contributed key reagents and expertise for the project. H.Ni and M.H. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.