|Home | About | Journals | Submit | Contact Us | Français|
Thrombin is a serine protease and regulator of hemostasis that plays a critical role in the formation of obstructive blood clots, or thrombosis, that is a life-threatening condition associated with numerous diseases such as atherosclerosis and stroke. To detect thrombi in living animals, we design and conjugate thrombin-sensitive peptide substrates to the surface of nanoparticles. Following intravenous infusion, these “synthetic biomarkers” survey the host vasculature for coagulation and, in response to substrate cleavage by thrombin, release ligand-encoded reporters into the host urine. To detect the urinary reporters, we develop a companion 96-well immunoassay that utilizes antibodies to bind specifically to the ligands, thus capturing the reporters for quantification. Using a thromboplastin-induced mouse model of pulmonary embolism, we show that urinary biomarker levels differentiate between healthy and thrombotic states and correlate closely with the aggregate burden of clots formed in the lungs. Our results demonstrate that synthetic biomarkers can be engineered to sense vascular diseases remotely from the urine and may allow applications in point-of-care diagnostics.
Urine analysis has a rich and long-standing clinical history as a tool for monitoring health and disease and remains an integral component of a medical examination.1−3 Well over 100 tests can be performed to indicate conditions as diverse as pregnancy,4,5 diabetes,6−9 kidney diseases,10−12 metabolic disorders,13,14 and others. Recently, the discovery of urinary analytes that were previously thought to be present mainly in circulating blood because of their large hydrodynamic radii (e.g., enzymes, exosomes and others) has expanded the diagnostic repertoire to include urinary biomarkers against diseases of distant organs such as breast and brain cancer.15,16 Inspired by the elegant physiology of the renal system, which has evolved the capacity to selectively filter liters of blood to remove byproducts of biological processes within minutes, we recently developed a class of protease-sensitive nanoparticles, called “synthetic biomarkers”, that in response to dysregulated protease activity at the sites of disease, release reporters into the circulation that are then concentrated into the host urine for noninvasive monitoring.17 In murine models of liver fibrosis and cancer, we showed that synthetic urinary biomarkers have the potential to noninvasively monitor solid organs and improve early stage detection of cancer compared to tumor-secreted blood biomarkers.
Here we hypothesized that synthetic biomarkers could be tailored to survey intravascular sites for acute thrombosis, the activation of a cascade of protease activity that orchestrates the formation of obstructive blood clots within vessels (Figure Figure11A). Thrombi are a critical pathophysiological feature of numerous vascular diseases including acute coronary syndrome, stroke, and venous thromboembolism.18 The most important serine protease in the coagulation cascade is thrombin, which not only catalyzes the conversion of fibrinogen to fibrin that serves as the structural scaffold of a clot, but also regulates hemostasis through positive and negative feedback circuits.19,20 To date, a number of studies have described the use of near-infrared fluorogenic probes to detect thrombin activity in the setting of thrombus formation as well as other thrombin-dependent diseases such as atherosclerosis.21−23 More recently, these probes have been modified to include cell penetrating mechanisms that are activated after cleavage to improve the retention of the imaging agent and maintenance of the detection signal.23,24 In the clinic, blood biomarkers such as prothrombin fragment 1.2 (a byproduct of prothrombin cleavage into thrombin) and D-dimer (a byproduct of fibrin degradation) are often used as indicators of thrombosis; however, these tests are highly susceptible to artifacts introduced by a blood draw, have poor specificity, and more accurately reflect upstream or downstream cleavage events (i.e., Factor Xa activation of prothrombin or plasmin activity during fibrinolysis, respectively) rather than thrombin activity.25−27 In this report, we engineer nanoparticles that survey the host vasculature for thrombi and,17,28 in response to thrombin activity, release reporters into the urine as an integrated measure of the aggregate burden of systemic clots. We describe a method to encode these reporters with structurally distinct ligands that allow antibody-based detection by enzyme-linked immunosorbent assay (ELISA) in standardized 96-well plates that makes this platform readily amenable for use in clinical laboratories.
The construction of synthetic biomarkers for thrombosis involves modifying the surface of iron oxide nanoworms (NW), a nanoparticle formulation previously developed by our collaborators,29,30 with substrate-reporter tandem peptides that are cleavable by thrombin and detectable by ELISA (Figure Figure11A). NWs were chosen for their safety profile and large hydrodynamic diameter (~40 nm, Figure S1A, Supporting Information), which prevents surface-conjugated peptides from filtering directly into the urine (~5 nm glomerulus size-exclusion limit) before cleavage.17,28 To first develop a suitable substrate, we extended the thrombin cleavable sequence fPR–x–S (x = site of cleavage, kcat/Km ~ 9.33 × 106)31−34 to include glycine spacers and a C-terminal cysteine to allow coupling to NWs via sulfhydryl chemistry.30 To test substrate specificity, we conjugated fluorophore-labeled derivatives onto NWs (sequence = (K-Flsc)GGfPRSGGGC, Figure S2A, Supporting Information) at a valency (~40 peptides per NW by absorbance spectroscopy, Figure S1B, Supporting Information) sufficient to reduce fluorescence by over 90% via homoquenching (Figure S1C, Supporting Information) and then incubated the NWs (200 nM by peptide, 5 nM by NW) with purified thrombin (2 μM) or a panel of blood clotting proteases (FXa (160 nM), APC (60 nM), FIXa (90 nM), FVIIa (10 nM), FXIa (31 nM)), each present at its maximal physiological concentration (Figure Figure11B). Freely emitting peptide fragments that were released by thrombin activity rapidly increased sample fluorescence (red, Figure Figure11C). By contrast, negligible proteolysis was observed from the panel of noncognate proteases, as well as by thrombin in the presence of bivalirudin (Bival), a clinically approved direct thrombin inhibitor. To further investigate the ability to sense thrombin activity from blood, we spiked NWs into human plasma samples inactivated with sodium citrate (an anticoagulant that chelates the cofactor calcium) and monitored plasma fluorescence after the addition of excess calcium chloride (CaCl2) to trigger coagulation, or phosphate buffered saline (PBS) as a control. Aligned with our previous observations with purified enzymes, plasma fluorescence markedly increased upon activation of the clotting cascade but not in control samples or in the presence of bivalirudin (Figure Figure11D). To test stability, we incubated fluorogenic NWs in 10% serum at 37 °C overnight and did not detect any significant differences in size (Figure S1D, Supporting Information) that would indicate precipitation or increases in sample fluorescence (Figure S1E, Supporting Information) that would indicate nonspecific substrate cleavage. Collectively, these results established the ability of our NWs to specifically sense the proteolytic activity of thrombin within the complex milieu of plasma, consistent with previously described thrombin-specific fluorogenic probes.22
We next set out to build a system of ligand-encoded reporters that would allow quantification of protease activity in a 96-well format by ELISA, the primary detection platform for many clinical tests. Conventional ELISAs detect a target analyte via a sandwich complex composed of two affinity agents that bind to distinct epitopes on the analyte (Figure Figure22A). To build a synthetic reporter, we modified the protease-resistant peptide Glutamate-Fibrinopeptide B (Glu-fib, sequence = eGvndneeGffsar, lower case = d-isomer), which we selected for its high renal clearance efficiency,35 at the termini with structurally distinct ligands (i.e., Flsc or AF488) and biotin (labeled R1 and R2 respectively; Figure Figure22A). To test the immunoassay, these reporters were then spiked into urine and applied to 96-well plates precoated with capture antibodies (α-Flsc or α-AF488) before the presence of R1 or R2 was detected by the addition of neutravidin-horseradish peroxidase (HRP) and its catalytic development of 3,3′,5,5′-tetramethylbenzidine (TMB). As predicted from the specificities of the antibodies, a significant change in color appeared only in wells containing matched antibody-ligand pairs (+/– or −/+ wells, Figure Figure22B) and was not affected by the presence of noncognate reporters (+/+ wells). Identical trends were observed at the limits of detection for both capture antibodies (~3 pM, Figure Figure22C, Figure S3, Supporting Information), indicating that our synthetic reporters were detected with high specificity and sensitivity comparable with protein-based ELISAs.36 With an optimized thrombin substrate and a reporter system in place, we then incubated NWs (100 nM by peptide, 2.5 nM by NW) decorated with our final tandem peptide construct (sequence = biotin-eGvndneeGffsar(K-Flsc)GGfPRSGGGC, Figure S2B, Supporting Information) with increasing levels of thrombin and found that the amount of cleavage products released into solution (isolated by size filtration) was dose dependent, reaching a plateau likely due to cleavage of all available substrates and establishing our ability to monitor thrombin activity by ELISA (Figure Figure22D). Collectively, these results indicate that the specificity of ligand–antibody interactions can be used to build panels of orthogonal reporters for monitoring protease activity by standardized 96-well assays.
We next investigated the ability of our synthetic biomarkers to detect thrombosis in living mice induced via intravenous (i.v.) administration of thromboplastin. This model has been used in the hematology literature to explore the role different vascular receptors play in host susceptibility to thrombosis and to probe the efficacy of new antithrombotic agents.37−39 Thromboplastin triggers the clotting cascade through the extrinsic pathway via complexation of tissue factor and factor VII, and blood clots embolize to the lungs in this model, recapitulating the life-threatening clinical condition of pulmonary embolism (PE). To quantify PE formation, we coinjected mice with thromboplastin and the clot precursor fibrinogen labeled with the near-infrared fluorophore VT750 so that the formation of fibrin clots by thrombin-mediated proteolysis of fibrinogen could be quantified by fluorescent analysis of whole organs (Figure Figure33A). Within 30 min of administration, we observed a more than 6-fold increase in the level of fibrin(ogen) deposited within the lungs and significant decreases in the kidneys and liver (P < 0.005 by Student’s t-test, n = 3 mice; Figure Figure33B), consistent with venous blood flow patterns that transport thrombi formed upon i.v. administration directly to the lungs from the heart, leading to depletion of VT750-fibrinogen in the other organs. Histochemical analysis of tissue sections corroborated these findings by revealing the presence of blood clots in lung sections (blue arrow, Figure Figure33C) that were absent in the other major organs (brain, heart, kidney, liver and spleen; Figure S4, Supporting Information) and in control animals. Animals given escalating but sublethal doses (observed LD50 ~ 3 μL per g b.w.) of thromboplastin accumulated fibrin(ogen) in the lungs in proportion to the dosage, and PEs were readily prevented in animals pretreated with bivalirudin (P < 0.005 by one-way ANOVA with Tukey post-test, n = 3–5 mice; Figure Figure33D, Figure S5, Supporting Information), confirming that clot formation is largely driven by the activity of thrombin. Altogether, these results established our ability to precisely control total clot burden in a model that resembles the clinical pathology of venous thrombosis.38,40
Next, we characterized the pharmacokinetics of our synthetic biomarkers in the context of thrombosis. We injected mixtures of VT750-labeled NWs and thromboplastin into mice and observed no significant differences in NW distribution between the thromboplastin and control groups in all of the excised organs, including the lungs, indicating that thrombosis did not alter the biodistribution of the NW scaffold (P > 0.05 by Student’s t-test, n = 3 mice; Figure Figure44A, Figure S6, Supporting Information). To monitor peptide cleavage and trafficking of the cleaved fragments, we coadministered NWs conjugated with fluorescently quenched substrates and observed significant increases in fluorescence in the lungs and kidneys by ~1.8 and ~2.5 fold over healthy animals, respectively (P < 0.01 by Student’s t-test, n = 3 mice; Figure Figure44B, Figure S7, Supporting Information). Paired with our earlier observations showing that thromboplastin did not alter the biodistribution of the NWs and induced blood clots that were localized to the lung (i.e., clots were not found in the kidneys), this finding provided evidence of peptide cleavage in the lungs and kidney accumulation of freely emitting fluorescent fragments. Immunofluorescent staining of lung sections further showed NW (green) localization with fibrin (red) at the sites of coagulation, which was absent in control sections (Figure S8, Supporting Information), supporting our hypothesis that circulating NWs can access local thrombi. To visualize the clearance efficiencies of the peptide fragments, we monitored mice by in vivo fluorescence imaging and observed a strong increase in fluorescent signal that was localized to the bladder of thrombotic mice relative to controls (Figure Figure44C). Taken together, our data illustrated that our synthetic biomarkers can systemically survey the vasculature for thrombin activity and release reporters at sites of thrombosis, which are then cleared efficiently into the host urine.
In considering clinical translation, we sought to develop a method to account for variations in the production rate of urine expected in individuals that could affect the urine concentration of our reporters. Urinary production rates are mainly dependent on the hydration state of the host (ranging from 50–1200 mOsm/kg of H2O in humans)41 and affected by many external factors (e.g., circadian rhythm, diet, activity, and others). Approaches to determine the concentration of urine include measuring the level of creatinine,42,43 a byproduct of muscle metabolism that filters into the urine at a steady state when at rest, or i.v. administration of inulin,44−46 a polysaccharide that is not actively absorbed or secreted by the kidneys and whose appearance in urine is directly related to the rate of urine production. Motivated by the clinical precedent set by inulin, we hypothesized that because our free reporters (R1, R2) are built from Glu-fib, which is likewise biologically inert,35 their filtration into urine following i.v. administration would be indicative of the concentration of urine. To test this, we excessively hydrated a cohort of mice with a subcutaneous bolus of saline equivalent to 10% of their body weight followed by i.v. administration of free R2. Compared to control mice infused with R2 only, hydrated mice produced over 2.5 fold more urine within 2 h (P < 0.005 by Student’s t-test, Figure S9A, Supporting Information) and their urinary concentration of R2 decreased by ~50% (P < 0.005 by Student’s t-test, Figure S9B, Supporting Information), showing that our free reporters could be used to monitor the hydration state and urine concentration of the animals. We next sought to monitor thromboplastin-induced PEs by urine analysis of the response of our synthetic biomarkers to thrombin activity. To simulate serial monitoring that frequently occurs in inpatient settings, we first determined the basal activity in healthy cohorts of animals each receiving thrombin-sensitive NWs and a free reporter (R2) for urine normalization (Figure Figure44D). After five days to allow NWs to fully clear (half-life ~6 h),17 we administered a mixture of thromboplastin, NWs, and R2 into the same mice and quantified reporter levels by ELISA. When compared to their healthy state (day 0), the induction of PEs (day 5) resulted in significant elevations in the level of urinary cleavage fragments by up to 3-fold (P < 0.005 by two-way ANOVA with Bonferroni post-test, n = 5 mice; Figure Figure44D). In mice treated with bivalirudin prior to thrombotic challenge (dose = 2 μL per g of b.w.), reporter levels were abrogated, consistent with our earlier findings showing the ability of bivalirudin to inhibit thrombin activity and prevent the formation of PEs. When the urinary biomarker marker levels from thromboplastin-challenged mice were directly compared to the amount of fibrin(ogen) deposited at identical doses of thromboplastin (Figure Figure33D), we found a striking correlation to the disease burden with a correlation coefficient of 0.99 (Pearson’s r, Figure Figure44E). Collectively, our findings showed that synthetic biomarkers can monitor thrombin activity in living mice and quantitatively measure the aggregate burden of sublethal PEs from the urine by ELISA.
By harnessing the capacity of peptide-decorated NWs to circulate and sense their local vascular microenvironment, we have engineered synthetic biomarkers that can detect thrombin activity in vivo and noninvasively quantify the aggregate amount of active clots. Unlike other nanoparticle sensors that function by producing a localized signal,21−24,47−49 our NWs sense protease activity by releasing reporters locally at the sites of thrombus formation but are then filtered and detected remotely from the urine. Interestingly, in imaging studies using fluorogenic thrombin-cleavable probes, this “washout” of the cleaved fragments was also directly observed by monitoring the attenuation of the strength of the detection signal localized at the thrombi.21 Similar to circulating biomarkers, our approach can reveal thrombosis at sites deep within the body, such as the lungs, that are difficult to detect with fluorogenic probes because of tissue absorption and scattering of light.50 This property allows urine analysis to integrate and quantitatively assess the burden of vascular clots, which would otherwise require systemic exploration by imaging. In addition, we developed a panel of heterobifunctional reporters that can be detected by standardized 96-well plate assays, removing the need for mass spectrometry as described in our previous study.17 This reporter system is readily extensible by incorporating additional ligand-capture agent pairs and is amenable for detection by other methods including paper-based tests at the point of care.51−53 Potential improvements to this platform include the use of new thrombin-sensitive substrates that are significantly more specific to reduce background activities from other plasma proteases,34,54 and further functionalizing NWs, which are superparamagnetic,30 with fibrin-targeted ligands to allow contrast-enhanced magnetic resonance imaging (MRI) of individual clots simultaneously with urine analysis.55−57 To allow clinical translation, we chose to use NWs because we previously showed that they are well-tolerated by mice, and similar FDA-approved formulations of iron oxide nanoparticles (e.g., Ferridex) are already used in patients;17,58−63 however, thrombin substrates may also be attached to other long-circulating nanoparticles, such as dextran or liposomes, to prevent peptide filtration into urine until cleavage by proteases.
Looking forward, several clinical applications warrant further investigation with this approach. Because sensing thrombin activity requires NWs to access the sites of coagulation, the local architecture of the vessels, clotting kinetics of the thrombi, and degree of occlusion would all likely influence the rate of peptide cleavage and clearance efficiencies of the reporters.21,23,24 Therefore, additional studies that utilize specific clinical models, such as deep vein thrombosis (DVT), would be important to determine the type of clots this approach could be used to detect. Further, whereas MRI or ultrasound can resolve anatomical features of clots, they cannot discriminate stable from extending thrombi without serial imaging. Related studies in atherosclerosis showed that thrombin activity could be used to differentiate stable from severe plaques, highlighting the potential benefits of an activity-based measurement compared to imaging alone.23 In summary, we believe this work further broadens the repertoire of nanomedicines that could be used for noninvasive monitoring of disease, and we anticipate generalization to additional clinical settings in which dysregulated thrombin activity is prominent.
Aminated iron oxide NWs were synthesized according to previously published protocols.30 Peptides (biotin-eGvndneeGffsar(K-Flsc)GGfPRSGGGC, lower case = d-isomer) were synthesized by the Tufts University Core Facility peptide synthesis service. To conjugate peptides to NWs, NWs were first reacted with succinimidyl iodoacetate (Pierce) to introduce sulfhydryl-reactive handles. Cysteine terminated peptides and 20 kDa polyethylene glycol–SH (Laysan Bio.) were then mixed with NWs (95:20:1 molar ratio) for one hour at room temperature (RT) and purified by fast protein liquid chromatography. Stock solutions were stored in PBS at 4 °C. The number of fluorescein-labeled peptides per NWs was determined by absorbance spectroscopy using the absorbance of fluorescein (490 nm) and its extinction coefficient (78000 cm–1 M–1). For pharmacokinetic studies, NWs were first reacted with NHS-VT750 (PerkinElmer) prior to PEGylation as above. For fluorogenic assays, thrombin substrates were synthesized with a terminal fluorescein or VT750 in lieu of a reporter.
NWs (1 μM by peptide, 25 nM by NW) were incubated in 10% fetal bovine serum at 37 °C. At selected time points, the particle size was measured by dynamic light scattering (Malverin Zetasizer Nano Series) and the fluorescence intensity was measured by microplate reader (SpectroMax Gemini EM).
NWs (200 nM by peptide, 5 nM by NW) were mixed with human thrombin (2 μM), FVIIa (10 nM), FIXa (90 nM), FXa (160 nM), FXIa (31 nM), and activated protein C (60 nM), all purchased from Haematologic Technologies, in a 384-well plate at 37 °C in activity buffers according to the manufacturer’s instructions and monitored with a microplate reader (SpectroMax Gemini EM). For plasma studies, NWs were mixed with 50 μL of control human plasma (Thermo Scientific) and 50 μL of 80 mM CaCl2 (Sigma) or PBS. For thrombin inhibition experiments, bivalirudin (Anaspec) was added to a final concentration of 5 mg/mL and preincubated for 2 min prior to addition of NWs. For the ELISA studies, NWs (100 nM by peptide, 2.5 nM by NW) were incubated with thrombin for 10 min at 37 °C, and cleaved reporters (R1) were purified from NWs by centrifugal size filtration (3 kDa MWCO).
The bottom of 96-well plates (Thermo Scientific) were coated with either 0.8 μg/mL of anti-Flsc (GeneTex, GTX19224) or 0.4 μg/mL of anti-Alex Fluor 488 (Invitrogen, A11094) diluted in PBS overnight at 4 °C. Plates were blocked with 1% w/v bovine serum albumin (Sigma) in PBS for 1 h before 100 μL of samples were added. Reporters captured on the plate were then detected by adding 100 μL of 0.2 μg/mL neutravidin-HRP (Pierce), developed with 50 μL TMB solution (Thermo Scientific) for 5–15 min, and quenched with 50 μL of HCl before the absorbance of the wells was determined by microplate analysis (SpectraMax Plus, Molecular Devices) at 450 nm. Plates were washed 3× with PBST between each step, and incubation occurred at RT unless otherwise stated.
Each vial of thromboplastin containing 3–4 mg (from rabbit brain, Sigma) was dissolved in 2 mL of PBS. To quantify fibrin deposition, bovine fibrinogen (Sigma) was reacted with 3-fold molar excess of VT750 for 1 h at RT and purified by centrifugal size filtration (100 kDa MWCO, Millipore). Swiss Webster mice (Taconic) were lightly anesthetized with isofluorane and administered mixtures of VT750-fibrinogen (1 nmol by VT750) with thromboplastin (n = 3 mice per dose) via tail vein injections. After 30 min, mice were euthanized by CO2 asphyxiation, and organs were scanned on the LI-COR Odyssey Infrared Imaging System. Fluorescence in each organ was quantified using Image J software (NIH). To test thrombin inhibition, mice were intravenously administered bivalirudin (10 mg/kg) 5 min prior to coinjection of thromboplastin. For histology, lungs were inflated with 4% paraformaldehyde, while all other organs were incubated in 4% paraformaldehyde for 1–2 h at RT. All organs were stored in 70% ethanol until paraffin-embedding, sectioning, and staining (Koch Institute Histology Core).
To analyze NW and peptide pharmacokinetics, mice were given either VT750-labeled NWs (5 μM by VT750) or NWs conjugated with VT750-labeled peptides (600 nM by peptide, 15 nM by NW) in conjunction with thromboplastin. To analyze tissue sections by immunostaining, NWs (600 nM by peptide, 15 nM by NW) and thromboplastin (2 μL/g of b.w.) were administered to mice, and major organs were harvested after 30 min. Representative lung sections were stained for NWs (anti-Flsc primary, Invitrogen, A11090), fibrin (Nordic, GAM/Fbg/Bio) and Hoechst (Invitrogen, H3569) before analysis by fluorescence microscopy (Nikon Eclipse Ti).
The free reporter R2 (biotin-eGvndneeGffsar(K-AF488)) was synthesized by the Tufts University Core Facility peptide synthesis service. Mice (n = 5 mice) were anesthetized and injected subcutaneously with a PBS bolus equivalent to 10% of their body weights. After two hours, R2 (125 nm) was administered to mice via a tail vein injection. Mice were placed over 96-well plates surrounded by cylindrical sleeves for 30 min post-NW injection to allow mice to void. Urine samples were stored at −80 °C until ELISA analysis.
Experiments were conducted in a paired setup. Thrombin-sensitive NWs (600 nM by peptide, 15 nM by NW) and R2 (125 nM) were coinjected into healthy mice (n = 5–10 mice) to determine background protease activity and placed over 96-well plates to collect urine. Five days later, mice were again dosed with NWs, R2, and thromboplastin before urine was collected from mice 30 min post-NW injection. For thrombin inhibition experiments, mice were intravenously administered bivalirudin (10 mg/kg) 5 min prior to NW/R2 injections. Urine samples were stored at −80 °C until ELISA analysis.
ANOVA analyses and Student’s t-test were calculated with GraphPad 5.0 (Prism). Pearson’s r coefficient was calculated with Excel (Microsoft Office).
All animal work was approved by the committee on animal care (MIT, protocol #0411-036-14).
We thank the Swanson Biotechnology Center (MIT) for use of their animal imaging facilities and assistance with tissue sectioning. We thank Dr. Heather Fleming (MIT) for critical readings of the manuscript. This work is supported by a grant from the Koch Institute Frontier Research Program through the Koch Institute Frontier Research Fund and the Kathy and Curt Marble Cancer Research Fund, the Mazumdar-Shaw International Oncology Fellows Program, and the MIT Deshpande Center Innovation Grant. K.Y.L. acknowledges support from CCNE (5 U54 CA151884-03). Dr. G.A.K. acknowledges support from the Ruth L. Kirschstein National Research Service Award (F32CA159496-02) and holds a Career Award at the Scientific Interface from the Burroughs Wellcome Fund. Dr. S.N.B is an HHMI Investigator. The authors wish to dedicate this paper to the memory of Officer Sean Collier, for his caring service to the MIT community and for his sacrifice.
Supplementary figures. This material is available free of charge via the Internet at http://pubs.acs.org.
K. Y. Lin and G. A. Kwong contributed equally.
The authors declare no competing financial interest.