Resolving the multifaceted mechanisms of the ferric chloride thrombosis model using an interdisciplinary microfluidic approach

In vivo animal models of thrombosis are used extensively to study clot formation and propagation, of which the ferric chloride (FeCl₃) thrombosis model is the most commonly used. In this model, filter paper saturated with FeCl₃ is applied to vessel adventitia of small mammals, typically mice, resulting in occlusive thrombosis.^1-5  As the FeCl₃ diffuses into the vascular space, clot formation is commonly monitored by Doppler flow and intravital microscopy. Although the FeCl₃ injury model is popular due to ease of implementation and robust thrombus formation, the underlying mechanisms of FeCl₃-induced clotting remain unclear.

Clot initiation in this model is historically attributed to free iron-induced denudation of endothelial cells and the subsequent exposure of the subendothelium that triggers platelet adhesion/aggregation and activation of the coagulation cascade.⁶  However, studies using mice deficient in collagen receptor glycoprotein VI have produced contradictory results, reporting both normal and altered thrombus formation.^7,8  Eckly et al observed that although endothelial cells appear denuded and platelets are present in the occlusive thrombus, there is limited collagen exposure in the vessel.⁹  Furthermore, an elegantly designed scanning electron microscope (SEM) analysis revealed that endothelial cells are not denuded, and surprisingly, that erythrocytes mediate the adherence of platelets to the endothelium by an unknown mechanism.¹⁰  In addition to the uncertain status of the endothelium, the effects of FeCl₃ vary based on the vessel to which it is applied: in mice lacking the extracellular domain of glycoprotein Ibα (GPIbα), the FeCl₃ model results in occlusion in mesenteric venules but not in the carotid artery.^7,11  These inconsistencies, in conjunction with the wide range of published FeCl₃ concentrations (10 mM-2 M; for percentage, see supplemental Table 1, available on the Blood Web site) and application times (1-20 minutes) used to elicit a clotting response, raise questions about the direct chemical effects of FeCl₃ on blood components and the role that mass transfer plays in this thrombosis model.¹² 

These questions inspired us to systematically investigate the mechanisms of FeCl₃-induced thrombosis via an in vitro reductionist approach: interrogating the effect of FeCl₃ on individual blood components in the presence and absence of endothelial cells. To this end, we designed a microfluidic device that enables precise control over FeCl₃ influx while allowing for direct visualization and quantification of FeCl₃-induced aggregation of isolated blood components. Our laboratory has shown that vascular phenomenon relevant to hematologic processes can be studied in vitro by culturing endothelial cells to a confluent monolayer in microchannels.^13,14  These “endothelial-ized” microfluidics allow us to recapitulate the in vivo endothelium-blood-FeCl₃ interface and monitor the status of the endothelium during FeCl₃ application, while controlling the flow conditions and cellular and molecular “inputs” to the system. We find that all isolated blood components and molecules aggregate in a FeCl₃ concentration-dependent manner, with and without endothelial cells. We therefore posit a novel 2-stage mechanism for the action of FeCl₃ in thrombus formation. The first stage is based on colloidal chemistry principles whereby cells and plasma proteins adhere and aggregate due to the binding of negatively charged proteins to positively charged iron species. The extent of this stage is dependent on effective FeCl₃ concentration, which is governed by intraluminal mass transport. In the second-stage, clot formation proceeds via platelet aggregation and activation of the coagulation cascade in response to both endothelial cell death and arrested blood components. Experimentally separating the first and second phases is nontrivial, as they are visually indistinguishable and occur on a similar time scale. This novel multifaceted mechanism reconciles conflicting findings in the literature and cautions the use of the FeCl₃ model, especially in interpreting the initial stages of clotting. Our 2-phase description of the underlying mechanisms of this model complements and further explains Barr et al’s findings that red blood cells (RBCs) mediate FeCl₃-induced thrombosis.¹⁰ 

Human blood was drawn according to institutional review board–approved protocols per the Declaration of Helsinki into sodium citrate (Becton Dickinson). To recapitulate the endothelium-blood-FeCl₃ interface, a polydimethylsiloxane-based T-shaped microfluidic was designed with a 250-μm-width main channel, 50-μm-width side channel, and overall device height of 50 μm. When appropriate, channels were endothelialized as previously described.¹⁵  To study whole blood, RBCs were diluted in platelet-rich plasma (PRP) to 12.5% to allow for improved microscopy and visualization of clot components. To recalcify blood, CaCl₂ (Sigma) was added to a final concentration of 0.01 M directly prior to every experiment, unless otherwise noted. To analyze drug treatment data, we used a 2-tailed Student t test with a 95% confidence interval (with statistical significance when P < .05). See supplemental Methods for blood separation protocol, in vivo iron measurements, diffusion calculations, and flocculation experiment and drug treatment/mouse experiment protocol.

To recapitulate the in vivo FeCl₃ model in a controlled in vitro system, the FeCl₃ concentration that the endothelial cells and blood are exposed to was determined. Although the FeCl₃ concentration at the vessel adventitia is that applied to the filter paper, the luminal endothelial cell concentration (after the FeCl₃ diffuses through the vessel layers) is the unknown, but calculable, value of interest (Figure 1A). As diffusion of FeCl₃ through the vessel wall is complex and difficult to experimentally validate, we instead measured the bulk iron concentration downstream of the application site in vivo and used Fick’s laws of diffusion to calculate the concentration at the endothelial cell/blood interface. Postapplication of 1 M FeCl₃ to the adventitia of a mouse carotid artery, the bulk iron concentration in blood was found to be ∼9 μg/mL or 160 μM. Steady-state FeCl₃ flux into the vessel is reached by 5 minutes and maintained at 10 minutes; the patch concentration is thus assumed to be constant over the relevant time scales. By equating flux of FeCl₃ into the vessel to flux out of the vessel, a conservative estimate (that does not account for system losses) of concentration at the wall of the lumen is calculated to be ∼50 mM (supplemental Methods). This FeCl₃ concentration can be introduced to blood components via perfusion into an in vitro microfluidic system, thereby giving us spatiotemporal control over FeCl₃ concentration, and the ability to monitor its effects on blood and the vasculature.

To visualize the effects of FeCl₃ on blood components in a reductionist manner, microfluidic channels were designed to introduce FeCl₃ to individual blood components in a fixed shear environment. In vivo, FeCl₃ continuously enters the vessel from the patch and interfaces with endothelial cells and blood. As convection dominates in this system, the FeCl₃ concentration remains high near the vessel wall and flows downstream with the blood. The flow environment is aptly recapitulated using a T-shaped microfluidic; the influx of FeCl₃ through a small side channel gives spatiotemporal control over the concentration, ensuring that the blood cells and endothelial cells at the streams’ interface are exposed to an exact input concentration. The 2 solutions interface near the channel wall, as it occurs in vivo (Figure 1C). The main channel of the microfluidic can be seeded with endothelial cells, which are cultured to a confluent monolayer on the channel surfaces, further recapitulating in vivo physiologic conditions (Figure 1B-C), though endothelial cells do not span the gap where the main channel joins the side channel. Endothelialized microfluidics enable us to concurrently visualize the effects of FeCl₃ on blood and endothelial cells using standard bright-field and fluorescence video microscopy.

We recreated the in vivo FeCl₃ model by introducing FeCl₃ to blood in an endothelialized microfluidic system. When exposed to 50 mM FeCl₃, endothelial cells remain adherent and whole blood forms an occlusive aggregate (Figure 2A; supplemental Video 1). Platelets and fibrinogen are present within the effective “clot,” but interestingly, erythrocytes constitute the bulk of the aggregate (Figure 2A). Furthermore, when individually introduced, washed RBCs, platelets, platelet-poor plasma (PPP), PRP, and even 5% bovine serum albumin (BSA) form aggregates at the FeCl₃ interface (Figure 2A; supplemental Figure 1A, supplemental Video 2). The effect of FeCl₃ on endothelial cell viability is concentration-dependent: endothelial cells are viable at low concentrations, but as concentrations increase from 5 mM to 1 M, endothelial cell death is more pervasive (Figure 2B). Dead endothelial cells remain adherent to the channels at all tested concentrations and shear stresses (up to 20 dyne/cm²), corroborating the results of Barr et al (data not shown).¹⁰ 

To characterize the effects of FeCl₃ on blood, independent of the endothelium, blood cells and proteins were individually exposed to FeCl₃ in a “bare,” nonendothelialized, microfluidic device (Figure 1B), and time to aggregate initiation (adherence of components) and stabilization (complete or unchanging occlusion) were recorded. In the absence of endothelial cells, blood components form aggregates in a FeCl₃ concentration–dependent manner, shown at 0.05, 0.5, and 1 M, as did 5% BSA (Figure 3; supplemental Figure 1B, supplemental Videos 3-4). At 50 mM, washed erythrocytes and platelets are slowest to aggregate, and proteins form sheet-like structures that appear to enhance cell binding and aggregation (Figure 3B). As the concentration of FeCl₃ increases, the time to aggregate initiation decreases, and the time to aggregate stabilization generally decreases (Figure 3A). Full channel occlusion and flow cessation occur at 1 M FeCl₃ in the absence of endothelial cells for all blood components.

The universal aggregation effect of FeCl₃ on blood cells and proteins in the absence of endothelial cells led us to posit a charge-based mechanism for aggregation (Figure 4A). Blood is a colloidal suspension of negatively charged cells and proteins; the stability, or resistance to aggregation, of colloidal particles in suspension is determined by the particle ζ potential, where a higher ζ potential corresponds to increased stability.¹⁶  In aqueous solutions, ferric ions undergo continuous hydrolysis, complexation, polymerization, and precipitation.¹⁷  At equilibrium, over a range of pH and FeCl₃ concentrations, a number of species are present, predominantly Fe³⁺, Fe(OH)²⁺, and Fe₂(OH)₂⁴⁺.^18,19  Fe³⁺ ions can bind directly to negatively charged proteins (both soluble and cell surface localized), resulting in destabilization of those proteins with respect to flocculation, and furthermore resulting in cell aggregation.²⁰  The cationic, amorphous hydroxide precipitates also deposit on negatively charged surfaces to form a large particulate matrix that entraps cells.^18,21  Aggregation of colloidal components in the presence of metal salts proceeds by a combination of factors: an increase in ionic strength that reduces repulsion between similarly charged particles; adsorption of ions that neutralize charge and allow for particle bridging; and cell entrapment in a hydroxide precipitate.^17,21  Interestingly, both FeCl₃ and aluminum (III) chloride (AlCl₃) are used in wastewater treatment to induce aggregation of colloidal suspensions of negatively charged toxins and biologics.^21,22 

In support of our hypothesized mechanism, introduction of Al³⁺, an ion similar to Fe³⁺ with respect to binding,²⁰  results in erythrocyte aggregation in microchannels (Figure 4B). A recent publication likewise showed that the in vivo application of AlCl₃ to a mouse carotid produces occlusive thrombosis.²³  On the other hand, erythrocyte suspensions do not aggregate when exposed to Cr³⁺ because the ion has slow substitution kinetics and preferentially binds water instead of negatively charged species (Figure 4C).^16,20  This suggests that to induce aggregation, a metal ion must be able to form inner sphere complexes (bind directly without intervening water molecules) with negatively charged moieties to form stable aggregates. Furthermore, exposure to both AlCl₃ and chromium (III) chloride (CrCl₃) results in endothelial cell death, suggesting that cell death alone is insufficient to initiate aggregation (Figure 4B-C).

Blood components aggregate in static conditions when exposed to FeCl₃. From 0.5 mM to 50 mM FeCl₃, FeOH precipitates are visible and RBC membranes appear deformed; at concentrations >50 mM, proteins form sheet-like structures and RBCs form large aggregates (Figure 5A). Both FeCl₃ and AlCl₃ cause aggregation of recalcified PRP and PPP in the presence of FeCl₃ as measured by dynamic light scattering, with aggregation generally increasing with increasing concentrations (Figure 5B-C; supplemental Figure 2). Higher concentrations of AlCl₃, relative to FeCl₃, are necessary to initiate aggregation in all solutions; this is expected as AlCl₃-induced colloid aggregation occurs over a narrow concentration and pH range.²¹  At concentrations of FeCl₃ > 500 mM and AlCl₃ > 1 M, the suspensions form a solid mass (supplemental Figure 2). Light-scattering measurements at these conditions are complicated by the nonergodic nature of the solution; that is, the size of the aggregates is large compared with the incident beam, so as the aggregates precipitate out, the solution is not temporally or spatially homogenous, and light scattering measurements are inaccurate.²⁴  Measurements of aggregation of washed RBCs are likewise hampered by the solution’s nonergodicity, but microscopic and macroscopic examination show definitive aggregation, with solid masses forming at >50 mM FeCl₃ (supplemental Figure 2B). As expected, CrCl₃ does not induce aggregation, and there is no increase in light scattering over the controls at any tested concentration (Figure 5B-C).

The presence of physiologic levels of divalent cations, Mg²⁺ and Ca²⁺, does not have a pronounced effect on aggregation parameters, as there are minimal changes in light scattering between samples with and without these cations (supplemental Figure 3A). After 20 minutes, recalcified PRP forms a stable, intrinsic coagulation pathway-derived clot with a sixfold increase in light scattering over the control. Light scatter of recalcified PRP solution exposed to 500 μM FeCl₃ increases a further twofold after 20 minutes, increasing the scatter to that of the physiological clot (supplemental Figure 3B). This suggests that physiological clotting can occur in addition to metal ion–based aggregation.

Our findings suggest that initial adherence and aggregation of blood components in this model is due to inner-sphere complexation of Fe³⁺ to plasma proteins and cell membrane–localized proteins, as well as binding of cells to FeOH precipitates. Sheet-like protein aggregates also act as Fe³⁺ linkers that enhance cell binding (Figure 4A).

Although the underlying mechanisms of the FeCl₃ thrombosis model remain in question, its utility as an in vivo model to study clotting is well established. Numerous publications have shown that FeCl₃-induced vascular occlusion is mitigated in knockout mice^10,25,26  and in the presence of antithrombotic agents.^27-31  Multiple platelet receptor defects have also been shown to modulate thrombus formation in the FeCl₃ injury model.³²  We monitored occlusion in both bare and endothelialized channels when blood was treated with aspirin (platelet inhibitor), heparin (thrombin inhibitor), hirudin (thrombin inhibitor), and eptifibatide (GPIIb/IIIa inhibitor), as well as blood reconstituted with von Willebrand factor (VWF)–free plasma obtained from a patient with type 3 von Willebrand disease. After 15 minutes, in both endothelialized and bare channels, all inhibitors significantly reduced occlusion (P < .05) when blood was exposed to 50 mM FeCl₃; alternatively, when 500 mM FeCl₃ was used, aspirin, heparin, and hirudin could not maintain vessel patency (Figure 6A-B; supplemental Figure 3). In control conditions, RBCs adhere at the FeCl₃/blood interface, platelets stick to adherent RBCs, and a protein matrix forms; additional cells are entrapped within the growing protein/cell matrix and the dense aggregate expands across the full width of the channel. The aggregate is finally bordered by a layer of fibrinogen that “seals” the clot from blood flow, and flow within the channel ceases within 3 minutes (supplemental Figure 4). Cells enter the aggregate mass, but do not traverse it, and blood flow is restricted to the portion of the channel distal to the FeCl₃ inlet. Alternatively, at 50 mM FeCl₃, all inhibitors except aspirin prevent the extensive growth of aggregates past the interfacial line; aggregates do not have a large protein component and cells traverse the loose aggregate matrix, adhere transiently, and then resume transit through the channel. In one heparin experiment, aggregates initially span the channel, but blood flow dislodges the blockage and the vessel remains patent (supplemental Video 5). Aspirin-exposed blood, on the other hand, gives rise to disjointed, unstable protein/cell aggregates that readily embolize in response to fluid shear and lodge downstream (Figure 6C). When the concentration of FeCl₃ is increased to 500 mM, aggregate formation is extensive in all conditions. Microchannel patency is partly retained in eptifibatide and VWF-free conditions because of the loose aggregate matrix, though blood flow is visibly decreased. At increasing FeCl₃ concentrations, pharmacologic inhibition does not effectively prevent channel occlusion because, although the clotting cascade is compromised, the FeCl₃ concentration is sufficiently high across the channel to induce charge-based aggregation.

Furthermore, FeCl₃-induced aggregation in blood from wild-type mice was compared with that of mice lacking the extracellular domain of GPIb-α (IL4-R/Iba mice). As expected, wild-type blood exposed to 50 mM FeCl₃ forms aggregates that fully occlude the channel in a manner similar to human blood (Figure 6D). GPIb-α–deficient blood results in significantly reduced occlusion with vessel patency maintained after 15 minutes. Unlike the extensive, continuous dense cell aggregate in the wild-type condition, IL4-R/Iba mouse blood forms disjointed aggregates that comprise regions of dense RBCs and sparse RBCs, allowing blood flow to continue through and around the aggregates (Figure 6D; supplemental Videos 6-7). The aggregates are primarily composed of RBCs as previously shown in vivo.¹⁰ 

The marked effects that mouse genotype and pharmacologic agents have on FeCl₃-induced thrombus formation are congruous with our proposed mechanism of FeCl₃ action, and our findings help to explain inconsistencies in thrombus formation between injury models, as well those between studies using the same injury model. Although the initial adhesion and aggregation of blood proteins and cells is mediated by concentration-dependent, charged-based interactions between FeCl₃ and blood components, the progression to complete occlusion occurs via conventional clotting mechanisms involving platelet activation and the coagulation cascade. Similarly, Eckly et al’s findings suggest that initial platelet adhesion in the FeCl₃ model is atypical of the clotting cascade as there is not extensive collagen exposure in the vessel.⁹ 

The extensive use of the FeCl₃ model in hematologic research and the lack of a well-defined mechanism of action led us to take a reductionist, yet interdisciplinary, microfluidic approach to study FeCl₃-induced thrombosis. Our study reveals that FeCl₃ has multiple dose-dependent effects on blood, in the presence and absence of endothelial cells, leading us to a novel 2-phase mechanism of action. In the first phase, the charge-based complexation of positively charged iron species and negatively charged proteins (cell surface bound and in solution) initiates FeCl₃-induced thrombosis independent of biological ligand-receptor interactions. RBCs are the first adherent blood component, and form the bulk of the initial aggregate. Platelets and proteins are visible within the initial aggregate to a lesser degree (supplemental Figure 4A). We also recreate the findings that the extent of thrombus formation in this model is dependent on physiological clotting processes by introducing anticlotting agents and blood with phenotypic clotting deficiencies. This leads to the second phase of our proposed mechanism in which endothelial cell death and iron-bound, adherent blood components initiate the conventional biological clotting cascade. The final clot is RBC, platelet, and protein rich and expands across the width of the channel (supplemental Figure 4B). The inability to tightly control the FeCl₃ concentration in vivo makes the distinction between chemically induced aggregation and physiological clotting impossible to assess. Likewise, there may be multifactorial effects from drugs such as heparin, whose high negative charge density likely affects the activity of FeCl₃.

RBCs, platelets, PRP, and PPP all aggregate when exposed to FeCl₃, tending toward increased aggregation as concentration increases. Aggregation is visible both macroscopically and microscopically, independent of endothelial cells, and with and without shear or mixing. FeCl₃ is known to cause aggregation of colloidal solutions by ζ potential reduction, by directly binding to negatively charged particles, and by forming amorphous cationic hydroxide precipitates that can entrap and bind negatively charged particles.²¹  Furthermore, the binding of FeCl₃ to negative proteins on blood cell surfaces, known as the “colloidal iron method,” was historically used to determine the distribution of anionic cites on cells, such as RBCs and endothelial cells.^33-35  Aqueous iron species (hydroxides) precipitate on the surface of endothelial cells, as seen in our in vitro studies, and as corroborated by an in vivo FeCl₃ mechanism study in which SEM revealed endothelial cell localized hydroxide deposits.³⁶  We propose that RBCs are the first and most numerous cell species in the initial FeCl₃-induced aggregate due to their membrane’s high negative charge density (sialic acid on the surface) and the concentration of erythrocytes in blood. Barr et al’s findings that RBC binding in the FeCl₃ model occurs at arterial, venous, and mesenteric shear rates, is well explained by charge-based binding.¹⁰  In the absence of RBCs, platelets too form aggregates on the channel surface, though interactions are more transient and occlusion proceeds more slowly. FeCl₃ is an oxidizing agent, and at high concentrations is a strong Lewis acid; this may contribute to the concentration-dependent endothelial cell death seen in our system. The acidity alone, however, does not cause cell aggregation: introducing acetic acid at pH 1 in endothelialized devices causes visible endothelial cell death but not RBC aggregation (not shown). The “charge-based” mechanism we present is further supported because an ion with similar binding activity, AlCl₃, also causes cell aggregation, and in vivo thrombus formation.²³  Inert CrCl₃, however, causes endothelial cell death, but not cell aggregation.

At low concentrations of FeCl₃ and high drug concentrations, extent of aggregation is mitigated. Aspirin-treated blood gives rise to an unstable, frequently embolizing thrombus; heparin and hirudin reduce the area of the channel covered by aggregates; eptifibatide leads to a loose matrix; and VWF deficiency increases time to aggregation and results in unstable occlusion. As seen in vivo, increasing the FeCl₃ concentration can overcome the protective effect of these agents: there is a narrow range of FeCl₃ concentrations for which there are differences between the treated and untreated conditions.

The dependency of thrombus initiation and extent of occlusion on FeCl₃ concentration is well explained by our proposed mechanisms: there is a critical concentration of FeCl₃ required at the lumen for aggregate initiation and propagation. Furthermore, the differences between occlusion in varying injury models and vessel types³²  is well explained, as the sensitivity of the system to FeCl₃ concentration is in part due to the intraluminal mass transport of FeCl₃, which is a function of blood velocity and vessel diameter; simulations show that intraluminal flow conditions determine the percentage of the vessel that is exposed to FeCl₃ concentrations that induce aggregation (>0.5 mM) (supplemental Figure 5). The role that mass transfer plays in the FeCl₃ thrombosis model explains the variability in application techniques and results.

Full venous occlusion by FeCl₃ in mice lacking the extracellular domain of GPIbα led researchers to propose that there is another ligand for VWF at venous shear rates; and similarly unusual RBC/platelet/ligand interactions have been proposed to explain unanticipated findings when using FeCl₃.^10,11,25,37  Our findings suggest that initial blood aggregation is independent of biological ligand/receptor interactions. Furthermore, in small vessels, the effective concentration of FeCl₃ is higher over a larger area of the channel and charge-based binding is likely more extensive (supplemental Figure 5). We surmise that vessel wall composition and thickness, which vary by vessel type as well as by sex, age, weight, and genotype of mouse, will also affect the intraluminal FeCl₃ concentration. However, investigation of these variables is beyond the scope of this article.

Our concentration-dependent mechanisms suggest that researchers should take caution when using the FeCl₃ injury model and that the use of this model is not appropriate for studies in clot initiation, namely platelet-endothelium interactions. Likewise, FeCl₃ concentration should be expressed as a molarity, as opposed to a percentage, for accurate comparisons between studies using hexahydrate and anhydrous FeCl₃ species. FeCl₃ should be also be reconstituted/diluted directly before use as spontaneous FeOH formation may change the diffusion properties of FeCl₃, and thus its thrombotic activity, especially at low concentrations (high pH) where FeCl₃ is unstable.

The controlled reductionist environment afforded by our in vitro endothelialized microfluidic system allows for precise control over FeCl₃-induced thrombosis and led us to propose a novel, 2-phase mechanism for the action of FeCl₃ in thrombus formation. The first phase is independent of typical biological “clotting,” and consists of the FeCl₃ concentration-dependent binding of negatively charged blood cells and proteins to positively charged iron species. The adherent cells from the first phase initiate the second phase, the standard biological clotting cascade. Where one phase ends and the other begins is nontrivial to determine, as the effective FeCl₃ concentration in the channel (which determines the extent of the first phase) is dependent on mass transfer phenomena. Researchers should take caution that this will affect their findings, especially in small vessels, and that aggregation phenomena that seemingly suggest a novel ligand/receptor interaction may in fact be due to nonspecific charge-based binding effects.

This work was supported by National Science Foundation (NSF) CAREER award 1150235 (W.A.L), and National Institutes of Health, National Heart, Lung, and Blood Institute grants U54-HL112309 (J.C.C. and W.A.L.), U01-HL117721 (W.A.L.), and R01-HL121264 (W.A.L.).

Contribution: J.C.C., R.L., L.A.L., S.M., C.G., and W.A.L. contributed to the conception, design, analysis, and interpretation of the research; J.C.C. and W.A.L. wrote the manuscript; J.C.C., Y.S., M.E.F., D.R.M., and B.H. performed research and collected and analyzed data; J.B.D. performed diffusion calculations; and all authors approved the final version of the manuscript.

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

Correspondence: Wilbur A. Lam, Georgia Institute of Technology, 345 Ferst Dr NW, Atlanta, GA 30318; e-mail: wilbur.lam@emory.edu; wilbur.lam@bme.gatech.edu.