Abstract
Fluorescence-based sensing systems offer potential for noninvasive monitoring with implantable devices, but require carrier technologies that provide suitable immobilization, accessibility, and biocompatibility. Recent developments towards this goal include a competitive binding assay for glucose that has been encapsulated in semipermeable microcapsule carriers. This paper describes an extension of this work to increase the applicability to in vivo monitoring, wherein two significant developments are described: (1) a near-infrared resonance energy transfer system for transducing glucose concentration, and (2) novel hybrid organic-inorganic crosslinked microcapsules as carriers. The quenching-based assay is a competitive binding (CB) system based on apo-glucose oxidase (AG) as the receptor and dextran as the competitive ligand. The encapsulated quencher-labeled dextran and near infrared donor-labeled glucose receptor showed a stable and reversible response with tunable sensitivity of 1–5%/mM over the physiological range, making these transducers attractive for continuous monitoring for biomedical applications.
1. Introduction
It is well known that frequent monitoring of glucose concentrations and appropriate countermeasures can help in achieving euglycemia and minimizing secondary complications of diabetes [1–5]. Optical sensors enabling glucose testing without any need for blood extraction are an attractive alternative to the current standard of painful finger-stick testing procedures [6–8]. In particular, implantable fluorescent sensors, which can be embedded in the highly vascularized dermis (<1 mm from the skin surface) and remotely monitored using light, would be an ideal solution [9–11]; such sensing mechanisms have been intensely pursued in recent years using techniques based on fluorescence spectroscopy due to the inherent high sensitivity [10–18]. A viable approach to noninvasive yet specific monitoring of clinical biomarkers would also be useful in the management of a large number of conditions where dosage of therapeutic agents is adjusted based on biomarker levels.
While several approaches for fluorescence-based glucose sensing have been developed using Concanavalin A [12, 19–25] and boronic acid receptors [26–32], their use remains complicated by concerns over toxicity, nonspecific and irreversible binding, and response ranges that do not match physiological levels. We recently developed a novel competitive-binding (CB) assay that employs an inactive form of the enzyme glucose oxidase (apo-GOx) as the glucose-binding protein and uses fluorescence resonance energy transfer (RET) to transduce binding [33]. Apo-GOx is highly selective to -D-glucose (5–10X, relative to -D-glucose, D-mannose, and sucrose), suspensions of capsules respond in less than 1 minute to glucose titrations, and the CB assay is responsive in the clinically relevant 0–30 millimolar (~0–600 mg/dL) range [15, 33–35]. In addition, GOx has been designated as “GRAS,” generally regarded as safe [36].
We also developed a carrier system that provides accessibility to the analyte while preventing movement of the molecules outside the region of intended use [15, 37–41]. This general concept of entrapping glucose sensing chemistry in biocompatible microspheres is an attractive approach to exploiting the sensitivity of fluorescence spectroscopy and the specificity of molecular recognition agents such as proteins for in vivo use. We pioneered the use of semipermeable polyelectrolyte microcapsules [42–45] for this purpose, because they are particularly well suited for the requirements of multimolecular assays. In this approach, the use of microcapsules enables the free movement of the ligand, receptor, and analyte during the process of competitive displacement, while the semipermeable capsule walls entrap the larger sensing assay elements in the capsule interior, yet allow free glucose transport through capsule walls. The feasibility of using the encapsulated CB assay as a method of glucose monitoring has been demonstrated using multiple RET pairs, including green/orange [15] and orange/red [35] systems; the ability to translate the transduction system to different wavelength regions is particularly advantageous to the current study. The encapsulation was shown to be stable [15] and thus the useful operating lifetime of the microcapsule assays will be determined by photobleaching and protein denaturation kinetics; investigation of these processes, along with quantitative assessment of response time, is currently underway. It is noteworthy that this assay has also recently been demonstrated to function within polyelectrolyte-coated alginate microspheres [46].
While these previous reports represent necessary steps toward a viable monitoring system, a number of additional obstacles to clinical acceptance remain. Three improvements that would help move toward the ultimate goal are an increase in the operating wavelengths, a more facile capsule fabrication and encapsulation process, and a stable capsule formulation. Longer wavelengths will enhance penetration of light through tissue. However, while a large selection of organic dyes and inorganic materials have been identified and used in RET systems [47], the broad excitation spectrum of available NIR dyes makes it difficult to find an efficient RET pair for the NIR region. We note that it is also desirable to match the donor excitation spectrum to the emission of an inexpensive light source, such that low-cost monitoring technology may be employed. The red spectral region is attractive for this purpose, due to the wide availability of excellent red LED and laser diode sources.
These problems were solved in this work by (1) converting the transduction scheme from a system in which the acceptor emits photons to one in which the acceptor is deexcited through nonradiative pathways, where a near-infrared dye (Alexa Fluor 647) is attached to the glucose receptor and a broadband quencher (QSY21) is attached to the competitive ligand, and (2) encapsulating the assay in hybrid polymer-silica microshells. A number of different microcapsules and construction strategies with controllable shell thickness and properties have been proposed based on the versatile LbL self-assembly process [44, 48–51]. Variations of these have been demonstrated for the encapsulation of various macromolecules, such as proteins [39, 52, 53], proteins/polysaccharide complexes [15], and enzymes [39, 41, 54–58]. While several encapsulation methods were based on controlling the permeability of the capsule wall [51, 59, 60], precipitation techniques [61, 62], or polymerization of monomers in microcapsules [41], we have shown that microcapsules with walls comprising photosensitive polymers can be used for highly efficient and stable encapsulation of macromolecules [39]. The disadvantages of this technique, however, include the use of toxic polymers, inconvenient assembly conditions that require protection from UV light prior to UV-initiated crosslinking, and relatively low stability of the polymeric shells during manipulation. While inorganic silica microcapsules and smart organic/inorganic microcapsules have also been proposed [42, 63, 64], the processes required to realize these are harsh and tedious, requiring exposure to strong acidic/basic conditions, high temperatures, and UV light, all of which pose potential hazards to biological molecules. Therefore, in this work, we explored the possibility of forming hybrid organic/inorganic microcapsules and encapsulating established sensing reagents into the capsules with milder processes. Specifically, we describe the modification of poly(allylamine hydrochloride) (PAH) with glycidyl-silane, such that the resulting PAH-silane conjugate could be alternately adsorbed with poly(styrene sulfonate) (PSS), to obtain PSS/PAH-silane microcapsules that slowly crosslink. When the capsules are suspended in a solution containing the macromolecular sensing reagents, the macromolecules diffuse inside and then are trapped as the crosslinking occurs.
This paper describes the investigation of an NIR glucose sensor comprised of competitive binding assay encapsulated in hybrid microcapsules, and compares the performance characteristics with those obtained for the FITC/TRITC and TRITC/Cy5 energy transfer pairs. We also report a novel method for the encapsulation of macromolecules into silane-modified PSS/PAH microcapsules that eventually form organo/inorgano hybrid microcapsules. The hollow silane-based microcapsules slowly form an interpenetrating silica network due to the hydrolysis and subsequent condensation of silane. While the thorough characterization of the hybrid microcapsules will be described in a separate report, the details of the materials and methods used to fabricate these unique structures are provided here for completeness.
Assay Description
A schematic of a
glucose sensor based on this quenching mechanism is shown in Figure 1. The dyes used
in this sensing mechanism are Alexa Fluor (AF) 647, QSY21, and AF750. AF647 emits in the range of 650–720 nm, which
overlaps with the QSY21 absorbance spectrum. Therefore, when AF647 and QSY21
are in close proximity, QSY21 significantly quenches the fluorescence of AF647.
As QSY21 is not fluorescent, there is only one fluorescent peak, preventing
ratiometric analysis of the data. A second near-infrared dye (Alexa Fluor 750)
is integrated into the walls of the microcapsules, acting as a reference
fluorophore for ratiometric monitoring; AF750 is weakly excited at 640 nm but is
not quenched by QSY21. Thus, when apo-GOx tagged to AF647 is exposed to QSY21-dextran
(QSY-dex), they will be in close proximity due to the binding affinity between
apo-GOx and dextran. This results in low emission intensity at 664 nm due to the
quenching of AF647 by QSY21. The addition of glucose results in the
displacement of dextran from apo-GOx, decreasing the quenching of AF647 and
increasing the emission at 664 nm (Figure 1). Thus, the
glucose sensitivity of the AF647-apo-GOx (AF-AG)/QSY-dex complexes entrapped in
AF750-labeled microcapsules can be determined by measuring the changes in
fluorescence intensities of AF647 relative to AF750.
Theory
The homogeneous
energy transfer assay described above can be mathematically described by
expressions for the competitive binding reactions between the immobilized
receptor () and ligand ().
The corresponding association constants ( & )
are given as where and is the ligand/receptor complex. Similarly, with analyte (), the binding reaction is described by
where and is the analyte/receptor complex. By
considering the concentrations of free (unbound) acceptor-labeled receptor,
donor-labeled ligand, and analyte, respectively (subscript ‘‘’’ denotes total concentrations), a
final equation to describe the entire system is ,
which can also be written in simplified, dimensionless form as: , where and give the ratio of free
to total ligand and dimensionless analyte concentration, respectively [65]. Dimensionless analyte concentration depends
upon the analyte concentration and receptor-analyte affinity compared to the
concentration and affinity of the competing ligand. Since the amount of unquenched receptor () is of interest
in this study, this quantity can be found as the sum of the free receptor and
analyte-bound receptor concentrations . This allows solving for the measured
quantity, relative amount of fluorescent receptor .
A key aspect of
this system is the dependence of the response on different concentrations of
assay components, in both absolute and relative senses. Using the equations above,
the graphs in Figure 2 were produced to illustrate the effect of these
parameters. Clearly, the response shifts
dramatically when the ratios of ligand to receptor or absolute reagent concentrations are altered; specifically,
as is increased, the predicted response magnitude increases while responding over
the same range. With dropping values,
the response range shifts to lower concentrations and a loss of sensitivity is
observed, due to the incompletely quenched receptor fluorescence in the absence
of analyte resulting from insufficient quenching ligand. Furthermore, the range over which the
response is observed varies directly with absolute assay concentration. From these models, the concentrations of the
assay elements needed for any application can be estimated based on the
required detection range if association constants are known, and the effects of
inhomogeneity in capsule populations on the sensor response can be predicted.
The response of the sensing system can be
characterized by the effective dissociation constant of the sensing assay, also
referred to as the concentration at which half of the full response is observed
[66]. Clearly, this quantity shifts to
larger values with increasing assay concentration. To predict the effects of
increasing assay concentration on the response to analyte binding, realistic
values for ligand and receptor concentrations and their respective association
constants were used ( nM; nM; ; )
to calculate the percentage change in unquenched receptor concentration versus
analyte concentration. It is even more
obvious from this linear plot (see Figure 3) that the magnitude of the response is greatly
enhanced with an increase in assay concentration, and the range over which the
response changes appreciably due to additional analyte increases as well.
(a)
(b)
2. Experimental Details
2.1. Materials
Glycidyl 3-(trimethoxysilyl)propyl ether (glycidyl-silane), GOx (G-2133), sodium poly(styrene sulfonate) (PSS, MW ~1 MDa), poly(allylamine hydrochloride) (PAH, MW 70 kDa), -D-glucose, sodium bicarbonate, dimethyl formamide, ammonium sulfate, and sodium acetate buffer were obtained from Sigma. Succinimidyl ester forms of Alexa Fluor 647 (AF647), Alexa Fluor 750 (AF750), and QSY 21 (QSY21) (Invitrogen) were used to label apo-GOx, PAH, and amino-dextran (500 kDa, Invitrogen), respectively, using a standard amine labeling procedure (Invitrogen). All reagents were used as received. (5 μm) particles were prepared as previously described [40].
2.2. Instrumentation
A UV-Vis spectrometer (Perkin Elmer Lambda 45) was used to collect absorbance spectra. The slit size (4 nm) and scanning speed (480 nm/min) were held constant throughout all the experiments. A scanning fluorescence spectrometer (QM1, Photon Technology International) with an extended-wavelength PMT (R928) was used to collect fluorescence emission spectra while exciting the sample at 640 nm. Confocal images were taken with a Leica TCS SP2 microscope equipped with a 63X oil immersion objective and a red He-Ne excitation laser. Counts and sizes of microcapsules were obtained with a Beckman Coulter counter (Z2) using a 100 μm aperture. The assembly of polyelectrolyte multilayers on colloidal templates was monitored on template particles by measuring the surface potential after the deposition of each polyion; electrophoretic mobility measurements were performed using a ZetaPlus Zeta Potential Analyzer (Brookhaven Instrument Corp.).
2.3. Methods
2.3.1. Solution-Phase Experiments
Assay Optimization
The
glucose assay concentration was optimized by performing a simple titration
experiment to obtain maximum change in signal for a given analyte
concentration. The concentrations of the sensing assay elements were optimized
based on the concentration-dependent quenching of AF647 by QSY21. To observe the
quenching behavior and confirm RET, aliquots of 700 nM
QSY21-dextran (QSY-dex) solution were titrated into 17.5, 35, and 70 nM AF647-AG
(AF-AG) in a stepwise manner, followed by measurements of fluorescence emission.
Glucose Sensitivity
The apo-GOx and
amino-dextran used in all experiments were labeled with AF647 and QSY21 at ratios
of 3.2 and 2.4, respectively. The quenching process between AF-AG and QSY-dex
was monitored using the fluorescence spectrometer by exciting the sample at 640 nm and collecting the emission across the range of 650–750 nm. Initially, ~16.5 picomoles of AF647-AG were added to 0.4 mL of DI water. Based on quenching studies,
the initial concentration of QSY-dex for 41 nM AF-AG was selected to be 143 nM. To assess the relative
affinity of apo-GOx for glucose and QSY-dex, recovery in the AF647 emission was
observed during the stepwise addition of aliquots of 500 mM -D-glucose solution
into the sample containing QSY-dex/AF-AG complexes. The effect of total QSY-dex
concentration on the displacement behavior was investigated by repeating the
same procedure at different concentrations of QSY-dex. Fluorescence changes in AF647 peak were
corrected for dilution effects by titrating QSY-dex/AF-AG complexes with DI
water in parallel experiments, then subtracting the change due to water from
that due to glucose. For all the experiments, sensitivity curves were
constructed by plotting the dilution-corrected emission intensity at 664 nm versus glucose
concentration.
2.3.2. Fabrication of Hybrid Microcapsules and Encapsulation of Sensing Reagents
Microcapsule Fabrication
The PSS and PAH solutions used for
building multilayer films were prepared in DI water at 50 mM (concentration based
on monomer). For LbL assembly of PSS/PAH-silane, 20 μL of glycidyl-silane was added into 1 mL of PAH
solution. Multilayer films of PSS/PAH-silane films were assembled on the
surface of 5 μm particles
(Figures 4(a) and 4(b)) by suspending the particles in polymer
solutions for 15 minutes, followed by three centrifugation and rinse steps to
remove unbound polymer [40]. The film architecture on the surface
of the templates at the completion of the assembly was /PSS. To confirm charge reversal, surface
potential measurements were completed after deposition of each polymer
layer. Following the completion of polymer layer deposition, core particles were dissolved (Figure 4(c)), yielding 5 μm inner diameter silane-modified
hollow microcapsules.
Loading of Sensing Reagents
Microcapsules
with the /PSS architecture were dispersed in a mixture
of AF-AG and QSY-dex (230 and 990 pM, resp.) (Figure 4(d)). The microcapsules
were incubated in the loading solution for two days to allow sufficient time for
the hydrolysis and condensation of silane, facilitating the formation of a denser
and less permeable interpenetrating network (Figure 4(e)). The
capsules were rinsed in DI water to remove unencapsulated assay molecules,
yielding microcapsules loaded with AF647-AG
and QSY21-dextran (Figure 4(f)). Finally,
the reference dye was incorporated into the microcapsule walls by coating the
outer layer (PSS) of microcapsules with PAH-AF750 via electrostatic self-assembly.
Confocal microscopy was then used to
assess the loading of assay elements into microcapsules. Leaching of the loaded assay elements was determined
by measuring absorbance and fluorescence from capsule supernatant five weeks after
encapsulation. A standard solution of 50 nM AF-AG was used to
estimate the supernatant concentration as a percentage of a standard.
2.3.3. Encapsulated Sensor Response Tests
Sensor Response in Microcapsules
The AF750-labeled microcapsules loaded with AF-AG and QSY-dex were
suspended in DI water, and an initial fluorescence spectrum was collected. Glucose
solution (500 mM) was then titrated into the microcapsule suspension, as was performed for the solution-phase studies.
The change in the relative emission intensities at 664 and 780 nm was calculated
for each spectrum and plotted with respect to glucose concentration. To confirm
theoretical prediction on the effect of receptor/competitive ligand
concentration and to evaluate the possibility of tailoring the sensor response,
these experiments were repeated with four different concentrations of capsules,
(, ,
, capsules/mL). All experiments were repeated three times,
and calculated changes in ratio values were pooled for statistical analysis.
Reversibility
To test the reversibility of the
microcapsule sensors, exposure to glucose in random order with respect to
concentration was performed. To achieve
this, microcapsules were dispersed in DI water ( capsules/mL)
and the glucose concentration of the suspension was increased by addition of
glucose stock solution. The glucose level was decreased by soaking the capsules
in DI water to remove the glucose after each measurement. In each case, after adding
glucose to the desired concentration and measuring the fluorescence, the
suspension was centrifuged three times, the supernatant (glucose solution) was
removed, and an equal volume of fresh DI water was added. The change in AF647/AF740
peak ratio was obtained at each step by collecting a fluorescence scan. This
procedure was repeated to cover the concentration range of 0–60 mM, in the
order of 0, 5.3, 0, 13.5, 20, 3.31, 26.6, 32.5, 5.5, 44.6, 6.3, 61.6, 10 mM. The experiment was repeated three times, and
results pooled for statistical analysis.
3. Results and Discussion
3.1. Solution Phase Experiments
Assay Optimization
AF-AG samples
exhibited decreasing emission intensity during the titration of QSY-dex,
following a pattern predicted by equilibrium binding (Figure 5). This
suggests that the quenching is due in part to RET and possibly also due to radiative
processes. By fitting the data collected
from three experiments performed at different concentrations of AF-AG, the for QSY21-dextran was
estimated to be between 50 and 100 nM. These findings prove the efficiency of the RET-based quenching process
due to apo-GOx:dextran association and provide insight into the binding
affinity, which is useful in assay design.
Glucose Response Tests
Taking the optimum assay concentrations obtained from the quenching experiments
into consideration, an assay mixture containing 142 nM QSY-dex and 42 nM AF-AG was
prepared in DI water and subjected to sequential glucose additions. As hypothesized, the added glucose displaced
dextran from apo-GOx, resulting in decreased quenching and correspondingly
enhanced fluorescence emission at 664 nm (Figure 6(a)). Thus, the shift of the assay to longer
wavelengths via the RET-based transduction was successful, and the sensitivity
of the response (0.7%/mM) is comparable to the original system that used
visible dyes [33]. The dilution-corrected emission peak ratio changed
a total of 25% (Figure 6(b)), with a linear response in the physiological range
of 0–30 mM. Figure 6(b) also
contains response curves for varying apo-GOx:dextran combinations, confirming the
change in sensitivity and range for different receptor and competitive ligand
concentrations that was predicted by theory (Figures 2 and 3). These
findings prove that the glucose sensitivity of the assay can be varied over a
wide concentration range (0–200 mM) by
controlling the quencher concentration; however, it is noteworthy that the
sensitivity in the region of interest (0–30 mM) is similar in all the
experiments, because the same relative number of quencher molecules are
displaced due to the same glucose concentration. Thus, the main parameter that
can be varied by controlling quencher concentration is the effective dissociation
constant of the sensing assay, also referred to as the concentration at which
half of the full response is observed [66]. In
these experiments, the was estimated to be 17, 25, 50,
and 83 mM for the assays with 71, 142.5, 285, and 427.5 nM QSY21-dex,
respectively.
(a)
(b)
Encapsulation of AF-AG/QSY-dex Complexes in Microcapsules
The surface
potential of microparticles alternated with the addition of each layer, (Figure 4(g)). This demonstrates
the successful fabrication of the shell on the template particles. Microcapsules
with PSS shell architecture were prepared by core
dissolution and used for encapsulating AF-AG/QSY-dex complexes, followed by
incorporation of the reference dye (AF750). Postloading supernatant
concentrations and capsule counts were used to calculate encapsulation amounts
of and moles/capsule for AF-AG and
QSY-dex, respectively. Release of AF-AG
into supernatant over five weeks was only 0.12%, which represents a 40-fold
decrease in leaching over previous approaches and indicates a stable entrapment
of the receptor and competing ligand. By
confocal microscopy and titration experiments, the fraction of apo-GOx in the
capsule interior was estimated to be 27% of the total, while the other 73% was
immobilized in the capsule walls and believed to be unable to respond to
glucose changes. It will be
desirable to improve the encapsulation process to remove unwanted background
signals and avoid waste; we believe this may be accomplished by, for example,
“presaturating” the nonspecific binding sites in the walls of the capsules with
albumin or other blocking reagent. However, we acknowledge that this may raise other difficulties and will
leave solution of this problem to future studies.
Sensor Response to Glucose Addition in Microcapsules
The intensity of fluorescence at 664 nm
increased with each addition of -D-glucose
solution to the microcapsules loaded
with AF-AG/QSY-dex complexes, as
shown in Figure 7(a). The corresponding percentage change in
the normalized intensity at 664 nm was calculated and plotted versus glucose
concentration, as shown in Figure 7(b). Again, there was no significant drop in sensitivity
over the region of interest (0–30 mM) when
compared to the visible dye systems [15, 35]. The decreased sensitivity
with increased capsule concentration matches the theory and solution-phase
measurements, confirming that the sensitivity can be tailored by changing the
competitive binding assay concentration. There was an increase in the sensitivity when comparing microcapsules to
the solution-phase response for comparable assay concentrations, on the order
of two fold. This is the result of the low concentration of
mobile sensing assay elements, due in large part to the entrapment of ~72% of
the encapsulated molecules in the capsule walls. The immobilized apo-GOx and
dextran molecules may not dissociate in the presence of glucose molecules, and
therefore contribute only a constant background signal to the emission
spectrum.
We note that this
behavior is consistent with previous studies with visible dyes [15, 35], though the total amount and
relative fraction of “available” assay reagents is different due to the
different capsule structure. Specifically, the current system encapsulates ~10X the amount of apo-GOx per capsule compared to
previous systems, which were more efficient (~50% inside/50% in walls) [15]; so, even with the apparently
low efficiency, a larger loading was accomplished with the new approach. This
is further supported by the relative “saturation” point for comparable numbers
of microcapsules: for the new system, this is reached around 35 mM for capsules, while for the previous versions, this was around 20 mM for the same
number [15].
Given that physiological
sensing is the goal, and that pathophysiological glucose levels experienced by
diabetics may range from over 0–30 mM (0–600 mg/dL),
matching the sensor response to this range requires proper selection of reagent
concentration. As can be seen in Figure 7 and Table 1, the range of response can be tuned to match the application by
controlling the amount of receptor and competing ligand in the sample. For example, to match the 0–30 mM range with a
linear response, 37 pmol QSY-dex and 46 pmol AF-AG should be encapsulated.
(a)
(b)
3.2. Demonstration of Reversible Sensor Response
Exposing the microcapsules ( capsules/mL) to random glucose concentrations confirmed the completely reversible nature of the assay without any loss in sensitivity. The changes in relative intensity plotted in Figure 8 also illustrate the consistency and linearity of the response over a wide concentration range. The 0.83%/mM sensitivity matches that observed for the same concentration of capsules exposed to a stepwise increase in glucose concentrations (Figure 7, Table 1).
4. Conclusions
An efficient RET quenching approach was developed as the transduction mechanism for a dextran/apo-GOx competitive binding assay, extending the operating region into the near-infrared. The sensing assay elements (a quencher-labeled dextran, fluorophore-tagged apo-glucose oxidase, and a fluorescent reference dye) were entrapped in microcapsules using a facile and efficient silane-based encapsulation procedure. Microcapsules containing labeled dextran/apo-GOx complexes showed glucose sensitivity of ~1–5%/mM, which is comparable to the original assay operating in the visible region and is among the most sensitive ever reported. Thus, this system is superior to previous iterations, as it possesses the following desirable qualities: (1) the receptor is nontoxic [36], (2) the silane-based encapsulation procedure employed in this system is simple and free of toxic materials; (3) the sensitivity of this system is equivalent or superior to previous embodiments of the microcapsule-based glucose sensors; and (4) a significant increase in the signal levels (signal to noise ratio) will be obtained when used in biological media, due to decreased scattering at the longer wavelengths of light. These improvements make this system for glucose sensing even more attractive as an implantable technology for continuous glucose monitoring.
Acknowledgments
The authors gratefully acknowledge the National Science Foundation (NIRT Grant no. 0210298) and NIH (R01EB000739). Any opinions, findings, and conclusions/recommendations expressed in this material are those of the authors and do not necessarily reflect the view of the National Science Foundation.