Two Etomidate Sites in α1β2γ2 γ-Aminobutyric Acid Type A Receptors Contribute Equally and Noncooperatively to Modulation of Channel Gating

• Etomidate binds to two sites on the γ-aminobutyric acid receptor type A (GABA^A) receptor, but the contribution of each site to etomidate effects remains untested

• Etomidate interactions at each of its GABA^Areceptor sites produce equal and additive gating modulation effects

IN central nervous system neurons, γ-aminobutyric acid type A (GABA^A) receptors are inhibitory receptors, and major targets of general anesthetics such as etomidate and propofol. GABA^Areceptors contain five homologous subunits arranged around a chloride-conducting pore. Each subunit includes a large extracellular domain and four transmembrane domains (M1–M4).1The isoform containing two α1, two β2, and one γ2 is the most common, and displays physiology and pharmacology similar to synaptic receptors.2 

Etomidate is a potent sedative/hypnotic drug that acts by potentiating GABA^Areceptor activity.3,–,5Etomidate enhances α1β2γ2 receptor currents elicited by low GABA concentrations, whereas high concentrations of etomidate directly activate these channels.6,7Structural and pharmacological studies show that both of these effects are mediated by a single class of etomidate binding sites,8and quantitative model analysis suggests that two equivalent etomidate sites are present on each GABA^Areceptor.7A photo-activatable etomidate analog, [³H]azi-etomidate, labels affinity-purified bovine brain GABA^Areceptors at both αM236 (located in the M1 domain) and βM286 (located in the M3 domain).9Etomidate inhibits photolabeling at αM236 and βM286 in parallel, suggesting that both residues abut a common etomidate site. Homology modeling based on the Torpedo  nicotinic acetylcholine receptor structure suggests that both photolabeled residues face intersubunit clefts,9,10and subunit stoichiometry studies indicate that each receptor contains two α/β interfacial pockets.11,12GABA^Areceptor mutations to bulky hydrophobic tryptopan (W) at either α1M236 or β2M286 mimic etomidate's effects by reducing GABA EC⁵⁰, increasing spontaneous activity, and increasing maximal GABA efficacy, while reducing modulation by etomidate.13Together, these results support the hypothesis that etomidate binds within transmembrane interfacial pockets between α-M1 and β-M3 domains, and that there are two etomidate sites on each α1β2γ2 receptor (fig. 1).

Current models of etomidate action assume that the two binding sites have identical structure and contribute equivalently and independently to modulation of receptor gating,7but these features remain untested. Asymmetry could be caused by the different spatial relationship of the γ2 subunit to each α/β interfacial etomidate site, which could induce unequal etomidate effects at each site. Furthermore, either positive or negative cooperative interactions may occur between etomidate sites.

In this study, we used constrained receptor assembly to test whether the individual etomidate binding sites produce equivalent and independent effects on receptor function. Concatenated-subunit wild-type dimer (β2-α1 = D^wt) and trimer (γ2-β2-α1 = T^wt) polypeptides or concatemers containing α1M236W mutations (β2-α1M236W = D^αM236W; γ2-β2-α1M236W = T^αM236W) were coexpressed in Xenopus  oocytes to produce functional GABA^Areceptors with five subunits arranged β2-α1/γ2-β2-α1 counterclockwise when viewed from the synaptic cleft (fig. 1).12,14We compared multiple functional characteristics of all four receptors formed with concatenated dimers and trimers (D^wtT^wt, D^αM236WT^αM236W, D^αM236WT^wt, and D^wtT^αM236W), with particular focus on whether the two single-site mutants, D^αM236WT^wtand D^wtT^αM236W, displayed equivalent versus  nonequivalent properties. To determine if etomidate sites interact cooperatively, we also tested whether adding together D^αM236WT^wtand D^wtT^αM236Weffects fully accounted for the large differences in GABA and etomidate sensitivities between D^wtT^wtreceptors and D^αM236WT^αM236Wreceptors.

Oocytes for electrophysiology were harvested from Xenopus laevis  frogs, as described previously.7,13Animals were housed in a veterinarian-supervised facility and used with approval of the Massachusetts General Hospital Subcommittee on Research and Animal Care, in accordance with federal and institutional guidelines.

Plasmids encoding the concatenated rat dimer (β2-α1) and trimer (γ2-β2-α1) subunit proteins were a gift from Erwin Sigel, Ph.D. (Professor, Institute for Biochemistry & Molecular Medicine, University of Bern, Bern, Switzerland).12,14These plasmids encode polypeptides that insert a 26-residue linker (QQQQQAAAPAQQAAAPAAQQQQQ) between the C-terminus of β2 and the N-terminus of the mature α1 (without its leader sequence). The trimer also contains a 23-residue linker (QQQQQAAAPAQQQAQAAAPAAQQQQQ) between the C-terminus of γ2 and the N-terminus of the mature β2. DNA mutations encoding a tryptophan substitution at residue 236 of the mature α1 sequence (α1M236W) were introduced into plasmids using Quickchange (Stratagene, La Jolla, CA), as described previously,13and confirmed by complete DNA sequencing of the coding regions. The αM236W mutation was chosen because it impacts multiple receptor function parameters (spontaneous activity, GABA EC⁵⁰, GABA efficacy, etomidate modulation, and etomidate direct activation), and we aimed to determine how single etomidate site mutations affect each of these parameters. R-etomidate was obtained from Bedford Laboratories (Bedford, OH) as a 2 mg/ml clinical formulation in 35% propylene glycol:water (v/v). GABA and other chemicals (more than 99% pure) were purchased from Sigma-Aldrich (St. Louis, MO).

Messenger RNA was synthesized on linearized coding DNA templates using commercial kits for both DNA transcription and polyadenylation (Ambion, Austin, TX). After purification, mixtures of messenger RNAs (dimer:trimer molar ratio = 1:1) were injected into Xenopus  oocytes, which were then incubated at 18°C for 1–3 days. All four receptor types formed by combining wild-type or mutant dimers and trimers, D^wtT^wt, D^αM236WT^wt, D^wtT^αM236W, and D^αM236WT^αM236W(fig. 1), were studied electrophysiologically.

Receptor-mediated currents were measured in oocytes at 21–22°C using the two-electrode voltage clamp (Model OC-725, Warner Instruments) technique. Perfusion and data acquisition were computer-controlled using Clampex8.1 software via  a Digidata 1440 interface (both from Molecular Devices, Cupertino, CA). During electrophysiological experiments, oocytes were placed in a 30 μl flow-chamber and perfused at a rate of 3 ml/min with recording buffer ND96 (96 mM NaCl, 2 mM KCl, 0.8 mM MgCl², 1.8 mM CaCl², and 5 mM HEPES, pH 7.5). GABA or etomidate solutions in ND96 were applied until currents reached a steady maximum or displayed desensitization. Wash-out times ranged from 1 min (low GABA concentrations) to 5 min (high GABA or low etomidate concentrations), or 10 min (high etomidate concentrations). In different sets of oocytes, GABA concentration-responses were measured either in the absence or presence of etomidate (3.2 μM, for comparison with previous studies of receptors formed from free subunits).7,13Picrotoxin (2 mM in ND96), a potent GABA^Areceptor inhibitor, was used to assess spontaneous channel activity. Maximal GABA efficacy was estimated in two ways. For etomidate-sensitive channels (D^wtT^wt, D^αM236WT^wt, and D^wtT^αM236W), the response to maximal GABA (1–3 mM) was compared with responses with maximal GABA plus etomidate (3.2 μM). For D^αM236WT^αM236Wchannels that displayed very weak etomidate modulation, alphaxalone (10 μM) was shown to enhance (at least fivefold) currents elicited with low (EC^10–20) GABA. D^αM236WT^αM236Wreceptor currents activated with maximal GABA (0.1 mM) were compared in the same cells with currents elicited with maximal GABA plus alphaxalone. We presume that maximal GABA plus the positive modulators activate all receptors.

Peak currents were digitally filtered (lowpass Gaussian 5–10 Hz), baseline corrected, and measured offline using Clampfit software (Molecular Devices). All currents were normalized to maximal GABA response in the same cell, which was measured at the beginning and intermittently throughout experiments on each oocyte. At least four measurements in different cells were made for each type of experiment and experimental condition. Normalized current data from all cells in each type of experiment were combined for nonlinear least-squares method analysis. Agonist concentration-response curves were fitted to logistic (Hill) functions with variable slope (eq. 1) using Graphpad Prism 5.1 (La Jolla, CA):

Agonist is either GABA or etomidate, EC⁵⁰is the concentration eliciting a current halfway between minimum and maximum, and nH is the Hill slope.

For the three mutated channels, apparent Gibbs free energy shifts (ΔG) for GABA-induced gating were calculated from fitted log (GABA EC⁵⁰) values associated with mutant versus  wild-type (D^wtT^wt) concatenated channels (eq. 2):

R is the universal gas constant (8.314 J/mol · K) and T is absolute temperature.

For all four channel types, etomidate modulation of GABA currents was assessed as the ratio of GABA EC⁵⁰in the presence of 3.2 μM etomidate to control GABA EC⁵⁰. Etomidate modulation was also quantified as changes in Gibbs free energy, using the fitted GABA EC⁵⁰values in the absence and presence of etomidate (eq. 3):

Maximal direct activation by etomidate was assessed as response to 300 μM etomidate, normalized to maximal GABA response. Higher etomidate concentrations produce a second, inhibitory effect on GABA^Areceptors.

Results are reported as mean ± SD unless otherwise noted. Standard errors for fitted log(EC⁵⁰) values were used to calculate 95% CIs for EC⁵⁰s. Comparisons of concentration-responses in the presence and absence of etomidate and for different receptors were performed using the comparison function in the nonlinear fitting module of Graphpad Prism. Statistical comparison of etomidate-induced shifts for different types of channels were performed by calculating the z-statistic from log(EC⁵⁰) differences. Other statistical analyses were performed in Graphpad Prism or Microsoft Excel (Redmond, WA) using one-way ANOVA with Tukey post hoc  test. For pairwise comparisons (e.g. , GABA concentration-responses in the presence vs.  absence of etomidate) we used P < 0.05 to establish significance. For multiple comparisons of each receptor type to three others, we applied Bonferroni correction, and used P < 0.015 to establish significance.

All four messenger RNA combinations of dimers (D^wtor D^αM236W) and trimers (T^wtor T^αM236W) produced functional GABA-responsive receptors in oocytes, with maximal currents typically over 2 μA at a clamping potential of −50 mV (figs. 2A–D). Oocytes injected with only dimer or only trimer messenger RNA produced maximal currents in response to GABA that were at most 10-fold smaller than those from oocytes coexpressing both dimers and trimers. The wild-type dimer and trimer combination, D^wtT^wt, displayed a GABA EC⁵⁰of 36 μM (fig. 2E; table 1). Receptors with mutations in both etomidate sites, D^αM236WT^αM236W, showed a much lower GABA EC⁵⁰of 1.0 μM, (fig. 2F; table 1; P < 0.0001 vs.  D^wtT^wt). The two single-site mutant receptors, D^αM236WT^wtand D^wtT^αM236W, were characterized by indistinguishable GABA EC⁵⁰values: 5.1 and 5.5 μM, respectively (figs. 2G and 2H; table 1; P = 0.53). These values differed significantly from GABA EC⁵⁰s for both D^wtT^wtand D^αM236WT^αM236W(P < 0.0001 for all pairwise comparisons). Plotting all four GABA concentration responses on a single semilogarithmic plot reveals that the midpoints (EC⁵⁰) for both single-site mutant channels are approximately halfway between those for D^wtT^wtand D^αM236WT^αM236Wchannels (fig. 3A). Calculating the Gibbs free energy for GABA EC⁵⁰shifts associated with the different mutated channels relative to D^wtT^wt(eq. 2) reveals that the single-site mutants, D^αM236WT^wtand D^wtT^αM236W, contribute −4.9 ± 0.48 and −4.7 ± 0.48 kJ/mol, respectively, each about half of the energy shift (−8.9 ± 0.52 kJ/mol) calculated for D^αM236WT^αM236Wchannels (fig. 3B).

To assess etomidate modulation of GABA^Areceptor activation, we measured GABA-dependent responses in the presence of 3.2 μM etomidate (figs. 2E–H, open symbols) and for each channel type calculated the ratios of GABA EC⁵⁰s measured in the absence and presence of anesthetic (Left Shift Ratios, table 1). In D^wtT^wtreceptors, etomidate lowered the GABA EC⁵⁰19-fold, from 36 to 1.9 μM (P < 0.0001), whereas D^αM236WT^αM236WGABA EC⁵⁰was reduced only 1.5-fold, from 1.0 to 0.69 μM (P = 0.0065). The EC⁵⁰ratios in D^wtT^wtversus  D^αM236WT^αM236Wwere significantly different (P < 0.0001). Etomidate shifted GABA EC⁵⁰s 3.9-fold, from 5.1 to 1.3 μM, in D^αM236WT^wtreceptors (P < 0.0001), and 4.5-fold, from 5.5 to 1.2 μM, in D^wtT^αM236Wreceptors (P < 0.0001). The two single-site mutant left-shift ratios were not significantly different from each other (P = 0.21), but they differed significantly from values for both D^wtT^wt(P < 0.0001) and D^αM236WT^αM236W(P < 0.0001). Figure 4shows etomidate left-shifts expressed as free energy changes, calculated with eq. 3. This display reveals that adding the gating free energy change associated with etomidate modulation in D^αM236WT^wtand D^wtT^αM236Wreceptors (−3.4 ± 0.49 and −3.7 ± 0.38 kJ/mol, respectively) sum to a value (−7.1 ± 0.62 kJ/mol) that is not significantly different (P = 0.11) from the sum of free energy changes for D^wtT^wtplus D^αM236WT^αM236W(−8.3 ± 0.75 kJ/mol).

In the absence of GABA, etomidate directly activated all four receptors (fig. 5; table 1). Maximal etomidate currents in D^wtT^wtchannels, elicited with 100–300 μM anesthetic, were 16% of currents stimulated with maximal GABA, and the fitted half-maximal activation (EC⁵⁰) etomidate concentration was 61 μM. D^αM236WT^αM236Wreceptors displayed greater sensitivity to etomidate direct activation, with half-maximal activation at 20 μM (P < 0.0001 vs.  D^wtT^wt) and maximal efficacy relative to GABA of 53% (P < 0.0001 vs.  D^wtT^wt). Etomidate activation studies in D^αM236WT^wtand D^wtT^αM236Wreceptors revealed similar etomidate EC⁵⁰s (52 vs.  41 μM; P = 0.38) and similar maximal etomidate efficacies (27 vs.  31%; P = 0.16). The maximal etomidate activation efficacies of the D^αM236WT^wtand D^wtT^αM236Wchannels were both significantly higher than that of D^wtT^wtand significantly lower than that of D^αM236WT^αM236W(P < than 0.01 for all pairs by one-way ANOVA). Etomidate EC⁵⁰s for D^αM236WT^wtand D^wtT^αM236Wchannels were similar to that for D^wtT^wt(P = 0.84 and 0.21, respectively), but significantly higher than that for D^αM236WT^αM236W(P < 0.001).

Spontaneous activity in GABA^Areceptors was assessed using the noncompetitive antagonist picrotoxin. When present, spontaneously open channels produce a resting-state current “leak,” which is inhibited by picrotoxin, resulting in an apparently outward current.7,13,15None of the concatenated GABA^Adimer and trimer combinations displayed measurable picrotoxin-induced outward currents. Given a maximal GABA signal-to-noise ratio of about 1,000 in these experiments, these results indicate that spontaneous activity is less than 0.1% of maximal GABA response (table 1).

The maximal GABA efficacy (table 1) represents the fraction of activatable receptors that open when all agonist sites are bound to GABA. GABA efficacy in D^wtT^wtreceptors was 0.79, based on etomidate enhancement of 3 mM GABA responses (fig. 2E). In contrast, etomidate did not significantly enhance maximal (1 mM) GABA responses in either D^αM236WT^wtor D^wtT^αM236Wreceptors (figs. 2G and 2H), indicating maximal GABA efficacies near 1.0. In D^αM236WT^αM236Wreceptors, which displayed low sensitivity to etomidate, alphaxalone was a strong allosteric enhancer at low GABA, but did not significantly enhance responses to maximal (0.1 mM) GABA (not shown). Thus, we infer that the maximal GABA efficacy in D^αM236WT^αM236Wchannels is near 1.0. All mutant channels displayed GABA efficacies significantly higher than D^wtT^wt(P < 0.001 for all pairs by one-way ANOVA).

We coexpressed concatenated GABA^Asubunit β2-α1 dimers and γ2-β2-α1 trimers that formed functional α1β2γ2 l receptors containing α1M236W mutations in neither, either, or both of the two etomidate sites formed between α1-M1 and β2-M3 domains. The major finding of this study is that a single α1M236W mutation in either the dimer or trimer etomidate binding site produces receptors with equivalent functional characteristics. With respect to both GABA sensitivity and etomidate modulation, each site contributes about half the effect observed in wild-type receptors with two etomidate sites.

Implicit in our experimental design is the assumption that subunit concatenation itself does not significantly alter the interaction of etomidate with GABA^Areceptors. Concatenated wild-type control receptors assembled from D^wtT^wtdisplayed etomidate sensitivity similar to that previously reported in wild-type α1β2γ2 channels formed with free α1, β2, and γ2 subunits. Etomidate at 3.2 μM reduced GABA EC⁵⁰19-fold in D^wtT^wtreceptors, comparable with shifts ranging from 11-fold to 23-fold previously reported for α1β2γ2 l receptors.7,13,16Etomidate direct agonism in D^wtT^wtwas characterized by EC⁵⁰and efficacy parameters similar to those reported for α1β2γ2 channels by Rüsch et al.  7 

The effects of two α1M236W mutations in the concatenated receptors (D^αM236WT^αM236W) also paralleled effects previously reported in α1M236Wβ2γ2 l GABA^Areceptors formed from free subunits.13In D^αM236WT^αM236Wreceptors, GABA EC⁵⁰was reduced 35-fold relative to wild-type (D^wtT^wt), and the GABA EC⁵⁰ratio was 1.5-fold at 3.2 μM etomidate. In receptors formed from free subunits, α1M236W mutations reduced GABA EC⁵⁰20-fold, and GABA EC⁵⁰ratio with 3.2 μM etomidate was 1.7-fold.13Another similarity to studies in GABA^Areceptors with free subunits is that etomidate direct activation in D^αM236WT^αM236Wchannels was more potent and efficacious than in D^wtT^wt. These effects are consistent with previous analysis7,13suggesting that α1M236W mutations produce a loss-of-function with respect to etomidate efficacy, combined with increased basal gating. The former explains the reduced etomidate left shift, whereas the latter underlies the increased sensitivity to orthosteric agonism by GABA and allosteric agonism by etomidate.13In brief, tryptophan mimics the presence of etomidate, increasing GABA sensitivity and efficacy, while also reducing etomidate's effectiveness as a modulator. Although photolabeling indicates that αM236 is near bound etomidate, we do not know precisely how the tryptophan mutation alters etomidate binding interactions. Based on the relative size of methionine and tryptophan sidechains, a steric component is postulated.

Calculating the gating free energy changes associated with one versus  two α1M236W mutations reveals that each mutation contributes equally and additively. Both single-mutant receptors (D^αM236WT^wtand D^wtT^αM236W) displayed indistinguishable GABA EC⁵⁰s and each mutation accounted for approximately half of the GABA gating energy change calculated for D^αM236WT^αM236W(fig. 3; table 1). Similarly, D^αM236WT^wtand D^wtT^αM236Wdisplayed indistinguishable etomidate modulation, with each site contributing approximately half of the total energy change associated with etomidate modulation in D^wtT^wtreceptors (fig. 4; table 1). These experiments also demonstrate that a single etomidate site is sufficient for receptor modulation to occur.

In addition to the results discussed above, we found that etomidate direct activation, spontaneous activation, and maximal GABA efficacy were all indistinguishable in receptors with one α1M236W mutation in either the dimer or trimer (fig. 5; table 1). Thus, the two sites have indistinguishable structures and interactions with etomidate molecules, and the γ2 subunit does not produce significant asymmetrical effects on the etomidate sites.

We can infer from our free energy calculations whether cooperativity between the two etomidate sites is present. If positive cooperativity between two equivalent sites exists, each single-site mutant (with one intact site) would display significantly less than half of the etomidate-induced energy shift observed in wild-type receptors, whereas negative cooperativity would produce converse results. The energy additivity found for both GABA gating and etomidate shift effects indicates that cooperativity between etomidate sites is negligible.

Previous studies using GABA^Asubunit concatemers and loss-of-function mutations for various ligands provide additional evidence for symmetry in transmembrane modulator sites, whereas studies of the extracellular agonist sites reveal some asymmetrical effects.17,18Neuroactive steroids potentiate GABA currents and directly activate receptors at high concentrations, effects that are reduced by mutations at α¹Q241.19In concatenated γ2-β2-α1 trimers and β2-α1 dimers, single α1Q241 mutations produced similar functional effects when introduced into either peptide.17Orthosteric agonists and competitive antagonists bind at two extracellular α/β interfacial sites.1,18Single βY205S GABA-site mutations, which produce profound loss of agonist function, produced asymmetrical effects on GABA and muscimol agonist EC⁵⁰s, and symmetrical effects on bicuculline IC⁵⁰.18These studies also showed that for efficient channel activation, both GABA sites must be occupied, whereas antagonist binding to either GABA site effectively blocks activation.

Some of our studies characterizing D^αM236WT^αM236Wreceptor function revealed differences from α1M236Wβ2γ2 GABA^Areceptors formed from free subunits, which displayed both spontaneous activity and full agonism by etomidate.13Several factors may contribute to these differences. First, our concatenated GABA^Areceptors contain rat subunits, whereas previous studies on α1M236W mutations were performed using human GABA^Asubunits with slightly different amino acid sequences. Second, receptors formed from concatemers are constrained to include a γ subunit, whereas evidence suggests that γ subunits are not consistently incorporated into receptors when free subunits are coexpressed in Xenopus  oocytes.20,21Finally, GABA EC⁵⁰s in concatenated GABA^Areceptors have been reported to be higher than receptors formed from identical free subunits, suggesting that concatenation itself may reduce basal gating.12,22 

In studies using concatenated channel subunit assemblies it is important to verify that linked subunits are not enzymatically cleaved into free subunits, and in the present case, to demonstrate that expression of dimers alone or trimers alone does not lead to significant formation of alternative channel assemblies that might influence results.22Our control experiments showed that expression of dimers alone or trimers alone produced small GABA-responsive currents, with amplitudes less than 10% of those recorded when dimers and trimers were coexpressed. Thus, alternative concatemer channel assemblies are at most a minor contributor to our electrophysiological measurements. Concatemer cleavage leading to assembly of receptors from free subunits also appears to be insignificant, based on these observations. The high Hill slopes observed in GABA concentration responses for D^αM236WT^wtand D^wtT^αM236Wreceptors (figs. 2, 3) is further evidence against concatemer breakdown. If free subunits were produced, mixtures of receptors containing zero, one, or two α1M236W mutations would form, resulting in shallow (low Hill slope) GABA concentration-response curves reflecting the dose-dependent gating effects of the mutation.11 

The experiments reported here did not address whether a single etomidate site is capable of directly activating wild-type GABA^Areceptors. We observed etomidate direct activation in D^αM236WT^wtand D^wtT^αM236Wreceptors, but this is likely because of the combined effects of one intact etomidate site together with basal gating enhancement associated with one α1M236W mutation, which contributes gating energy similar to partial etomidate occupancy. A related study using concatemers combining etomidate-sensitive β2 and etomidate-insensitive β1 subunits in the same GABA^Areceptors suggests that a single β2 subunit is not sufficient for direct activation with etomidate.23This question could be further addressed with concatemer studies using point mutations that selectively and profoundly reduce etomidate sensitivity.

In conclusion, our results show that etomidate interactions at each of its two GABA^Areceptor sites produce equal and additive gating modulation effects. These results support functional models wherein the two etomidate sites in α1β2γ2 l receptors are equivalent and noncooperative.7In terms of clinical relevance, these studies add important details to our understanding of anesthetic mechanisms at their molecular targets, a key step toward developing improved anesthetic drugs.24Of more direct relevance, heteromeric GABA^Areceptors containing both β1 and β2/3 subunits are postulated to exist in the nervous system.23In the presence of etomidate, our results predict that such receptors will experience approximately half the gating energy effect of receptors containing two β2/3 subunits. Future studies using concatenated GABA^Asubunit assemblies will be useful in investigating the linkages between the etomidate and GABA sites on the same subunit, and assessing the mechanisms of other anesthetics, which also likely bind to multiple sites.25 

The authors thank Erwin Sigel, Ph.D., Professor, University of Bern, Bern, Switzerland, for plasmids encoding concatenated GABA^Asubunits, and Aiping Liu, M.S., Senior Laboratory Technologist, Massachusetts General Hospital, Boston, Massachusetts, for technical support on this project.