γ-Aminobutyric Acid Type A Receptor Modulation by Etomidate Analogs

ETOMIDATE is a highly selective intravenous anesthetic agent that is widely believed to produce hypnosis by binding to a site (or class of sites) on the γ-aminobutyric acid type A (GABA_A) receptor.^1–5  Although structural information regarding this site is quite limited, there is growing evidence that it is located at the interface between the GABA_A receptor’s α and β subunits in the transmembrane domain.^6–8  The result of such binding is an enhancement of GABA_A receptor function.^9,10  Specifically, etomidate binding increases agonist potency for activating (i.e., opening) GABA_A receptors, a process termed “agonist potentiation.” In the absence of agonist, etomidate binding also directly activates GABA_A receptors. Although the agonist potentiating and direct activating actions are measurable using electrophysiologic techniques over distinct etomidate concentration ranges (low and high, respectively), it has been proposed that the underlying receptor mechanism for these two enhancing actions is the same (stabilizing the open channel state) and that they reflect etomidate binding to the same receptor site(s).^10–12  In addition to enhancing GABA_A receptor function, etomidate, etomidate analogs, and other anesthetics can inhibit the receptor’s function.^13–15  This inhibitory action typically occurs only at very high anesthetic concentrations and is likely mediated by a site that is distinct from that which produces enhancement.

Studies of etomidate’s two enantiomers indicate that the structural requirements for GABA_A receptor binding and enhancement can be quite specific. Although etomidate’s enantiomers have identical physical properties, they differ by 1 to 2 orders of magnitude in their in vitro affinities and potencies for the GABA_A receptor and, consequently, their in vivo hypnotic potencies.^3,13,16  Other small structural modifications around the chiral center similarly impact GABA_A receptor and hypnotic potencies, implying that this region of etomidate’s molecular scaffold is a critical determinant of pharmacologic activity.¹³  The pharmacologic importance of other regions is unknown.

Over the past several years, our laboratory has developed multiple novel etomidate analogs that contain structural modifications in various parts of etomidate’s molecular scaffold and exhibit unique pharmacology.^13,17–20  These compounds include etomidate esters that are ultrarapidly metabolized, pyrrole analogs devoid of adrenocortical side effects, and achiral analogs with reduced hypnotic and adrenocortical inhibitory potencies. Although it seems likely that only a handful of these analogs have potential value as clinical anesthetic agents, together they can be used as pharmacologic tools for structure–activity relationship studies to better define the structural determinants of etomidate’s enhancing action on the GABA_A receptor. Such information may also provide clues regarding the nature of the etomidate-binding site and ultimately suggest ways of making more potent analogs.

In this study, we quantified the potencies with which etomidate analogs directly activate the heteromeric GABA_A receptors. We incorporated a channel mutation into the receptor that stabilizes its open state, thus increasing anesthetic sensitivity and allowing more complete anesthetic concentration–response curves to be generated (1) before reaching aqueous saturation at high anesthetic concentrations and (2) without the potentially confounding effects of a coadministered agonist.^10,13,21  We also defined the potencies with which etomidate analogs produce hypnosis in rats. We then used computational techniques to build statistical and graphical models that relate the GABA_A receptor and hypnotic potencies of these etomidate analogs to their physical structures (i.e., steric and electrostatic properties) in a three-dimensional space.

All studies were conducted with the approval of and in accordance with the regulations of the Institutional Animal Care and Use Committee at Massachusetts General Hospital (Boston, MA). Xenopus laevis adult female frogs were purchased from Xenopus One (USA). Adult male Sprague-Dawley rats (300 to 450 g) were purchased from Charles River Laboratories (USA).

Figures 1 and 2 show the structures of etomidate and the etomidate analogs used in this study and indicate their names and compound numbers (1 to 23). The structures of etomidate and metomidate are shown in the top row of figure 1. Etomidate was purchased from Bachem Americas (USA), and metomidate was synthesized by Aberjona Laboratories (USA) using the general approach previously reported for etomidate.²²  The second row in figure 1 shows the structures of S-etomidate, dihydrogen etomidate, and cyclopropyl etomidate. These etomidate analogs, which have chiral center modifications, were synthesized as previously described.¹³  The third row in figure 1 shows the structures of the pyrrole etomidate analogs carboetomidate and 4-pyridine carboetomidate. Both compounds were synthesized by Aberjona Laboratories. The last row in figure 1 shows the structures of racemic pentafluoroetomidate, S-pentafluoroetomidate, and R-pentafluoroetomidate. The first two compounds were synthesized by Aberjona Laboratories using the approach previously reported for etomidate using racemic 1-pentafluorphenylethanol or R-1-pentafluorphenylethanol, respectively, in the place of 1-phenylethanol.²²  R-pentafluoroetomidate was not synthesized because the necessary chiral alcohol reagent (S-1-pentafluorphenyl ethanol) was not commercially available. However, potency values for this enantiomer were estimated from those determined for racemic pentafluoroetomidate and S-pentafluoroetomidate. The structures of the etomidate esters used in this study are shown in figure 2. They were synthesized as previously described.^17,19 

Oocytes were harvested from frogs and injected with mRNA encoding the α₁(L264T), β₃, and γ₂ subunits of the human GABA_A receptor (5 ng of mRNA total at a subunit ratio of 1:1:3). After RNA injection, oocytes were incubated in ND96 buffer (96 mM NaCl, 2 mM KCl, 1.8 mM CaCl₂, 1 mM MgCl₂, 5 mM HEPES, pH = 7.4) containing 0.1 mg/ml ciprofloxacin and 0.05 mg/ml gentamicin for 18 to 48 h at 18°C before electrophysiologic study. Electrophysiologic recordings were performed at room temperature using the whole cell two-electrode voltage clamp technique as described previously.²⁰  In each experiment, the peak amplitude was defined as the difference between the baseline current before etomidate analog infusion and the single highest point in the current trace during infusion. This amplitude was then normalized to the peak amplitude measured in control currents elicited by 100 µM GABA (a concentration of GABA that maximally activates GABA_A receptors) in the same oocyte. For each etomidate analog, the concentration–mean response data were fit to a Hill equation with minima and maxima constrained to 0 and 100%, respectively, using equation 1 as previously described¹³ :

where EC₅₀ is the etomidate analog concentration that evokes a peak current amplitude that is half that evoked by 100 µM GABA, [Analog] is the etomidate analog concentration, and n is the Hill coefficient.

The hypnotic potencies of etomidate analogs were assessed in rats using a loss of righting reflexes (LORR) assay.¹⁷  In brief, the desired dose of analog in dimethyl sulfoxide vehicle (0.1 to 0.3 ml) was rapidly injected through an intravenous catheter placed in a tail vein followed by a normal saline flush. Immediately afterward, rats were turned supine. A rat was judged to have LORR if it failed to turn itself back onto all four paws within 5 s after drug administration. For each analog, the median effective dose (ED₅₀) for LORR was determined from a data set of at least 24 separate doses using the method of Waud.²³  For in silico modeling studies, each ED₅₀ value expressed in milligram per kilogram was converted to units of micromolar per kilogram using the analog’s molecular weight.

The octanol:water partition coefficient of each etomidate analog was determined as previously described.¹³ 

The low-energy conformer of etomidate and each etomidate analog was defined using MMFF94 force field utilizing an energy gradient convergence criterion of 0.01 kcal/mol with SybylX2.1.1 (Certara, USA). Partial electronic charges were calculated by the software using the Gasteiger–Hückel method.²⁴  By using etomidate’s structure as the template, each etomidate analog was then automatically aligned in silico to overlay its molecular shape, hydrogen-bonding capacities, and electrostatic properties with Surflex-Sim (Certara L.P.). When an imidazole ring was present in an etomidate analog, these alignments were further optimized by manually aligning this ring with that of etomidate.

Classic comparative molecular field analysis (CoMFA) interaction fields (steric and electrostatic fields) were calculated with SybylX2.1.1 using the default settings. The steric and electrostatic interaction energies were calculated on grid points of a regularly spaced three-dimensional lattice with a sp³ carbon probe atom having a charge of +1 and a van der Waals radius of 1.52Å. The grid size had the software’s default resolution of 2Å. Cutoffs were applied to both the steric and electrostatic interactions with energies at 30 kcal/mol. CoMFA region focusing was used to enhance the resolution of the model and improve its predictive power. In this procedure, the weightings of lattice points in the grid are modified by the software in an iterative fashion to enhance or reduce each point’s contribution to subsequent analysis.

The partial least-squares approach was used to derive the three-dimensional quantitative structure–activity relationship, and cross-validation was performed using the leave-one-out method.²⁵  The optimum number of components (NC) that produced the lowest SE of predictions (SDEP) was determined, and the cross-validation correlation coefficient (q²) was calculated. A coefficient of determination (r²) was then defined from the relationship between the predicted and the experimental potencies using linear regression.

The five molecules with the lowest GABA_A receptor EC₅₀s (i.e., the five most potent GABA_A receptor direct activators) were extracted from the alignment prepared for the GABA_A receptor CoMFA analysis, and a pharmacophore model was constructed with unity (Certara L.P.). Such models describe the common features present in these ligands and the molecular interactions that they make with the protein that are important determinants of potency. For this analysis, we required that such interactions be present in at least four of the five high-potency molecules in the data set. We used a maximum volume constraint of 1 Å around each pharmacophore point. Thus, in some cases, a pharmacophore point on the ligand could not be associated with one at the receptor-binding site because the latter’s volume exceeded the 1 Å cutoff (i.e., there was too much uncertainty regarding the location of the receptor pharmacophore point). Aromatic and hydrophobic ligand domains were identified using rules in SybylX2.1.1 software. We then used MOLCAD (Cetara L.P.) to visualize the surface features and physical properties required for molecular recognition.

All data are reported as mean ± SD. Sample sizes were chosen based on our previous experience.¹³  Hypothesis testing was two tailed, and a P value less than 0.05 was considered to be statistically significant. Curve fitting to define GABA_A receptor EC₅₀s and rat LORR ED₅₀s and their respective SDs were performed using Igor Pro 6.1 (Wavemetrics, USA). Linear regressions of logarithm-transformed data (GABA_A receptor EC₅₀s and rat LORR ED₅₀s vs. hydrophobicity) were performed using Igor Pro 6.1. Statistical differences between the EC₅₀s of two compounds were assessed using an extra sum-of-squares F test with Prism 6 (Graphpad, USA). The sample sizes for data shown in all figures are indicated in the text. CoMFA statistics (NC, SDEP, q², and r²) were calculated by SybylX2.1.1 software. There was no lost or missing data.

All etomidate analogs directly activated α₁(L264T)β₃γ₂ GABA_A receptors expressed in oocytes in a concentration-dependent manner. Figure 3A shows representative electrophysiologic traces recorded on perfusing a single oocyte with the diastereomeric etomidate ester pair R-isopropyl-methoxycarbonyl metomidate and S-isopropyl-methoxycarbonyl metomidate at concentrations of 1 (top) and 10 µM (bottom). At 1 µM, the R- and S-diastereomers evoked currents having peak amplitudes of 0.041 and 0.29 μA, respectively. At 10 µM, they evoked respective currents having peak amplitudes of 0.36 and 0.90 µA, respectively. For the two diastereomers, figure 3B plots the concentration–response relationship for peak current activation (mean ± SD, n = 6 oocytes per data point). For both diastereomers, the normalized peak current response increased with concentration. At a near saturating aqueous concentration (100 µM), S- (but not R-) isopropyl-methoxycarbonyl metomidate evoked currents that approximated those evoked by a maximally activating concentration (100 µM) of GABA. A fit of these concentration–response curves to equation 1 yielded EC₅₀s for direct activation of 2.6 ± 0.3 and 46 ± 6 µM for S- and R-isopropyl-methoxycarbonyl metomidate, respectively (P < 0.0001).

Figure 4A plots the concentration–response relationships for peak current activation by etomidate and carboetomidate, a pyrrole etomidate analog that does not suppress adrenocortical function. The normalized peak current amplitude increased with both etomidate and carboetomidate concentrations. However, similar to R-isopropyl-methoxycarbonyl metomidate, carboetomidate’s relatively low potency and aqueous solubility limited our studies to carboetomidate concentrations that activated (at most) only 68 ± 6% of GABA_A receptors. From these concentration–response curves, we determined that carboetomidate’s potency for directly activating GABA_A receptors is approximately one eighth that of etomidate with EC₅₀s of 13.8 ± 0.9 µM for carboetomidate and 1.83 ± 0.28 µM for etomidate (P < 0.0001).

Figure 4B plots the concentration–response relationships for peak current activation by the fluorinated etomidate analog S-pentafluoroetomidate. This figure also plots this relationship for racemic pentafluoroetomidate, which contains equal quantities of the R- and S-enantiomers. The EC₅₀ for the S-enantiomer was 22-fold higher than that of the racemic mixture (166 ± 25 vs. 7.6 ± 0.6 µM, respectively; P < 0.0001). This implies that essentially all of the directly activated current recorded during application of the racemic mixture was attributable to the R- enantiomer and that R-enantiomer’s EC₅₀ is approximately 3.8 µM (i.e., one half the EC₅₀ of the racemic mixture).

Table 1 gives the GABA_A receptor EC₅₀s for etomidate and all of the etomidate analogs that we have characterized to date, along with their physical properties (i.e., molecular weights, molecular volumes, and octanol:buffer partition coefficients) and their hypnotic ED₅₀s measured in Sprague Dawley rats. We have reported some of these values previously and provided the references for these published values in the table.

Although hydrophobicity has long been considered to be an important determinant of in vitro and in vivo anesthetic potency, figure 5 shows that the correlation between the GABA_A receptor potencies of etomidate and etomidate analogs and their octanol:buffer partition coefficients is rather poor.^26–28  A linear fit of the logarithm-transformed values of this relationship yielded a slope that is not significantly different from 0 (0.13 ± 0.20, P = 0.5271) and an r² of only 0.019. An analogous linear fit of the logarithm-transformed relationship between hypnotic potency in rats and octanol:buffer partition coefficient similarly yielded a slope that was not significantly different from 0 (0.09 ± 0.30, P = 0.7562) and an r² of only 0.005 (data not shown).

We used CoMFA to identify the structural elements in our etomidate analogs (fig. 6) that account for their widely ranging (91-fold) GABA_A receptor potencies. Figure 7 shows the final alignment of etomidate and the 21 different etomidate analogs along with the resulting CoMFA contour maps. Figure 7A shows the steric contour map depicting where bulky substituents increase (green) or decrease (yellow) GABA_A receptor potency. Figure 7B shows the electrostatic contour map depicting where electronegative substituents (red) or electropositive substituents (blue) increase GABA_A receptor potency. The model had good predictive ability as assessed by leave-one-out cross-validation with values of q² = 0.458, NC = 4, and SDEP = 0.458. Figure 8 shows the correlation between the GABA_A receptor EC₅₀ values predicted by the CoMFA model and the experimentally derived GABA_A receptor EC₅₀ values. The model explained 91.2% of the variance in the observed activities of the etomidate analogs with values of r² = 0.912, NC = 4, and SDEP = 0.227.

The identical alignment was used to identify the structural features of our etomidate analogs that account for their widely ranging (53-fold) hypnotic potencies in rats. Figure 9A shows the steric contour map depicting where bulky substituents increase (green) or decrease (yellow) hypnotic potency. Figure 9B shows the electrostatic contour map depicting where electronegative substituents (red) or electropositive substituents (blue) increase hypnotic potency. Figure 10 shows the correlation between the hypnotic ED₅₀ values predicted by the CoMFA model and the experimentally derived hypnotic ED₅₀ values. The CoMFA model explained 97.3% of the variance in the observed hypnotic potencies (r² = 0.973, NC = 5, and SDEP = 0.093) and showed good predictive capability with leave-one-out validation (q² = 0.521, NC = 5, and SDEP = 0.377).

Visual comparison of the contour maps for in vitro GABA_A receptor potency (fig. 7) and in vivo hypnotic potency (fig. 9) reveals certain similarities. For example, in both steric contour maps (figs. 7A and 9A), there is a yellow contour near the chiral carbon located between the phenyl and the imidazole rings indicating that the S-form is sterically disfavored. There is a similar yellow contour near the spacer located between the two ester moieties of etomidate esters indicating that the R-form is sterically disfavored. The “steric penalty” for being in the disfavored enantiomeric form at either of these chiral centers was a reduction in receptor and hypnotic potencies of as much as 1 to 2 orders of magnitude (table 1). For both steric contour maps, there is also a green contour around the alkyl groups of the etomidate ester spacer (represented as R₂ in fig. 6) indicating that bulky substituents in this region increase both receptor and hypnotic potencies. In both electrostatic contour maps (figs. 7B and 9B), there is a red contour around the conserved carboxylate ester showing favored negative charge related to the carbonyl oxygen and a blue contour around the distal ester of etomidate esters showing favored positive charge relating to the carbonyl carbon.

To better understand the drug–receptor interactions that mediate high GABA_A receptor potency, we built a pharmacophore model using the five most potent GABA_A receptor direct activators shown on table 1. Because four of five of these compounds have an additional carboxylate ester moiety (they are etomidate esters), we allowed a 20% miss rate (one of five molecules) to allow optional pharmacophore features to be included in the final model. This model, along with the five most potent compounds, is shown in figure 11. The model identified four common features and one optional feature in these compounds. The first feature is the basic nitrogen in the imidazole ring, which was identified as a hydrogen bond acceptor for a donor in the receptor. This interaction explains why etomidate’s potency is 1 to 2 orders of magnitude higher than those of the pyrrole analogs of carboetomidate and 4-pyridine carboetomidate. The second feature is the phenyl ring, which is hydrophobic and explains the 24-fold higher potency of carboetomidate compared with 4-pyridine carboetomidate. The third and fourth features are the two conserved oxygens that form the ester adjacent to the imidazole ring. Similar to the imidazole nitrogen, the carbonyl oxygen of this ester was identified as a hydrogen bond acceptor for a donor in the receptor. The distance between these two donors in the receptor was determined to be 9.36 Å by the model. The final feature is the distal ester found in the etomidate esters, which was identified as optional hydrogen bond acceptor. The van der Waals surface of these five most potent compounds was then calculated, and their lipophilic potentials were mapped onto that surface. Figure 12 shows that surface, along with the structures of those five most potent compounds, the pharmacophore model, and the CoMFA contours for GABA_A receptor potency. As the yellow contours denote regions where bulky substituents reduce potency (presumably because steric hindrance reduces binding affinity), they are considered to represent (part of) the lining of the etomidate-binding pocket. Conversely, the green contour denotes the only region in our receptor model where bulky substituents increase potency. As this bulk is in the form of hydrophobic alkyl groups in all of our compounds (R₂ in fig. 6), we conclude that this part of the binding pocket is relatively hydrophobic and spacious as it can even accommodate an isopropyl group, which was the largest alkyl group that we studied.

Etomidate is widely believed to produce its hypnotic effects by binding to the GABA_A receptor and enhancing its function. This conclusion is most strongly supported by studies showing that (1) etomidate’s R- and S-enantiomers produce hypnosis with potencies that correlate with their GABA_A receptor affinities and potencies; and (2) an amino acid mutation that abolishes etomidate’s ability to enhance GABA_A receptor function (N265M on the β subunit) significantly reduces the etomidate sensitivities of transgenic mice containing that mutation.^3,8,13,16,29  This amino acid, along with others, may contribute to an etomidate-binding site on the GABA_A receptor’s open state that is located in the membrane-spanning receptor domains between the α and β subunits.^5–7 

There have been few studies to define the anesthetic structural features required for high-affinity binding to the GABA_A receptor’s etomidate-binding site and/or modulation of the receptor’s function. Competition studies using an etomidate photoaffinity label indicate that propofol and barbiturates bind to that site with lower affinity than R-etomidate, and the steroid anesthetic alphaxalone does not bind to that site at all.⁸  Thus, this binding site can distinguish among classes of general anesthetics. The in vitro electrophysiologic data presented in this study demonstrate that even within a single anesthetic class (i.e., etomidate), this site exhibits considerable selectivity as evidenced by the widely (91-fold) ranging GABA_A receptor potencies of our etomidate analogs that could not be simply explained by their different hydrophobicities. The widely ranging GABA_A receptor potencies of our analogs were matched by a similarly large (53-fold) range in their hypnotic potencies, and a logarithmic plot of GABA_A receptor potency versus hypnotic potency (fig. 13) showed a significant correlation (r² = 0.72) between these two potency measurements. The slope of this relationship was significantly different from 0 (P < 0.0001) and near unity (1.22 ± 0.18), indicating that (in general) when an etomidate analog’s GABA_A receptor potency doubled, its hypnotic potency also doubled. These findings are consistent with a direct cause-and-effect relationship between GABA_A receptor enhancement by etomidate analogs and the production of in vivo hypnosis.

Our computational studies provide additional details regarding the specific anesthetic structural features present in high-potency modulators of the GABA_A receptor function and define the important interactions that etomidate analogs make with amino acids that form the etomidate-binding site on the receptor. For example, our studies show that the basic nitrogen in the imidazole ring and the carboxylate ester adjacent to that ring are important hydrogen bond acceptors for donors in the receptor. Although we do not know the identity of these amino acids, it is tempting to speculate that one of these donor amino acids is β₃N265. This could explain the results of allosteric modeling studies indicating that mutating this asparagine to a methionine (which is not a hydrogen bond donor) reduces etomidate’s affinity for open-state GABA_A receptors by orders of magnitude, whereas mutating it to a serine (which can be hydrogen bond donor) reduces etomidate’s affinity by only half.^30–32  It was concluded from those studies that the reduced etomidate sensitivity of the serine-containing mutant (β₃N265S) measured in electrophysiologic studies is primarily because of a reduction in etomidate’s efficacy rather than its binding affinity. Our computational studies also indicated that (1) the phenyl ring engages in a hydrophobic interaction with hydrophobic amino acid residues in the binding site; (2) there is chiral selectivity around the carbon located between the phenyl and imidazole rings; and (3) in the case of etomidate esters, there is a hydrogen bond interaction between the distal carboxylate ester and a donor on the protein, and chiral selectivity around the spacer that links the distal ester to the etomidate pharmacophore.

The computational modeling approach utilized in the present studies (CoMFA) to define ligand–protein interactions are distinct from—but complimentary to—computational docking studies.^33–36  In the CoMFA approach, there are no a priori assumptions regarding the nature of the protein-binding site. Instead, the molecular (i.e., steric and electrostatic) interactions between a set of ligands and their protein-binding site are identified from the relationship between ligand molecular structure and binding affinity, and the latter often quantified experimentally using a functional assay.^37–39  This technique allows one to construct a pharmacophore model that explains the observed biologic activity and provides information about the molecular forces involved in binding. There are several important limitations of this approach. These limitations include potential uncertainty regarding the alignment of ligands, particularly when the structures of ligands in a data set vary widely. In addition, the conformation of a ligand may change on protein binding. Consequently, the low-energy states calculated by CoMFA may differ from the actual protein-bound states. Such inherent limitations may explain why the correlations that we observed between model-predicted and experimental potency values were good, but not perfect. In contrast, docking studies attempt to fit a ligand into a defined protein-binding site in silico. They are typically agnostic to ligand potency and can utilize a single ligand but require a high-resolution structural model of the binding site to calculate the interaction energy between ligand and protein. Unfortunately, the structures of open-state heteromeric GABA_A receptors have not yet been defined at high resolution. However, progress in that direction is being made through the construction of homology models that seek to approximate the receptor’s structure by utilizing structurally similar proteins.^33,34  With further refinement and validation of these models or the production of a high-resolution heteromeric GABA_A receptor crystal structure, it may be possible to unambiguously identify amino acids at the etomidate-binding site that are the important determinants of receptor (and therefore hypnotic) potency and account for our pharmacophore model.

In summary, we have measured the GABA_A receptor potencies of a series of etomidate analogs and found that their potencies range by 91-fold. The receptor potencies of these analogs correlated with their hypnotic potencies but not with their hydrophobicities. CoMFA indicated that there are multiple structural elements in these etomidate analogs that define their GABA_A receptor potencies. These include the imidazole nitrogen and carboxylate ester (which act as hydrogen bond acceptors), the phenyl ring (which can engage in hydrophobic interactions), the chiral carbon located between the phenyl and the imidazole rings (the R configuration has higher potency), and the chiral carbon located between the etomidate pharmacophore and the distal ester in etomidate esters (the S configuration has higher potency). Modifying any of one of these structural elements can alter receptor potency by an order of magnitude or more.

This study was funded by grant R01-GM087316 from the National Institutes of Health, Bethesda, Maryland, and the Department of Anesthesia, Critical Care, and Pain Medicine, Massachusetts General Hospital, Boston, Massachusetts.

The Massachusetts General Hospital, Boston, Massachusetts, has published and submitted patent applications for certain etomidate analogs. Drs. Raines and Husain are inventors of this technology. They and their laboratories have received income related its development. This conflict is being managed by the Partners Healthcare Office for Interactions with Industry. The other authors declare no competing interests.