Physicochemical Features of the hERG Channel Drug Binding Site
2004; Elsevier BV; Volume: 279; Issue: 11 Linguagem: Inglês
10.1074/jbc.m310683200
ISSN1083-351X
AutoresDavid Fernández, Azad Ghanta, Gregory W. Kauffman, Michael C. Sanguinetti,
Tópico(s)Receptor Mechanisms and Signaling
ResumoBlockade of hERG K+ channels in the heart is an unintentional side effect of many drugs and can induce cardiac arrhythmia and sudden death. It has become common practice in the past few years to screen compounds for hERG channel activity early during the drug discovery process. Understanding the molecular basis of drug binding to hERG is crucial for the rational design of medications devoid of this activity. We previously identified 2 aromatic residues, Tyr-652 and Phe-656, located in the S6 domain of hERG, as critical sites of interaction with structurally diverse drugs. Here, Tyr-652 and Phe-656 were systematically mutated to different residues to determine how the physicochemical properties of the amino acid side group affected channel block by cisapride, terfenadine, and MK-499. The potency for block by all three drugs was well correlated with measures of hydrophobicity, especially the two-dimensional approximation of the van der Waals hydrophobic surface area of the side chain of residue 656. For residue 652, an aromatic side group was essential for high affinity block, suggesting the importance of a cation-π interaction between Tyr-652 and the basic tertiary nitrogen of these drugs. hERG also lacks a Pro-Val-Pro motif common to the S6 domain of most other voltage-gated K+ channels. Introduction of Pro-Val-Pro into hERG reduced sensitivity to drugs but also altered channel gating. Together, these findings assign specific residues to receptor fields predicted by pharmacophore models of hERG channel blockers and provide a refined molecular understanding of the drug binding site. Blockade of hERG K+ channels in the heart is an unintentional side effect of many drugs and can induce cardiac arrhythmia and sudden death. It has become common practice in the past few years to screen compounds for hERG channel activity early during the drug discovery process. Understanding the molecular basis of drug binding to hERG is crucial for the rational design of medications devoid of this activity. We previously identified 2 aromatic residues, Tyr-652 and Phe-656, located in the S6 domain of hERG, as critical sites of interaction with structurally diverse drugs. Here, Tyr-652 and Phe-656 were systematically mutated to different residues to determine how the physicochemical properties of the amino acid side group affected channel block by cisapride, terfenadine, and MK-499. The potency for block by all three drugs was well correlated with measures of hydrophobicity, especially the two-dimensional approximation of the van der Waals hydrophobic surface area of the side chain of residue 656. For residue 652, an aromatic side group was essential for high affinity block, suggesting the importance of a cation-π interaction between Tyr-652 and the basic tertiary nitrogen of these drugs. hERG also lacks a Pro-Val-Pro motif common to the S6 domain of most other voltage-gated K+ channels. Introduction of Pro-Val-Pro into hERG reduced sensitivity to drugs but also altered channel gating. Together, these findings assign specific residues to receptor fields predicted by pharmacophore models of hERG channel blockers and provide a refined molecular understanding of the drug binding site. Long QT syndrome (LQTS) 1The abbreviations used are: LQTS, long QT syndrome; hERG, human ether-a-go-go related gene; Kv, voltage-gated K+; V½, voltage half-point for channel activation; VHSA, van der Waals hydrophobic surface area; WT, wild type; Mes, 2-[N-morpholino]ethanesulfonic acid; EAG, ether-a-go-go. is a disorder of ventricular repolarization that predisposes affected individuals to cardiac arrhythmia and sudden death. Inherited LQTS is caused by mutations in K+ or Na+ ion channel genes or ankyrin-B (1Keating M.T. Sanguinetti M.C. Cell. 2001; 104: 569-580Abstract Full Text Full Text PDF PubMed Scopus (862) Google Scholar, 2Mohler P.J. Schott J.J. Gramolini A.O. Dilly K.W. Guatimosim S. DuBell W.H. Song L.S. Haurogne K. Kyndt F. Ali M.E. Rogers T.B. Lederer W.J. Escande D. Marec H.L. Bennett V. Nature. 2003; 421: 634-639Crossref PubMed Scopus (839) Google Scholar). Acquired LQTS is more common and can be induced as an unintended and rare side effect of treatment with many structurally diverse medications. In the past few years, several commonly used drugs (e.g. terfenadine, cisapride, sertindole, thioridazine, grepafloxacin) were withdrawn from the market, or their approved use was severely restricted, when it was discovered that they caused arrhythmia or were associated with unexplained sudden death, albeit very infrequently (3Pearlstein R. Vaz R. Rampe D. J. Med. Chem. 2003; 46: 2017-2022Crossref PubMed Scopus (167) Google Scholar). The molecular basis of drug-induced LQTS is block of human ether-a-go-go related gene (hERG) channels that conduct IKr, the rapid delayed rectifier K+ current important for repolarization of cardiac action potentials (4Sanguinetti M.C. Jiang C. Curran M.E. Keating M.T. Cell. 1995; 81: 299-307Abstract Full Text PDF PubMed Scopus (2161) Google Scholar, 5Trudeau M. Warmke J.W. Ganetzky B. Robertson G.A. Science. 1995; 269: 92-95Crossref PubMed Scopus (1101) Google Scholar). A reduction in IKr prolongs the action potential duration of ventricular myocytes, lengthens the QT interval and increases dispersion as measured by ECG recordings, and increases the risk of torsades de pointes, a ventricular tachyarrhythmia that can degenerate into fibrillation and cause sudden death. In a laboratory setting, it is possible to induce arrhythmia in animals with drugs that block voltage-gated K+ (Kv) channels other than hERG. However, in clinical practice, drug-induced LQTS is always attributable to direct or indirect (via interference with metabolism of a co-administered medication) block of hERG channels (6Redfern W.S. Carlsson L. Davis A.S. Lynch W.G. MacKenzie I. Palethorpe S. Siegl P.K. Strang I. Sullivan A.T. Wallis R. Camm A.J. Hammond T.G. Cardiovasc. Res. 2003; 58: 32-45Crossref PubMed Scopus (1329) Google Scholar). This understanding has prompted intense efforts to quantify hERG channel activity of new chemical entities during an early stage of the drug development process (6Redfern W.S. Carlsson L. Davis A.S. Lynch W.G. MacKenzie I. Palethorpe S. Siegl P.K. Strang I. Sullivan A.T. Wallis R. Camm A.J. Hammond T.G. Cardiovasc. Res. 2003; 58: 32-45Crossref PubMed Scopus (1329) Google Scholar, 7Fermini B. Fossa A.A. Nat. Rev. Drug Discov. 2003; 2: 439-447Crossref PubMed Scopus (428) Google Scholar). A better understanding of the molecular basis of hERG channel block could facilitate computer-assisted drug design and enable presynthetic, virtual screening of compounds for hERG activity. Moreover, a description of the physicochemical features of the drug binding site would complement pharmacophore models (8Cavalli A. Poluzzi E. De Ponti F. Recanatini M. J. Med. Chem. 2002; 45: 3844-3853Crossref PubMed Scopus (381) Google Scholar, 9Ekins S. Crumb W.J. Sarazan R.D. Wikel J.H. Wrighton S.A. J. Pharmacol. Exp. Ther. 2002; 301: 427-434Crossref PubMed Scopus (270) Google Scholar) of drugs that block hERG channels and define the molecular basis for receptor fields predicted by these models. Toward this goal, we have used site-directed mutagenesis and voltage clamp analysis of mutant channels expressed in Xenopus oocytes to elucidate the molecular mechanisms of hERG channel block by structurally diverse drugs, including MK-499, cisapride, terfenadine, vesnarinone, chloroquine, and quinidine (10Mitcheson J.S. Chen J. Lin M. Culberson C. Sanguinetti M.C. Proc. Natl. Acad. Sci. U. S. A. 2000; 97: 12329-12333Crossref PubMed Scopus (861) Google Scholar, 11Kamiya K. Mitcheson J.S. Yasui K. Kodama I. Sanguinetti M.C. Mol. Pharmacol. 2001; 60: 244-253Crossref PubMed Scopus (132) Google Scholar, 12Sanchez-Chapula J.A. Navarro-Polanco R.A. Culberson C. Chen J. Sanguinetti M.C. J. Biol. Chem. 2002; 277: 23587-23595Abstract Full Text Full Text PDF PubMed Scopus (175) Google Scholar, 13Sanchez-Chapula J.A. Ferrer T. Navarro-Polanco R.A. Sanguinetti M.C. Mol. Pharmacol. 2003; 63: 1051-1058Crossref PubMed Scopus (120) Google Scholar). These studies identified 2 aromatic residues, Tyr-652 and Phe-656, located in the S6 domain and predicted to face the central cavity of the channel (Fig. 1a) that are critical for high affinity binding of these drugs. Kv channels of the Kv1–4 families have an Ile or Val residue in the positions equivalent to Tyr-652 or Phe-656 of hERG (Fig. 1b). This suggests a plausible explanation for why hERG and not Kv1–4 channels are readily blocked by structurally diverse drugs: aromatic residues in S6 are required for high affinity binding. In addition, Kv1–4 channel α-subunits have a Pro-Val(Ile)-Pro motif in the S6 domain that was suggested to cause a bend in the α-helix and alter the shape of the central cavity as compared with that predicted for KcsA, a K+ channel that lacks these Pro residues. The S6 domain of hERG also lacks the Pro-Val-Pro motif and instead has the residues Ile-Phe-Gly in the equivalent positions. Thus, we previously proposed that aromatic residues (2 on each subunit, 8 per channel) that face the inner cavity, and the lack of the Pro-Val-Pro motif, were important determinants of the high affinity hERG channel binding site (10Mitcheson J.S. Chen J. Lin M. Culberson C. Sanguinetti M.C. Proc. Natl. Acad. Sci. U. S. A. 2000; 97: 12329-12333Crossref PubMed Scopus (861) Google Scholar). EAG channels also lack a Pro-Val-Pro motif and have a Tyr and Phe in the S6 domain but are relatively insensitive to drugs that block hERG. However, EAG channels were made sensitive to block by cisapride by moving the Tyr residue by one position in the S6 domain (14Chen J. Seebohm G. Sanguinetti M.C. Proc. Natl. Acad. Sci. U. S. A. 2002; 99: 12461-12466Crossref PubMed Scopus (177) Google Scholar). Pharmacophore models predict that important features of potent hERG channel blockers are 1) a basic nitrogen that is usually protonated at physiological pH and 2) three hydrophobic centers of mass (centroids) arranged in a specific spatial pattern around the centrally located nitrogen (8Cavalli A. Poluzzi E. De Ponti F. Recanatini M. J. Med. Chem. 2002; 45: 3844-3853Crossref PubMed Scopus (381) Google Scholar, 9Ekins S. Crumb W.J. Sarazan R.D. Wikel J.H. Wrighton S.A. J. Pharmacol. Exp. Ther. 2002; 301: 427-434Crossref PubMed Scopus (270) Google Scholar). For many potent hERG blockers, these centroids are aromatic groups. We and others have postulated that π -stacking between aromatic groups of the drug and Phe-656 and Tyr-652 of hERG are required for high affinity binding and channel block (10Mitcheson J.S. Chen J. Lin M. Culberson C. Sanguinetti M.C. Proc. Natl. Acad. Sci. U. S. A. 2000; 97: 12329-12333Crossref PubMed Scopus (861) Google Scholar, 15Lees-Miller J.P. Duan Y. Teng G.Q. Duff H.J. Mol. Pharmacol. 2000; 57: 367-374PubMed Google Scholar). In addition, because most hERG blockers contain a basic nitrogen, it has been suggested that cation-π interactions with Tyr-652 or Phe-656 might also be required for high affinity binding to the channel. In this study, we systematically investigated these hypotheses by mutating Tyr-652 and Phe-656 to several other amino acids and determined the sensitivity of the resulting mutant channels to block by cisapride, MK-499, and terfenadine. A quantitative comparison between the IC50 values for the mutant channels and the physicochemical properties of the side chain of the mutant residues was used to provide further insights into the molecular features of the hERG binding site. Molecular Biology—Point mutations of hERG subcloned into the pSP64 plasmid expression vector (Promega, Madison, WI) were made as described (10Mitcheson J.S. Chen J. Lin M. Culberson C. Sanguinetti M.C. Proc. Natl. Acad. Sci. U. S. A. 2000; 97: 12329-12333Crossref PubMed Scopus (861) Google Scholar). Constructs were confirmed with restriction mapping and DNA sequencing. Complementary RNAs for injection into oocytes were prepared with SP6 Cap-Scribe (Roche Applied Science) following linearization of the expression construct with EcoRI. Oocytes were injected with 5–10 ng of hERG cRNA 2 days before use in voltage clamp experiments. Voltage Clamp of Oocytes—Isolation and maintenance of Xenopus laevis oocytes and cRNA injections were performed as described (16Sanguinetti M.C. Xu Q.P. J. Physiol. (Lond.). 1999; 514: 667-675Crossref Scopus (142) Google Scholar). Currents were recorded at room temperature (22–24 °C) with a GeneClamp 500 amplifier (Axon Instruments, Burlingame, CA) using standard two-microelectrode voltage clamp techniques (17Stuhmer W. Methods Enzymol. 1992; 207: 319-339Crossref PubMed Scopus (262) Google Scholar). Oocytes were bathed in a low Cl– solution containing 96 mm NaMes, 2 mm KMes, 2 mm CaMes2, 5 mm HEPES, 1 mm MgCl2, adjusted to pH 7.6 with methanesulfonic acid. Current-voltage (I-V) relationships were determined with test pulses applied to voltages ranging from –60 to +50 mV, at a frequency of 0.06 Hz and from a holding potential of –90 mV. Deactivating (tail) currents were measured at –70 mV. To quantify drug-induced block of hERG, currents were elicited with 5-s pulses to 0 mV, applied repetitively at 0.166 Hz from a holding potential of –90 mV. Drug-induced block of current was assessed by measuring the outward currents at the end of the 5-s pulse to 0 mV, a test potential that minimized the contribution of leak or endogenous currents. Cisapride (Research Diagnostics Inc., Flanders, NJ), terfenadine (Sigma), and MK-499 (Merck Research Laboratories) were prepared as 10 mm stock solutions in Me2SO and then dissolved in the external solution immediately before use to obtain desired concentrations. Data Analysis—Analyses of the kinetics of current activation and deactivation were performed using curve fitting routines of Clampfit software (Axon Instruments). The amplitudes of hERG tail currents were fit with a Boltzmann function to estimate the half-point (V ½) and slope factor (k) for the voltage dependence of channel activation using Origin software (OriginLab, Northampton, MA). Concentration effect data were fit to the Hill equation to determine the drug concentration required for 50% inhibition (IC50) and the Hill coefficient (h). Data are presented as mean ± S.E. (n = number of cells), and statistical comparisons between experimental groups were performed using the Student's t test. Differences were considered significant at p < 0.05. Theoretical and physicochemical descriptors were calculated for each of the amino acids substituted at Phe-656 using the Molecular Operating Environment, MOE (Molecular Operating Environment 2003, Chemical Computing Group, Inc., Montreal, Quebec). MOE is a general purpose computational chemistry software application capable of calculating 146 two-dimensional descriptors. Two-dimensional descriptors do not require geometric information about the molecules of interest, making them a suitable choice when modeling the properties of fragments of macromolecular structures. Examples of these descriptors include atom, bond, and fragment counts, graph theoretical connectivity and shape indices, hydrophobicity and hydrophilicity measures, topological charge measures, and two-dimensional van der Waal approximations of surface areas and volumes. Of the 146 descriptors calculated, 115 had non-zero values for the amino acids being modeled. The search for predictive subsets of descriptors was performed using a simulated annealing feature selection algorithm (18Sutter J.M. Jurs P.C. J. Chem. Inf. Comput. Sci. 1995; 35: 77-84Crossref Scopus (197) Google Scholar) included in the ADAPT software package (19Jurs P.C. Chou J.T. Yuan M. Olsen E.C. Christoffersen R.E. Computer Assisted Drug Design. American Chemical Society, Washington, D. C.1979: 103-129Google Scholar, 20Stuper A.J. Jurs P.C. J. Chem. Inf. Comput. Sci. 1976; 2: 99-105Crossref Scopus (74) Google Scholar). Simulated annealing is an optimization algorithm that has found widespread use in the area of variable selection in computational correlation studies. The algorithm attempts to identify subsets of descriptors that best correlate with the experimental end point of interest. In many cases, the direct relevance of descriptors selected to the property being studied is observed. The optimal descriptor subset, or model, is identified as that which minimizes the root-mean-square error between the predicted and experimental log(fold-change in IC50) values and possesses the highest correlation coefficient (R2). MOE and the simulated annealing calculations were performed on an SGI Octane 2 Work station. All other data analysis and manipulation was done using Microsoft Excel under the Windows 2000 operating system. Biophysical Properties of Phe-656 Mutant hERG Channels— The residue equivalent to Phe-656 in hERG is conserved in hEAG (Fig. 1b). However, in most other Kv channels, this amino acid is a Val or Ile and is the middle residue of a Pro-Val(Ile)-Pro motif that has been proposed to produce a bend in the S6 domain near the site of activation gating (21del Camino D. Holmgren M. Liu Y. Yellen G. Nature. 2000; 403: 321-325Crossref PubMed Scopus (322) Google Scholar). We determined the biophysical and pharmacological consequences of mutating Phe-656 to 12 other residues, including Ile and Val. Seven of the mutant channels, produced by mutation of Phe-656 to Val, Ile, Leu, Met, Thr, Trp, or Tyr, retained relatively normal biophysical properties. Currents elicited by 2-s pulses to a variable voltage from a holding potential of –90 mV are shown in Fig. 2. The I-V relationships for these mutant channels were bell-shaped and peaked at a voltage near –10 mV when currents were activated with 2-s pulses (Fig. 2b). These properties are similar to WT hERG channels, and because the rectification of the I-V relationship results from rapid inactivation, they indicate that the mutant channels retained a relatively normal voltage dependence of inactivation. The voltage dependence of channel activation of WT hERG had a half-point (V½) of –17.8 ± 1.2 mV and a slope factor (k) of 7.8 ± 0.2 mV (n = 5). Most of the mutations caused relatively minor shifts in the V½ for activation (all within ±8 mV) or changes in the time constant for the onset of activation at –20 mV (Table I). An exception was the F656W mutation that shifted V½ by –25 mV. Most of the mutations induced significant changes in the rate of deactivation, but K+ selectivity was not altered based upon the lack of effects on reversal potential of the current (Table I). There was no correlation between the kinetics of activation or deactivation and physicochemical parameters such as hydrophobicity, non-polar surface area, or volume of the side group for the amino acid substituted for Phe (data not shown). Mutation of Phe-656 to Gly, Ser, Arg, or Glu reduced functional expression and disrupted closure of channels (Fig. 2c), and therefore, the voltage dependence and kinetics of gating were not determined for these mutant channels.Table IBiophysical parameters of mutant hERG channelsHERG channelActivationDeactivationErevNumberV½kττslowτfastArefmVmVmsmsmsmVnWT-17.8 ± 1.27.8 ± 0.21386 ± 94743 ± 39156 ± 80.30-105 ± 1.25F656W-43.3 ± 1.1**6.5 ± 0.4*409 ± 28**1145 ± 58**140 ± 50.16**-107 ± 1.66F656Y-23.8 ± 0.9**9.4 ± 0.4**877 ± 49**290 ± 14**60 ± 1**0.68**-105 ± 1.68F656M-21.5 ± 0.8*7.8 ± 0.41301 ± 44675 ± 55109 ± 4**0.40-106 ± 1.66F656L-14.8 ± 1.08.6 ± 0.71362 ± 95237 ± 11**63 ± 3**0.66**-105 ± 0.36F656I-11.6 ± 0.8**7.9 ± 0.41116 ± 57*144 ± 5**47 ± 2**0.83**-100 ± 2.15F656V-22.1 ± 1.1*6.6 ± 0.1**1020 ± 66**265 ± 13**93 ± 4**0.67**-111 ± 1.15F656T-23.0 ± 0.7**7.2 ± 0.41223 ± 44579 ± 24**120 ± 9**0.43*-107 ± 1.86Y652W-20.4 ± 0.78.2 ± 0.1940 ± 22**461 ± 34**77 ± 3**0.60**-102 ± 1.26Y652F-27.6 ± 0.7**10.1 ± 0.6**916 ± 19**622 ± 14**78 ± 2**0.30-104 ± 1.06Y652IaDeactivation fit with monoexponential function; Erev determined using 20 mM [K+]e.+7.3 ± 1.27**11.2 ± 0.2**574 ± 8.6**—6.1 ± 1.6**0.89**-41 ± 0.85Y652V-5.9 ± 0.2**7.9 ± 0.3555 ± 32**29 ± 1.2**13 ± 0.2**0.57**-106 ± 0.45Y652T-12 ± 0.8**8.5 ± 0.2931 ± 25**122 ± 4**38 ± 0.6**0.77**-103 ± 1.28Y652Q-3.6 ± 0.3**8.4 ± 0.2596 ± 17**40 ± 8.8**13 ± 1.0**0.80**-102 ± 0.95Y652E-9.5 ± 1.2**7.1 ± 0.2680 ± 38**99 ± 10**37 ± 6**0.57-103 ± 1.75a Deactivation fit with monoexponential function; Erev determined using 20 mM [K+]e. Open table in a new tab Hydrophobicity of Residue 656 Determines Potency of hERG Channel Blockers—The sensitivity of Phe-656 mutant channels to drugs was determined by measuring the decrease in current activated by 5-s pulses applied repetitively to 0 mV. Examples of steady-state block of WT and five mutant Phe-656 hERG channels achieved with 1 or 10 μm MK-499 are shown in Fig. 3a. Concentration-response relationships for MK-499 for these same channels are plotted in Fig. 3b. As expected, if aromaticity is important for drug sensitivity, mutation of Phe-656 to the other natural aromatic amino acids Trp or Tyr altered block by MK-499 only slightly (IC50 increased by a factor of 1.8 and 3.3, respectively). In contrast, the IC50 was increased by over 3 orders of magnitude when Phe-656 was mutated to Gly, Glu, or Arg, mutations that also severely disrupted channel gating (Fig. 2c). Unexpectedly, the IC50 for MK-499 was only moderately increased when Phe-656 was mutated to the non-aromatic residues Met, Leu, or Ile. For these residues, the IC50 was increased by a factor of 3.7, 7.8, and 11, respectively (Table II). We also determined the sensitivity of all 12 mutant channels to block by cisapride (Fig. 3c) and terfenadine (Fig. 3b). Similar to MK-499, the IC50 values for these drugs were only slightly changed when Phe-656 was mutated to Tyr or Trp and increased the most by mutation to Gly or Arg (Table II). Again, mutation of Phe-656 to the hydrophobic residues Met, Leu, or Ile had little or no effect on channel sensitivity to block by cisapride and terfenadine.Table IISummary of IC50 values and Hill coefficients determined from concentration-response relationships for MK-499, cisapride, and terfenadine block of Phe-656 and Tyr-652 mutant hERG channelsTable IISummary of IC50 values and Hill coefficients determined from concentration-response relationships for MK-499, cisapride, and terfenadine block of Phe-656 and Tyr-652 mutant hERG channels Block of hERG by some drugs (e.g. quinidine and chloroquine) is voltage-dependent (12Sanchez-Chapula J.A. Navarro-Polanco R.A. Culberson C. Chen J. Sanguinetti M.C. J. Biol. Chem. 2002; 277: 23587-23595Abstract Full Text Full Text PDF PubMed Scopus (175) Google Scholar, 13Sanchez-Chapula J.A. Ferrer T. Navarro-Polanco R.A. Sanguinetti M.C. Mol. Pharmacol. 2003; 63: 1051-1058Crossref PubMed Scopus (120) Google Scholar), and therefore, the test potential chosen to determine IC50 can be important. However, MK-499 (22Spector P.S. Curran M.E. Keating M.T. Sanguinetti M.C. Circ. Res. 1996; 78: 499-503Crossref PubMed Scopus (274) Google Scholar) and terfenadine (23Roy M.-L. Dumaine R. Brown A.M. Circ. Res. 1996; 94: 817-823Crossref Scopus (272) Google Scholar) do not block hERG in a voltage-dependent manner, and cisapride exhibits only modest voltage dependence (24Potet F. Bouyssou T. Escande D. Baro I. J. Pharmacol. Exp. Ther. 2001; 299: 1007-1012PubMed Google Scholar). The only Phe-656 mutation that significantly altered the voltage dependence of gating was F656W; however, the IC50 values for this mutant channel were not significantly altered. We also tested for voltage dependence of block for F656T and F656M. Block was voltage-independent for both mutant channels by all three drugs (n = 4–6 for each drug, data not shown). Thus, it is appropriate to measure block of hERG by these drugs at 0 mV to determine IC50 values. The logarithmic value for the -fold change in IC50 (relative to WT hERG) for each mutant channel was regressed as a function of a variety of physical and calculated properties of the amino acids substituted for Phe-656. The parameters chosen for regression analysis were three physical parameters, the side chain surface area, free energies of transfer from octanol to neutral aqueous solution, and buried residue non-polar surface area, and the 115 calculated two-dimensional descriptors from MOE. For all parameters investigated, the correlation was highest for MK-499 and weakest for terfenadine. The calculated descriptor that was best correlated with logarithmic -fold change in IC50, for all three compounds, was the two-dimensional van der Waals hydrophobic surface area (VHSA). VHSA is calculated by summing the approximate van der Waals surface areas of atoms deemed hydrophobic in character. VHSA accounts for the hydrophobicity of the sulfur atom of Met and was thus superior in predictive power to the non-polar surface area physical parameter, which does not take this into account. Table III summarizes the correlation coefficients (R2) for the linear regression plots of physicochemical descriptors versus experimental log(fold change in IC50). Fig. 4 shows a plot of the predicted versus experimental log(fold change in IC50) for the three drugs based on the VHSA values using ADAPT as described under "Experimental Procedures." Predicted values for additional residues can be determined by inserting the VHSA value for the residue into each of the three equations presented in the legend of Fig. 4.Table IIISummary of linear correlation coefficients (R2) for relationships between physicochemical properties and drug sensitivity of Phe-656 mutant channelsMK-499CisaprideTerfenadineAllNo ArgNormal gatingAllNo ArgNormal gatingAllNo ArgNormal gatingSide chain surface area0.382*0.706†0.857†0.3020.591†0.726†0.2950.648†0.669*Non-polar surface area0.846†0.870†0.743†0.747†0.763†0.585*0.698†0.728†0.509*Octanol → water0.842†0.849†0.763†0.847†0.851†0.734†0.786†0.739†0.613*VHSA0.881†0.970†0.958†0.811†0.897†0.858†0.768†0.887†0.794†VHSA + nN0.980†0.976†0.974†0.927†0.917†0.904†0.930†0.928†0.881† Open table in a new tab The correlations between VHSA and experimental values were further improved if the data points for residues that significantly altered the channel gating were removed from the analysis. The correlation coefficients for analyses without Arg, and without Arg, Glu, Gly, and Ser are included in Table III. Other calculated descriptors related to hydrophobicity demonstrated high correlations as well, including the Crippen fragment-derived logP estimator (SlogP) (25Wildman S.A. Crippen G.M. J. Chem. Inf. Comput. Sci. 1999; 39: 868-873Crossref Scopus (839) Google Scholar), a linear logP model estimator (logP (octanol/water), from P. Labute in MOE), and a simple count of the number of hydrophobic atoms. SlogP and logP (octanol/water) differ from VHSA in that they account for all atoms in each residue, rather than just the hydrophobic atoms. In addition to hydrophobicity measures, descriptors encoding the sums of two-dimensional van der Waals surface areas of partially positive atoms and partially negative atoms, separately, were among the more highly correlated metrics, but not as highly correlated as the hydrophobicity measures. When considering the full data set (i.e. all 3 compounds and all 13 amino acids), VHSA was the most highly correlated descriptor and possessed the most statistical significance (F-test, t test, p value). SlogP and logP (octanol/water) were always the second and third most highly correlated descriptors, respectively, indicating a clear consensus that hydrophobicity is a driving force behind the observed data. The simulated annealing algorithm was used to search for optimal two-descriptor models and identified the combination of VHSA and nN, the number of nitrogen atoms, as a superior predictor of log(fold change in IC50) for all three compounds. In this context, nN is merely a correction factor for residues that contain more than one nitrogen atom (i.e. Trp and Arg). For example, R2 for the plot of predicted versus experimental IC50 values was 0.88 (root-mean-square error = 0.4) when VHSA was the sole descriptor, but adding the nN term to the quantitative structure activity relationship fit enhanced R2 to 0.98 (root-mean-square error = 0.163) for MK-499. The equations for the linear regression analyses evaluating VHSA + nN for MK-499 (Equation 1), cisapride (Equation 2), and terfenadine (Equation 3) were as follows. y=3.674-0.0389*VHSA+0.446*nN(Eq. 1) y=2.762-0.033*VHSA+0.437*nN(Eq. 2) y=1.707-0.022*VHSA+0.357*nN(Eq. 3) The R2 values obtained from these regression analyses are listed in Table III. A better fit could also be obtained by omitting Arg from the analysis (Table III, No Arg), perhaps because this residue repels the basic nitrogen of the drug and diminishes binding affinity even more than expected based simply on hydrophobic surface area of the side chain. Together, these data indicate that the important physicochemical feature of Phe-656 is hydrophobic volume, not aromaticity per se. Aromatic Residue at Position 652 Required for Potent hERG Channel Block—The residue equivalent to Tyr-652 in hERG is conserved in EAG; however, in Kv1–4 channels, this amino acid is Ile (Fig. 1b). Tyr-652 was mutated to 8 other residues (Phe, Trp, Ala, Val, Glu, Gln, Ile, Thr) to determine the effect of changing the volume, size, polarity, and hydrophobicity of the side group on drug sensitivity. Mutations of Tyr-652 were well tolerated, and the mutant channels retained K+ selectivity and relatively normal biophysical properties (Fig. 5 and Table I). Drugs might interact by H-bonding to the hydroxyl group or via cation-π or π -stacking with the phenol of Tyr-652. Mutation of Tyr-652 to Phe or Trp reduced or did not significantly change the sensitivity to MK-499 (Fig. 6, a and b), and like WT hERG, block of
Referência(s)