hERG subunit composition determines differential drug sensitivity

Abstract

BACKGROUND AND PURPOSE

The majority of human ether-a-go-go-related gene (hERG) screens aiming to minimize the risk of drug-induced long QT syndrome have been conducted using heterologous systems expressing the hERG 1a subunit, although both hERG 1a and 1b subunits contribute to the K⁺ channels producing the repolarizing current I_Kr. We tested a range of compounds selected for their diversity to determine whether hERG 1a and 1a/1b channels exhibit different sensitivities that may influence safety margins or contribute to a stratified risk analysis.

EXPERIMENTAL APPROACH

We used the IonWorks™ plate-based electrophysiology device to compare sensitivity of hERG 1a and 1a/1b channels stably expressed in HEK293 cells to 50 compounds previously shown to target hERG channels. Potency was determined as IC₅₀ values (µM) obtained from non-cumulative, eight-point concentration–effect curves of normalized data, fitted to the Hill equation. To minimize possible sources of variability, compound potency was assessed using test plates arranged in alternating columns of cells expressing hERG 1a and 1a/1b.

KEY RESULTS

Although the potency of most compounds was similar for the two targets, some surprising differences were observed. Fluoxetine (Prozac) was more potent at blocking hERG 1a/1b than 1a channels, yielding a corresponding reduction in the safety margin. In contrast, E-4031 was a more potent blocker of hERG 1a compared with 1a/1b channels, as previously reported, as was dofetilide, another high-affinity blocker.

CONCLUSIONS AND IMPLICATIONS

The current assays may underestimate the risk of some drugs to cause torsades de pointes arrhythmia, and overestimate the risk of others.

Introduction

During the past 15 years, major efforts to reduce the incidence of sudden cardiac death as an adverse drug reaction have focused largely on assays using the K⁺ channel (K_v11.1) (channel nomenclature follows Alexander et al., 2009) encoded by the human ether-a-go-go-related gene (hERG), now a standard component of preclinical safety testing (http://www.ich.org/products/guidelines/safety/article/safety-guidelines.html) (Pollard et al., 2010). In this assay, drugs are tested for the potency with which they block K⁺ currents encoded by a variant of the hERG1 (KCNH2) gene expressed in a cell line. A proxy for the repolarizing current cardiac I_Kr (Sanguinetti et al., 1995; Trudeau et al., 1995), hERG currents are inhibited to a degree that predicts whether a compound will increase action potential duration, prolong the QT interval of the surface electrocardiogram and trigger catastrophic torsades de pointes (TdP) arrhythmias associated with acquired long QT syndrome (LQTS). The assay has been adopted by pharmaceutical companies and regulatory agencies because most drugs known to trigger TdP arrhythmias are ‘hERG blockers,’ and thus it is hoped such danger posed by new lead compounds will be averted early in preclinical studies (Redfern et al., 2003).

Increasing evidence indicates that the composition of native I_Kr channels is more complex than previously appreciated. The hERG1 gene encodes at least two transcripts: hERG 1a, the original isolate, and hERG 1b, an alternate transcript in which the first five exons of 1a are substituted with a single, distinct exon (Lees-Miller et al., 1997; London et al., 1997). The encoded subunit is identical to 1a throughout the transmembrane domains and carboxy terminus, but has a much shorter amino (N) terminus unique in primary structure. The two subunits associate in native myocardium (Jones et al., 2004), and in heterologous systems they form heteromeric channels with properties distinct from those of the hERG 1a homomers (London et al., 1997; Larsen et al., 2008; Sale et al., 2008). Homomeric 1b channels are produced with extremely low efficiency (Lees-Miller et al., 1997; London et al., 1997). Compared with hERG 1a homomers, 1a/1b heteromers are faster to activate and recover from inactivation, and, as a consequence, contribute 80% more repolarizing charge during a ventricular action potential (Sale et al., 2008). Thus, loss of 1b subunits should prolong repolarization, a prediction borne out in a LQTS patient with a 1b-specific missense mutation that drastically reduces hERG protein expression (Sale et al., 2008).

Could subunit composition affect sensitivity to drugs? The two subunits are identical in pore sequences shown to bind drugs (Lees-Miller et al., 2000; Mitcheson et al., 2000a). However, gating processes also contribute to the efficacy of inhibition. Drugs diffuse cross the plasma membrane and enter open hERG channels from the cytosol (Snyders and Chaudhary, 1996; Spector et al., 1996; Zhou et al., 1998). After the drug binds, the activation gate closes in response to repolarization, trapping the drug in the vestibule and contributing to use-dependent block (Mitcheson et al., 2000b). In addition, the uniquely voltage-dependent inactivation characteristic of hERG channels is important, as almost all mutants that disrupt inactivation also dramatically reduce efficacy of block (Wang et al., 1997; Ficker et al., 1998; 2001; Herzberg et al., 1998; Lees-Miller et al., 2000). Previous studies have shown that two hERG ‘activators’ exhibit similar efficacies for homomeric hERG 1a and 1a/1b currents (Larsen et al., 2010), but inhibition by E-4031 is fourfold more potent for 1a versus 1a/1b channels (Sale et al., 2008). In this study, we have expanded this analysis by comparing the sensitivity of hERG 1a and 1a/1b channels to block by 50 compounds using planar patch technology. Although these compounds are diverse in terms of both their physicochemical properties and their structure, they are also different in terms of their primary pharmacology, disease indication, range of maximum effective free therapeutic plasma concentration (EFTPC_max) and risk of TdP (Redfern et al., 2003; Männikköet al., 2010). Significant differences observed in the potencies of some drugs for the two targets may have implications for drug safety.

Methods

Stable hERG 1a and 1a/1b cell lines

hERG 1a and 1b cDNAs were subcloned into the EcoRI and BamHI sites of expression vectors pcDNA3.1 and pcDNA3.1zeo (Invitrogen, Carlsbad, CA, USA). Constructs were verified by double stranded DNA sequencing. HEK-293 cells (ATCC) were stably transfected using lipofection reagent (Mirus Bio LLC, Madison, WI, USA) and clonally selected as described earlier (Zhou et al., 1998). Briefly, clonal HEK-293 cells stably expressing hERG1a were selected based on their resistance to neomycin, where hERG1a/1b cells were selected for resistance to neomycin and zeocin. To confirm the presence of hERG proteins, clonal cell lines were screened by Western blot analysis using hERG1 isoform-specific antibodies, as described earlier (Jones et al., 2004), for the presence of hERG1a and 1b proteins. One cell line was selected for use based on robust and roughly equivalent levels of 1a and 1b, as demonstrated by Western blot analysis using a pan-hERG antibody (Figure 1). The levels of 1b protein have been observed to decline with passage number and, for this reason, we used only cells from low-passage number and periodically verified expression of 1b protein by Western blot analysis.

Figure 1.

Western blot showing hERG 1a and 1b protein stably expressed in HEK-293 cells and visualized with a C-terminal pan-hERG antibody (Novus Biologicals, Littleton, CO, USA). From the top, the signals represent mature (Golgi-glycosylated) hERG 1a at 155 kDa, immature (ER-resident) 1a at 135 kDa (Zhou et al., 1998), mature hERG 1b at 95 kDa, and immature 1b at 85 kDa (Jones et al., 2004).

Cell culture

For electrophysiological recordings, both hERG 1a and 1a/1b cell lines were grown to semi-confluence at 37°C in a humidified environment (5% CO₂) in minimum essential medium (Invitrogen, Paisley, UK), 10% foetal calf serum and 100 mg·mL⁻¹ G418 (PAA laboratories GmbH, Austria) to maintain selection for hERG 1a; in addition, for hERG 1a/1b cells, 100 µg·mL⁻¹ zeocin (Invitrogen, Paisley, UK) was added to the media to maintain selection for the hERG 1b subunit. The monolayer of either cell culture was washed using a pre-warmed free Ca²⁺/Mg²⁺ Dulbecco's phosphate-buffered saline (PBS; Sigma-Aldrich Company Ltd, Poole, UK) and trypsinized with 0.05% trypsin EDTA solution (Invitrogen, Paisley, UK) at 37°C for 2 min. Cells were detached from the bottom of the flask by gentle tapping and 10 mL of free Ca²⁺/Mg²⁺ PBS was added to the flask and aspirated into a 15 mL centrifuge tube prior to centrifugation (261 × g, for 2 min). The resulting supernatant was discarded and the pellet gently re-suspended in 5 mL of PBS (Invitrogen, Paisley, UK) containing Ca²⁺ and Mg²⁺. An aliquot of cell suspension was removed, and the number of viable cells was manually counted using FastRead Counting slides (Immune Systems Ltd, London, UK) so that the cell re-suspension volume could be adjusted with PBS containing Ca²⁺/Mg²⁺ to give the desired final cell concentration of 1 x 10⁶ cells·mL⁻¹.

Pharmacological comparison

We determined the pharmacological profile of each hERG cell line based on direct assessment of channel function using the IonWorks™ automated, plate-based electrophysiology device (Molecular Devices Inc, Sunnyvale, CA, USA; Schroeder et al., 2003), according to the method described by Bridgland-Taylor et al. (2006). In brief, for each experimental ‘run’ of IonWorks™, the device made perforated whole-cell recordings at ∼21°C, usually from more than 250 of the 384 wells in a standard PatchPlate™. After attainment of the whole-cell configuration, a single voltage pulse was applied once to obtain a control recording and again after compound application (cells were incubated for approximately 3.5 min) to measure inhibition of hERG currents. For each pulse, a holding potential of −70 mV was applied for 20 s, followed by a 160 ms step to −60 mV (allowing an estimated leak current to be measured), and a 100 ms step back to −70 mV. The voltage was then stepped to +40 mV for 1 s and a steady-state current was observed. A 2 s step down to −30 mV, inducing the tail current, was then followed by a 0.5 s step to −70 mV. Voltage was not controlled between pulses. The current signal was sampled at 2.5 kHz. hERG current magnitude was measured automatically from the leak-subtracted traces by the IonWorks™ software by taking a 40 ms average of the baseline current measured at −70 mV and subtracting this from the peak of the tail-current response measured during the voltage step at −30 mV. The acceptance criteria for the currents evoked in each well were: pre-scan seal resistance >60 MΩ, pre-scan outward current amplitude >150 pA and post-scan seal resistance >60 MΩ. The degree of inhibition of the hERG current was assessed by dividing the post-scan hERG current by the respective pre-scan hERG current for each well. Data were normalized to vehicle and 100% blocking levels, and the resulting data fitted to the Hill equation using a custom-written program in Origin. Potency values are represented as IC₅₀ (µM) with 95% confidence intervals or pIC₅₀. Great care was taken to ensure that sources of variability were minimized; for example, fresh cells were prepared immediately before the start of each run and compound potency was determined in pair-wise combinations for the 1a and 1a/1b targets.

Compound selection and preparation of compound plates

A panel of 50 compounds was selected based on the data reported by Männikköet al. (2010) and Redfern et al. (2003). These compounds were diverse in terms of their physicochemical properties, structure and torsadogenic risk. Compound preparation was based on the method described by Bridgland-Taylor et al. (2006) with some variations. Briefly, compounds were dissolved to their stock concentration (300 times higher than the highest test concentration) in DMSO (Sigma-Aldrich Company Ltd). These solutions were then aliquoted into wells H01-H10 of a 96 well ‘V’ bottomed plate, the remaining two columns containing controls (Cisapride). An eight-point half log serial dilution in DMSO was then performed by Tecan Genesis Workstation 200 (Männedorf, Switzerland). A 1:100 dilution was then performed by Platemate 2 × 2 using PBS containing Ca²⁺/Mg²⁺ (Invitrogen, Paisley, UK) into two more 96 well ‘destination plates’, which were run on the machine. IonWorks™ diluted the compound a further threefold, resulting in the top test concentration. Additionally, the two hERG toxins (ErgToxin and rBeKm-1) were diluted directly into PBS containing Ca²⁺/Mg²⁺. The solutions used for IonWorks™ were the same as described by Bridgland-Taylor et al. (2006).

Drugs were either synthesized by and/or on behalf of AstraZeneca (Macclesfield, UK) or sourced from either Sigma-Aldrich Company Ltd or Apin Chemicals (Abingdon, UK). BeKm-1 and rErgtoxin-1 toxins were purchased from Alomone Laboratories Ltd (Bucks, UK).

Electrophysiological comparison

The preparation of the cells is as described above, and the same solutions were used as for the pharmacology runs, but the voltage protocols and set-up varied as follows: the IonWorks™ machine used for the biophysics assays had a 48-channel fluidics head (compared with 12 channels on the pharmacology IonWorks™ machines), and because of this, a 24-well boat was used for the cell suspension, one half with hERG 1a and the other with hERG 1a/1b. The position of each cell line was alternated for each run to ensure there were no edge/position effects. No compound plates were used.

Voltage dependence of activation was estimated using the activation I-V protocol as shown in the inset of Figure 2A. Potential was stepped to test potentials ranging from −80 mV to +60 mV for 1 s, followed by a 1 s step to −50 mV to evoke a tail current. The peak of this tail current was normalized to the maximum tail current as a proxy for relative conductance and plotted against test voltage; the data were fitted to a single Boltzmann function: I/I_max = {1 +exp[(V_1/2−V)/k]}⁻¹, where V_1/2 is the half-activation potential, V is the test voltage and k is the slope factor.

Figure 2.

Activation and deactivation properties of hERG 1a and 1a/1b channels measured using IonWorks™. (A and B) Typical traces of hERG 1a and 1a/1b currents elicited by the voltage protocol shown in the upper inset of (A). (C) Steady state activation plots. The V_1/2 and the slope factor for 1a channels are +8.4 ± 1.3 mV and +8.7 ± 1, respectively (n = 76); and for 1a/1b channels, +4.8 ± 1.1 mV and +8.9 ± 1, respectively (n = 108). (D) Time course of activation. Apparent activation is faster for hERG 1a/1b compared with hERG 1a currents. Time constants of activation were 1262 ± 134 ms and 1104 ± 82 ms for hERG 1a (n = 28–278) and 1a/1b (n = 41–748), respectively. (E) Deactivation is faster for hERG 1a/1b versus 1a currents. Scaled tail currents recorded at −105 mV are shown. (F) Time constants of fast and slow components from double exponential fits to deactivating tail currents are plotted for comparison between 1a and 1a/1b channels. **P < 0.001 versus hERG 1a values.

Currents evoked using an envelope of tails protocol were measured to determine time course of activation. Peak inward tail currents were evoked by a step to −105 mV following a prepulse of increasing duration to 0 mV. Holding potential was −80 mV. The peak amplitudes of the tail currents were plotted against test pulse duration and fit to a single exponential function. Because voltage protocols are limited to a maximum of 10 steps resulting in five depolarization steps, it was not possible to assess the full pulse duration range (Figure 2D), and only five different durations were measured in one run. This is believed to be a limit of the IonWorks™ machine and/or software.

We determined deactivation kinetics from tail currents evoked at −100 mV for 1 s subsequent to a step to +60 mV for 0.5 s (Figure 2E). Fitting the deactivating component of the tail current with the double exponential function y = A₀+A_fe^–t/τf+A_se^–t/τs revealed fast and slow time constants (τ) (Figure 2F).

hERG currents elicited by the three-pulse protocol were used to measure the time course of inactivation. A 0.5 s pulse to +60 mV to activate and then inactivate hERG is followed by a 2-ms pulse to −100 mV to remove the inactivation. In the third pulse, varying the potential between −20 and +60 mV allowed the inactivation time course to be measured as a function of voltage (Figure 3A). The time constants for onset of inactivation were estimated by fitting the decay of the currents in the third pulse to a single exponential function and plotted as a function of test potential.

Figure 3.

Inactivation properties of hERG 1a and 1a/1b channels measured using IonWorks™. (A and B) Typical traces of hERG currents elicited by the three-pulse protocol (upper inset of Figure 2A) to measure the time course of inactivation of 1a and 1a/1b channels. (C) There were no significant differences in the time constants of inactivation for hERG 1a (n = 20–24) and hERG 1a/1b (n = 17–24) channels (P > 0.05 versus hERG 1a values). (D) Plot quantifying data showing recovery from inactivation is faster in hERG 1a/1b (n = 29) compared with hERG 1a (n = 21) channels (P < 0.0001 vs. hERG 1a values). Tail currents were evoked by the protocol shown in the upper insert.

Time constants of recovery from inactivation was measured using a two-pulse protocol, in which the cells were depolarized to +60 mV for 1 s to activate and inactivate hERG channels, and then was repolarized to potentials between −100 and −20 mV to give a tail current (Figure 3D). Time constants were measured as the single exponential fit to the rising phase (more than −80 mV) or as the fast time constant of a double exponential fit (less than or equal to −80 mV) to the tail current.

Statistical analysis

Results are expressed as mean ± SEM. Differences were tested for statistical significance using the unpaired (two sample assuming unequal variances; one cell line vs. another) Student's t-test. A value of P < 0.05 was considered significant.

Results

Electrophysiological properties of hERG 1a and 1a/1b channels using planar patch clamp recordings

The overall objective of this study was to compare the sensitivity of hERG 1a and hERG 1a/1b channels to inhibition by a relatively large number of different compounds (Redfern et al., 2003; Männikköet al., 2010). We used planar patch electrophysiology to facilitate throughput and to ensure a large sample size and sufficient statistical power. To validate this approach, we analysed currents evoked by standard protocols and quantified the macroscopic gating transitions among the closed, open and inactivated states at room temperature. In general, the hERG 1a and 1a/1b currents exhibited characteristics broadly consistent with those previously reported using traditional patch clamp techniques. Most notably, hERG 1a/1b currents exhibited less rectification at positive voltages and faster deactivation upon repolarization compared with hERG 1a currents (Figure 2A and B).

Quasi-steady-state distributions of channels between the open and closed state were determined by measuring tail currents at −50 mV, following a series of 1-s steps to different voltages and fitting the data with a Boltzmann function (Figure 2C). The half-maximal voltage and slope factors were similar for the two channel types. V_1/2 is sensitive to pulse duration (Subbiah et al., 2004) and the values we obtained are consistent with previous reports using similar protocols (Trudeau et al., 1995); longer pulses required to reach steady state, especially at more negative voltages where activation is extremely slow, were not feasible for this study. The time course of current activation was measured using an envelope of tails protocol, with peak tail currents plotted as a function of duration of the preceding depolarizing pulse to 0 mV in sequences of three pulses (see Methods; Figure 2D). Single-exponential fits gave time constants reflecting an approximately 15% faster activation for hERG 1a/1b compared with hERG 1a channels. Similarly, the time course of deactivation, determined from double-exponential fits to currents elicited at −100 mV following a step to +60 mV (Figure 2E), was also faster for hERG 1a/1b, as reflected in 24% and 35% decreases in the fast and slow time constants respectively (Figure 2F). Compared with previous measurements obtained at near-physiological temperatures (Sale et al., 2008), the differences in activation and deactivation kinetics between hERG 1a and hERG 1a/1b currents were qualitatively similar but diminished in magnitude.

Inactivation was evaluated using a three-pulse protocol to isolate the transition between the open and inactivated states (Figure 3A and B). Inactivation time constants, measured from exponential fits to the current decay, showed the intrinsic voltage dependence characteristic of hERG channels (Figure 2C), and were similar for hERG 1a and hERG 1a/1b currents. Recovery from inactivation, measured from fits to the resurgent current as voltage was stepped from +60 mV to a range of more negative voltages, was nearly twofold faster for hERG 1a/1b compared with hERG 1a currents. The similar inactivation time courses, and the approximately twofold difference in the time course of recovery from inactivation (Figure 3D), mirror the results obtained at higher temperatures in a previous study (Sale et al., 2008).

Differences in drug potencies for hERG 1a and 1a/1b channels

Having characterized key electrophysiological differences of the two cell lines on the IonWorks™ platform, we compared their sensitivities to block by 50 diverse compounds (Redfern et al., 2003; Männikköet al., 2010). To determine drug potency, we extracted IC₅₀ values from non-cumulative, eight-point concentration–effect curves, with data normalized to vehicle and 100% blocking levels, and fitted to the Hill equation. We confirmed the previous finding that the hERG blocker E-4031 is more potent for hERG 1a versus hERG 1a/1b channels (Sale et al., 2008). In addition, we observed a surprising range of potencies for other compounds, which fell into three categories: most drugs, such as encainide (Figure 4A), blocked hERG 1a and 1a/1b with similar potency; others, like E-4031 (Figure 4B) and dofetilide, were more potent blockers of hERG 1a compared with hERG 1a/1b currents. Still, others like fluoxetine (Figure 4C) inhibited hERG 1a/1b currents with greater potency.

Figure 4.

Concentration-response curves to encainide (A), E-4031 (B) and fluoxetine (C) against hERG 1a () or 1a/1b channels () measured using IonWorks™. The IC₅₀ values for these compounds are shown in Table 1. Although encainide exhibited similar potency for 1a and 1a/1b, E-4031 and fluoxetine exhibited greater potency for 1a (as previously reported by Sale et al., 2008) and 1a/1b respectively.

A summary of the drug potencies for 50 compounds is provided in Table 1 and shown in a Bland Altman-type plot in which the difference in potency (as difference of log IC₅₀) is plotted versus mean log IC₅₀ (Figure 5A). Drugs showing equal potency for the two targets align with the x-axis (y = 0). More positive y-axis values represent relatively greater potency for hERG 1a, whereas more negative values reflect relatively greater potency for hERG 1a/1b. Although there is a clustering near the y = 0 axis, noteworthy divergence is observed, particularly for ebastine, fluoxetine and desipramine, which exhibit enhanced potency for hERG 1a/1b. In two separate experiments ebastine exhibited IC₅₀ differences corresponding, respectively, to fourfold and eightfold greater potency for hERG 1a/1b over hERG 1a. Across three experiments, fluoxetine exhibited a twofold to fivefold greater potency for hERG 1a/1b channels. Two hERG toxins, BeKm-1 and Erg-Tx, were somewhat more potent for hERG 1a.

Table 1.

Summary of the pharmacology of the human ether-a-go-go-related gene (hERG) homomeric 1a versus heteromeric 1a/1b channels

Compound	Compound abbreviation	EFTPCmax (µM)	hERG	IC50 (µM)	LCL	UCL	pIC50	Hill coefficient	Margin	n
Ajmaline	Ajm	1.6–6.7	1a	5.96	4.82	7.38	5.22	1.23	0.88	7–12
			1a/1b	4.98	4.21	5.89	5.30	1.10	0.74	14–19
Alfuzosin	Alf	0.000015	1a	53.46	46.6	61.4	4.27	1.11	3564000	17–22
			1a/1b	42.26	37.3	47.9	4.37	1.07	2817333	40
Almokalant	Alm	0.15	1a	3.66	3.20	4.19	5.44	1.16	24.4	8–19
			1a/1b	5.25	4.61	5.98	5.28	1.0	34.99	16–30
Amiodarone	Amio	1.5–2.9	1a	7.99	6.36	10	5.09	0.88	2.76	40
			1a/1b	7.27	5.8	9.12	5.14	0.86	2.51	13–20
Amitriptyline	Amit	0.16–0.95	1a	8.04	7.55	8.58	5.09	1.96	8.47	17–25
			1a/1b	5.45	5.07	5.85	5.26	1.53	5.73	17–26
			1a	5.89	5.41	6.41	5.23	1.47	6.20	22–31
			1a/1b	5.12	4.71	5.57	5.29	1.31	5.39	19–29
Amlodopine	Amlo	0.00034–0.00175	1a	11.65	11	12.4	4.93	1.64	6657	27–32
			1a/1b	8.05	7.56	8.58	5.09	1.43	4601	27–34
Astemizole	Ast	0.00026	1a	0.24	0.22	0.26	6.62	1.96	913	20–25
			1a/1b	0.22	0.21	0.24	6.65	2.50	861	22–31
Azimilide	Azi	0.07	1a	1.29	1.22	1.38	5.89	1.37	18.5	40
			1a/1b	1.11	1.04	1.17	5.96	1.45	15.8	29–37
Bepridil	Bep	0.33	1a	1.62	1.52	1.73	5.79	2.01	4.92	21–28
			1a/1b	1.25	1.19	1.32	5.90	2.29	3.80	40
Cetirizine	Cet	0.056	1a	240	205	251	3.62	1.0	4282	14–25
			1a/1b	233	212	246	3.63	1.18	4157	22–30
Chlorpheniramine	Chl	0.011	1a	3.58	3.35	3.82	5.45	1.10	325	23–28
			1a/1b	3.10	2.93	3.29	5.51	1.11	282	40
Cisapride	Cis	0.0049	1a	0.41	0.39	0.44	6.38	1.81	84	28–32
			1a/1b	0.34	0.33	0.36	6.47	2.14	70	23–28
D,L-Sotalol	D,L-S	14.73	1a	705	615	771	3.15	0.93	48	20–25
			1a/1b	747	672	817	3.13	0.93	51	20–31
Desipramine	Des	0.108	1a	16.7	15.3	18.2	4.78	1.58	154	16–22
			1a/1b	7.43	6.92	7.97	5.13	1.62	68	16–25
			1a	13.1	11.6	15.0	4.88	1.75	121	4–9
			1a/1b	7.88	7.27	8.88	5.10	1.42	73	12–20
Diltiazem	Dil	0.122	1a	21.9	20.1	24.1	4.66	1.13	180	16–24
			1a/1b	22.6	20.69	24.9	4.65	1.19	186	16–23
			1a	36.3	27.0	39.8	4.44	1.0	298	40
			1a/1b	26.6	23.2	26.8	4.58	1.0	218	24–30
Diphenhydramine	Dip	0.034	1a	8.76	8.15	9.65	5.06	1.41	258	25–31
			1a/1b	5.98	5.32	6.30	5.22	1.30	176	30–35
d-Norpropoxyphene	d-Nor	0.455–0.652	1a	27.6	25.5	29.9	4.56	1.32	42	19–25
			1a/1b	23.8	21.8	26.1	4.62	1.18	37	18–23
Dofetilide	Dof	0.002	1a	0.097	0.085	0.11	7.01	1.45	48	40
			1a/1b	0.146	0.128	0.17	6.83	2.13	73	40
Domperidone	Dom	0.019	1a	0.72	0.69	0.76	6.14	1.92	38	25–31
			1a/1b	0.70	0.67	0.74	6.15	2.22	37	40
E-4031	E4031	NA	1a	0.084	0.069	0.102	7.07	1.33	NA	40
			1a/1b	0.119	0.09	0.156	6.93	1.21	NA	40
			1a	0.075	0.067	0.084	7.13	1.0	NA	15–31
			1a/1b	0.124	0.114	0.135	6.91	1.59	NA	19–28
			1a	0.068	0.06	0.076	7.16	1.25	NA	40
			1a/1b	0.123	0.11	0.136	6.91	1.29	NA	40
			1a	0.055	0.049	0.061	7.26	1.21	NA	40
			1a/1b	0.101	0.09	0.113	6.99	1.24	NA	40
			1a	0.064	0.051	0.079	7.20	1.0	NA	20
			1a/1b	0.123	0.102	0.149	6.90	0.99	NA	20
			1a	0.061	0.056	0.068	7.21	1.32	NA	16–26
			1a/1b	0.133	0.118	0.15	6.88	1.0	NA	17–26
			1a	0.089	0.051	0.108	7.05	1.83	NA	4–14
			1a/1b	0.171	0.133	0.194	6.77	1.48	NA	11–17
			1a	0.077	0.070	0.084	7.11	1.28	NA	18–21
			1a/1b	0.117	0.105	0.129	6.93	1.37	NA	18–24
			1a	0.097	0.072	0.127	7.01	1.47	NA	20
			1a/1b	0.166	0.117	0.191	6.74	1.30	NA	20
Ebastine	Eba	0.0051	1a	11.47	10.1	13.8	4.94	1.09	2249	12–20
			1a/1b	2.68	2.50	2.92	5.57	1.97	525	21–31
			1a	32.5	24.5	38.7	4.49	1.05	6380	24–28
			1a/1b	4.3	3.70	4.59	5.37	1.80	846	27–34
Encainide	Enc	0.061	1a	5.42	5.08	5.79	5.27	1.0	89	22–27
			1a/1b	5.38	4.99	5.8	5.27	0.97	88	18–29
ErgToxin	Erg	NA	1a	0.017	0.11	0.027	7.77	0.59	NA	20
			1a/1b	0.029	0.027	0.041	7.52	0.71	NA	20
Fexofenadine	Fex	0.345	1a	501	417	603	3.29	1.0	1453	25–33
			1a/1b	531	436	647	3.27	0.97	1538	25–30
Flecainide	Flec	0.753	1a	2.63	2.3	3.0	5.58	1.12	3.5	40
			1a/1b	2.55	2.19	2.96	5.59	1.0	3.38	40
Fluoxetine	Flu	0.46–1.44	1a	2.41	2.24	2.58	5.61	1.47	1.67	22–31
			1a/1b	0.46	0.44	0.49	6.33	2.06	0.32	20–31
			1a	6.34	5.50	7.22	5.20	1.92	4.41	2–10
			1a/1b	2.89	2.68	3.21	5.54	1.45	2.0	25–32
			1a	5.76	4.72	7.11	5.24	1.48	3.99	4–12
			1a/1b	3.03	2.80	3.33	5.52	1.44	2.10	15–22
Fluvoxamine	Fluv	0.34–0.57	1a	8.39	7.68	9.18	5.08	1.41	15	17–26
			1a/1b	7.22	6.61	7.89	5.14	1.08	13	16–23
Haloperidol	Hal	0.013–0.66	1a	0.30	0.28	0.32	6.52	1.59	0.46	40
			1a/1b	0.31	0.29	0.33	6.50	1.75	0.48	40
Impramine	Imp	0.106	1a	10.13	9.43	10.9	4.99	1.39	96	22–29
			1a/1b	6.83	6.37	7.32	5.17	1.47	64	17–26
Ketanserin	Keta	0.018	1a	0.75	0.69	0.87	6.13	1.24	41	9–19
			1a/1b	0.44	0.42	0.48	6.36	1.36	24	17–26
Ketoconazole	Keto	0.177	1a	4.61	4.29	4.96	5.34	1.53	26	19–34
			1a/1b	3.74	3.45	4.04	5.43	1.44	21	17–23
Loratadine	Lor	0.00045	1a	10.9	10.0	11.8	4.96	1.68	24222	40
			1a/1b	10.3	9.44	11.1	4.99	1.83	22800	40
Mefloquine	Mef	0.0952	1a	14.1	12.7	15.7	4.85	1.0	149	23–31
			1a/1b	10.9	10.2	11.6	4.96	1.80	114	40
Mibefradil	Mib	0.012	1a	2.96	2.79	3.13	5.53	2.0	246	14–32
			1a/1b	2.53	2.39	2.67	5.59	2.04	210	9–28
Moxifloxacin	Mox	1.87–3.11	1a	165	145	187	3.78	0.95	53	18–25
			1a/1b	128	119	138	3.89	0.86	41	25–35
Nitrendipine	Nit	0.003	1a	7.98	6.52	9.75	5.09	0.80	2640	22–33
			1a/1b	7.41	6.38	8.6	5.13	0.93	2452	40
Olanzapine	Ola	13.63	1a	9.99	9.39	10.6	5.00	1.23	0.73	22–32
			1a/1b	6.61	6.08	7.17	5.18	1.0	0.48	20–28
Pimozide	Pim	0.001	1a	0.38	0.35	0.41	6.42	2.43	381	15–23
			1a/1b	0.49	0.46	0.53	6.31	2.38	491	23–32
Procainamide	Proc	54.186	1a	418	361	482	3.38	1.10	8	14–26
			1a/1b	381	336	431	3.42	1.11	7	15–24
Propafenone	Prop	0.241	1a	1.31	1.21	1.41	5.88	1.38	5	19–29
			1a/1b	0.79	0.73	0.87	6.09	1.47	3	6–18
			1a	1.26	1.15	1.39	5.89	1.17	5	20
			1a/1b	0.84	0.77	0.92	6.08	1.23	4	20
rBeKm-1	rBeKm-1	NA	1a	0.009	0.0069	0.0122	8.04	0.72	NA	20
			1a/1b	0.016	0.0146	0.0183	7.79	1.0	NA	20
Risperidone	Ris	0.00181	1a	0.69	0.61	0.76	6.17	1.0	375	19–30
			1a/1b	0.62	0.57	0.67	6.21	1.34	342	16–27
Sertindole	Ser	0.00159	1a	0.56	0.49	0.65	6.25	1.69	352	11–14
			1a/1b	0.74	0.67	0.81	6.13	1.98	466	12–18
Sulpiride	Sul	0.149–1.171	1a	805	680	866	3.09	1.0	688	8–20
			1a/1b	1260	1050	1400	2.89	0.93	1076	13–25
Terfenadine	Terf	0.009	1a	0.95	0.86	1.04	6.02	2.40	105	8–15
			1a/1b	1.34	1.23	1.47	5.87	1.94	149	7–14
Terodiline	Tero	0.012	1a	2.13	1.97	2.29	5.67	1.13	177	17–33
			1a/1b	1.51	1.38	1.65	5.82	1.07	126	16–25
Thioridazine	Thi	0.24–4.91	1a	1.99	1.85	2.13	5.70	1.55	0.4	17–25
			1a/1b	2.65	2.33	3.01	5.58	1.0	0.5	40
Verapamil	Ver	0.081	1a	1.55	1.44	1.68	5.81	1.61	19	40
			1a/1b	1.62	1.51	1.74	5.79	1.44	20	40
Ziprasidone	Zip	0.0041	1a	0.45	0.38	0.53	6.35	2.23	109	7–14
			1a/1b	0.43	0.38	0.48	6.37	2.65	104	13–16
			1a	0.84	0.80	0.99	6.08	1.26	204	5–12
			1a/1b	1.11	0.93	1.31	5.96	1.05	270	20

EFTPC_max, maximum effective free therapeutic plasma concentration (Redfern et al., 2003; Schulz and Schmoldt, 2003); IC₅₀, concentration producing 50% inhibition of the hERG channel; LCL, lower 95% confidence limit; UCL, upper 95% confidence limit; pIC₅₀ = -Log10 of the IC₅₀; Margin, hERG IC₅₀ divided by the EFTPC_max or the upper end of the EFTPC_max range; n, number of cells tested; NA, not applicable. Note that reproducibility of IC₅₀ data was assessed by testing some compounds in multiple runs.

Figure 5.

(A) The potency Bland-Altman plot to illustrate the difference in potency (pIC₅₀) as a function of the average potency of a compound for hERG 1a versus 1a/1b channels. The axis y = 0 represents no difference in the potency for the two hERG channels. Negative deviations represent greater potency for hERG 1a/1b versus 1a, whereas positive deviations represent greater potency for hERG 1a versus 1a/1b. 49 hERG blockers, previously selected for their diversity (Männikköet al., 2010), were tested. (B) Safety margin Bland-Altman plot illustrates the difference in margin as a function of the average margin of a compound for hERG 1a versus 1a/1b channels. The axis y = 0 represents no difference in the safety margin for the two hERG channels. Negative deviations represent lower margins for hERG 1a versus 1a/1b, whereas positive deviations represent lower margins for hERG 1a/1b versus 1a channels. The safety margin was defined as hERG IC₅₀ (µM)/EFTPC_max (µM).

Analysis of the safety margin for each compound, represented as the ratio of the IC₅₀ for hERG inhibition and the EFTPC_max, is shown in Table 1 and Figure 5B. Many compounds, like encainide and ziprasidone, show no difference in the safety margin for hERG. However, compounds such as ebastine and fluoxetine have smaller safety margins for hERG 1a/1b versus 1a channels, whereas other compounds like sulpiride and dofetilide have larger safety margins for hERG 1a/1b versus 1a channels.

Discussion and conclusions

The hERG block assay is one component of the ‘integrated risk assessment,’ a process intended to ensure that drugs entering the market are less likely to cause QT prolongation, TdP arrhythmias and sudden cardiac death (Pollard et al., 2010; Valentin et al., 2010). A double-edged sword, the assay has decreased the fraction of lead compounds entering the pipeline because so many of these compounds exhibit hERG block. Essential to the combined goals for safety and availability of new drugs is the development of proxies for hERG block that faithfully represent the native channel. Building on more recent studies showing two isoforms expressed in heart tissue (Jones et al., 2004; Wang et al., 2008), we sought to characterize pharmacological sensitivities of heteromeric hERG channels to a range of diverse compounds known to block hERG. We used the IonWorks™ platform to increase throughput and facilitate the screening of 50 diverse compounds. Despite its shortcomings (Kiss et al., 2003; Wang and Li, 2003; Bridgland-Taylor et al., 2006), the IonWorks™ platform was shown to yield data on voltage dependence of activation that were comparable to data generated using conventional electrophysiology (this study; Schroeder et al., 2003; Tao et al., 2004; Bridgland-Taylor et al., 2006; Harmer et al., 2008). Additionally, this study reports that currents produced using this platform replicated those produced by conventional electrophysiology with respect to qualitative differences between hERG 1a and 1a/1b gating kinetics (Sale et al., 2008), with high reliability and strong statistical power based on the large number of samples afforded by this approach compared with manual patch clamp studies (Figures 2 and 3). Surprisingly, some drugs more potently blocked hERG 1a channels and others more potently blocked the hERG 1a/1b heteromers; up to eightfold difference in IC₅₀ values were observed (Figures 4 and 5).

Given that the 1a and 1a/1b channels are identical in primary sequence throughout the hydrophobic core comprising the drug-binding sites (Lees-Miller et al., 2000; Mitcheson et al., 2000a), it is likely that differences in drug sensitivity arise from differences in gating. Compared with hERG 1a homomers, 1a/1b channels spend more time in the open state because of faster activation and a less stable inactivated state (Sale et al., 2008). Thus, we can speculate that drugs like fluoxetine, with greater potency for 1a/1b channels, have a higher affinity for the open state, whereas drugs like dofetilide showing greater potency for 1a may have a higher affinity for the inactivated state. The role of inactivation in hERG block is still a matter of debate because pulse protocols designed to facilitate entry into the inactivated state do not promote inhibition by dofetilide (Snyders and Chaudhary, 1996). However, nearly all mutations disrupting inactivation also reduce apparent affinity (Wang et al., 1997; Ficker et al., 1998; 2001; Herzberg et al., 1998; Lees-Miller et al., 2000). In addition, closely related channels in the Eag family share critical drug-binding residues, but do not exhibit high-affinity drug block unless residues critical for inactivation in hERG are substituted in the outer pore region, simultaneously conferring inactivation and drug block (Ficker et al., 1998; 2001; Herzberg et al., 1998). Thus, some drugs may promote inactivation gating by an allosteric mechanism expected to be more pronounced in hERG 1a compared with 1a/1b channels. Moreover, given the effect of temperature on hERG gating transitions (Zhou et al., 1998; Sale et al., 2008), it will be important to determine whether the differences in drug potencies for 1a versus 1a/1b are enhanced at physiological temperatures compared with the lower temperatures used in this study.

The automated plate-based electrophysiology platform provides a rapid method for comparing a large number of compounds in pair-wise tests that would have been prohibitively labour-intensive using the manual patch clamp technique, but several limitations must be considered. One such limitation is greater between-run variability. For example, the data in Figure 5A indicate some real differences in sensitivity when using a tightly controlled experimental design (i.e. 1a and 1a/1b cells in the same test plate). Based on a variation of ±0.1 log₁₀ units, 38% of compounds showed comparable IC₅₀s, whereas 29% and 32% of them were found to be more potent against 1a and 1a/1b channels respectively. However, over a 1-year period, the potency of cisapride in our standard IonWorks™ hERG 1a assay varied by ±0.3 log₁₀ units (i.e. ±2.0-fold; sample size of 1660 IC₅₀ values per year) (unpublished observation), not much less than the difference in potency of some drugs for 1a versus 1a/1b. At this level of variability, only 1.5% and 8% of compounds were found to be more potent against 1a and 1a/1b channels, respectively. This suggests only a very small percentage of compounds could confidently be considered to have a different potency in hERG 1a/1b cells when running longitudinal drug studies rather than side-by-side determinations of IC₅₀ values.

Other considerations relating to our use of the planar patch clamp approach include pulse protocols and the plate materials. To maximize the stability of recordings, we used relatively short pulse durations and did not maintain clamp between pulses. We do not know the membrane potential between pulses, but we expect it would be somewhat depolarized due to leak through the seal, and this in turn would increase the number of channels in the open state and potentially accessible to drug. We observed IC₅₀ values for fluoxetine in the low micromolar range, as found in manual patch clamp studies using longer pulses, suggesting that drug binding was at steady state in both cases (Thomas et al., 2002; Kirsch et al., 2004; Hancox and Mitcheson, 2006; Rajamani et al., 2006). If there is an error attributable to the shorter pulses, it is likely to result in an underestimate, rather than an overestimate, of the differences in drug potencies for the 1a and 1a/1b targets: for E-4031 we observed at most twofold differences in IC₅₀ using the IonWorks™ compared with fourfold differences observed with manual patch (Sale et al., 2008). Future studies will be required to determine the effects of pulse duration and frequency used in studies such as this on drugs with different on rates, use dependence and potential for trapping. There are also some discrepancies in IC₅₀ measurements attributable to adsorption of some drugs to the plate matrix (Guo and Guthrie, 2005; Sorota et al., 2005), but this concern does not apply when comparing potencies for the 1a versus 1a/1b targets within a given platform. The differences we have identified in drug potencies thus provide strong justification for screening many more compounds; depending on channel composition, compounds exhibiting selectivity for the 1a or 1a/1b channels may ultimately be found to have specific safety profiles or even direct therapeutic applications.

It is important to ask how faithful a proxy the hERG 1a channel is for I_Kr, and whether more drugs would be developed were we able to reconstitute the native channel more accurately. Previous studies have identified interacting proteins, both cytosolic and membrane-resident, affecting hERG drug sensitivity (McDonald et al., 1997; Abbott et al., 1999; Kupershmidt et al., 2003). Our current study shows, in addition, that the α-subunit composition of the channel can affect its sensitivity to block. Expression of 1a and 1b subunits are developmentally regulated in the mouse, where both 1a and 1b subunits contribute to foetal and neonatal cardiac I_Kr, whereas in adult the 1a subunit predominates (Wang et al., 2008). Selective knockout of the 1b isoform causes a corresponding reduction of I_Kr in neonates and perturbation of rhythmic activity (Lees-Miller et al., 2003). If a similar developmental expression profile holds for humans, then infants and children are more likely than adults to be adversely affected by drugs with higher potency for 1a/1b compared with 1a channels. Such a prediction is consistent with our identification of a 1b-specific mutation in an eight-year-old LQTS patient (Sale et al., 2008). Indeed, our findings raise the question whether QT prolongation reported for free plasma drug concentrations associated with only low levels of hERG 1a inhibition in vitro (Valentin and Hammond, 2008) might reflect the drug's more potent action on native channels comprising hERG 1a and 1b subunits or other components. However, at this time we only have limited information on expression of the 1b subunit in human heart (Jones et al., 2004), and we do not know the stoichiometry of channels in the 1a/1b cell line. Future experiments aimed at determining the scope of I_Kr subunit composition as a function of biological variables like age and sex, and the extent to which the stable cell line reflects the native state, may aid in the urgent task of stratifying risk for acquired LQTS and safely developing more drugs.

Glossary

Abbreviations

EFTPC_max

maximum effective free therapeutic plasma concentration

hERG

ether-a-go-go-related gene

LQTS

long QT syndrome

TdP

torsades de pointes

Conflicts of interest

The authors state no conflict of interest.