hERG S4-S5 linker acts as a voltage-dependent ligand that binds to the activation gate and locks it in a closed state

Delayed-rectifier potassium channels (hERG and KCNQ1) play a major role in cardiac repolarization. These channels are formed by a tetrameric pore (S5–S6) surrounded by four voltage sensor domains (S1-S4). Coupling between voltage sensor domains and the pore activation gate is critical for channel voltage-dependence. However, molecular mechanisms remain elusive. Herein, we demonstrate that covalently binding, through a disulfide bridge, a peptide mimicking the S4-S5 linker (S4-S5L) to the channel S6 C-terminus (S6T) completely inhibits hERG. This shows that channel S4-S5L is sufficient to stabilize the pore activation gate in its closed state. Conversely, covalently binding a peptide mimicking S6T to the channel S4-S5L prevents its inhibiting effect and renders the channel almost completely voltage-independent. This shows that the channel S4-S5L is necessary to stabilize the activation gate in its closed state. Altogether, our results provide chemical evidence that S4-S5L acts as a voltage-controlled ligand that binds S6T to lock the channel in a closed state, elucidating the coupling between voltage sensors and the gate in delayed rectifier potassium channels and potentially other voltage-gated channels.

results, hERG-mimicking peptide effects were moderate. In a previous study on this channel, a disulfide bond was created between S4-S5 L and S6 T , stabilizing it in a closed state 6 . Here, after the first step of peptide identification, we used the same cysteine approach to increase the peptide-channel interaction and to target the peptide to its hypothetical receptor (S6 T for S4-S5 L and vice-versa). We observed that covalently binding a peptide mimicking S4-S5 L to the endogenous S6 T could completely inhibit hERG. On the other hand, covalently binding a peptide mimicking S6 T to the endogenous S4-S5 L rendered the channel almost voltage-independent, most likely by preventing the inhibitory effect of S4-S5 L on S6 T in the channel.
Altogether, these results suggest that the voltage dependence of these two delayed rectifier channels, KCNQ1 and hERG, follows a ligand/receptor model in which S4-S5 linker acts as an inhibitor, locking the activation gate under the control of the voltage sensor S4.

Results
One S4-S5 L peptide inhibits hERG channels. To identify a potential inhibitory S4-S5 L peptide, three S4-S5 L encoding plasmids were designed based on sequence alignment with KCNQ1 (Fig. 1B), in which the ligand/receptor model was first suggested, the S4-S5 linker (S4-S5 L ) being the inhibiting ligand and the S6 C-terminus (S6 T ) acting as the receptor 11 . To study peptide effects on channel activity, each plasmid coding for one of the S4-S5 L peptides was co-transfected with a plasmid encoding the hERG channel.
If the endogenous S4-S5 L acts like a ligand to inhibit the activation gate, then a peptide mimicking endogenous S4-S5 L should decrease hERG channel activity. As expected, the S4-S5 L (+3) peptide decreased the current density ( Fig. 2A and B, Supplemental Table 1). Western blot experiments revealed that this decrease was not due to a Upon membrane depolarization, S4 pulls S4-S5 L out of the S6 T receptor, allowing channel opening. The S4-S5 L peptide (red) mimics endogenous S4-S5 L , locking the channel in its closed conformation. Contrarily, S6 T peptide (green) binds to the endogenous S4-S5 L and limits its locking effect, leading to channel up-modulation. (B) Alignment used to design hERG peptides from previously identified KCNQ1 S4-S5 L and S6 T peptides (based on the multiple alignment obtained using Clustal Omega, presented in Supplemental Fig. 4). In red are represented the basic residues, in yellow acidic residues, and in purple the position of the narrowest part of the bundle crossing, also named the gating residue (see methods). The color boxes represent the transmembrane segments. Grey lines represent the peptides tested in KCNQ1 while red lines and green lines represent the designed hERG S4-S5 L and S6 T peptides, respectively. A check sign (✓) indicates that the KCNQ1 S4-S5 L peptide inhibits the channel (red) and that the KCNQ1 S6 T peptide activates the channel (green).
Scientific RepoRts | 7: 113 | DOI:10.1038/s41598-017-00155-2 lower expression of hERG protein in the presence of this peptide, as compared to control conditions ( Fig. 3A and C). Thus, these results are consistent with the ligand/receptor model. However, the S4-S5 L (−3) peptide did not show any effect on current density and led to a 5-mV negative shift in the activation curve ( Fig. 2A-C, Supplemental Table 1). The S4-S5 L (−3) peptide presents 5 charged amino-acids, 3 positively charged on one face of the helix and 2 negatively charged on the other face. The absence of any effect on current density suggests a non-specific effect such as membrane charge screening, locally changing the potential detected by the voltage-sensor.
The S4-S5 L (0) peptide induced a paradoxical increase in hERG current density without any effect on the activation curve ( Fig. 2A-C, Supplemental Table 1). Western blot experiments revealed that this peptide led to a doubling of hERG protein as compared to control conditions ( Fig. 3A and C). Surprisingly, this peptide also induced an increase in GFP expression (by a factor of 7.1, p < 0.01), as shown in Fig. 3A. However expression of endogenous control GAPDH was not altered by S4-S5 L (0) ( Fig. 3A and C).
This set of experiments shows specific effects of only one of the S4-S5 L . The S4-S5 L (+3) peptide decreased the current density without modifying the activation curve, similar to the effects of the S4-S5 L inhibiting peptides on KCNQ1 channel 11 . This is consistent with the ligand/receptor model. and in the presence of S4-S5 L or S6 T peptides (0.6 µg hERG plus 1.4 µg peptide plasmids). Left: schemes of the hypothetical effects of S4-S5 L inhibiting or S6 T activating peptide; activation voltage protocol used (one sweep every 8 s) located above the WT hERG currents; right: data from experiments performed in the presence of various S4-S5 L or S6 T peptides. (B) Mean hERG tail-current density at −40 mV after a prepulse at +60 mV in the presence of S4-S5 L peptides. (C) Activation curve, obtained from tail currents using the protocol shown in A, in the presence of S4-S5 L peptides (n = 21-65). (D) Inactivation curve in the presence of S4-S5 L peptides (n = 6-24). Inset, Inactivation voltage protocol used (one sweep every 5 s, first pulse = 1 s, second pulse = 15 ms, third pulse = 0.5 s). (E) Through (G) same as B through D, in the presence of S6 T peptides (F, n = 28-65; G, n = 15-24). *p < 0.1, ***p < 0.01 versus control hERG, Mann-Whitney test.
Scientific RepoRts | 7: 113 | DOI:10.1038/s41598-017-00155-2 One S6 T peptide activates hERG channels. If S6 T is the receptor for S4-S5 L , then a peptide mimicking endogenous S6 T should act as a decoy to the endogenous S4-S5 L , preventing its binding to the endogenous S6 T and increasing the channel activity. Indeed, the S6 T (−3) peptide up-regulated the WT channel and shifted the activation curve (and inactivation curve) by 5 mV toward negative potentials ( Fig. 2A,E-G, Supplemental Table 1) without any change in the hERG expression level ( Fig. 3B and C), which is similar to the effects of the S6 T -activating peptides on KCNQ1 channel 11 and therefore also consistent with the ligand/receptor model. The other peptides, namely S6 T (0) and S6 T (+3) did not produce any effect.
Altogether, the effects of one inhibiting and one activating peptide were similar for both hERG and KCNQ1 channels 11 . Moreover, both hERG inhibiting and activating peptides are aligned with those of the KCNQ1 channel ( Fig. 1). In conclusion, these two similarities suggest that the ligand/receptor model applies to both channels.
A cysteine disulfide bond reinforces the effect of the S4-S5 L (+3) peptide, leading to full inhibition of hERG channel. Although consistent with the ligand/receptor model, the S4-S5 L (+3) effect is moderate, as in KCNQ1, likely because the S4S-5 L /S6 T interaction has to be loose, as explained in the introduction. In Xenopus oocytes, introduction of a cysteine pair in the hERG channel (D540C in S4-S5 L with L666C in S6 T ) has been previously shown to stabilize the channel in a closed conformation through the formation of a disulfide bond 6 . We reasoned that this disulfide bond could be exploited to covalently bind the mimicking S4-S5 L (+3) peptide to the S6 T channel, which should enhance its inhibitory effect.
First, we evaluated the effects of oxidizing the native cysteines on WT hERG activity. We observed that in COS-7 cells, as previously shown in Xenopus Oocytes, WT currents were unaffected after 2 hours of incubation with 0.2 mM tert-butylhydroperoxide (tbHO 2 ; Fig. 4A,C and D, Supplemental Table 2), showing that oxidizing the native cysteines does not have any effect on channel gating and trafficking. Such a tbHO 2 treatment led to a permanent closure of the D540C-L666C hERG mutant and this inhibition was reversed by subsequent incubation with 10 mM DTT ( Fig. 4B and E, Supplemental Table 2). Finally, the current density of the single mutants D540C and L666C was not altered by incubation with tbHO 2 (Supplemental Fig. 2), confirming that the D540C-L666C mutant inhibition by tbHO 2 is specifically due to the disulfide bond formed between D540C and L666C. The inhibiting S4-S5 L (+3) peptide does not contain the D540 residue ( Fig. 1), but the E544 residue, located on the same side of the S4-S5 L helix as D540, is negatively charged as well. Both D540 and E544 may interact electrostatically with R665 (next to L666) in S6 T when the channel is in the closed state 12,13 . Therefore, we tested if a cysteine introduced in E544 hERG instead of D540 and another cysteine introduced in L666 hERG could also form a disulfide bridge, which would lead to a permanent closure of the channel. COS-7 cells were transfected with the E544C-L666C double mutant hERG, and the generated current was indeed inhibited by incubation with 0.2 mM tbHO 2 , while the E544C and L666C single mutants were not (Fig. 5A, Supplemental Fig. 2). Since E544C/ L666C can form a disulfide bond within the channel, we tested if this same cysteine pair could also covalently link the E544C mutated S4-S5 L (+3) peptide onto the S6 T of L666C single mutant hERG channel. Cells were co-transfected with the E544C S4-S5 L (+3) peptide and the L666C channel. Two hours of incubation with 0.2 mM tbHO 2 almost completely inhibited the current (Fig. 5B, Supplemental Table 3), and this effect was specifically due to the presence of both cysteines ( Fig. 5C and D, Supplemental Table 3). In conclusion, covalently binding S4-S5 L to S6 T inhibited the channel, strongly suggesting that S4-S5 L acts as a ligand binding to S6 T and locking the channel in a closed state. hERG inhibition provoked by E544C S4-S5 L (+3) peptide covalent binding to L666C mutant channel is reversible. Cells co-transfected with the E544C S4-S5 L (+3) peptide and the L666C channel were incubated for 2 hours with tbHO 2 , followed by 2 hours incubation with 10 mM DTT. This double treatment reversed the effect induced by tbHO 2 incubation, as current densities were comparable to the control condition (Fig. 6). To exclude that the current recovery is due to de novo arrival of hERG at the plasma membrane during the 2 hours DTT incubation, we tested two additional conditions: 1) 10 mM DTT was applied for only 10 minutes or 2) we measured the current in the presence of 10 mM DTT in the pipette tip. In both conditions, current densities were similar to control, consistent with DTT application leading to peptide unbinding. hERG activation by S4-S5 L (0) peptide is not reinforced by its covalent binding through a disulfide bond. S4-S5 L (0) peptide increased WT hERG current density ( Fig. 2A and B), but this effect is due to an increased expression of hERG protein, suggesting no effect of S4-S5 L (0) on channel gating. Consistent with this hypothesis, the current generated by COS-7 cells co-transfected with E544C S4-S5 L (0) peptide and L666C hERG channel was not increased upon incubation with tbHO 2 (Fig. 7A, Supplemental Table 4). Since S4-S5 L (0) also includes the D540 residue ( Fig. 1), we also tested the effect of D540C S4-S5 L (0) peptide on L666C single mutant hERG channel. Again, the current was not increased upon tbHO 2 application (Fig. 7B, Supplemental Table 4). Altogether, these observations confirmed that the paradoxical activation effect initially observed when S4-S5 L (0) peptide was co-transfected with WT hERG channel was not due to an interaction of this peptide with the S6 T region of hERG.
A cysteine disulfide bond reinforces the effect of the activating S6 T peptide, rendering the channel almost voltage independent. Since the S6 T peptide contains the L666 residue, we mutated this residue to a cysteine and tested the effect of this mutated peptide on the D540C single mutant hERG channel,   both in the absence and in the presence of tbHO 2 . As shown in Fig. 8C, the D540C channel presents an atypical activation curve, with a double Boltzmann and a voltage-independent component, even at very negative potentials, unlike the WT channel. This suggests that the mutation D540C by itself alters the role of endogenous S4-S5 L as an inhibiting ligand. This voltage-independent component is similar to the one induced by some mutations in KCNQ1 that were interpreted as decreasing the interaction between S4-S5 L and S6 T 11,14 . Covalent binding of the L666C S6 T (−3) peptide to the D540C hERG channel dramatically increased the voltage-independent component ( Fig. 8B and D). Such a tbHO 2 -dependent increase in the voltage-independent component was not observed in absence of the peptide (Fig. 8A and C). The voltage-independent component is potassium selective (Fig. 8E). These observations further confirm the ligand/receptor mechanism by demonstrating that S4-S5 L is necessary to lock the channel gate in a closed state.
Covalent binding of the L666C S6 T (−3) peptide to the channel D540C also leads to a 12-mV negative shift in the inactivation curve ( Fig. 8G and Supplemental Table 5). In the hERG channel, inactivation is not directly coupled to activation 15,16 , suggesting that the observed effect of the L666C S6 T (−3) peptide on the inactivation curve is not a consequence of a modification of activation. Of note, S6 T peptides bind to hERG S4-S5 L (Y542-F557) which is surrounded by regions that are involved in inactivation: double mutant cycle analysis suggests that interaction between L529, L532, V535 in S4 and I560, L564, I567 in S5 play a role in inactivation gating 17 . Thus, covalent binding of S6 T to S4-S5 L may alter this interaction leading to the observed effect on inactivation.

Discussion
In a previous study on KCNQ1, and in the present study on hERG, we proposed a ligand/receptor model of voltage dependence. In this model, S4-S5 L would act as a ligand locking the S6 T gate in the closed state. To test this, we used S4-S5 L and S6 T mimicking peptides. Our results are in favor of this hypothesis. However, the effects were moderate. Here, we used another approach to reinforce the peptide effects on the hERG channel by introducing a cysteine pair covalently linking the mutated peptide to the single mutant channel. We observed that covalently linking the E544C S4-S5 L (+3) peptide to the L666C hERG almost completely inhibits the channel (Fig. 5B), by presumably preventing the outward splaying of the S6 helices. Covalently linking the L666C S6 T (−3) peptide to D540C hERG renders the channel almost completely voltage independent (Fig. 8D). We postulate that the covalently bound S6 T (−3) peptide blocks the normal interaction between endogenous S4-S5 L and S6 T which stabilizes the closed state, and thus the channel remains locked in the open state. For both peptides, either S4-S5 L or S6 T , covalent link to the hERG channel clearly reinforces the respective effects of these two peptides when first tested on WT hERG channel without any introduced cysteine (as in Fig. 2). Altogether, these observations further reinforce the ligand/receptor model of voltage dependence. hERG S4-S5 L and S6 T peptides that produce the most potent effects are aligned with the most potent peptides of KCNQ1, strengthening the notion of a similar model for KCNQ1 and hERG activation gating (Fig. 9).
Cysteine introduction either in S4-S5 L or S6 T endogenous segments leads to alteration of the single mutant channel activity, as shown in Supplemental Fig. 3. This alteration is indicative of a change in the channel conformation. This observation may give rise to an alternative hypothesis: an artificial S4-S5 L /S6 T interaction is only due to the cysteine introductions and does not exist in WT conditions. However, (i) S4-S5 L (+3) and S6 T (−3) peptides altered WT channel current density, suggesting that the ligand/receptor model is relevant for the WT situation even without the disulfide bond, (ii) despite the alteration of the channel voltage dependence due to the cysteines introduction, covalently linking the peptide to its target leads to effects which are consistent with the ligand/receptor model. Indeed, we observed, on one hand, a complete inhibition by E544C S4-S5 L (+3) on L666C hERG, and, on the other hand, a loss of voltage dependence by L666C S6 T (−3) on D540C hERG. In conclusion, two independent sets of experiments (with or without cysteines) are consistent with the ligand/receptor model. Interestingly, the same combination of cysteine introduction in S4-S5 L (0) peptide (E544C) and hERG (L666C), led to a complete loss of the current increase which was demonstrated to be nonspecific to the channel activity change but rather due to the channel expression increase when probed by western blot experiments. Altogether, our results suggest that cysteine introduction in the channel preserved the ligand/receptor mechanism that was originally observed in WT hERG (and also KCNQ1).
In the absence of structural data of K v channels in the closed state, molecular mechanisms underlying the coupling between voltage-sensor movement and pore opening are still under debate. This coupling interaction has been suggested to be electromechanical and categorized as "attractive" or "repulsive", depending on the force exerted on the gate by the voltage sensors via the S4-S5 L and S6 T interaction 18 . In the "attractive" coupling, the S4 voltage sensors pull the gate to open it, requiring tight binding between S4-S L and S6 T. In the "repulsive" coupling, the voltage sensors push the gate to constrict it. In the present work on hERG channel, the ligand/receptor model corresponds to a third mechanism in which there is no need of a force directly linking the voltage-sensor and the gate, neither "attractive" nor "repulsive". Rather, S4-S5 L seems to act as a ligand that binds S6 T to lock the channel in a closed state. When S4 segments are in the "down" state at negative potential, S4-S5 L binds to S6 T , locking the channel in the closed state. When S4 segments are in the "up" state, they pull S4-S5 L out of its binding pocket (S6 T ), unlocking the channel. This molecular mechanism is consistent with previous observations made on hERG channels and other K v channels: . hERG inhibiting (S4-S5 L (+3)) and activating peptides (S6 T (−3)) are aligned with KCNQ1 inhibiting and activating peptides. Same as in Fig. 1B, onto which the results obtained with hERG have been added. A check sign (✓) indicates that the S4-S5 L peptide inhibits the channel and that the S6 T peptide activates the channel.
• Ferrer and collaborators showed that introduction of two cysteines, one in S4-S5 L and one in S6 T , in the hERG channel, locks it in a closed state via a disulfide bond 6 . This is consistent with our hypothesis that S4-S5 L acts as an endogenous inhibitor, as confirmed by the exogenous peptide approach. • Split hERG channels expressed as two distinct proteins, one containing the voltage sensor domain and one containing the pore, keep the same voltage dependence as the full-length channel 19 . In accordance with our model, where there is no mechanical force transduction from S4 movement to S6 T opening, there is no need for physical continuity between the two parts of the channel for activation transduction. • The ligand/receptor model also helps in understanding why some hERG mutants, such as D540K, are open by both membrane depolarization and membrane hyperpolarization 20 . Since specific interactions between S4-S5 L and S6 T are needed to keep the channel closed, a charge reversal mutation on S4-S5 L may destabilize the interaction and lead to channel opening when S4-S5 L is pulled out of its S6 T binding pocket, either toward the cytosol (hyperpolarization) are toward the membrane (depolarization). • Of note, it has been recently shown that the structure of the K v 10.1 channel, which is 65% homologous to hERG, is incompatible with the mechanical lever mechanism 21 . In the open state structure of K v 10.1, S4-S5 L is a short linker that is not domain swapped and, thus, is not in a position to function as a mechanical lever, leaving the possibility that the K v 10.1 follows the same ligand/receptor model as hERG and KCNQ1.
This ligand/receptor model implies that when S4-S5 L is absent, the hERG activation gate should be in the open state, independent of the potential, as observed in one potassium 22 and one sodium 23 bacterial channels. In other words, this gate is intrinsically more stable in the open state. Surprisingly, hERG pore domain construct (S5-S6) alone, isolated from the voltage sensor domain, gives rise to a nonfunctional channel 19 . One particularity for hERG, compared to the aforementioned bacterial channels, is that the cytosolic N-terminus deletion greatly accelerates deactivation [24][25][26] . Thus, the N-terminus stabilizes the channel in the open state, and ablation of the hERG voltage sensor domain, which also deletes the N-terminus, may render hERG nonfunctional.
Of note, this ligand/receptor model is completely in frame with the allosteric model of voltage-dependence suggested for KCNQ1 and hERG, in which voltage-dependent movement of S4 is allosterically coupled to pore-opening transitions 27,28 . In such a model, the gating of KCNQ1 can be well described by an allosteric model where the channel can open before all S4 voltage sensors have been activated 29,30 . Our ligand/receptor model suggests that allosteric coupling is realized through the interaction between S4-S5 L and S6 T (leading to channel closure), which is favored when S4 segments are in the "down" state at resting potential and not favored when S4 segments are the in the "up" state.
In conclusion, the results that we obtained in the hERG channel with S4-S5 L and S6 T mimicking peptides, with or without cysteine introduction, reinforce the ligand/receptor model originally observed in KCNQ1 and suggest that this model applies to other voltage-gated channels. In Shaker channels, sequence complementarity between S4-S5 L and S6 T has also been shown to be critical to coupling voltage sensor movement and pore opening 4,5 . Testing peptides mimicking S4-S5 L and S6 T in Shaker-like channels would confirm whether the ligand/receptor model applies to Shaker-like channels.

Methods
Plasmid constructs. Oligonucleotides encoding hERG peptides were synthesized by TOP Gene Technologies and contained a XhoI restriction enzyme, a methionine (ATG) for translation initiation, and a glycine (GGA) to protect the ribosome binding site during translation and the nascent peptide against proteolytic degradation 31 . A BamHI restriction enzyme site was synthesized at the 3′ end immediately following the translational stop codon (TGA). These oligonucleotides were then cloned into pIRES2-EGFP (Clontech) and sequenced.
Three different S4-S5 L and S6 T plasmids were designed with the same length (16 amino-acids for S4-S5 L and 13 for S6 T ) but different starting positions in order to compensate for the inaccuracy of the alignment due to the poor homology between hERG and KCNQ1 sequences ( Fig. 1 and Supplemental Fig. 4). Multiple alignment was realized with Clustal Omega 32 . This program aligned the predicted transmembrane domains (Uniprot Q12809 for hERG, P51787 for KCNQ1) and the predicted position of the narrowest part of the bundle crossing, also named gating residue 33,34 , of hERG and KCNQ1. We first designed a hERG S4-S5 L peptide based on the most potent KCNQ1 S4-S5 L (L251-L266 11 ). Since mutagenesis studies suggested a close proximity of D540 in S4-S5 L and R665 in S6 T 6, 13 , we chose two peptides shifted by 3 and 6 amino acids to the N-terminus to include the D540 residue. Names of the peptides were given according to their position along the sequence: S4-S5 L (−3) is A536-F551, S4-S5 L (0) is L539-M554, and S4-S5 L (+3) is Y542-F557. For S6 T peptides, we first selected the hERG sequence which aligned with the most potent peptide of KCNQ1 (I346-K358). Since in KCNQ1 the effects of the peptides were increased as the sequence was shifted to the C-terminus, we tried two other peptides position-shifted by 3 and 6 amino acids towards the C-terminus. S6 T (−3) is G657-G669, S6 T (0) is S660-R672, and S6 T (+3) is I663-T675.
Cell culture and transfection. The African green monkey kidney-derived cell line COS-7 was obtained from the American Type Culture Collection (CRL-1651) and cultured in Dulbecco's modified Eagle's medium (GIBCO) supplemented with 10% fetal calf serum and antibiotics (100 IU/ml penicillin and 100 µg/ml streptomycin) at 5% CO 2 and 95% air, maintained at 37 °C in a humidified incubator. Cells were transfected in 35-mm Petri dishes when the culture reached 50-60% confluence, with DNA (2 to 4 µg total DNA as described below) complexed with Fugene-6 (Roche Molecular Biochemical) according to the standard protocol recommended by the manufacturer. In the screening experiments used to identify the most potent peptides (Fig. 2), COS-7 cells were co-transfected with 0.6 µg pSI-hERG and 1.4 µg pIRES2-EGFP (Clontech) plasmids either encoding or not encoding a peptide. In the experiments with cysteine introduction, some mutations are associated with a decrease in current amplitude (D540C, E544C) or an increase in current amplitude (L666C). Plasmid quantities were thus optimized (i) to maximize the quantity of peptides, as assessed by the amount of florescence, and (ii) to keep the average current amplitude in the same range, in order to avoid undetectable currents or an incorrect voltage clamp. As a result, experiments on channels containing a D540C, E544C or L666C mutation were done with 1 µg mutated pSI-hERG and 3 µg pEGFP. Experiments on double mutant hERG channels (D540C/L666C or E540C/L666C) were performed with 3.6 µg pSI-hERG and 0.4 µg pEGFP. In pIRES2-EGFP plasmids, the second cassette (EGFP) is less expressed than the first cassette, guaranteeing high levels of peptides expression in fluorescent cells 11 . Cells were re-plated onto 35-mm Petri dishes the day after transfection for patch-clamp experiments.
Western blot. 20  Stain free gel technology (Bio-Rad) was used as loading control for protein normalization: band intensities were first normalized to the intensity of the corresponding stain free membrane lane, and ratios were then normalized to control hERG condition 35 . Electrophysiology. The day after splitting, COS-7 cells were mounted on the stage of an inverted microscope and constantly perfused by a Tyrode solution (cf. below) at a rate of 1-3 ml/min. The bath temperature was maintained at 22.0 ± 2.0 °C. Stimulation and data recording were performed with Axon pClamp 10 through an A/D converter (Digidata 1440A) using an Axopatch 200B (all Molecular Devices). Patch pipettes (tip resistance: 2-3 MOhms) were pulled from soda lime glass capillaries (Kimble-Chase). Currents were recorded in the whole-cell configuration. hERG activation and inactivation curves were obtained from the tail currents and fitted by Boltzmann equations.
Statistics. All data are expressed as means ± S.E.M. Statistical differences between samples were determined using Student's t-tests, rank-sum tests (when data were not normally distributed), and two-way analysis of variance associated with a Bonferroni post-hoc test when needed. In the screening experiment ( Fig. 2 and Supplemental Table 1), a value of p < 0.10 was considered significant. In the rest of the study, a value of p < 0.05 was considered significant.