The hiPSC-CM (“iCell-cardiomyocytes”, catalog #CMC-100-110-001), the iCell-cardiomyocytes Plating Medium and the iCell-cardiomyocytes Maintenance Medium were from Cellular Dynamics International (Madison, WI). The Calcium-4 kit was from Molecular Devices (Sunnyvale, CA). The commercially-available set of 640 drugs (“FDA-Approved Drug Library”, catalog #BML-2841) was from Enzo Life Sciences (Lausen, Switzerland); all compounds were tested at a single identical concentration (1 µM) to mimic the flow of standard HTS. All other chemicals, including those used for follow-up concentration-response curves (CRCs), were from Sigma-Aldrich (Buchs, Switzerland).

On day-0, the hiPSC-CM were thawed, plated in 384-well assay plates (catalog #781091, Greiner-Bio-One, Frickenhausen, Germany) at 10,000 cells per well in 25 µL Plating Medium and maintained at 37˚C in a humidified atmosphere containing 5% CO₂. On day+1, the medium was replaced with 50 µL maintenance medium and the plates were covered with an adhesive gas-permeable membrane (catalog #676051, Greiner-Bio-One). Thereafter, half of the medium was renewed every 2 - 3 days and it was renewed entirely on day+21.

On day+22, the 640 compounds were prepared in two 384-well microplates (catalog #781280, Greiner-Bio-One) from 2-mM DMSO stock solutions by diluting 1 µL in 94 µL Maintenance Medium (1.05% DMSO). The positive controls, forskolin and N6-cyclopentyl-adenosine,were prepared similarly and added to both plates (4 wells each) while the negative control, DMSO alone, was added to the remaining 56 wells of each plate. The two compound plates were loaded into a FDSS7000 fluorescent plate reader (Hamamatsu Photonics, Massy, France) heated to 37˚C.

The following protocol was repeated on four cell plates to test the two compound plates in duplicate. 20 µL medium was replaced with 30 µL medium containing the Ca-sensitive fluorescent dye Fluo-4 and a quencher (component A from the Calcium-4 kit reconstituted in 50 mL Maintenance Medium) and the cell plate was loaded into the FDSS7000. After 60-min in dye-containing medium, a 2-min segment of Ca waves was recorded to define the baseline beating rate for each well. Then the compounds were added by pipetting well-to-well 3 µL from the compound plate into the cell plate. After addition, the concentration was 1 µM and DMSO was 0.05% (v/v). Thereafter, 2-min segments of Ca waves were recorded starting 5, 15, 30, 45, and 60 min after compound addition. For this study, we used a 10-Hz sampling frequency since we only evaluated changes in beating frequency.

For follow-up CRCs, an identical methodology was used with DMSO stock solutions prepared from dry compound powders such that the final DMSO concentration was also 0.05% (v/v).

Beating frequency was calculated with Igor Pro software (Wavemetrics, Portland, OR) and a custom analysis routine using the Lomb Periodogram function (based on the Lomb-Scargle method of least-squares spectral analysis). For each well of the cell plate, the beating frequency at each time point after compound addition was normalized to baseline and the time-matched mean DMSO effect for the corresponding cell plate was subtracted. Results were averaged for the two cell plates receiving the same compound plate. For compounds that changed frequency (see Results for the definition), a visual examination of the beating pattern evaluated whether the compounds produced arrhythmias.

Among the 640 drugs, 95 have a therapeutic target that plays a role in heart rate regulation at the car-

diomyocyte level. They comprise 13 agonists and 14 antagonists of β-adrenergic receptors, 6 agonists and 8 antagonists of muscarinic receptors, 19 phosphodiesterase (PDE) inhibitors, 1 Na/K pump inhibitor, 6 Class I or Class III antiarrhythmics, 17 Ca channel blockers (CCB). These compounds all produce their expected effect on the beating rate of the hiPSC-CMs.

The β-agonists except bambuterol increase frequency (non-selective or β1- or β2-selective: albuterol, bambuterol, cimaterol, clenbuterol, procaterol, salmeterol, tulobuterol, xamoterol, dobutamine, ractopamine, epinephrine, norepinephrine, isoproterenol; Figure 3(a)). Bambuterol is a pro-drug of terbutaline

and has little inherent agonistic activity, which explains the absence of effect; indeed, terbutaline produced an effect similar to albuterol when tested in a separate experiment (see next section). All compounds accelerate >20% without arrhythmias; the effect develops within ~5 min then decreases with time. Four classes of compounds emerge when comparing the time course of the frequency increase: 1) for six compounds (albuterol, cimaterol, clenbuterol, salmeterol, tulobuterol, ractopamine), the effect decays slowly with kinetics similar to forskolin (<50%/60 min); 2) for four compounds (dobutamine, epinephrine, norepinephrine, isoproterenol), the effect decays rapidly (100%/60 min); 3) for procaterol, we see intermediate kinetics (~90%/60 min); and 4) for xamoterol, the frequency increases to an intermediate level then decays minimally (<20%/60 min). Remarkably, all slowly-decaying agonists are β2-selective whereas all fully-decaying agonists are β1-selective or non-selective. The β2-agonist albuterol has an EC₅₀ of ~3 nM in our assay (separate experiment, not shown); therefore activation of β2-adrenergic receptors increases frequency in these cells. Procaterol is particular as the only β2-agonist with an effect that decays faster than forskolin. However, procaterol is unstable in buffered solutions and degradation explains the rapid decay [ 8 ]. Finally, the β1-selective partial agonist xamoterol is notable because it increases frequency partially (~18%) and its effect barely decays; this is similar to what is observed in humans [ 9 ].

When β-blockers are applied, most produce no effect indicating that there is no basal activation of β-adrenergic receptors (non-selective or β1- or β2-selective: alprenolol, atenolol, betaxolol, bopindolol, carvedilol, esmolol, metoprolol, pindolol, practolol, pronethalol, propranolol, (S)-propranolol, sotalol, timolol). However, six compounds accelerate the beating frequency slightly (alprenolol, bopindolol, pindolol, practolol, pronethalol, timolol; ~5% - 10%) without arrhythmias and with kinetics similar to xamoterol (Figure 3(b)). These six β-blockers possess intrinsic sympathomimetic activity (ISA) [ 10 - 12 ], indicating that they are in fact, like xamoterol, partial β-adrenergic agonists. The remaining compounds do not possess ISA [ 12 - 15 ], demonstrating that the preparation is extremely sensitive in distinguishing compounds with or without ISA.

Among muscarinic agonists, the more potent, slightly M2-selective compounds carbachol and oxotremorine transiently decrease beating frequency (6% - 7%) but there is no effect of the weaker acetylcholine and butyrylcholine and the M1/M3-selective bethanechol and pilocarpine. Carbachol has an EC₅₀ of ~250 nM and is antagonized by atropine (separate experiment, not shown), consistent with an M2-mediated effect. Among the eight muscarinic antagonists, atropine, butylscopolamine, ipratropium, pirenzepine, scopolamine, telenzepine, tropicamide have no effect but tiotropium accelerates transiently (~7%): no described property of tiotropium can explain this specificity.

Among PDE inhibitors, nine compounds (denbufylline, etazolate, IBMX, ibudilast, pentoxifylline, rolipram, trequinsin, vardenafil, zardaverine) produce a sustained acceleration (6% - 17%) without arrhythmias. The remaining compounds (aminophylline, anagrelide, cilostamide, dipyridamole, doxofylline, irsogladine, milrinone, siguazodan, sildenafil, zaprinast) have no effect. Except vardenafil, which produces a small sustained acceleration (7% - 9%), all active compounds are non-selective PDE inhibitors or PDE4-selective inhibitors. When we repeated the experiment with new compound and higher concentrations, milrinone and sildenafil also accelerated starting at 3 - 10 µM. Thus, PDE4 is of particular importance in controlling hiPSC-CM beating frequency but PDE3 and PDE5 may play a small role.

Following application of the Na/K pump inhibitor ouabain, there is no effect after 5 min, a strong reduction in frequency after 15 min (−40%) and a complete arrest of beating after 30 min. In contrast, the laxative bisacodyl, which was also described as inhibiting the Na/K pump [ 16 ], strongly accelerates at 5 min (+30%) but not later, suggesting that it does not inhibit the Na/K pump in this preparation.

Ion channel modulators are also compounds expected to affect cardiac rhythm. Ten of the eleven dihydropyridine L-type CCBs (amlodipine¹, cilnidipine, felodipine, manidipine, nicardipine, nifedipine, niguldipine, nimodipine, nisoldipine, nitrendipine) accelerate strongly (150% - 250%) with arrhythmias while lacidipine stops the beating immediately; such arrest is observed later with niguldipine and nisoldipine. The other CCBs produce diverse effects: verapamil provokes a rapid arrest, diltiazem a sustained acceleration (~100%) without arrhythmias, lomerizine and flunarizine a transient acceleration (13% - 25%), whereas gabapentin and bepridil have no effect. This can be expected of the neuronal Ca channel-selective gabapentin but it is more surprising with bepridil. However, bepridil is a weaker CCB than the dihydropyridines and it has much stronger plasma protein binding (99.8%) than verapamil (91%) or diltiazem (82%). Since our assay is performed in culture medium containing serum (~10%), the free fraction of bepridil in a 1-µM solution may be too low to block Ca channels: we repeated the experiment with new compound and no effect was seen again at 1 µM whereas complete arrest was observed at 10 µM (also with lomerizine and flunarizine).

The actions of Class I and Class III antiarrhythmics are predictable. Dofetilide causes immediate arrest, with minimal recovery after 30 min; nifekalant does not stop beating but slows frequency (15% - 50%) with arrhythmias; and amiodarone produces a small acceleration (~12%) without arrhythmias. Although amiodarone blocks the cardiac rapid delayed-rectifier K current (IKr) potently, it is strongly bound to plasma proteins (>99.9%), which may explain its relatively weak effect. The Class Ic antiarrhythmic flecainide slows the rhythm (~13%) whereas propafenone or the Class Ia procainamide have no effect. Since Class I antiarrhythmics are generally not very potent drugs [ 17 ], we retested Na channel blockers at higher concentrations and we consistently observed a reduction in beating rate (see next section).

In the previous section, compounds were tested following a HTS methodology, using one single identical concentration. Results are therefore somewhat qualitative as they do not provide a threshold effect concentration or a defined maximal effect. In order to better qualify some of the mechanisms, we performed CRCs with a few characteristic compounds; in the category “Drugs with a therapeutic target playing a direct role in heart rate regulation”, we selected the β-agonist terbutaline, the PDE inhibitors rolipram (PDE4) and milrinone (PDE3), the CCBs amlodipine and lomerizine, the K channel blocker E-4031, and the Na channel blocker tetrodotoxin (TTX).

Figure 4 shows the corresponding CRCs for all the measurement times. Terbutaline (Figure 4(a)) accelerates the rhythm concentration-dependently but efficacy decreases with time of application as already observed for other β-agonists (Figure 3); however, potency does not seem affected. Acceleration starts around 2 - 10 nM and can be well-fit using standard concentration-response curves with a Hill slope of 1 and EC₅₀ of 9.1 nM (5 min) or 25 nM (60 min); this is consistent with potent activation of adrenergic receptors and significant receptor reserve as observed in bronchial smooth muscle [ 18 ]. Rolipram and milrinone (Figure 4(b)) also accelerate the rhythm concentration-dependently but with different concentration- and time-dependencies. Rolipram, a PDE4 inhibitor, is already active at 2 nM and its effect changes little during the 60-min application; its EC₅₀ of ~5 nM is consistent with very potent inhibition of PDE4 and increased intracellular cAMP. In contrast, milrinone, a PDE3 inhibitor, is only active initially at ≥3 µM but it becomes more potent and less efficacious during the 60-min application; the slow appearance of an effect may result from cAMP diffusion from subcellular compartments dependent on PDE3 (see Discussion).

Amlodipine and lomerizine (Figure 4(c)) produce very similar patterns: they accelerate the rhythm concentration-dependently, up to a maximal concentration; then the rhythm stops suddenly (dashed lines). Acceleration becomes significant at 14 nM (amlodipine) or 140 nM (lomerizine), consistent with potent block of cardiac voltage-gated Ca channels. It is also apparent that drug activity increases during the 1‑hour application: acceleration becomes more pronounced with time and arrest occurs at lower concentrations; this could conceivably result from positive feedback through the well-described use- or frequency-dependent block of Ca channels by these CCBs (blockðaccelerationðmore blockðmore acceleration…). E-4031 and TTX (Figure 4(d)) slow the rhythm concentration-dependently but with different concentration- and time-dependencies. The IKr blocker E-4031 is already active at 10 - 50 nM and its effect changes little during the 60-min application; its EC₅₀ of 400 - 600 nM is consistent with very potent block of cardiac voltage-gated K channels. In contrast, the INa blocker TTX is only active at ≥1 µM

consistent with weaker block of cardiac voltage-gated Na channels. In addition, TTX tends to produce an all-or-none response as it slows beating rate minimally (≤15%) up to a concentration when it suddenly stops beating completely (dashed lines); this limit concentration decreases during the 60-min application. Aside from this last aspect, we observed a very similar pattern with other Na channel blockers and the limit concentration was clearly related to Na channel block potency (50 µM lidocaine & disopyramide; 10 µM mexiletine; 2 µM quinidine, bepridil, flecainide & propafenone; not shown).

In conclusion, “drugs with a therapeutic target playing a direct role in heart rate regulation” produce a well-defined response on the beating rate of the hiPSC-CMs, which can be associated with their primary (i.e., most potent) mode-of-action.

The actions of dofetilide, nifekalant, and E-4031 suggest that this assay detects IKr blockers as shown previously with a slightly different technology [ 5 ]. To evaluate this possibility, we collated published data on IKr-blocking potency and plasma-protein-binding for the 640 compounds. As could be expected for older drugs, such information does not exist for all but IKr-blocking IC₅₀s (K₅₀) could be found for 138 compounds and plasma-protein-binding percentages (PPB) for 365 compounds. Furthermore, for 119 compounds (36 with published K₅₀ and 83 without), we evaluated IKr block experimentally by measuring block of K_V11.1 (hERG) channels expressed in CHO cells, using planar patch-clamp (Table 1). Finally, for 39 compounds with K₅₀ but no published PPB, we measured it in the test medium (MPB) and, since this medium contains only 10% serum, we calculated PPB as 100% − 10%/((MPB/(100% − MPB))+10%). This corrected value is reported in Table 1 along with published PPB values.

For the 36 drugs with published K₅₀, the correlation with measured values is very good for 30 (within a factor of 3) and good for 2 (factor of 10); however, flunarizine is notably weaker in our measurements whereas nifedipine, nitrendipine and estradiol are more potent. Among 83 compounds without published K₅₀, 11 block IKr with potency between haloperidol and clozapine (published K₅₀ of 27 nM & 320 nM [ 19 ]; azaperone, loperamide, methiothepin, naftopidil, naltriben, naltrindole, nicergoline, nifekalant, spiperone, trifluperidol, triprolidine). Thirty-six other drugs block with lower potency between clozapine and propranolol (published K₅₀ of 10,000 nM [ 20 ]; alprenolol, benzamil, benzydamine, bopindolol, butaclamol, cilnidipine, cirazoline, cyclosporine, cyproheptadine, dilazep, disulfiram, dorzolamide, felodipine, gestrinone, harmine, hydroxytacrine, imiquimod, lacidipine, levallorphan, lofexidine, lomerizine, manidipine, naloxone, naltrexone, nicardipine, niguldipine, nimodipine, nisoxetine, phenamil, phentolamine, quinacrine, strychnine, tacrine, telmisartan, tropisetron, vinorelbine).

In order to correct for drug bound to proteins in the test medium, we calculated a “corrected IC₅₀” as corrK₅₀ = K₅₀/(1 − MPB), where MPB is calculated from PPB by reversing the above equation. We then compared the effect of compounds on hiPSC-CM rhythm and their corrK₅₀ (Figure 5). Interesting rules emerge from this figure: first, compounds that stop rhythm are all strong IKr blockers and/or CCBs, with the exception of ouabain. Second, IKr blockers with corrK₅₀ < 1 µM (the test concentration) stop rhythm or slow it markedly (>20%) with few exceptions: the dopamine agonist pergolide accelerates (12% - 20%) whereas the histamine antagonist terfenadine, the potassium-sparing diuretic benzamil and the opioid antagonist naltrindole slow <20%. The action of pergolide is surprising but, in the published study where it blocked IKr with K₅₀~120 nM, pergolide shortened rather than lengthened action potential duration in canine Purkinje fibers [ 70 ]. This could indicate that pergolide is a CCB although we did not observe block of expressed Ca_V1.2 channels (not shown). The weaker effects of terfenadine, benzamil and naltrindole also suggest activities additional to IKr block. Conversely, 3 other compounds slow beating strongly but are not potent IKr blockers (the adrenergic agonist cirazoline, the MAO inhibitor harmine and the microtubule destabilizing agent vinorelbine). Like ouabain, the effect of vinorelbine developed after 5 min, suggesting a cytotoxic mechanism. On the other hand, the effects of cirazoline and harmine developed before 5 min then remained stable.

Forty-five drugs have a cardiovascular therapeutic target that is not expected to play a role in heart rate regulation at the cardiomyocyte level. They comprise 9 α-adrenergic agonists and 12 antagonists, 7 angiotensin-converting-enzyme inhibitors (ACEI), 5 angiotensin-1 receptor antagonists (ARA), 1 endothelin receptor antagonist (ERA), 4 activators and 7 blockers of K_ATP channels. Most of these compounds produce no effect on hiPSC-CM beating.

None of the 8 selective α₂-adrenergic agonists (clonidine, guanabenz, guanfacine, lofexidine, methyldopa, naphazoline, tizanidine, xylazine) changes frequency but the α₁-selective agonist cirazoline decreases it markedly (21% without arrhythmias). Among the 12 α-adrenergic antagonists (non-selective or α₁- or

a: [ 21 ]; b: [ 22 ]; c: [ 23 ]; d: [ 24 , 25 ]; e: [ 26 ]; f: [ 27 ]; g: [ 28 ]; h: [ 19 ]; i: [ 29 ]; j: [ 30 ]; k: [ 31 ]; l: [ 32 ]; m: [ 33 ]; n: [ 34 ]; o: [ 35 ]; p: [ 36 ]; q: [ 37 ]; r: [ 38 ]; s: [ 39 ]; t: [ 40 ]; u: [ 41 ]; v: [ 42 ]; w: [ 43 ]; x: [ 44 ]; y: [ 45 ]; z: [ 46 ]; aa: [ 47 ]; bb: [ 48 ]; cc: [ 49 ]; dd: [ 50 ]; ee: [ 51 ]; ff: [ 52 ]; gg: [ 53 ]; hh: [ 54 ]; ii: [ 55 ]; jj: [ 56 ]; kk: [ 57 ]; ll: [ 58 ]; mm: [ 59 ]; nn: [ 60 ]; oo: [ 61 ]; pp: [ 62 ]; qq: [ 63 ]; rr: [ 64 ]; ss: [ 65 ]; tt: [ 66 ]; uu: [ 67 ]; vv: [ 68 ]; ww: [ 69 ]; xx: [ 84 ]; yy: [ 85 ].

α₂-selective: doxazosin, idazoxan, efaroxan, naftopidil, nicergoline, oxymetazoline, phenoxybenzamine, phentolamine, prazosin, tamsulosin, terazosin, yohimbine), only nicergoline and phentolamine decrease frequency markedly (>40% with arrhythmias) whereas doxazosin and naftopidil decrease transiently (−12% and −7%). These actions were all confirmed with new compound and major effects were seen at 10 µM and above (frequency reduction of >20% up to arrest). This is consistent with the observation that nicergoline and phentolamine block cardiac IKr with corrK₅₀ below 1 µM, while doxazosin and naftopidil block have corrK₅₀ below 10 µM (Figure 5). The particular case of cirazoline may indicate that α₁-selective agonism can directly slow beating in these cells. This would need to be verified with additional selective agonists.

The 7 ACEIs (captopril-disulfide¹, enalapril, enalaprilat, fosinopril, lisinopril, quinapril, ramipril), the 5 ARAs (candesartan, eprosartan, losartan, olmesartan, telmisartan), the ERA bosentan, the 4 activators and 7 blockers of K_ATP channels (diazoxide, minoxidil, nicorandil, pinacidil and gliclazide, glimepiride, glipizide, glyburide, nateglinide, tolazamide, tolbutamide) have no effect on the beating rate of the hiPSC-CM.

In addition to the compounds discussed above, 511 drugs have a therapeutic target that should not play a role in heart rate regulation at the cardiomyocyte level. Among these, 417 compounds have no effect at 1 µM on the rhythm of the hiPSC-CM². Among the 94 compounds that affect rhythm, 23 are strong IKr blockers (corrK₅₀ ≤ 1 µM, Figure 5) while two weaker blockers slow beating strongly (>20%: harmine, vinorelbine). There are no published properties of the chemotherapeutic alkaloid vinorelbine that explain its strong actions whereas the MAO inhibitor harmine depresses the automaticity of sinoatrial myocytes with an unknown mechanism [ 71 ].

Many different drugs slow rhythm moderately or accelerate it but, in most cases, there is insufficient published information to propose a mechanism. The frequency is slowed 11% - 20% by 6 compounds³ and 6% - 10% by 10 compounds⁴ whereas it is accelerated 11% - 42% by 11 compounds⁵ and 6% - 10% by 40 compounds⁶.

Among antibacterial agents, only azithromycin and tylosin produce a notable effect (acceleration of 10% - 11%, confirmed with new compound and exceeding 50% at 10 µM). The cardiac safety of azithromycin has been under question for many years and this compound was recently shown to increase the risk of cardiovascular death [ 72 ]. Similarly, the veterinary antibiotic tylosin increases ventricular arrhythmic risk in dogs [ 73 ]. None of the quinolone antibiotics (ciprofloxacin, clinafloxacin, enoxacin, enrofloxacin, fleroxacin, gatifloxacin, levofloxacin, lomefloxacin, nadifloxacin, norfloxacin, ofloxacin, rufloxacin, sara- floxacin, sparfloxacin) affect beating at 1 µM. It is likely that this concentration is too low and, when we retested some quinolones at higher concentrations, moxifloxacin and sparfloxacin slowed beating (>20% at 100 µM) whereas ciprofloxacin, enrofloxacin, fleroxacin, levofloxacin, lomefloxacin, norfloxacin, ofloxacin, and pefloxacin do not. This is consistent with a PPB of 20% - 50% for quinolones and with their published K₅₀s (~18 µM for sparfloxacin, 129 µM for moxifloxacin, 915 µM for levofloxacin, 966 µM for ciprofloxacin, 1420 µM for ofloxacin) [ 64 ].

Four histamine antagonists decrease frequency (astemizole, clemastine, terfenadine, triprolidine); this is likely due to IKr block as discussed above. The weaker IKr blocker mepyramine [ 44 ] slows beating transiently (~8% for <15 min); whether this results from IKr block is not certain. It is worth noting that hiPSC-CM respond to histamine with a very strong acceleration (EC₅₀ ~100 nM), which is prevented by the H₂-selective antagonist famotidine and not by the H₁-selective antagonist pyrilamine (separate experiments, not shown); however, these antagonists have no effect in the absence of histamine, indicating that there is no constitutive activation of histamine receptors or paracrine secretion of histamine. Finally, the weak IKr blockers cyproheptadine, olopatadine and promethazine accelerate rhythm (~25% for 60 min for cyproheptadine, minimally for the others). Cyproheptadine has high plasma protein binding (96%) and strongly increases frequency, suggesting that it acts through amechanism distinct from IKr block.

Many hormone receptor agonists and antagonists produce sustained or transient acceleration (estradiol, estriol, estrone, ethisterone, fulvestrant, gestrinone, mifepristone, norethindrone, progesterone, raloxifene) but no pattern is obvious. Diethylstilbestrol and many progesterone receptor agonists produce no effect, suggesting that activation of ERα, ERβ, or PR receptors does not accelerate beating. Estradiol, estriol, estrone, fulvestrant, and raloxifene are agonists of the G-protein-coupled estrogen receptor, which may play a role in their effect, but such activity has not been reported for ethisterone, gestrinone, mifepristone, norethindrone, or progesterone.