Leak current, even with gigaohm seals, can cause misinterpretation of stem cell-derived cardiomyocyte action potential recordings Open Access

Abstract

Aims

Human-induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) have become an essential tool to study arrhythmia mechanisms. Much of the foundational work on these cells, as well as the computational models built from the resultant data, has overlooked the contribution of seal–leak current on the immature and heterogeneous phenotype that has come to define these cells. The aim of this study is to understand the effect of seal–leak current on recordings of action potential (AP) morphology.

Methods and results

Action potentials were recorded in human iPSC-CMs using patch clamp and simulated using previously published mathematical models. Our in silico and in vitro studies demonstrate how seal–leak current depolarizes APs, substantially affecting their morphology, even with seal resistances (R_seal) above 1 GΩ. We show that compensation of this leak current is difficult due to challenges with obtaining accurate measures of R_seal during an experiment. Using simulation, we show that R_seal measures (i) change during an experiment, invalidating the use of pre-rupture values, and (ii) are polluted by the presence of transmembrane currents at every voltage. Finally, we posit that the background sodium current in baseline iPSC-CM models imitates the effects of seal–leak current and is increased to a level that masks the effects of seal–leak current on iPSC-CMs.

Conclusion

Based on these findings, we make recommendations to improve iPSC-CM AP data acquisition, interpretation, and model-building. Taking these recommendations into account will improve our understanding of iPSC-CM physiology and the descriptive ability of models built from such data.

Graphical abstract

Open in new tabDownload slide

Introduction

Human-induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) are a renewable and cost-effective model for studying genetic disease mechanisms,^1,2 drug cardiotoxicity,³ and inter-patient variability.⁴ Computational approaches have been developed to translate experimental results from iPSC-CMs to make predictions in adult cardiomyocytes.⁵ Such work attempts to bridge the critical gap that remains between the physiology of iPSC-CMs and excised adult human cardiac cells.

Whilst iPSC-CMs have transformed many areas of cardiac arrhythmia research, phenotypic heterogeneity and immaturity continue to stymie their potential impact.^6,7 Investigating sources of these limitations and their biological implications is important as iPSC-CMs (and mechanistic models describing their behaviour) are used to inform increasingly complex clinical decisions.^8,9 Studies of iPSC-CMs in a single-cell patch clamp context have indicated that their depolarized, highly varying resting membrane potential is primarily due to decreased inward rectifier potassium current (I_K1) and increased funny current (I_f) compared with adult cardiomyocytes.¹⁰

Recently, findings from Horváth et al.¹¹ and Van de Sande et al.¹² indicate that the heterogeneous and depolarized resting membrane potential is also due, far more than previously thought, to a simple seal–leak current (I_leak). Relative to electrically coupled iPSC-CMs, they show a substantial depolarization in the resting membrane potential in isolated iPSC-CMs despite some cells having similar I_K1 densities to human adult cardiomyocytes.¹¹ These findings indicate that I_leak plays an important role in iPSC-CM AP morphology during single-cell patch clamp experiments.

I_leak is inversely proportional to the seal resistance (R_seal) formed between the micropipette tip and cell membrane during patch clamp experiments. A sufficiently large R_seal is expected to limit I_leak’s effect on AP morphology. Upon reviewing single-cell electrophysiological iPSC-CM studies, including those used to build iPSC-CM computational models,^13–15 we found that studies do not report either an R_seal,^10,16–19 a >1 GΩ R_seal acceptance criteria,²⁰ or an average R_seal < 3 GΩ.^11,12

In this study, through in vitro experiments and computational modelling, we show that I_leak affects iPSC-CM AP morphology, even above the R_seal values usually deemed acceptable in the literature. We show that R_seal cannot be easily compensated because it cannot be accurately measured during an experiment. Additionally, we posit that the background sodium current (I_bNa) in iPSC-CM models may be overestimated and mimic the effects of leak on AP morphology. Ultimately, we argue that leak current should be considered when interpreting, analysing, and fitting iPSC-CM AP data.

Methods

Modelling I_leak

We added a leak equation to the Kernik¹³ and Paci¹⁴ iPSC-CM and ToR-ORd²¹ adult cardiomyocyte models. Knowing that leak acts as a depolarizing current in iPSC-CM studies and lacking information about specific charge carriers, we modelled I_leak as having a reversal potential of zero:^22,23

where R_seal is the seal resistance and V denotes the membrane potential. The inverse of R_seal is the conductance, g_seal. Note that more complicated equations for leak current (non-linear, and/or with a non-zero reversal potential) may be required in experiments where CaF₂ seal enhancer is used.²⁴

The effect of I_leak on the evolution of V was modelled as follows:

where I_ion represents the sum of transmembrane currents and C_m is the membrane capacitance. C_m was set to 50 pF (the experimental average from the cells used in the present study) for the Kernik and Paci simulations, and for ToR-ORd, a value of 50 or 153 pF (the ToR-Ord baseline capacitance) was used unless specified otherwise.

Electrophysiological setup and data analysis

Perforated patch clamp experiments were conducted following a previously described protocol (see Supplementary Methods for more details).²⁵

After contact was made with a cell and a seal of >300 MΩ was formed, the perforating agent slowly decreased the access resistance to the cell (usually 10–15 min). This low R_seal acceptance criterion was selected because we wanted to explore seal–leak effects above and below 1 GΩ. A series resistance (R_s) of 9–50 MΩ was maintained for all experiments. In this study, we used all cells from Clark et al.²⁵ with membrane resistance (R_m) and R_s measurements acquired before and after current clamp recordings and that did not produce spontaneous alternans (n = 37 out of 40 cells). R_m, C_m, and R_s values were measured at 0 mV within 1 min prior to the acquisition of current clamp data.

All action potential (AP) features were calculated using a 10-s sample of current clamp data. The minimum potential (MP) was taken as the minimum voltage during this 10-s span. Maximum upstroke velocity (dV/dt_max), AP duration at 90% repolarization (APD₉₀), and cycle length (CL) were averaged over all APs in the 10 s sample.

R_in as an estimate of R_seal

We calculate R_seal using a small test pulse in voltage clamp mode:²⁶

Here, ΔV_cmd is the applied voltage step, and ΔI_out is the difference in recorded current from before to during the step. Once access is gained to a cell, it can be difficult to estimate R_seal, as the measured input resistance (R_in) depends on both R_m and R_seal [Eq. (4); Figure 1]. The effect of patch clamp series resistance on R_in measures was excluded from Eq. (4).

The smallest R_seal considered was 300 MΩ, whilst R_s values ranged from 9 to 50 MΩ. An increase of R_s from 9 to 50 MΩ (a worst-case scenario we never observed) for a cell with a 300 MΩ R_seal would change R_in by 13%. So, whilst R_s can change in these experiments, it is unlikely to affect R_in by more than a few per cent, and R_seal is likely the predominant parameter affecting changes of R_in.

Additional methods

Additional methods can be found in the Supplementary material.

Results

Leak affects human-induced pluripotent stem cell-derived cardiomyocytes action potential morphology even at seal resistances above 1 GΩ

To investigate the effects of leak current on AP morphology, we simulated the addition of I_leak in the Kernik¹³ and Paci¹⁴ iPSC-CM models (Figure 2). Simulated AP recordings show that I_leak substantially alters AP morphology, even when R_seal ≥ 1 GΩ, a common threshold used in cardiac patch clamp experiments.²⁰ For both models, decreases in R_seal depolarize the MP and cause a decrease in the dV/dt_max, likely due to an incomplete recovery of sodium channels at these depolarized MPs. Indeed, the Kernik model shows a transition to a small amplitude oscillation with very low upstroke velocity when R_seal < 3 GΩ and then depolarized quiescence when R_seal < 2 GΩ. I_leak effects on the APD₉₀ differ for the two models—decreases to R_seal cause AP prolongation in the Paci model and AP shortening in the Kernik model. There are also differences in the effect of R_seal on CL: in the Kernik model, decreases in R_seal lead to a gradual decrease in CL, whilst in the Paci model, decreasing R_seal initially has limited effect on CL but then causes shortening as R_seal decreases below 5 GΩ.

Leak effects on adult cardiomyocyte action potentials are moderated by different current densities and increased ionic currents

The ToR-ORd adult cardiomyocyte model is also susceptible to I_leak effects, but the extent depends on cell capacitance (Figure 3). Simulations with C_m set to the average iPSC-CM capacitance (50 pF) result in substantial AP morphological changes when R_seal is between 1 and 2 GΩ. However, when C_m is set to a value in the range of adult human ventricular cardiomyocytes (153 pF), I_leak has little effect on AP morphology when R_seal is ≥1 GΩ (Figure 3B).

R_seal is not stable

Unlike voltage clamp recordings, the effects of I_leak on AP morphology (measured in current clamp mode) cannot be corrected in post-processing. Current clamp leak compensation is a potential solution to the issue^22,23 but requires an accurate measure of R_seal throughout the experiment.

R_seal cannot be accurately determined after access is gained because measures are contaminated by R_m; such resistance measures are a composite of these two resistances that we nominally refer to as R_in (see Figure 1 and Methods). It is, therefore, tempting to measure the value before gaining access and assume it remains unchanged for the duration of an experiment. To investigate this, we considered in vitro R_in measures taken two times during iPSC-CM experiments. R_in was measured with 5 mV steps from a holding potential of 0 mV (i.e. the leak reversal potential) before and after acquiring current clamp data. The data are skewed, with a mean of R_in = 2.71 GΩ and median of R_in = 0.82 GΩ.

The relative change in R_in from the first to the second time point was calculated and is plotted against the time elapsed between R_in measurements in Figure 4B. The median change of R_in is −15%. Because positive and negative changes cancel each other out in these statistics, we also inspected the absolute change, where we found a median of 20%. These data illustrate that R_in measurements often change over time. If we assume R_m is stable during experiments, this change in R_in should be attributed to R_seal and suggests that the average cell’s R_seal decreases (and therefore I_leak increases) over time.

R_in is not a good approximation of R_seal at any holding potential

A holding potential of −80 mV is a common choice for approximating R_seal with R_in measures. At this potential, sodium, calcium, and several potassium currents are expected to be largely inactive, but contributions from both I_K1 and I_f must still be considered. Whilst I_K1 is perhaps close to its reversal potential (and therefore small), I_f is not and can play a large role at this voltage.

We recently showed that I_f is present in at least some of the iPSC-CMs used in this study.²⁵I_f is also present in both the Kernik and Paci models, and we found the dynamics of the Kernik I_f model to be quite similar to the in vitro data in this study (Figure 5A and B). Figure 5A shows an example cell’s response to an I_f-activating hyperpolarizing step before and after treatment with quinine, at a concentration expected to lead to 32% I_f block (these data are taken from a section of a larger protocol—see Clark et al.²⁵Figure 6A). A change in total current of nearly 2 A/F is observed after holding at −120 mV for 1 s (Figure 5A). In Clark et al.,²⁵ nine cells were treated with quinine, and the average change during the I_f-activating segment was 1.34 A/F. We found that these nine cells could be sorted into three triplets based on the amount of quinine-induced I_out change during the I_f segment: no/little sensitivity (ΔI_out of 0–0.2 A/F), moderate sensitivity (Δ I_out of 0.7–1.2 A/F), and large sensitivity (Δ I_out of >1.9 A/F). Simulations using the Kernik model with 32% block of I_f show a change of 1 A/F (i.e. moderate change) in I_out (Figure 5B).

To illustrate the effect of I_f on leak calculations, we compared simulations from Kernik + leak models with R_seal = 1 GΩ and with g_f set to zero (i.e. not sensitive to quinine during hyperpolarizing step), the Kernik baseline value (g_f = 0.0435 nS/pF, i.e. moderate sensitivity), or twice its baseline value (g_f = 0.087 nS/pF, i.e. large sensitivity) (Figure 5C). We also reduced g_K1 in these models to 10% of the baseline value to highlight the effects of I_f on R_in measures independent of I_K1. The calculated R_in values for these models at −80 mV are 2.03 GΩ for g_f = 0 nS/pF (little change), 1.50 GΩ for g_f = 0.0435 nS/pF (moderate change), and 1.16 GΩ for g_f = 0.087 nS/pF (large change) (Figure 5C). These simulations show that, at −80 mV, I_f contributes to I_out and affects measures of I_leak.

Using these same models, we then calculated R_in values at multiple holding potentials between −90 and +30 mV to determine whether we could find a potential where R_in is close to R_seal, thereby minimizing the prediction error (Figure 5D). The model predicts that 20 mV (R_in = 0.96 GΩ) minimizes the error in our approximation of R_seal. This does not mean that R_in measurements at 20 mV will always produce the best estimate of R_seal. Instead, it indicates the size of I_ion does not change much when taking a 5 mV step from this potential. There is, however, a considerable amount of total current present, making this R_seal prediction sensitive to variations in the predominant ionic currents at this potential. Moreover, I_leak will be small and therefore more difficult to measure as 10 mV is close to the leak reversal potential (0 mV). It is also worth noting that the complex voltage- and time-dependent behaviour of transmembrane currents make R_in measures sensitive to both the duration and size of the voltage step (e.g. see supplement to Clerx et al.²⁷). In summary, it is difficult to find a holding potential where R_seal can be measured without contamination from any transmembrane currents (i.e. where I_leak = I_out).

Taken together, these findings provide evidence to the claim that R_seal cannot be reliably measured in iPSC-CMs once access is gained.

Next, we compared the effect of I_f on R_m and investigated the error in assuming R_seal ≈ R_in, at both a 0 mV (i.e. I_leak reversal) and −80 mV holding potential. At 0 mV, the Kernik + leak model is not sensitive to changes in g_f, as I_f is largely non-conductive (Figure 6A). However, due to an increased relative contribution of inward currents at 0 mV, the Kernik + leak model predicts a R_in with a large overestimation of R_seal (Figure 6B). This error increases as the true value of R_seal increases. Figure 6B also illustrates the sensitivity of the model to variations in g_f at −80 mV, with R_seal estimation errors decreasing as g_f increases; these errors also increase as R_seal increases. The improved prediction accuracy of the 0.087 nS/pF model at −80 mV is a coincidental side effect of doubling g_f: with a different distribution of ion current densities or a larger baseline g_f value, the same doubling could just as easily worsen R_seal predictions. For example, the R_in of an iPSC-CM with a large I_K1 current may slightly underestimate R_seal at −80 mV—doubling g_f in this case would result in a greater underestimation, increasing the error of the estimate.

C_m and R_in(0 mV) correlate with minimum potential

The iPSC-CMs used in this study displayed a heterogeneous phenotype (Figure 7), producing both spontaneously firing (n = 25) and non-firing (n = 12) current clamp recordings. Figure 7A shows three cells with very different baseline current clamp recordings: non-firing and depolarized (green), spontaneously firing with a short AP (teal), and spontaneously firing with a long AP (red). Non-firing cells (MP = −42 ± 8 mV) and cells with spontaneously firing APs were depolarized (MP = −54 ± 7 mV)—the spontaneously firing cells also had a shorter AP duration (APD₉₀ = 128 ± 71 ms) (Figure 7B) relative to adult cardiomyocytes²⁸ and iPSC-CM models.^13,14

We used linear regression analyses to determine if there is a correlation between g_in/C_m and AP biomarkers. Here, we use g_in (instead of R_in), as it reduces the spread of this variable and positively correlates with I_leak providing a more interpretable comparison with AP morphology. The values of each cell’s g_in and C_m are shown in Figure 7C. I_leak’s effect on AP morphology is expected to scale directly with g_in and inversely with C_m. This is because g_in, even if a poor estimate, is expected to correlate with g_seal (Figure 6B)

A given g_leak will cause a smaller contribution in larger cells (i.e. cells with larger C_m), because the ionic currents are expected to scale with the size of the cell. For this reason, four AP biomarkers (MP, APD₉₀, CL, and dV/dt_max) were compared with g_in/C_m (Figure 8). The MPs of spontaneously firing (R = 0.44, P < 0.05) and non-firing (R = 0.76, P < 0.05) cells are positively correlated with g_in/C_m (Figure 8A). This finding is in agreement with our in silico studies showing that increasing g_seal, thereby increasing g_in, will depolarize the cell (Figure 2). The other three biomarkers failed at least one of the assumptions required when conducting a linear regression analysis (see Supplementary Methods). There are no obvious trends when comparing g_in/C_m with CL or dV/dt_max. The APD₉₀ plot, however, indicates there may be some AP shortening as g_in/C_m increases. Due to under-sampling and a lack of linearity, we cannot make any claims of significance between these two measures. Leak simulations with the models, though correlated, did not predict a linear relationship between g_seal and these biomarkers (Figure 2C and D). However, the MP vs. g_in/C_m relationship passes all tests of linear regression assumptions and trends in the same direction as the Kernik and Paci simulations in Figure 2.

Fitting background currents in human-induced pluripotent stem cell-derived cardiomyocyte models can absorb and imitate I_leak

We used optimization to study the potential of linear background currents (e.g. sodium and calcium) to imitate leak effects (see Supplementary Methods). We fit the baseline Kernik model to a Kernik + leak model with R_seal = 5 GΩ (Figure 9), allowing only the background sodium (g_bNa) and background calcium (g_bCa) conductances to vary. These currents were selected because they were incorporated into the Kernik model without independent iPSC-CM experimentation or validation. The best-fit model had an increased g_bNa (×7.0), whilst g_bCa (×1.0) did not change much relative to the baseline model (Figure 9A). Whilst not a perfect match, the best-fit trace reproduced qualitative features of the baseline + leak trace, showing a depolarized MP and a smaller amplitude (Figure 9B). This indicates that increased I_bNa can affect the AP in a fashion similar to I_leak such that mathematical iPSC-CM models may absorb I_leak effects by erroneously increasing background currents.

Discussion

Leak current is a common and unavoidable experimental artefact that affects patch clamp recordings. In this study, using both model predictions and experimental data, we show that leak current: (i) affects iPSC-CM AP morphology, (ii) can vary during experiments, (iii) cannot be accurately estimated after access is gained to an iPSC-CM, and (iv) may be absorbed by linear equations for background currents when iPSC-CM models are fit to experimental AP data. During iPSC-CM current clamp studies, leak consideration often starts with a pre-rupture seal measurement (with a 1 GΩ threshold) and is ignored if the seal appears to remain stable throughout the study. Here, we argue leak effects should be quantitatively scrutinized during the acquisition, analysis, and fitting of experimental data. Furthermore, we believe cell-to-cell variation in seal resistance contributes to observed iPSC-CM AP heterogeneity—often attributed nearly entirely to variations in ionic current densities.

Leak affects action potential morphology

Simulations in chick embryonic cardiomyocytes, which are smaller than adult human cells (with model C_m = 25.5 pF), have previously shown that leak current substantially depolarizes the MP and shortens the CL, even with R_seal values of 5 GΩ.²⁹ More recently, it was shown that in vitro iPSC-CMs were significantly depolarized during single-cell experiments, but not when cells were clustered.^11,12 These results indicate that isolated iPSC-CMs are likely affected by leak current. Our in vitro and in silico findings support this conclusion and strengthen the argument that iPSC-CM AP morphology is strongly affected by leak current.

Our in silico work indicates that I_leak has a smaller effect on recordings of adult cardiomyocyte AP morphology when compared with iPSC-CMs (Figure 3B). This effect is strongly modulated by C_m, indicating the larger size of adult cardiomyocytes has a moderating effect on I_leak-induced AP changes. When the I_leak artefact in this adult model is normalized by the average iPSC-CM capacitance (50 pF, Figure 3A), I_leak substantially alters the AP shape at R_seal values above 1 GΩ. But the effects are much less than in the iPSC-CM model (Figure 2)—this indicates the ionic current expression profile of adult cardiomyocytes (e.g. greater I_K1 and lower I_f density), in addition to cell size, and moderates the effects of I_leak on adult AP recordings. Thus, differentiation strategies that aim to mature the iPSC-CM phenotype (both in size and ionic current expression) will likely produce cells that are affected less by I_leak artefact.

Human-induced pluripotent stem cell-derived cardiomyocytes have long been defined by their immature and heterogeneous electrophysiological phenotype.^10,30 Such features are due, at least in part, to the types of ion channels expressed and cell-to-cell variations in ionic current conductances.^10,30 In this study, differences in the I_f responses to nine quinine-treated cells are an example of how iPSC-CM ionic currents can vary from one cell to the next. Heterogeneity in AP morphology and ionic current expression is also seen in primary adult cardiomyocytes.^31–33

In this study, we show that I_leak also contributes to this immature and heterogeneous AP phenotype during single-cell patch clamp experiments. The relative importance of I_leak’s influence on AP shape varies amongst cells and depends on several factors, including R_seal, C_m, and the ionic current expression profile. Simulations indicate that the AP shape can be substantially altered (relative to non-patched cells), even when R_seal is equal to 10 GΩ, an unrealistically high acceptance criterion for iPSC-CM patch clamp studies. These factors, along with the potential for R_seal to change during an experiment, can confound drug and genetic mutation studies. For example, the irregular and depolarized phenotype (caused at least in part by I_leak) of iPSC-CMs in our recent cardiotoxicity study²⁵ made it impossible to measure consistent cell-specific changes in spontaneous AP morphology from pre- to post-drug application.

The AP-altering effects of I_leak can be effectively eliminated by patching cells whilst in engineered heart tissue or monolayer. The electrical coupling of cells in these conditions results in an enormous effective capacitance, rendering I_leak an infinitesimal contributor to total current. Whilst this eliminates the I_leak artefact, it also comes at a cost—this approach does not allow for the direct measure of APs in individual cells, limiting the ability to study iPSC-CM heterogeneity. In addition, it is not possible to acquire voltage clamp data from cells in these conditions—as such, one could not acquire both AP and descriptive data about individual currents, as we recently have done in isolated cells.²⁵

Predicting R_seal during experiments

R_seal can be well approximated prior to gaining access to a cell, but after perforation (or rupture), the presence of membrane currents makes it impossible to obtain an accurate measurement (Figure 5). Our in silico work shows that, even when currents such as I_f and I_K1 are reduced to <10% of their baseline values, R_in (measured at −80 mV) is still a poor approximation of R_seal (Figure 6, solid black line).

To address these difficulties, we believe it may be feasible to use the pre-rupture R_seal and post-rupture R_in measures to calculate estimates of R_seal during an experiment. This approach would require an accurate measure of R_in just after access is gained. Using R_seal and the initial R_in, it is possible to calculate R_m (Figure 1). An estimate of R_seal could then be made at any time during the experiment, assuming the calculated R_m stays constant, by re-measuring R_in and using Eq. (4). This approach relies on two major assumptions: (i) the perforation/rupture step does not affect the seal, and (ii) a protocol or procedure exists that can be used prior to each measurement of R_in to ensure that the contribution of R_m is consistent. We cannot say for certain that these assumptions will always be valid. However, we believe that recording frequent R_in measurements, estimating R_seal, and scrutinizing changes are important steps for the correct interpretation of iPSC-CM current clamp data.

Correcting for R_seal during experiments

We believe these R_seal estimates should be used in a dynamic clamp leak compensation setup to address the limitations caused by a depolarized and variable MP. The approach works by injecting simulated currents into a cell in a real-time continuous loop during current clamp experiments.³⁴I_K1 dynamic clamp has been used on iPSC-CMs to attain quiescence at a MP below −70 mV so the cells can be paced at a desired frequency.^25,35–37 A dynamically clamped leak compensation current has been implemented and used in manual patch clamp studies with neonatal mouse cardiomyocytes,²² demonstrating the potential of using such an approach with small cardiomyocytes. The effects of leak and the ability of leak compensation to recover adult cardiomyocyte behaviour have also been demonstrated in an in silico study.²³ Together, these investigations demonstrate the potential of dynamic clamp as an experimental tool to simultaneously address shortcomings of the cells (i.e. I_K1 density) and experimental setup (i.e. I_leak). This technique has the potential to improve the descriptive ability of iPSC-CMs when used in biophysical and drug investigations.

Inaccuracies in these estimates, however, will remain, resulting in the potential to under- or over-compensate. Over-compensation will hyperpolarize the MP and prolong Phases 1 and 2 of the AP, so we believe under-compensation is preferable. We suggest injecting a fraction of the full compensatory current to mitigate the risk of underestimating R_seal. The Nanion Dynamite⁸ sets the leak per cent compensation to 70%, which seems reasonable.³⁸

Models of background currents can incorporate leak artefacts

The Kernik and Paci iPSC-CM models took ion-specific background currents from the ten Tusscher et al.³⁹ model. These currents can trace their roots to the seminal work of Luo et al.,⁴⁰ where they were included to help maintain physiologically realistic intra-cellular concentrations.

Direct measurements of I_bCa and I_bNa in iPSC-CMs have not been reported. The Kernik and Paci iPSC-CM models both adopted the ventricular³⁹ formulation for I_bCa and I_bNa and then set the conductances of these currents by comparing model predictions of the AP with in vitro measurements in iPSC-CMs. We posit that I_bNa is overestimated and compensates for the explicit consideration of leak current artefacts, a source of discrepancy between these models and reality. We expect consideration of leak when constructing iPSC-CM models to reduce background sodium current and result in a more realistic model of intact iPSC-CMs.

Modelling experimental artefacts

Whilst the effects of experimental artefacts in single-cell studies are well-established, consideration of them whilst building ion channel and AP models has been limited.⁴¹In silico studies investigating series resistance effects on voltage clamp recordings have been done in fast-activating currents, such as I_Na and I_to,^42,43 but to our knowledge, artefact equations have not been included in the calibration process for widely used models of these currents—although the I_Na model by Ebihara et al.⁴² was incorporated directly into the widely copied I_Na model by Luo et al.⁴⁰ Recently, Lei et al.⁴⁴ demonstrated that coupling experimental artefact equations with an I_Kr mechanistic model improved predictions. These studies show that including experimental artefact equations in model fitting can improve the descriptive ability of the resulting electrophysiological models. As such, we believe experimental artefacts should be explicitly considered at the modelling phase and not ignored simply because a pre-determined minimum threshold is reached (e.g. 1 GΩ). Based on our findings, we believe cardiomyocyte models and especially iPSC-CM models should explicitly include leak currents when fitting to experimental current clamp data.

Recommendations

Our results provide important insights and recommendations for experimentalists and modellers alike:

1. Experimental: R_seal should be recorded before gaining access to a cell and R_in measured frequently during an experiment. It is important to measure R_in from a voltage that provides a consistent measure of R_m, such that any changes in R_in can be attributed to changes in R_seal.

2. Experimental: Dynamic injection of a leak compensation current can help a cell recover its native AP, including the MP. Because R_seal is difficult to measure during experiments and to avoid over-compensation, we advise under-compensation (e.g. 70%). Additionally, R_seal and R_in measures should be reported.

3. Modelling: Explicit inclusion of I_leak will improve the descriptive ability of iPSC-CM models. Whilst this may not always improve fits to AP data, it will take into account an important current affecting iPSC-CM recordings.

Limitations and future directions

This study has several limitations that should be considered during future investigations that may be affected by I_leak. First and foremost, when gathering these data for a previous study, we did not follow our new recommendation of recording the exact value of R_seal before gaining access and then measuring R_in just after perforation. Going forward, we hope to use these two values to predict R_seal at multiple time points during an experiment, as outlined in Section 3.2. Second, we only conducted these experiments in one cell line. Whilst our results appear similar to data from other labs,¹¹ it would be useful to conduct this study on multiple cell lines in the same lab. Third, we did not attempt dynamic injection of a leak compensation current—in future work, we would like to investigate this as an approach to reducing cell-to-cell heterogeneity. Finally, the iPSC-CM models have innumerable differences from the cells used in this study, which is evident when comparing AP morphologies of in vitro cells (Figure 7A) to in silico models (Figure 2). However, the agreement that we did see between simulations and our in vitro data demonstrates the potential of improving the descriptive ability of iPSC-CM models by including a leak current.

Conclusion

In this study, we demonstrate that leak current affects iPSC-CM AP morphology, even at seal resistances above 1 GΩ, and contributes to the heterogeneity that characterizes these cells. Using both in vitro and in silico data, we showed the challenges of estimating R_seal after gaining access to a cell and that R_seal is subject to change during the course of an experiment. We also posit that background sodium current in iPSC-CM models may be responsible for masking leak effects in in vitro data. Based on these results, we make recommendations that should be considered by anyone who collects, analyses, or fits iPSC-CM AP data.

Supplementary material

Supplementary material is available at Europace online.

Acknowledgements

We thank the members of our labs for their support and feedback at every step of the way.

Funding

This work was supported by the National Institutes of Health (NIH) National Heart, Lung, and Blood Institute (NHLBI) grants U01HL136297 to D.J.C. and F31HL154655 to A.P.C., the Wellcome Trust via a Senior Research Fellowship to G.R.M. (grant number 212203/Z/18/Z), Universidade de Macau via a UM Macao Fellowship and support from FDCT Macao (Science and Technology Development Fund, Macao S.A.R. (FDCT) reference number 0048/2022/A) to C.L.L., and the MKMD programme of the Netherlands Organization for Health Research and Development (grant number 114022502) to T.P.B.

Data availability

All data, code, and models can be accessed from GitHub (https://github.com/Christini-Lab/iPSC-leak-artifact).

Translational perspective

Human iPSC-CMs have emerged as a promising translational tool to study human cardiac physiology outside of the clinic. They have been particularly useful to investigate cell-level pro-arrhythmic substrates, including genetic mutations and ion channel-blocking drugs, and play a critical role as a model for validating drug effects on human whole-cell electrophysiology in the Comprehensive in vitro Proarrhythmia Assay (CiPA) initiative. However, the depth of insights from iPSC-CM data is often limited by inter- and intra-lab heterogeneity caused, at least in part, by patch clamp experimental artefact. In this manuscript, we show how the seal–leak current is an often-overlooked artefact that confounds studies with iPSC-CMs. Ultimately, the findings and recommendations within this manuscript will improve the use of iPSC-CMs as an in vitro model to study cardiac electrophysiological diseases and patient-specific treatment strategies.

References

Author notes