Human Atrial Cardiac Microtissues for Chamber-Specific Arrhythmic Risk Assessment

Abstract

Introduction

Although atrial fibrillation is the most prevalent disorder of electrical conduction, the mechanisms behind atrial arrhythmias remain elusive. To address this challenge, we developed a robust in vitro model of 3D atrial microtissue from human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes and evaluated chamber-specific chemical responses experimentally and computationally.

Methods

We differentiated atrial and ventricular cardiomyocytes (aCMs/vCMs) from GCaMP6f-expressing hiPSCs and assessed spontaneous AP activity using fluorescence imaging. Self-assembling 3D microtissues were formed with lactate purified CMs and 5% human cardiac fibroblasts and electrically stimulated for one week before high resolution action potential (AP) optical mapping. AP responses to the atrial-specific potassium repolarizing current I_Kur-blocker 4-Aminopyridine (4-AP) and funny current I_f-blocker Ivabradine were characterized within their therapeutic window. Finally, we expanded upon a published hiPSC-CM computational model by incorporating the atrial-specific I_Kur current, modifying ion channel conductances to match the AP waveforms of our microtissues, and employing the updated model to reinforce our experimental findings.

Results

High purity CMs (> 75% cTnT⁺) demonstrated subtype specification by MLC2v expression. Spontaneous beating rates significantly decreased following 3D microtissue formation, with atrial microtissues characterized by their faster spontaneous beating rate, slower AP rise time, and shorter AP duration (APD) compared to ventricular microtissues. We measured atrial-specific responses, including dose-dependent APD prolongation with 4-AP treatment and dose-dependent reduction in spontaneous activity post-Ivabradine treatment.

Conclusion

The presented in vitro platform for screening atrial-specific responses is both robust and sensitive, with high throughput, enabling studies focused at elucidating the mechanisms underlying atrial arrhythmias.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12195-021-00703-x.

Introduction

Atrial fibrillation (AFib) has become increasingly prevalent globally,²⁹ with the number of U.S. cases projected to double between the years 2010 and 2030.¹⁰ As the most common form of sustained cardiac arrhythmia, developing long-lasting treatments for AFib is critical for reducing the risk of stroke and heart failure in the aging population.^29,36 There are currently two accepted treatments for AFib: (1) radiofrequency ablation and (2) antiarrhythmic drugs. In the first approach, a catheter is used to ablate focal sources and reentry points of arrhythmias, in an attempt to terminate AFib.⁵⁰ Accurate identification of all ablation targets, however, remains challenging, and an annual recurrence rate of up to 10% has been reported post-ablation therapy.^14,37 The most common treatment for AFib, however, is the use of class I and III antiarrhythmic drugs, which respectively block Na⁺ and K⁺ channels to restore sinus rhythm.^15,61 While these drugs are moderately effective at suppressing the disease, they indiscriminately target both the atria and ventricles, increasing the patient’s susceptibility to developing potentially fatal ventricular arrhythmias via QT prolongation, particularly when class III K⁺ channel blockers are prescribed.⁶¹ Therefore, recent drug discovery efforts for AFib treatment have focused on developing atrial selective drugs that target ion channels primarily expressed in the atria, such as the ultrarapid delayed rectifier K⁺ current, I_Kur.^48,55 Even though several I_Kur blockers are under development, their dose-dependent response on action potential (AP) properties, safety, and efficacy require further investigation as experimental and clinical data remain scarce.^12,22

To fully guide the development of atrial-specific drugs and evaluate their safety and efficacy, there is a critical need for robust in vitro platforms for cardiotoxic assessment and mechanistic target evaluation. In 2013, the Comprehensive In Vitro Proarrhythmia Assay (CiPA) initiative emphasized the crucial need to integrate human platforms into the drug development process,⁴⁴ as in vivo animal models often fail to replicate the human drug response due to species-specific differences in ion channel expression levels.⁵⁶ Furthermore, off-target drug effects which may alter the activity of multiple ion channels must be thoroughly characterized to monitor for unexpected pro-arrhythmic hazards. To this end, many research groups have employed the use of human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) for toxicity screening of novel therapeutics as they recapitulate key physiological properties, including proper human ion channel expression profiles, contractility, and AP shape.^18,54

Many hiPSC-CM cardiotoxicity studies to date, however, rely on homotypic 2D monolayer cultures, which do not account for the complex 3D cell-to-cell interactions between multiple cell types, which are now known to modulate the electrophysiological behavior of tissues.^18,52 Only recently has there been development and validation of 3D micro/macrotissues to investigate drug effects on diverse pro-arrhythmic metrics, such as AP properties, calcium transients, and contractility, and challenges related to limited throughput suggest that these 3D models are best suited to later stage development and toxicity testing purposes. Furthermore, the field remains heavily focused on ventricular responses to potentially cardiotoxic drugs while their effects on atrial arrhythmogenicity are largely overlooked. Inherent differences in ion channel expressions between the atria and ventricles, as well as the presence of atrial-specific channels,^20,53,59 may result in chamber-specific arrhythmogenic responses to the same compound. While ventricular arrhythmias tend to be more deadly than their atrial counterparts, thorough characterization of chamber-specific responses is imperative in establishing high safety standards for drug testing across all patient populations. For example, patients with Wolff–Parkinson–White syndrome have diverse atrial arrhythmia presentation (from AFib to tachycardia) and are clinically described as having developed accessory electrical conduction pathways between the atria and ventricles.⁷ Taken together, there remains a critical need to accurately assess the effects of novel therapeutics in both atrial and ventricular tissues with high throughput for cardiotoxicity evaluation and establishing drug targeting efficacy.

To address these needs, we present a highly sensitive and predictive in vitro platform for evaluating atrial-specific arrhythmia responses with high throughput. In this study, we generated self-assembling 3D atrial and ventricular microtissues from hiPSC-CMs and cardiac fibroblasts, assessed their calcium transients, and used optical mapping to characterize cardiac subtype differences in AP properties. We then validated our model by evaluating chamber-specific responses to atrial specific compounds targeting the I_Kur and I_f channels and used our experimental findings to update a previously published hiPSC-CM AP computational model^41,42 to include the atrial specific channel, I_Kur, for understanding AP waveform perturbations.

Methods

Cardiomyocyte Differentiation

We differentiated atrial and ventricular cardiomyocytes (aCMs/vCMs) from GCaMP6f-expressing human induced pluripotent stem cells (hiPSCs; WTC human male iPSCs, Gladstone Institutes) using small molecule modulations of Wnt signaling, as described previously with slight modifications.^6,32 Briefly, hiPSCs were cultured on vitronectin coated plates in Essential 8 Medium (E8 media; Thermo Fisher Scientific). Prior to starting differentiation, hiPSCs were singularized, seeded onto Matrigel-coated plates in E8 media with 5 μM ROCK Inhibitor (RI; Fisher) and cultured to 80% confluency, at which point the cells were treated with 4.5 μM CHIR 99021 (Tocris), a glycogen synthase kinase 3 (GSK3) inhibitor, for 24 ± 1 h in cardiac differentiation medium with 3 components (CDM3) basal medium (RPMI 1640 supplemented with 213 μg/mL l-ascorbic acid and 500 g/mL human serum albumin).⁶ At differentiation day 3, cells were treated with 5 μM IWP2 (Tocris), a Wnt inhibitor, in a CDM3 media mixture containing half spent and half fresh media. Atrial-subtype differentiation was achieved by daily supplementation of 1 μM retinoic acid (RA; Sigma Aldrich) at days 3–6,¹¹ while cells for ventricular specification did not receive RA treatment. We differentiated both atrial and ventricular cardiomyocytes within each differentiation batch for direct comparison of chamber-specific responses. We removed IWP2 at day 5 of differentiation and replaced the CDM3 culture medium every other day. Following the first signs of contraction between days 9 and 13, hiPSC-CMs were maintained in RPMI 1640 with B27 supplement (RPMI/B27). We then harvested hiPSC-CMs between days 13–15 with 0.25% trypsin in 0.5 mM EDTA and replated the cells to Matrigel-coated plates for metabolic-based lactate purification.⁵⁷ At day 20, hiPSC-CMs were fed with lactate media composed of 4 mM sodium L-lactate (Sigma) in sodium pyruvate- and glucose-free DMEM (Thermo Fisher Scientific; Catalog # 11966025) for 4 days, with media changes every other day. Purified hiPSC-CMs were then cultured in RPMI/B27 and used for generating 3D microtissues between days 27 and 30. A timeline summarizing our cardiac differentiation protocol is shown in Fig. 1a.

Figure 1.

Study design. (a) Overview of the cardiomyocyte differentiation timeline to 3D microtissue formation. Cardiac-directed differentiation was achieved via timed modulation of the Wnt signaling pathway, and atrial-subtype specification was obtained by retinoic acid (RA) supplementation. Beating CMs were lactate purified before (b) being used to assess spontaneous action potential (AP) activity via GCaMP fluorescence imaging. (c, top) Self assembling 3D microtissues were generated by seeding hiPSC-CMs into agarose hydrogel molds, followed by electrical stimulation. (c, bottom) Immunohistochemistry confirmed cardiomyocyte subtype with atrial (MLC2a, red) and ventricular (MLC2v, green) markers. (d) AP traces were captured with a voltage-sensitive dye and a high-speed CMOS camera to assess changes in AP properties in response to different drug treatments.

Human Cardiac Fibroblast Maintenance

Primary human cardiac fibroblasts (hCFs, Sigma) were cultured in DMEM/F12 with 10% fetal bovine serum (FBS), 1% Pen/Strep, and 4 ng/mL basic fibroblast growth factor (Reprocell). hCFs between passage numbers P2-P4 were used to generate 3D microtissues to promote heterocellular crosstalk that aids tissue compaction and improves electrical conduction and excitability.^25,28,51

GCaMP Evaluation of Spontaneous Beating Rates

We utilized GCaMP fluorescence to characterize the automaticity of aCMs and vCMs in both 2D culture and 3D microtissues. The samples were imaged with an inverted fluorescent microscope (Olympus IX50) 1–3 days before and 6 days after 3D microtissue formation without electrical stimulation. Fluorescent images were acquired for 15 s for an accurate estimate of signal periodicity across multiple beats. Background fluorescence was removed and changes in fluorescence signal intensity corresponding to the intracellular calcium transients (CaTs) of the beating cardiomyocytes were plotted using a custom MATLAB script (Fig. 1b). An automated peak-detection algorithm was implemented to quantify the period between beats and the width of GCaMP signal, and their averages used to estimate the frequency of spontaneous activity.

3D Microtissue Generation

Sterile 2% (wt/vol) agarose in PBS was pipetted into 35-microwell negative molds with hemispherical bottoms (Fig. 1c; 3D Petri Dish^®, MicroTissues Inc.). Once casted and gelled, hydrogels were removed from the negative molds and allowed to equilibrate in RPMI/B27 media with 1% Pen/Strep overnight in an incubator. Lactate-purified hiPSC-CMs were then harvested, singularized, and suspended in RPMI/B27 media with 10% FBS and 1% Pen/Strep, collecting a subset of the cells for flow cytometry analysis of cTnT and MLC2v expression. We added 5% hCFs of the total number of hiPSC-CMs to the cell suspension and pipetted the cell mixture to the center of the hydrogel at a density of 500,000–700,000 cells/hydrogel. Cells settled by gravity and developed cell-cell adhesions, producing 35 individual self-assembled, scaffold free, and spheroidal microtissues consisting of 15,000–25,000 cells/microtissue. Cells were allowed to settle into the cylindrical recesses for 30 min before adding media supplemented with 5 μM RI. Culture medium was changed one day post-seeding and replaced every other day. The 3D microtissues were electrically field stimulated for 6–8 days with a 1 Hz, 10.0 V, and 4.0 ms duration bipolar pulse train (C-Pace EP, IonOptix) to maintain excitability and precondition the microtissues prior to optical mapping.⁴⁷ The 3D microtissues beat synchronously with 1 Hz field stimulation within two days, and significant tissue compaction was observed within the first three days post-seeding (Electronic Supplementary Material (ESM) Movie 1 and ESM Fig. 1).

Flow Cytometry

Samples for flow cytometry were fixed in 4% (vol/vol) paraformaldehyde for 10 min in the dark at room temperature and permeabilized with 0.75% saponin in PBS. Cells were stained with 1:100 mouse monoclonal IgG1 cTnT (Invitrogen; Catalog#: MA5-12960; Clone 13–11) and 1:10 monoclonal IgG1 myosin light chain 2v (MLC2v conjugated to APC; Miltenyi Biotec; Catalog#: 130-106-134) for 1 h. Secondary staining was performed with 1:200 goat anti-mouse IgG PE (Jackson; Catalog#: 115-116-072) for 1 h. cTnT⁺ cells were used to determine cardiomyocyte purity, and MLC2v^+/− cells were used to distinguish between ventricular and atrial subtypes, respectively, as previously described.³⁰ Samples were run on a BD FACSAria™ IIIu Flow Cytometer (BD Biosciences), and data were analyzed with FlowJo.

Optical Mapping of Cardiac Action Potential

We transferred hydrogels containing microtissues to a Petri dish on a temperature-controlled chamber (Dual Automatic Temperature Controller TC-344B, Warner Instrument) to maintain ambient temperatures of 35 ± 1 °C throughout imaging. Microtissues were gently perfused in a solution containing (in mM) 140 NaCl, 4.0 KCl, 1.0 MgCl₂, 1.25 CaCL₂, 0.33 NaH₂PO₄, 5.0 HEPES, 5.0 sodium pyruvate, and 7.5 glucose warmed with an inline heater. Microtissues were allowed to equilibrate in the perfusion solution for 30 min and subsequently labeled with a voltage-sensitive dye (5 μM di-4-ANEPPS; Fisher Scientific) for 5 min to enable membrane potential (V_m) recordings of AP. Excess residual dyes were washed out thoroughly before data collection. A CMOS camera (Ultima-L, Scimedia, Japan) was used to acquire fluorescence images at 1000 or 2000 frames-per-second, and AP traces were reconstructed from fluorescence intensity data. Batches of microtissues with signal-to-noise ratio greater than 30:1 and rapid upstroke with rise time below 20 ms during baseline recording were selected for drug testing. Semi-automated analyses of AP parameters were conducted using an in-house analysis software and included rise time, APD₃₀, APD₅₀, APD₈₀, APD to maximum repolarization rate (APD_MxR), and APD triangulation (APD_tri = APD_MxR − APD₅₀). We then normalized APD_tri to APD_MxR to account for significantly longer APDs in ventricular microtissues. An in-depth description on the post-processing of optical mapping data for AP analysis is detailed in our previous work²⁸ and summarized in Fig. 1d.

Screening of Arrhythmogenic Compounds

We studied the effects of 4-aminopyridine (4-AP; Sigma-Aldrich), an I_{Kur and} I_to blocker, and Ivabradine (Sigma-Aldrich), an I_f blocker, on the AP properties of 3D cardiac microtissues. Three molds of atrial or ventricular microtissues obtained from 3 unique differentiation batches were used to test the compound 4-AP under 1 Hz electrical stimulation with a custom platinum field electrode controlled by a Myopacer EP field stimulator (IonOptix, 10 V/cm strength, 4 ms biphasic stimulation). Baseline AP recordings were acquired before three increasing dosages of 4-AP (1, 3, and 100 μM) were introduced into the perfusion solution. At each dose, the microtissues were allowed to respond to the drug for 5–10 min before AP recordings were acquired. At least 10 s of data were acquired per microtissue at each dose. In another set of experiments, three molds of atrial microtissues from 3 unique differentiation batches were used to test the compound Ivabradine without electrical stimulation to investigate how AP properties and spontaneous activity were altered in response to the drug. We did not test Ivabradine on ventricular microtissues as they do not exhibit spontaneous APs. Similar to the drug administration for 4-AP, four increasing dosages of Ivabradine (0.5, 1, 3, and 5 μM) were introduced to the perfusion solution, with the microtissues given 5–10 min to respond to the drug before AP assessment of spontaneous activity.

Immunohistochemistry

Seven days post-tissue formation, 3D microtissues were fixed overnight inside the 35-microwell hydrogel with 4% (vol/vol) paraformaldehyde and 8% (wt/vol) sucrose in PBS at room temperature. Hydrogels were washed twice with PBS and allowed to equilibrate in 15% and then 30% sucrose in PBS for at least 12 h to preserve the structural integrity of the hydrogels during sectioning. Hydrogels containing microtissues were then processed into frozen blocks with OCT, sectioned into 10 μM sections, and stored at – 80 °C until ready for staining. In preparation for immunohistochemistry, frozen sections were rinsed 3 times in PBS, and non-specific binding was blocked with 1.5% normal goat serum in PBS for 1 h. Sections were incubated in primary antibodies against myosin light chain 2a (MLC2a, 1:400, Synaptic Systems, Catalog#: 311 011) and myosin light chain 2v (MLC2v, 1:400, ProteinTech, Catalog#: 10906-1-AP) or cardiac troponin I (cTnI, 1:100, Catalog#: ab47003) and vimentin (1:100, Sigma, Catalog#: V6630) overnight at 4 °C. After washing with PBS 3 times for 5 min, sections were incubated in secondary antibodies conjugated to either Alexa Fluor 488 or Alexa Fluor 594 (1:200, Invitrogen) for 1 h at room temperature. All sections were counterstained for cell nuclei with Hoechst 33342 (1:2000, Invitrogen) for 15 min and rinsed with PBS before coverslips were mounted with a Vectashield Antifade Mounting Medium (Vector Laboratories). Images were then taken on an inverted confocal microscope (Olympus FV3000) and processed with ImageJ.

Gene Expression Analysis of 3D Cardiac Microtissues

We isolated total RNA from two independent differentiation batches of both atrial and ventricular microtissues 7-days post tissue formation using a ReliaPrep RNA Miniprep Systems (Promega), according to manufacturer’s protocols. RNA yield and purity were quantified with a NanoDrop 1000 UV-Vis Spectrophotometer (Thermo Fisher Scientific), and reverse transcription reactions were performed in triplicate using random hexamers to synthesize ~ 150 ng of complementary DNA (cDNA) per sample (Applied Biosystems). We then employed TaqMan gene expression assays to amplify target genes during real-time quantitative polymerase chain reaction (RT-qPCR) for up to 40 cycles with a ViiA 7 Real Time PCR System. A thorough list of all selected target genes, which included atrial and ventricular marker genes, as well as general ion channel genes, are provided in ESM Table S1. Finally, RT-qPCR results were normalized to the 18S rRNA housekeeping gene and relative gene expression evaluated with the $2^{-\mathrm{\Delta }\mathrm{\Delta }{C}_{\mathrm{t}}}$ method.³³

Computational Modeling

We relied on computational modeling to recapitulate the effects of I_Kur inhibition by 4-AP on the AP properties of hiPSC-aCMs. Our model was based on the 2015 hiPSC-CM AP model equations established by Paci et al.^41,42 Briefly, the Paci model was derived from patch-clamp I-V curves and AP data of atrial and ventricular-like immature hiPSC-CMs,³⁴ and included the major currents (I_Na, I_to, I_CaL, I_K1, I_Kr, I_Ks, and I_f), pump/exchanger currents (I_NaK, I_pCa, and I_NaCa), Ca²⁺ dynamics and buffering in the sarcoplasmic reticulum, and background currents. However, the Paci model did not include the atrial-specific ultrarapid voltage gated repolarizing potassium current I_Kur, which we added based on the AP model equations published by Maleckar et al. for adult human atrial CMs.³⁵ Specifically, the following set of equations³⁵ for I_Kur were incorporated into the Paci model:

$$I_{\text{Kur}}=g_{\text{Kur}}\ast a_{\text{ur}}\ast i_{\text{ur}}\left(V-E_{\text{K}}\right)$$

$$\frac{\text{d}{a}_{\text{ur}}}{\text{d}t}=\frac{\left(a_{\text{ur},\infty }-a_{\text{ur}}\right)}{{\tau }_{\text{aur}}}$$

$$\frac{\text{d}{i}_{\text{ur}}}{\text{d}t}=\frac{\left(i_{\text{ur},\infty }-i_{\text{ur}}\right)}{{\tau }_{\text{iur}}}$$

$$a_{\text{ur},\infty }=\frac{1.0}{\left[1.0+{\text{e}}^{-\left(V+6\right)/8.6}\right]}$$

$$i_{\text{ur},\infty }=\frac{1.0}{\left[1.0+{\text{e}}^{\left(V+7.5\right)/10.0}\right]}$$

$${\tau }_{\text{aur}}=\frac{0.009}{\left[1.0+{\text{e}}^{\left(V+5.0\right)/12.0}\right]}+0.0005$$

$${\tau }_{\text{iur}}=\frac{0.59}{\left[1.0+{\text{e}}^{\left(V+60.0\right)/10.0}\right]}+3.05$$

where g_Kur is the maximum conductance, a_ur is the activation gating variable, i_ur is the inactivation gating variable, and τ_a/iur is the activation/inactivation time constant for I_Kur. Further, V corresponds to the membrane potential and E_K corresponds to the Nernst potential for K⁺. To appropriately scale I_Kur for immature hiPSC-CMs, we tuned the maximum conductance of the ultrarapid potassium channel, together with the other major ion channels, by minimizing the sum of square residuals between the modeled and experimentally measured averaged AP waveforms, without altering the activation/inactivation kinetics of the ion channels. All other parameters remained unchanged from the original Paci model. All modeled APs were simulated over 800 beats to ensure that steady-state behavior was reached under 1 Hz stimulation before APD measurements were taken and traces presented in the figures.

Statistical Analyses

All data were reported as mean ± standard deviation and tested for normality with the Shapiro-Wilk test. A student’s two-tailed unpaired t-test was used to study the effects of cardiac subtype on the different AP metrics, CaTs, and automaticity. For compound testing experiments, a one-way analysis of variance with Dunnett’s multiple comparison to baseline controls was performed for all AP metrics with normal distribution. Statistical analyses of non-normal data were performed with the nonparametric Kruskal–Wallis test. For relative gene expression data, a two-tailed unpaired t-test assuming unequal variance was used for the statistical analysis of individual target genes. All statistical tests were conducted in GraphPad Prism version 8.1.1 (GraphPad Software) with p < 0.05 representing statistical significance.

Results

Cardiomyocyte Subtype, Maturation State, and 3D Microenvironment Influence Spontaneous Beating Rates

RA supplementation at days 3–6 of differentiation yielded cardiomyocytes with significantly reduced MLC2v expression compared to those without RA supplementation (MLC2v⁺_aCM = 18 ± 5 vs. MLC2v⁺_vCM = 31 ± 0.2, p < 0.05; ESM Fig. 2), indicating successful subtype specification. This result agrees with immunohistochemistry data showing substantial MLC2a and MLC2v positive cells in atrial and ventricular microtissues, respectively (Fig. 1c), and relative gene expression data confirming the upregulation of atrial genes (NR2F2 and NPPA) and downregulation of ventricular genes (MYL2 and IRX4) in atrial microtissues (ESM Fig. 3A). We also consistently generated high-purity cardiomyocytes following metabolic-based lactate purification for both atrial and ventricular subtypes (cTnT⁺_aCM = 83 ± 5% vs. cTnT⁺_vCM = 89 ± 5%, p > 0.05; ESM Fig. 2). Fluorescence imaging of CaTs in 2D monolayer cultures without electrical stimulation showed that spontaneous activity varied with cardiac subtypes, with aCMs demonstrating faster automaticity than vCMs at day 28 of differentiation (freq_aCM,D28 = 0.64 ± 0.25 Hz vs. freq_vCM,D28 = 0.31 ± 0.06 Hz, p < 0.05; n = 8 wells per subtype across two differentiation batches; Figs. 2a and 2b and ESM Movie 2). Interestingly, in a single differentiation batch where we cultured cardiomyocytes for up to 45 days, we measured a significant increase in spontaneous beating rates in aCMs to 1.72 ± 0.49 Hz (n = 12 wells, p < 0.0001 compared to day 28), while that of vCMs remained relatively unchanged at 0.53 ± 0.28 Hz (n = 6 wells; p = 0.24 compared to day 28). At day 28, we noticed that slow automaticity in vCMs is associated with CaT prolongation (CaT_aCM = 0.94 ± 0.18 sec vs. CaT_vCM = 1.34 ± 0.17 sec, p < 0.001, Fig. 2d), which presented wider peaks in vCMs, with a prominent plateau phase (black arrows, Fig. 2a). Furthermore, we noted a decrease in spontaneous activity following 3D microtissue formation with the addition of 5% hCF, confirmed via immunohistochemical staining with vimentin (ESM Fig. 4). Specifically, the frequency of spontaneous activity significantly decreased to 0.41 ± 0.13 Hz (p < 0.05 and p < 0.0001 compared to monolayer culture at day 28 and day 45, respectively) in 3D atrial microtissues (n = 30 microtissues across 2 differentiation batches; ESM Movie 3), while spontaneous activity was absent in 3D ventricular microtissues (n = 30 microtissues across 2 differentiation batches; Fig. 2c). Taken together, these results suggest that the subtype, maturation state, and 3D microenvironment modulate cardiomyocyte electrophysiological behavior.

Figure 2.

Differences in spontaneous AP activity firing rates and Ca²⁺ transients between atrial and ventricular subtypes across 2D and 3D structures. (a) GCaMP traces of hiPSC-CMs from 2D culture showed (b) significant differences in spontaneous AP beating rates with subtype and age. Atrial hiPSC-CMs demonstrated faster automaticity than their ventricular counterpart and continued to exhibit faster pacing with longer culture times. The incorporation of 5% human cardiac fibroblast to generate 3D microtissues (c) reduced automaticity in atrial samples and ceased spontaneous activity in ventricular samples. (d) Ventricular hiPSC-CMs demonstrated longer Ca²⁺ transient duration than atrial hiPSC-CMs likely attributed to longer APD. Values are shown as mean ± standard deviation (*p < 0.05, ***p < 0.001, ****p < 0.0001). ^‡Denotes statistically significant differences with every other conditions.

Figure 4.

Dose-dependent effects of 4-Aminopyridine (4-AP) and Ivabradine on AP properties. (a) 4-AP treatment resulted in dose-dependent APD prolongation in atrial microtissues, as observed by the rightward shift of the APD₈₀ cumulative probability distribution. (b) These effects were not observed in the ventricular samples. (c) Treatment with Ivabradine across all tested dosages significantly reduced the cycle length of spontaneous AP events in a dose-dependent manner. Values are shown as mean ± standard deviation (*p < 0.05, **p < 0.01, ****p < 0.0001) and represent data from n = 3 differentiation batches.

3D Atrial Microtissues are Characterized by Slow AP Upstroke and Shorter Action Potential Duration Compared to 3D Ventricular Microtissues

Using optical mapping, we were able to capture highly repeatable action potential traces from 3D cardiac microtissues under 1 Hz pacing with minimal beat-to-beat and microtissue variability within the same batch, although greater variations in ventricular AP traces were observed during late repolarization (ESM Fig. 5A–D). While batch-to-batch differences in APDs were observed, these variations were small and remained distinct between atrial and ventricular microtissues (ESM Fig. 5E-F). Notably, atrial microtissues demonstrated slower rise time to peak (rise time_aCM = 12.2 ± 2.4 ms vs. rise time_vCM = 6.3 ± 1.8 ms, p < 0.0001). Atrial AP also lacked a prominent plateau phase, resulting in APD₃₀, APD₅₀, APD₈₀, and APD_MxR values that were on average 3x shorter than those of ventricular microtissues (Figs. 3a through 3f and ESM Movie 4). APD_tri normalized to APD_MxR is significantly higher in atrial microtissues, (Fig. 3g), suggesting that AP triangulation is greater in atrial tissue. Table 1 summarizes the chamber-specific differences in AP parameters measured via optical mapping in our 3D microtissues. Finally, we observed a sharp AP peak during early repolarization (black arrow, Fig. 3a) in 3D microtissues generated from atrial cardiomyocytes cultured for 45 days, suggesting that maturation of input cardiomyocytes produced microtissues with enhanced ion channel expression profile for adult atrial AP, including the I_to channel.

Figure 3.

Differences in AP properties between atrial and ventricular 3D microtissues under 1 Hz pacing. (a) Comparison of AP traces showed shorter AP duration (APD) in atrial microtissues, with longer culture resulting in a rapid phase 1 repolarization (green line, black arrow). Notably, atrial microtissues exhibited significantly (b) slower rise time and shorter (c) APD₃₀, (d) APD₅₀, (e) APD₈₀, and (f) APD_MxR when compared to ventricular microtissues, with (g) APD_Tri/APD_MxR showing increased AP triangulation. Values are shown as mean ± standard deviation (****p < 0.0001) and represent data from n = 3 differentiation batches.

Table 1.

Atrial and ventricular action potential differences.

AP parameter	Atrial	Ventricular	Statistics
Rise time (ms)	12.2 ± 2.4	6.3 ± 1.8	p < 0.0001
APD30 (ms)	70.7 ± 9.5	247.8 ± 29.4	p < 0.0001
APD50 (ms)	90.1 ± 11.7	289.6 ± 33.1	p < 0.0001
APD80 (ms)	114.6 ± 15.5	334.6 ± 39.8	p < 0.0001
APDMxR (ms)	118.3 ± 14.3	317.7 ± 35.6	p < 0.0001
APDTri/APDMxR	0.24 ± 0.04	0.09 ± 0.01	p < 0.0001

Data are presented as mean ± standard deviation and represent data from n = 3 differentiation batches

3D Microtissues Exhibit Dose-Dependent and Chamber-Specific Responses to Known I_Kur and I_f Channels Blockers

We tested our 3D cardiac microtissues with 4-AP, a drug which more sensitively inhibits the atrial-specific I_Kur channel at low doses while targeting I_to at higher doses,⁵ to investigate chamber-specific responses. As anticipated, we identified a response to low-dose 4-AP in atrial microtissues that was absent in ventricular microtissues (Figs. 4a and 4b). Specifically, APD₃₀, APD₅₀, APD₈₀, and APD_MxR in atrial microtissues were all significantly prolonged with increasing doses of 4-AP (ESM Fig. 6 and ESM Table 2), consistent with delayed repolarization and a rightward shift in the cumulative probability distribution plot shown for APD₈₀ in Fig. 4a. While not observed across all doses, we also found a significant increase in rise time in our atrial microtissues at the 100 μM 4-AP dose, though APD triangulation remained unchanged (ESM Fig. 6). On the contrary, no significant changes in APD₈₀ in ventricular microtissues were detected across all drug dosages of 4-AP (Fig. 4b). We did, however, observe a small decrease in APD₃₀, APD₅₀, and APD_MXR at 3 μM dose of 4-AP, which returned towards baseline values at 100 μM (ESM Fig. 6). We believe that this APD shortening is not driven by the effects of 4-AP, as the drug is known to prolong APD by blocking either I_Kur, I_to, or I_Kr with increasing dosages. Compared to atrial APD responses, which showed a distinct upward shift in mean values with diminishing overlap of the standard deviation bars at higher 4-AP doses, the standard deviation range in ventricular APD responses maintained a significant overlap across all 4-AP doses when compared to baseline (ESM Fig. 6). This observation is likely due to the larger variability observed in our ventricular AP traces during the repolarization phase, when compared to atrial traces (ESM Fig. 5C–D), and the large sample size that increased our statistical power and enabled high sensitivity. Further, the maturity state of the hiPSC-CMs, as further characterized by ion channel gene expression data (ESM Fig S3), is likely still not at the level of mature adult cardiomyocytes despite our advances to mature our cardiomyocytes with metabolic-based purification treatment and 3D culture; although we did identify significantly reduced KCNA5 gene expression, the gene that encodes atrial-specific I_Kur channel, in ventricular microtissues compared to atrial microtissues. The slight increase in APD₃₀ and APD₅₀ values towards baseline at the 100 μM dose, however, could potentially be driven by a partial I_to block in ventricular microtissues.

Table 2.

Major ion channel conductances.

Max conductance	Atrial	Ventricular
gNa	1.9939e3 (S/F)	2.3262e3 (S/F)
gCaL	5.1814e-5 (m3/(Fxs))	8.6357e-5 (m3/(Fxs))
gto	59.8077 (S/F)	14.9519 (S/F)
gKur	0.01875 nS	N/A
gKr	31.360 (S/F)	43.3067 (S/F)
gKs	2.041 (S/F)	3.0615 (S/F)
gK1	19.1925 (S/F)	36.5940 (S/F)
gf	75.2578 (S/F)	21.0722 (S/F)

The major ion channel conductance values were modified from the 2015 Paci model,⁴² with the conductance for I_Kur modified from the Maleckar model,³⁵ to match AP waveforms obtained experimentally from our atrial and ventricular 3D microtissues

We also tested the effect of Ivabradine, a hyperpolarization-activated cyclic nucleotide gated channel (HCN channel, I_f current) blocker, on the spontaneous AP activity of atrial microtissues. As shown in the middle panel of Fig. 4c, Ivabradine treatment resulted in a dose-dependent increase in spontaneous cycle length from 1071 ± 93 ms at baseline to 1393 ± 154 ms at a 5 μM dose (p < 0.0001). A small subset of atrial microtissues (14%), however, lost their spontaneous activity at the highest Ivabradine dose (ESM Fig. 7). Closer inspection of our voltage traces showed that these events appeared to have been driven by a series of failed depolarizations, as drifting baseline potential failed to reach the necessary threshold to produce successful depolarization (black arrow, ESM Fig. 7). These results suggest that multiple mechanisms may underlie automaticity in atrial microtissues, such as a Ca²⁺ clock underlying pacemaker activity,²⁴ in addition to HCN4 channel expression. However, we confirmed that cycle length prolongation and loss of spontaneous activity was a direct result of Ivabradine treatment as atrial microtissues remained responsive to and followed 1 Hz electrical pacing post-Ivabradine treatment (ESM Fig. 7). Interestingly, we found no difference in HCN4 expression between atrial and ventricular microtissues; instead, compared to ventricular microtissues, atrial microtissues presented significant downregulation of the KCNJ2 gene that encodes the I_K1 channel responsible for establishing resting membrane potential (ESM Fig. 3B).

Computational Modeling of Chamber Specific hiPSC-CMs Recapitulates Experimental Findings

We incorporated the atrial-specific I_Kur channel into a previously established hiPSC-CM AP computational model.^41,42 We adjusted the maximum conductance values for the major ion channels from the Paci model, summarized in Table 2, to best match the AP shape of our 3D atrial and ventricular microtissues (Figs. 5a and 5b) as the Paci model was derived from 2D culture of hiPSC-CM and minor adjustments are necessary to compensate for the effect of 3D environment on electrophysiology. The best-fitted parameters showed that atrial microtissues had reduced I_Na and I_CaL when compared to ventricular microtissues (Fig. 5c), explaining the longer rise time to peak and lack of a well-developed plateau phase leading to the triangular shape of the experimentally measured atrial AP traces. Furthermore, atrial microtissues had reduced I_Kr, I_Ks, and I_K1, but increased I_to and I_f when compared to ventricular microtissues (Fig. 5c), which together accounted for the narrowed AP peak and slowly depolarizing resting membrane potential that resulted in increased spontaneous activity in our 3D atrial microtissues. To explore the effects of I_Kur on the AP shape and duration of atrial microtissues, we conducted a large g_Kur sweep, relative to the optimized g_Kur parameter in our atrial model, and reported the corresponding changes in APD₃₀, APD₅₀, and APD₈₀ in Fig. 5d, with AP waveforms for representative g_Kur values presented in Fig. 5e. As expected, the observed changes in APD followed a behavior that resembled a dose-response curve, with decreased conductances, reflective of drug-induced I_Kur block, resulting in APD prolongation. Figure 5e also showed that changes in I_Kur altered the concavity of the AP repolarization phase. Next, we performed a sigmoidal fit on the APD response curve and used the modeled equation to relate how drug-induced APD prolongation by 4-AP correspond to changes in g_Kur in our model. A 1, 3, and 100 μM dose of 4-AP respectively reduced relative g_Kur to 0.75, 0.72, and 0.55, with computationally modeled responses represented in Fig. 5f showing similar APD prolongation behavior as experimentally obtained AP traces (Fig 4a). Closer inspection at individual ion channel changes in our experimental model showed a delayed inactivation of I_Kr, I_Ks, I_K1, and I_f, as seen by a small rightward shift in Fig. 5c, along with small compensatory increases in I_to and I_Kr current magnitudes, in the modeled AP responses to 100 μM dose of 4-AP.

Figure 5.

Computational modeling of hiPSC-CM action potentials. Comparison between modeled (orange) and experimentally obtained AP traces (gray) for (a) atrial and (b) ventricular microtissues demonstrate a strong fit. (c) Modeled traces of the major ion currents responsible for determining AP shape demonstrate variations in current intensities between the atrial and ventricular hiPSC-CMs. Notably, hiPSC-aCMs exhibit smaller I_Na, I_CaL, I_Kr, I_Ks, and I_K1, but more prominent I_to and I_f, similar to findings reported in the literature for adult CMs. Atrial specific I_Kur channel was incorporated into our computational model. Modeled current responses to the treatment of 100 μM of 4-AP resulted in delayed inactivation of I_Kr, I_Ks, I_K1, and I_f, along with small compensatory increases in I_to and I_Kr current magnitudes. (d) Modeled changes in APD₃₀, APD₅₀, and APD₈₀ values and (e) atrial AP waveforms in response to changes in I_Kur conductances, g_Kur. (f) Computationally modeled changes in AP waveforms in response to low-dose 4-AP treatment, with g_Kur values estimated from the sigmoidal curve fitted in panel d.

Discussion

The prevalence of arrhythmias in disease and in response to drugs has led to great interest in developing human in vitro models to study arrhythmia mechanisms and cardiotoxicity. While a focus on ventricular responses to drugs has been critical for reducing fatal consequences of drug-induced toxicity, drug effects on atrial electrophysiology remains understudied partly due to a lack of available atrial testing platforms with high throughput and validated responses to atrial-targeting drugs. In this study, we have demonstrated an in vitro platform for chamber-specific evaluation of arrhythmogenic risk using human atrial 3D cardiac microtissues. Using GCaMP fluorescence imaging and optical mapping, we highlighted important differences in the spontaneous activity, calcium handling, and AP properties of atrial and ventricular microtissues. We then detected dose-dependent and chamber-specific responses to the atrial-selective I_Kur-blocking drug 4-AP and the I_f-blocking drug Ivabradine, driven by the critical need to address both atrial and ventricular responses in cardiotoxicity evaluation and the development of novel therapeutics. Finally, we used our experimental AP traces to incorporate the atrial I_Kur current into an established hiPSC-CM AP computational model⁴² and investigate differences in the behaviors of major ion channels between the two microtissue subtypes to confirm the mechanistic alterations of individual ion currents observed in the composite AP waveform.

In vitro platforms using hiPSC-CMs are proving to be invaluable for cardiotoxic assessment of chemical compounds and drugs. However, biophysical and environmental cues, maturation of hiPSC-CMs, and measurement technology all shape the suitability of a platform to be “fit-for-purpose” to appropriately model human in vivo responses to drugs. As such, it is imperative when presenting a new screening platform that we thoroughly discuss the choices made behind its development and their implications on the interpretation of the results. An advantage of our atrial platform results from our decision to lactate-select our input cardiomyocytes for improved purity (ESM Fig. 2) and to generate 3D microtissue of hiPSC-CMs, interspersed with a low percentage of hCFs under scaffold-free conditions. Our rationale is to capture cellular interactions in 3D space and heterocellular crosstalk with fibroblasts.⁶⁴ 3D assembly has been shown to accelerate cardiomyocyte maturation rate⁴ while altering AP propagation due to increased cell-to-cell connectivity.⁵² On a similar note, the role that fibroblasts play in modulating the electrophysiological properties of tissues is becoming widely accepted,⁶³ and we have previously reported that the addition of 5% hCFs optimally improves electromechanical function and promotes compaction in our engineered microtissues.^28,51 Our 3D microtissues were cultured for a minimum of 6-days under electrical stimulation to improve electromechanical function,⁴⁷ and we showed in a small subset of our samples that increasing 2D culture times to 45 days prior to microtissue formation further promotes maturation in our aCMs, resulting in a pronounced phase 1 repolarization, indicating enhanced I_to and/or I_Kur expression that is more reflective of adult human aCMs (black arrow, Fig. 3a). This suggests that our platform can be tailored to better replicate various in vivo AP behaviors of adult humans, by longer 2D or 3D culture, providing flexibility beyond the current study parameters that prove sufficient in detecting dose-dependent and chamber specific responses to atrial-selective drugs. Finally, our optical mapping approach using a voltage sensitive dye and a high-speed camera allows us to capture AP shape with high signal-to-noise ratio (~ 70:1) and high frame rate (~ 1000 Hz), including rise time of action potential upstroke, phase 1 repolarization and plateau, and APD at different repolarization levels. This technological sensitivity enables reliable measurements of potential cardiotoxicity and presents a robust method to characterize arrhythmia responses, as we report quantitative AP metrics of bulk tissue, rather than individual CMs, that are highly reproducible with low variability.

The cardiomyocyte subtype differences in spontaneous beating rates, calcium handling, and AP properties measured in our cultures (Figs. 2 and 3) are consistent with those reported in the literature. In agreement with the findings of other groups,^11,45 hiPSC-aCMs exhibit reduced MLC2v expression (ESM Fig. 2), upregulation and downregulation of atrial (NR2F2 and NPPA) and ventricular (MYL2 and IRX4) specific genes respectively (ESM Fig. 3A), and faster spontaneous beating rates compared to hiPSC-vCMs. Gene expression data indicate that this difference in spontaneous activity is caused by decreased KCNJ2 expression in hiPSC-aCMs (ESM Fig. 3B),¹³ which is responsible for establishing stable resting membrane potential via the I_K1 current. This behavior is evident in our AP traces (Fig. 3a), where a slowly depolarizing baseline potential was observed in between 1 Hz pacing in atrial, but not in ventricular microtissues, and reinforced by the reduced I_K1 current of our atrial hiPSC-CM computational model (Fig 5c). Interestingly, we noted that longer culture times increased spontaneous activity in aCMs, while remaining unchanged in vCMs (Fig. 2). This result suggests differences in maturation progression hallmarks between cardiac subtypes, with increased I_K1 expression in vCMs suppressing automaticity at an earlier timepoint. Furthermore, similar to the literature on adult human cardiomyocytes, our atrial AP and CaT traces showed a triangulated profile, as opposed to the “spike-and-dome” shaped ventricular AP with a prominent plateau phase.¹³ These waveforms can be attributed to differences in Ca²⁺ handling between the two cardiomyocyte subtypes, with aCMs exhibiting smaller systolic CaTs that decayed more rapidly⁵⁹ while vCMs presenting CaTs with wider peaks and a plateau phase (Fig. 2a). Finally, the slow rise time in atrial microtissues corroborates findings of slowed upstroke velocity reported in atrial tissues by several groups¹⁹ and is likely driven by a more depolarized and drifting resting membrane potential partially inactivating Na⁺ channels available for AP generation.

While spontaneous beating is a common output metric of hiPSC-CM screening technologies such as microelectrode arrays,³⁹ we demonstrate that cardiomyocyte assembly in 3D significantly decreased spontaneous beating rates in atrial microtissues while eliminating spontaneous activity in ventricular microtissues (Fig. 2). Several factors may underlie the increased spontaneous activity of both atrial and ventricular hiPSC-CMs in 2D compared to 3D. It has been shown that ion channel expression profiles and amplitude in 3D are different and higher, promoting adult-like action potentials.²⁷ It is also worth noting that automaticity can easily arise in 2D environments where cell-cell coupling is much less prominent than in 3D, allowing CMs to act with greater autonomy. Meanwhile, the slow elevation of pacemaker potential can be easily dissipated to the neighboring cells in 3D, stabilizing resting membrane potential, in a phenomenon termed source-sink mismatch.⁶² As a result, a larger number of cells firing synchronously is required to initiate an action potential in 3D compared to 2D.^46,62 Another source of apparent automaticity in 2D is that reentry circuits can be formed in a planar 2D culture,²³ which may appear as a regular rhythmic behavior when quantification occurs in a limited area (e.g. field of view or electrode array). However, the small size of our 3D microtissues, with diameters of ~ 380 μm (ESM Fig. 1),²⁸ does not allow reentry formation. Further studies are required in the field in order to identify the molecular signature of differentiation, ion channel profiles, and cell-cell coupling effects on automaticity in 2D vs. 3D culture systems.

The drug 4-AP is a candidate therapeutic for the treatment of AFib as it blocks the atrial specific repolarizing current I_Kur at low doses, with an IC₅₀ of ~ 5 μM,^2,38 to prolong APD and reduce spontaneous excitation rate. 4-AP, however, is also known to indiscriminately target I_to across both chambers at higher doses, although the reported IC₅₀ for I_to in aCMs is one-third that reported in vCMs (~ 0.5 mM in atria vs. ~ 1.5 mM in the ventricles).^2,38 At even higher doses, 4-AP blocks the hERG channel that drives I_Kr (IC₅₀ ~ 4 mM),⁴⁹ raising concerns regarding its potential side effect of inducing Torsade de Pointes via QT prolongation. In our experiments, we noted dose-dependent prolongation of APD₃₀, APD₅₀, APD₈₀, and APD_MxR in atrial microtissues with increasing low-dose treatment of 4-AP (1, 3, and 100 μM) that was not observed in ventricular microtissues (Figs. 4a and 4b and ESM Fig. 6), indicating that we were successful in selectively inhibiting ion channels in the atria without targeting I_to or I_Kr in ventricular microtissues. We did not detect atrial changes in normalized APD_tri (ESM Fig. 6) across all tested doses, suggesting that APD prolongation was not driven by hERG channel block, although further study is necessary to explore how other repolarizing currents compensate to increased I_Kur block at higher 4-AP concentrations, as we observed small compensatory increases in I_to and I_Kr (green line, Fig. 5c) in response to low-dose I_Kur block. Taken together, these findings highlight that our atrial microtissue platform is highly sensitive in detecting low-dose responses to 4-AP, within the drug’s therapeutic window for AFib treatment.

Ivabradine is a drug proposed to suppress AFib by inhibiting the pacemaker current I_f to lower heart rate and maintain normal sinus rhythm, acting during the resting (polarized or diastolic) phase in contrast to 4-AP which changes depolarization/repolarization (or systolic) kinetics. Recent meta-analyses of clinical trial data, however, showed that use of Ivabradine is associated with increased AFib incidence,¹ emphasizing the need to better understand the role that I_f and Ivabradine play in atrial electrophysiology and in vivo suppression or initiation of AFib. We identified a dose dependent increase in spontaneous activity cycle length with a subset of atrial microtissues exhibiting a loss of spontaneous activity at high dosages of Ivabradine, likely attributed to complete or partial I_f block that resulted in a series of failed depolarizations (Fig 4c and ESM Fig. 7). These irregular patterns between failed and successful AP generation suggest that alternative mechanisms, including an intrinsic Ca²⁺ clock, may regulate pacemaker activity²⁴ and may be a source of triggered ectopic activity for arrhythmia initiation. These oscillatory V_m behavior without complete repolarization can inactivate Na⁺ channels, which may increase risks for conduction failure and precipitate reentrant arrhythmias. Therefore, targeting I_f alone may not be sufficient, but additional modulation of Ca²⁺ clock may be necessary to successfully prevent AFib.

The drug-induced AP changes to 4-AP and Ivabradine observed in our study matched the findings of other research groups that utilized hiPSC-CMs based platforms for drug screening. Despite significant differences in how the authors employed hiPSC-CMs to investigate drug responses, Gunawan et al.²¹ and Lemme et al.³¹ both detected dose-dependent APD prolongation to 4-AP in atrial hiPSC-CMs that was absent in ventricular hiPSC-CMs. While Gunawan et al. explored drug responses in monolayer cultures, Lemme et al. generated 3D engineered heart tissues (EHTs) to account for 3D cellular interactions. Similarly, low doses of Ivabradine have been shown to significantly reduce spontaneous beating rates in EHTs constructed from either atrial or ventricular hiPSC-CMs, with high dosages resulting in complete cessation of spontaneous activity.^3,8 However, none of the above studies employed purification methods or incorporated cardiac fibroblasts into their model. Complex heterocellular crosstalk between cardiomyocytes and other non-cardiomyocyte populations, notably fibroblast, endothelial cells, and vascular smooth muscle cells, is fundamental in governing the electromechanical behavior of tissues and should not be overlooked. For example, cardiomyocyte-fibroblast coupling has been shown to alter the conduction velocity and resting membrane potential of tissues,^26,58 while paracrine signaling from endothelial cells regulates cardiomyocyte response and remodeling in disease.⁴⁰ Giacomelli et al.^16,17 demonstrated the important contributions of noncardiomyocyte populations by investigating how 3D triculture of ventricular microtissues composed of hiPSC-CMs, cardiac fibroblasts, and cardiac endothelial cells impacted tissue maturation, AP shape, and Ca²⁺ transients. They showed that ventricular microtissues supplemented with both fibroblast and endothelial cells resulted in the most mature phenotype, although fibroblasts appeared to play a more significant role in electromechanical maturation. Additionally, their triculture microtissues exhibited larger fraction of APs with a prominent I_to notch, which we observed in our atrial microtissues with longer culture times, reflective of a more mature phenotype. Taken together, although future work will include thorough investigations of the role of endothelial cells, our atrial microtissues reproducibly captured drug responses that matched closely with the findings of others in the literature.

In order to better understand the complex interactions between ion currents in atrial and ventricular cardiomyocytes, combining experimental data with in silico approaches is useful for predicting potential cardiotoxic effects of pharmacological compounds beyond hERG-based QT prolongation. Since in silico studies can compensate for the deviation in responses between hiPSC-CMs and human adult cardiomyocytes, computational models can greatly assist the interpretation of our experimental results. We based our computational model off the work by Paci et al.,^41,42 who developed their model equations from extensive patch clamp data of hiPSC-CMs,³⁴ and tuned the maximum conductances of major ion channel⁹ to replicate our experimentally measured AP waveforms for both atrial and ventricular microtissues. Although our optimized conductances slightly differ from the original model (Table 2), these differences may arise from our experimental data coming from 3D microtissues compared to 2D measured APs, as we have described previously that 3D structure alters electrophysiological behaviors. The overall pattern of our optimized channel conductances, however, is similar to findings reported in the literature, with reduced g_Na, g_CaL, and g_K1, increased g_to and g_f, and the presence of g_Kur (absent in the vCM model) in our atrial model compared to the ventricular model. The customized atrial conductances recapitulates the AP profile seen in our atrial microtissues and enables differentiation of atrial versus ventricular specific proarrhythmic toxicity. Despite lacking patch clamp data for I_Kur in hiPSC-aCMs (which has not been published to date) and so using equations for I_Kur adapted from the work by Maleckar et al. on adult human atrial CMs,³⁵ our model helps explain many of our experimental findings, including explaining the differences in spontaneous beating rates and AP properties between microtissue subtypes. Strong I_K1 and reduced I_f currents helped stabilize resting membrane potential to reduce spontaneous activity, while a prominent I_CaL resulted in a “spike-and-dome” AP profile in ventricular hiPSC-CMs (Figs. 5a thorugh 5c). Furthermore, our computational model exhibited a dose-response behavior across a wide range of g_Kur, which accurately modeled APD prolongation in response to 4-AP at the tested dosages. Future work may focus on refining this model to better capture dose-dependent responses of microtissues across a wide range of doses and develop a robust, predictive computational model for drug responses.

Finally, one of the greatest strengths of any in vitro platform, including the one presented here, is the potential to develop models from patient cells for healthy populations with diverse backgrounds or for disease modeling and approach questions in personalized medicine for drug cardiotoxicity or individualized therapies.⁴³ The mechanisms underlying atrial arrhythmogenesis are complex and remain elusive to this date, with both triggered activity and substrate level reentry pathways capable of initiating and sustaining arrhythmias.⁶⁰ As such, establishing patient specific models that better represent a vulnerable population, such as those with Wolff–Parkinson–White syndrome, is critical not only for the development of novel anti-arrhythmic drugs but also to screen for cardiotoxic effects of existing drugs. We have shown in our previous publication that we can capture drug-induced triggered arrhythmogenic events including early after depolarizations in ventricular microtissues.²⁸ The ability to evaluate chamber-specific drug responses in atrial microtissues will advance our understanding of cell-triggered atrial arrhythmia mechanisms and provide an avenue for developing and evaluating novel therapeutics.

Conclusion

In summary, we have presented a highly predictive in vitro platform to evaluate arrhythmic responses in atrial cardiac microtissues generated from hiPSC-aCMs. Our platform, which incorporates cellular interactions in 3D space and heterocellular crosstalk with fibroblasts, exhibited distinct atrial and ventricular characteristics in terms of their spontaneous beating rates, calcium handling, and AP properties. Treatment with 4-AP at low doses, aimed at targeting atrial specific I_Kur channel, resulted in dose-dependent APD prolongation unique to atrial microtissues that was not observed in ventricular microtissues, highlighting the sensitivity of our platform in capturing chamber-specific responses. Meanwhile, Ivabradine treatment induced a dose-dependent increase in spontaneous activity cycle length with several occurrences of failed depolarization detected. We further corroborate our experimental findings with an updated hiPSC-CM computational model, which incorporates the atrial specific I_Kur channel. Taken together, our work presents a significant step in establishing complimentary experimental and computational models for advancing cardiotoxicity evaluation, precision medicine, and drug development.

Supplementary Information

Below is the link to the electronic supplementary material.