Drug Screening Platform Using Human Induced Pluripotent Stem Cell-Derived Atrial Cardiomyocytes and Optical Mapping

Current drug development efforts for the treatment of atrial fibrillation (AF) are hampered by the fact that many preclinical models have been unsuccessful in reproducing human cardiac atrial physiology and its response to medications. In this study, we demonstrated an approach using human induced pluripotent stem cell-derived atrial and ventricular cardiomyocytes (hiPSC-aCMs and hiPSC-vCMs, respectively) coupled with a sophisticated optical mapping system for drug screening of atrial-selective compounds in vitro. We optimized differentiation of hiPSC-aCMs by modulating the WNT and retinoid signalling pathways. Characterization of the transcriptome and proteome revealed that retinoic acid pushes the differentiation process into the atrial lineage and generated hiPSC-aCMs. Functional characterization using optical mapping showed that hiPSC-aCMs have shorter action potential durations and faster Ca2+ handling dynamics compared to hiPSC-vCMs. Furthermore, pharmacological investigation of hiPSC-aCMs captured atrial-selective effects by displaying greater sensitivity to atrial-selective compounds 4-aminopyridine, AVE0118, UCL1684, and vernakalant when compared to hiPSC-vCMs. These results established that a model system incorporating hiPSC-aCMs combined with optical mapping is well-suited for pre-clinical drug screening of novel and targeted atrial selective compounds.

The advent of human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-2 CMs) has revolutionized the field of cardiac research. It has enabled the study of cardiac 3 diseases in a patient-specific and human-relevant in vitro model system which provides a 4 unique opportunity for clinical translation 1 . Furthermore, the ability to differentiate chamber-5 specific cardiomyocytes allows for a more precise study of cardiac disease physiology and 6 pharmacology. 7 The cardiomyocytes of the lower (ventricles) and upper (atria) chambers have distinct 8 characteristics that arise from differential developmental pathways. Previous work in vivo has 9 shown that the expression patterns of retinoic acid and retinaldehyde dehydrogenase 2 10 (RALDH2) are important determinants of the atrial fate 2-5 . These results were later recapitulated 11 in a pivotal study by Lee & Protze et al. 6 who determined that atrial cardiomyocytes (aCMs) 12 differentiated from human embryonic stem cells (hESCs) originate from a unique mesoderm 13 characterized by robust RALDH2 expression. This study established an atrial differentiation 14 protocol that included the addition of retinoic acid. Retinoic acid has also been utilized to 15 selectively differentiate hESCs and hiPSCs into aCMs in other studies [6][7][8][9][10] . 16 The distinct properties of the atrial and ventricular cardiomyocytes are determined by the 17 differential expression of unique sets of ion channels and other proteins that optimize their 18 specific function. Drugs that target atrial ion channels selectively can therefore produce 19 differences in pharmacological function in the two chambers. This atrial-selective pharmacology 20 is of utmost interest in the study and treatment of atrial-specific diseases such atrial fibrillation 21 (AF), which is the most common heart rhythm disorder. Investigating atrial-selective 22 pharmacology can assist and guide novel cardiac drug development as well as improving both 23 safety and efficacy by avoiding potential toxic electrophysiologic effects on the ventricular 24 chambers. 25 The differential pharmacology of stem cell-derived aCMs was studied previously by 26 Laksman et al. 7 who showed that flecainide can rescue the AF phenotype in a dish. Other 27 studies have also studied the selective pharmacological effects of agents on hiPSC-derived 28 aCMs but have largely focused on being proof-of-concept studies using limited number of test 29 compounds and standard measurement systems that are low in throughput 9,10 . With a focus on 30 translation, a pre-clinical model platform that characterizes pharmacological activity must 31 capture the main cardiac functional signatures that most closely mimic and predict human 32 cardiac physiology and drug responses. As such, we established in this study an in vitro assay 1 platform by combining hiPSC-derived atrial cardiomyocytes (hiPSC-aCMs) and high-content 2 optical mapping, a non-invasive all-optical system that simultaneously measures membrane 3 potential (Vm) and Ca 2+ transients at a high-resolution in a monolayer tissue format. 4 We first demonstrate a selective hiPSC-aCM differentiation protocol by modifying the 5 well characterized GiWi protocol 11 through the controlled introduction of retinoic acid. The 6 recapitulation of the human atrial phenotype of the hiPSC-aCMs was validated with assays that 7 measure the expression of gene transcripts and proteins, as well as functional signatures. We Maintenance and Expansion of hiPSCs 3 hiPSCs (WiCell, IMR90-1) were maintained and expanded in mTeSR1 medium and 4 feeder-free culture using 6-well plates coated with Matrigel. Using Versene (EDTA), hiPSCs 5 were passaged every 4 days or ~85% confluency at 1:15 ratio. Passaged hiPSCs were cultured 6 with mTeSR1 supplemented with 10 µM Y27632 for the first 24 hours and the mTeSR1 was 7 exchanged daily during cell culture maintenance. 8 Directed Differentiation of hiPSCs into Atrial and Ventricular Subtypes 9 hiPSC-derived ventricular cardiomyocytes were differentiated by employing a modified 10 GiWi protocol 11 that we previously described 12 . In brief, hiPSCs were seeded at a density of 11 87,500 cells/cm 2 . At day 0, differentiation was initiated using 12 µM CHIR99021. At day 3, the 12 cells were incubated with 5 µM IWP-4. At day 5, the media were refreshed with RPMI-1640 13 supplemented with B27 minus insulin. At day 7, the medium was replaced with cardiomyocyte 14 maintenance media (RPMI-1640 supplemented with B27 with insulin). Thereafter, 15 cardiomyocyte maintenance medium was replaced every 4 days. For the atrial differentiation 16  Gene expression profiling was conducted using multiplexed NanoString and real time 29 quantitative PCR (qPCR). Pooled total RNA was used in both assays. The extracted RNA was 30 reverse transcribed into cDNA which was used in the qPCR assay. Oligonucleotide sequences 1 are described in Table S7. The multiplexed mRNA profiling was conducted using a NanoString 2 Technologies (Seattle, WA) platform with a custom code set containing 250 gene probes. 3 Analysis was performed on the Nanostring Sprint instrument and nSolver analysis software with 4 the Advanced Analysis module. 5 Atrial Natriuretic Peptide Measurement 6 The levels of atrial natriuretic peptide (ANP) of hiPSC-aCMs and -vCMs were measured 7 by a competitive enzyme-linked immunosorbent assay (ELISA) using a commercially available 8 kit (Invitrogen, CA). The assay was conducted according to the manufacturer's protocol and was 9 measured using a spectrophotometric plate reader. 10

11
For cardiac enrichment, hiPSC-aCMs and -vCMs at day 20-30 post-differentiation were 12 dissociated into single cells which were then enriched using a MidiMACS PSC-derived 13 Cardiomyocyte Isolation Kit (Miltenyi Biotec, Germany) according to the manufacturer's 14 protocol. Enriched hiPSC-CMs were seeded on Matrigel-coated 24-well plates at a seeding 15 density of 600,000 cells per well. 16

17
Single hiPSC-aCMs and -vCMs were plated on gelatin (0.1%) and Geltrex (1:10) at 18 30,000 cells per well. After 48 hours in culture, glass electrodes were used to achieve the 19 whole-cell configuration with single hiPSC-CMs and only cells with gigaohm seals were used for 20 further analysis. The formulation for internal and external recordings solutions are outlined in the 21 Supplemental Information. Current recordings were performed using an Axon Instruments 700B 22 amplifier and digitized at 20 KHz. All recordings were performed at 33-35 ºC as maintained. For 23 pacing at 1 Hz, gradually increasing amounts of current were injected with a 1 ms pulse width 24 until reliable action potentials (APs) were triggered. The maximal upstroke velocity was 25 determined by calculating the maximum derivative and the resting membrane potential was 26 measured during a 5 second epoch without spontaneous activity one minute after break-in. 27 Further details on data analysis are found in the Supplemental Information. Information). The hiPSC-CMs were loaded with RH-237, blebbistatin, and Rhod-2AM 2 sequentially before imaging as described 12,13 . Both RH-237 and Rhod-2AM were excited by 530 3 nm LEDs. Images were acquired at a frame rate of 100 frames/second by a sCMOS camera 4 (Orca Flash 4.0 V2, Hamamatsu Photonics, Japan) equipped with an optical splitter. The cells 5 were paced using programmable stimulation. Data collection, image processing, and initial data 6 analysis were accomplished using custom software. The multi-well optical mapping system was 7 custom engineered in the lab based on a system as described previously 17 . Further details are 8 found in the Supplemental Information. 9

10
The drugs used in this study are listed Table S8. Drug stocks were further diluted in Ca 2+ 11 Tyrode's solution prior to pharmacological testing with the final DMSO concentration in the 12 experimental solution not exceeding 0.03% (v/v). Drug effects were studied in serum-free 13 conditions (i.e. Ca 2+ Tyrode's and drug only) at four doses by sequentially increasing the drug 14 concentration in the same well with recordings at 20-minute intervals. 15

16
Further details on data and statistical analysis can be found in the Supplemental 17 Information. Unpaired t-tests were conducted to compare two groups (i.e. hiPSC-aCMs vs. 18 hiPSC-vCMs) in the analysis of qPCR, ELISA, patch clamp recordings, and optical mapping 19 (baseline condition and normalized drug effects). Analysis of dose-dependent effects were 20 performed using one-way ANOVA and Dunnett's post-hoc test. All data are presented as mean 21 ± SEM unless noted otherwise. Significance level for all statistical analysis was set at p < 0.05 22 with the following notation: *p < 0.05, **p < 0.01, ***p < 0.001.   Compared to hiPSC-vCMs, hiPSC-aCMs were found to have no significant difference in 12 the pan cardiac phenotype. Expression of the pan cardiac transcript NKX 2.5 measured by 13 qPCR was similar between hiPSC-aCMs and -vCMs ( Figure 1B), as was cardiac troponin T 14 (cTnT) protein expression measured by flow cytometry (Figures 1C and S1). The protein 15 expression of MLC-2v was reduced in hiPSC-aCMs compared to hiPSC-vCMs (8.0 ± 1.1 % vs. 16 57.0 ± 0.5 %; p<0.05) ( Figure 1C). Furthermore, hiPSC-aCMs displayed higher concentrations 17 (increased by 91%) of atrial natriuretic peptide (ANP) at 65 ± 2 compared to 34 ± 6 ng/mL in 18 hiPSC-vCMs as measured by ELISA (p<0.05). 19 The qPCR assay revealed that atrial-specific transcripts such as atrial natriuretic peptide 20 (NPPA), connexin 40 (GJA5), the L-type calcium channel Ca V1.3 (CACNA1D), and the K + 21 channels Kv1.5 (KCNA5) and Kir3.1 (KCNJ3) transcripts were all expressed at a significantly 22 higher levels in hiPSC-aCMs compared to hiPSC-vCMs (p<0.05, Figure 1B). Another ventricular 23 marker, IRX4, also had decreased expression in hiPSC-aCMs ( Figure 1B). Furthermore, 24 consistent with previous studies 8-10,14,15 , hiPSC-aCMs started beating at day 10 or earlier and 25 exhibited an increased beating frequency relative to hiPSC-vCMs, which started beating around 26 day 10-12 post-differentiation. 27 Gene Expression Analysis of hiPSC-aCMs 28 We performed an extensive gene expression analysis of hiPSC-aCMs and -vCMs using 29 NanoString technology in which each mRNA copy was digitally counted for accurate and 30 sensitive detection of gene expression 16 . Five independent differentiation batches of each 31 cardiac subtype were included in the analysis. The unsupervised hierarchical clustering analysis 1 showed clear grouping of hiPSC-aCM samples that were segregated relative to hiPSC-vCMs 2 ( Figure 2A). The gene expression profile of the hiPSC-vCM samples were more variable with 2 3 samples closer in distance to the hiPSC-aCMs while 3 samples displayed clear segregation 4 ( Figure 2A). The overall difference in global gene expression and lineage between hiPSC-aCMs 5 and -vCMs was also captured in the principal component analysis (PCA, Figure S4A). Out of the 6 250 transcripts analyzed, 200 genes were detected above background noise defined by a 7 threshold of 50 raw digital counts as determined by the negative controls of the assay. In the 8 hiPSC-aCMs, 14 and 27 genes were significantly upregulated and downregulated, respectively 9 ( Figure 2C). As expected, hiPSC-aCMs displayed significantly higher expression profiles of 10 atrial-specific markers including atrial-specific K + channel Kv1.5 (KCNA5) and transcription 11 factors (NR2F2 and TBX18) ( Figure 2C). Meanwhile, hiPSC-vCMs displayed higher expression 12 of ventricular-specific genes such as those encoding for contractile proteins MYL2, MYH7, and 13 the L-type Ca 2+ channel isoform Cav1.2 (CACNA1C) ( Figure 2C). The genes encoding for the 14 proteins in the sarcoplasmic reticulum complex such as TRDN, CASQ2, and RYR2 were 15 expressed in significantly lower amounts in the hiPSC-aCMs samples ( Figure 2C). Meanwhile, 16 pan-cardiac markers NKX2-5 and TNNT2 were expressed at similar levels in both hiPSC-aCMs 17 and -vCMs, further corroborating the efficacy of the differentiation protocol ( Figure S4B). 18

1
In this study, we were successful in efficiently differentiating hiPSCs into a monolayer of 2 cardiomyocytes with an atrial phenotype by modifying the GiWi protocol 11 . We used multiple 3 phenotypic approaches such as qPCR, digital multiplexed gene expression analysis with 4 NanoString technology, flow cytometry, ELISA, voltage measurements with current clamp 5 electrophysiology as well as simultaneous voltage and Ca 2+ transient measurements with optical 6 mapping to demonstrate a clear and distinct atrial phenotype. Unique to our study, we 7 completed an in-depth pharmacological analysis with simultaneous voltage and Ca 2+ 8 measurements to demonstrate the differential responses of these chamber-specific 9 cardiomyocytes, and their utility as a translational model in screening for the safety and efficacy 10 of novel atrial-specific compounds for the treatment of AF. 11 Our observations support previous data in showing that atrial specification is in part 12 mediated by RA 6, [8][9][10]15 . In our protocol, atrial differentiation was accomplished by adding 0.75 13 µM RA twenty-four hours after WNT inhibition, with a total exposure time of 72 hours. The 14 generated hiPSC-aCMs showed an atrial-specific phenotype as validated at both protein and 15 transcript levels with a decrease in ventricular-specific and an increase in atrial-specific 16 markers. These results suggest that RA, at the dose and temporal exposure used in this study, 17 maintains cardiac differentiation efficacy while pushing the differentiation process into an atrial 18

lineage. 19
As a complementary assay, we used the NanoString digital multiplexed gene expression 20 analysis to assess the expression of 250 genes custom-curated from the existing literature. We generating predominantly ventricular hiPSC-CMs. One study has shown a higher expression of 3 MLC-2a in hiPSC-aCMs analyzed at a later date (earliest at day 60) 6 . 4 Electrophysiological differences between atrial and ventricular cardiomyocytes, in terms 5 of voltage and Ca 2+ handling, define their function and are critical to the development and 6 determination of efficacy of atrial-specific compounds. As demonstrated by whole-cell patch 7 clamp and optical mapping measurements, the hiPSC-aCMs generated in this study exhibited 8 atrial-like AP and Ca 2+ handling properties. Namely, the AP of hiPSC-aCMs were significantly 9 shorter, along with a lack of a prolonged plateau phase as opposed to the AP of hiPSC-vCMs, 10 an observation that is aligned with native cardiomyocyte electrophysiology 28 . Similarly, the CaT 11 of hiPSC-aCMs had faster kinetics with a faster decay time as reflected by the differential 12 expression of Ca 2+ channel isoforms, further demonstrating the differential physiology between 13 hiPSC-aCMs and -vCMs. 14 In terms of APD measurements, we observed a good correlation between the patch 15 clamp and optical mapping recordings for hiPSC-aCMs. In hiPSC-vCMs, however, the optical 16 AP measurements were shorter overall than patch clamp recordings. This discrepancy may be 17 attributed to the heterogeneity of our current ventricular differentiation protocol which generated 18 predominantly ventricular cardiomyocytes but also contain a small proportion of non-ventricular 19 phenotypes (i.e. atrial myocytes and nodal cells). Thus, the optical AP signals in the rig used 20 represent an average from about 300,000 cells in each 1 cm 2 region of interest. 21 Another hallmark of cardiomyocyte function is rate-dependence, as described by the 22 electrical restitution curve 29 . We observed that the electrical restitution properties were different 23 between hiPSC-aCMs and -vCMs. Compared to hiPSC-vCMs, hiPSC-aCMs displayed a steady-24 state-like property by undergoing minimal APD80 shortening in response to the lower ranges of 25 the pacing protocol (cycle lengths of about 400 to 1000 ms) indicating full recovery of ion 26 channel kinetics at these pacing ranges. In contrast, the hiPSC-vCMs displayed consistent 27 APD80 shortening at the same pacing range. It is important to note that APD restitution curves 28 are likely different when using the standard steady-state extra stimulus protocol compared to 29 dynamic pacing, particularly in cardiomyocytes with immature Ca 2+ handling and memory 29 . In 30 relation to dynamic pacing protocol, hiPSC-vCMs exhibit a steeper maximum slope of the 31 restitution curve compared to hiPSC-aCMs as steady-state APD is the principal determinant of 1 the slope of the ventricular restitution curve 30 . 2 The presence of specific ion channel currents (i.e. IKur, IKAch, and ICaL) explain, in part, the 3 functional differences between the two cardiac chamber sub-types, the expressions of which 4 were already shown in our qPCR and NanoString assays. We used a series of compounds (4-5 aminopyridine, dofetilide, vernakalant, AVE0118, UCL1684, and nifedipine) to demonstrate the 6 function of atrial-specific ionic currents in our model system and were able to show the expected 7 chamber specific differences between hiPSC-aCMs and -vCMs. The compound 4AP has been shown to selectively block Kv1.4 (Ito) and Kv1.5 (IKur) 33 and 20 is therefore expected to elicit a response in hiPSC-aCMs at lower doses than in hiPSC-vCMs as 21 IKur (Kv1.5) is a strong functional indicator of atrial phenotype. Confirmation of the atrial 22 expression of IKur channels was demonstrated by the stronger dose-dependent hiPSC-aCM AP 23 prolongation to 4AP at all tested doses (10, 30, 50 and ,100 µM) suggesting selective sensitivity 24 of hiPSC-aCMs to 4AP due to a greater expression of Kv1.5. The inhibitory effects of 4AP were 25 observed at higher doses (50 and 100 µM) in hiPSC-vCMs which can be attributed to the 26 heterogenous population, potential off-target effects at these high doses, as well as baseline 27 expression of Kv1.4 (Ito). 28 Using nifedipine, we demonstrated the functional differences in Ca 2+ handling dynamics 29 between hiPSC-aCMs and -vCMs. Nifedipine elicited a dose-dependent response in hiPSC-30 vCMs demonstrating high sensitivity at 300 nM thereby confirming the functional presence of 31 Cav1.2. In contrast, hiPSC-aCMs were relatively insensitive to nifedipine showing no statistically 1 significant differences in APD at all tested doses. This finding is further corroborated by the 2 relatively decreased expression of CACNA1C (Cav1.2) in the hiPSC-aCMs. This suggests that 3 Ca 2+ handling in hiPSC-aCMs may be reliant on other voltage-gated Ca 2+ channels such as 4 Cav1.3, as this Ca 2+ channel is blocked less potently by nifedipine 34 . Moreover, our qPCR assay 5 confirmed that hiPSC-aCMs had higher expression of CACNA1D (Cav1.3). 6 AVE0118 is an experimental K + channel blocker (Ito, IKur, and IKr) that was predicted to 7 demonstrate targeted effects in hiPSC-aCMs. However, only a nuanced atrial specificity was 8 observed in our assay. Although the effects were proportionally larger in hiPSC-aCMs, 9 AVE0118 prolonged early repolarization of both hiPSC-aCMs and -vCMs in a similar fashion. 10 The drug prolonged mid-and late-repolarization at a lower dose (1 µM) in hiPSC-aCMs showing 11 minimal atrial specific effects. Interestingly, AVE0118 greatly affected Ca 2+ handling in hiPSC-12 aCMs compared to hiPSC-vCMs with larger proportional prolongation of CaTD50 at all doses. 13 These results were unexpected as AVE0118 is thought to be highly specific to hiPSC-aCMs due 14 to its IKur blocking component. Perhaps the observed mixed-effects in both cell types is due to 15 the drug binding to Ito (IC50: 3.4 µM) and IKr (IC50: 9.6 µM) 35 which prolongs APD at the tested 16 AVE0118 doses of 3 and 10 µM as genes encoding the channels producing the Ito (KCNA4) and 17 IKr (KCNH2) were expressed in our hiPSC-vCMs. The drug has also been shown to be effective 18 in terminating certain ventricular arrhythmias 36 which was predicted based on our results of 19 prolongation in the APD of hiPSC-vCMs. 20 Next, we used UCL1684, a highly specific SK channel pore blocker, to assess the 21 presence of functional SK channels in hiPSC-aCMs. The SK channel has 3 paralogs but the 22 SK3 channel variant (KCNN3) has been shown to be atrial-specific and has been implicated in 23 AF pathogenesis in several GWAS studies 37,38 . In this study, UCL1684 displayed high specificity 24 towards hiPSC-aCMs with a strong dose-dependent response. The drug confirmed the 25 presence of functional SK channels in hiPSC-aCMs at 3 µM with a positive dose-dependent 26 response while having no effect on hiPSC-vCMs at all tested doses (0.3, 1, 3, and 10 µM). This study has several limitations. One limitation in our findings is that we cannot directly 4 compare the results from qPCR and NanoString as both assays have fundamental differences 5 in technical principles and statistical methodologies. Taken together, however, both assays 6 show the global changes in cell type specific gene markers and further validate the role of 7 retinoic acid in directing the cardiac differentiation process towards an atrial lineage. The main 8 limitation in this field is the maturation state of the hiPSC-CMs as they have an overall immature 9 phenotype with some crucial differences compared to adult cardiomyocytes 40 . Nonetheless, we 10 were able to observe the stark differences in transcriptomic, protein, as well as functional 11 signatures of AP and CaT in the two generated chamber-specific cell types. Additionally, 12 maturation stage does not explain the differences in chamber-specific phenotype as parallel 13 batch differentiation and time-in-culture were incorporated in our study design. Most importantly, 14 we were able to capture effects of drugs that were expected to have atrial-specific properties in were added to the cells every 24 hours on days 4, 5, and 6 with media exchanged to RPMI1640 4 + B27 with insulin at day 7. Cells were harvested for analysis at day 20. B) qPCR analysis of 5 ventricular markers MYL2 and IRX4, cardiac marker NKX2.5, and atrial markers NPPA, GJA5, 6 CACNA1D, KCNA5, and KCNJ3. n = 3, unpaired t-test, *p < 0.05. C) Flow cytometric analysis of 7 cardiac troponin T (cTnT) and myosin light chain 2v (normalized to cTnT expression) in hiPSC-8 aCMs and -vCMs. n = 4, unpaired t-test, ***p<0.001. D) Average beating rates of hiPSC-aCMs 9 and -vCMs from the day they begin to beat until day 20. n = 4 independent differentiation 10 batches E) Atrial Natriuretic peptide (ANP) concentration between hiPSCs, and hiPSC-aCMs 11 and -vCMs determined by competitive ELISA. n = 3 and n = 2 hiPSC lines, unpaired t-test 12 *p<0.05, ** p<0.01, ***p<0.001. Data are presented as mean ± SEM and the n represents the 13 number of independent differentiation batches. 14  spontaneous (left) and/or paced at 1 Hz (right). C) Current clamp recording from a 31 spontaneously beating hiPSC-aCM. D) Single AP from hiPSC-aCM demonstrates shortened 1 action potential duration (APD) and lack of prolonged plateau phase, spontaneous (left), paced 2 at 1 Hz (right). E) The first differential of voltage recordings from hiPSC-aCMs and -vCMs were 3 used to calculate the maximal upstroke velocities. F) One minute after achieving the whole-cell 4 configuration, the average resting membrane potential was measured. G) Spontanous and 1 Hz 5 paced APs were assessed for duration at 50% of peak (APD50), and H) 90% of peak (APD90). 6 Statistics were performed by unpaired t-test. * p < 0.05, *** p <0.005. Data are presented as 7 mean ± SEM. Two differentiation batches were included in this analysis. 8