Animal procedures were approved by the Institutional Animal Care and Use Committee of the Children’s Research Institute and followed the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals. All animals were euthanized by exsanguination under anesthesia following heart excision, as previously described in detail (Jaimes et al., 2019a; Swift et al., 2019). Experimental data were acquired from isolated, intact hearts that were maintained on a temperature-controlled (37°C), constant-pressure (70 mmHg) Langendorff-perfusion system. To reduce motion artifact, Krebs-Henseleit perfusion buffer was supplemented with 10-15 μM (−/−) blebbistatin (Cayman Chemical) (Fedorov et al., 2007; Swift et al., 2012), or 10 mM 2, 3-butanedione monoxime (BDM; Sigma-Aldrich). Intact heart preparations were loaded with a calcium indicator dye (Rhod-2, AM: 100 μg rat, 100 μg guinea pig, 400 μg swine), which was added and allowed to stabilize for 10 min, followed by a potentiometric dye (RH237: 62 μg rat, 62 μg guinea pig, 248 μg swine). The epicardial surface was illuminated with filtered LED light (535 ± 25 nm), and emitted fluorescent light was split through a dichroic mirror (660 + nm) and directed through high-transmission filters to separate the V_m (710 nm long pass) and Ca²⁺ (585 ± 40 nm) signals. These simultaneous recordings were projected onto a single sensor with 2,016 × 2,016 pixel resolution and a pixel size of 11 × 11 μm (pco.dimax cs4), using an optosplit (Cairn Research) (Jaimes et al., 2019a; Swift et al., 2019). Adequate dye loading and signal detection was verified during the experiment using Fiji, an open-source image analysis software tool (Schindelin et al., 2012). Image stacks acquired from experimental data were saved in TIFF format (^∗.tif) for subsequent analysis.

Cryopreserved hiPSC-CMs (iCell cardiomyocytes; Cellular Dynamics International) were thawed and plated onto fibronectin-coated cover glass (∼60,000 cells/cm²), as previously described. Cells were maintained under standard cell culture conditions (37°C, 5% CO₂) in iCell cardiomyocyte maintenance medium (Cellular Dynamics International). Three days after plating, hiPSC-CMs formed a confluent monolayer, and were stained with either a potentiometric dye (FluoVolt; 1:1,000 dilution) or a calcium indicator dye (Fluo-4, AM; 10 μM) (Thermo Fisher Scientific) for 45 min at room temperature. Cell monolayers were then washed in dye-free Tyrode salt solution. Pace-induced action potentials and calcium transients were acquired at room temperature using a Nikon TiE microscope system, equipped with 470 nm excitation LED (SpectraX, Lumencor), 505–530 nm emission filter, and Photometrix 95B sCMOS back-illuminated camera. Cardiac monolayers were paced using field stimulation (monophasic, 10 ms pulse width, 1.5x threshold voltage, 0.3–0.7 Hz frequency).

KairoSight source code was written in the Python programming language (Python 3.8). Files related to the user interface elements and layout were generated in Qt Designer and converted to ^∗.py python script files, suitable for the PyQt library used to give interface elements their functionality. When available and appropriate, methods from the standard Python libraries were used for data handling and analysis. Steps were taken to minimize user input and error by taking advantage of existing optimized numerical libraries like NumPy, scikit-image, and SciPy, Python’s core scientific computing library. The software is freely available to download as source code from https://github.com/kairosight/kairosight-2.0, which can be run and edited in either the Spyder IDE or from the command prompt using the Anaconda distribution of Python. For convenience, an installation video, graphical user interface usage video, README file, and sample image stacks (spatially binned due to file size limitations) are provided on GitHub.

Kairosight processes optical data by first importing an image stack and then applying the following steps (Supplementary Figure 1): (1) Image Properties (e.g., acquisition speed, cropping), (2) Processing (e.g., masking, spatial and temporal filtering, detrending, and signal normalization), and (3) Analysis (e.g., isolate action potential or calcium transient signal, compute relevant timepoints and durations, export results as pixel-wise values or pseudo-color maps). Data can be exported after the Analysis step as either pixel signals (^∗.csv), or optical maps (^∗.png).

Images and text were exported through the user interface. A single frame or entire image stack was exported as a ^∗.tif and a single frame or any generated map was exported as a ^∗.png. Further, generated maps or corresponding temporal, spatial, or statistical results were exported as ^∗.csv. Signal data was exported to obtain an (x,y) plot of action potential or calcium transient values. An array of fluorescence data was exported for each ROI over time, including intensity signals, analysis results (all values or mean ± SD) or calculated/mapped values as ^∗.csv.

Results were reported as mean ± standard deviation. Differences between group means were determined using two-tailed Student’s t-test or analysis of variance (GraphPad Prism). Significance was defined as a p-value < 0.05, or an adjusted p-value (q < 0.1) after correction for multiple comparisons using two stage linear step-up procedure to control the false discovery rate (0.1), as described by Benjamini et al. (2006). Significance was denoted in the figures with an asterisk (^∗). Replicates were defined in each figure legend; independent optical signals were used from the same heart preparation (Figures 2A–C, 3A–C signal-to-noise data), and independent heart preparations were used for other measurements (Figures 4–8).

To provide a more accessible tool for optical mapping data analysis, we developed a novel, open-source software package with an intuitive user interface (Supplementary Figure 1) that directs users through a linear workflow (example shown in Figure 1). Image processing was accomplished by importing an image stack and applying the following steps: (1) Image Properties, (2) Processing, and (3) Analysis. The KairoSight interface used a workflow with very few input options, which minimized user error, maximized computational resources, and enabled new users to easily understand the methods to implement data analysis steps with accuracy. To illustrate the Image Properties step, cropping and masking were applied to an experimental dataset in Figure 1.

Multiparametric optical signals were acquired from Langendorff-perfused rat, guinea pig, and swine hearts, and hiPSC-CM, which were stained with fluorescent dyes. Using the cropping feature, the user isolated areas of interest and specified the type of signal (voltage—inverted signal, or calcium—upright signal) from the images loaded onto KairoSight. Activation maps, action potential, or calcium transient duration maps (Figures 2–9), as well as statistical results (e.g., action potential and calcium transient duration, amplitude, activation time, transient decay tau) were generated. The software identified and analyzed optical voltage and calcium signals from multiple species (Figures 4, 5, 7), with variations in signal shape and pacing frequency.

To confirm the consistency of the signal analysis software in the event of imperfect signals, experimental data was analyzed with varying degrees of noise. Optical signals were collected from hearts loaded with a voltage sensitive dye (RH237) and a calcium dye (Rhod-2, AM). The signal-to-noise ratio (SNR; calculated as the signal amplitude divided by the standard deviation of the baseline during the diastolic interval) was decreased by reducing LED illumination on the surface of the heart. Reducing LED illumination decreased voltage signal SNR (100% light: SNR 118.1 ± 19.1, 85% light: 104.9 ± 10.6, 70% light: 85.0 ± 4.5) and calcium signal SNR (100% light: SNR 118.9 ± 19.2, 85% light: 104.0 ± 9.8, 70% light: 82.3 ± 6.2). Kairosight was used to analyze corresponding voltage (Figures 2A–C) and calcium signals (Figures 3A–C) within the same heart, wherein only the degree of LED illumination was altered. Activation times for voltage (100% light: 17.4 ± 1.2, 85% light: 16.8 ± 1, 70% light: 16.3 ± 1.5 ms) and calcium (100% light: 18.2 ± 1.8, 85% light: 17.9 ± 1.9, 70% light: 18.1 ± 1.1 ms) were calculated from the middle of the epicardial surface, along with action potential duration time at 80% repolarization (APD₈₀ 100% light: 57.1 ± 1.9, 85% light: 53.8 ± 2.3, 70% light: 54.8 ± 0.3 ms) and calcium transient duration at 80% reuptake (CaD₈₀: 100% light: 90 ± 2.7, 85% light: 88.3 ± 1.3, 70% light: 91.7 ± 3.4 ms); values were computed from n = 6 optical signals within the same heart. Experiments were then repeated (n = 4 independent heart preparations) under varying degrees of illumination to alter the SNR (Figures 2D, 3D). We noted deviations in the activation time for the calcium signal (“calcium release”) when SNR < 50, but calcium transient duration measurements remained consistent.

To test the flexibility of the software in analyzing various signal morphologies, tissues from four source species were analyzed. Transmembrane voltage signals were collected from the hearts of rats, guinea pigs, and swine, as well as cultured hiPSC-CM layers. In comparing the morphology of the electrical and calcium traces between species, varying differences in the optical signals were detected (Figures 4, 5). Rat action potential and calcium transient optical signals aligned with examples in the literature (Walton et al., 2013; Jaimes et al., 2016b; Zasadny et al., 2020). The corresponding electrical restitution curve showed little change when the dynamic pacing cycle length (PCL) was decremented from 200 to 100 ms (Figure 4A), however, the CaD₈₀ shortened by 18.8% at the faster pacing rate (Figure 5A). Guinea pig signals and restitution values mirrored those reported in the literature (Guo et al., 2008; Lee et al., 2012), as well as those modeled by Edwards and Louch (2017). As expected, duration times were longer in the guinea pig heart, compared to the rat. Specifically, the APD₈₀ and CaD₈₀ shortened by approximately 15.2 and 17.4%, respectively, as the PCL was decremented from 200 to 140 ms (Figures 4B, 5B). Likewise, swine APD₈₀ and CaD₈₀ exhibited even slower dynamics, in agreement with previous studies (Bourgeois et al., 2011; Walton et al., 2013; Lee et al., 2017). Decrementing the PCL from 350 to 200 ms shortened the APD₈₀ by 25.5% (Figure 4C) and the CaD₈₀ by 30.3% (Figure 5C). Although not directly comparable to an intact heart preparation, hiPSC-CM were also imaged and analyzed as proof of concept. Decrementing the PCL from 3,330 to 1,430 ms shortened the APD₈₀ by 16.1% (Figure 4D) and CaD₈₀ by 27.6% (Figure 5D).

Action potential and calcium transient activation maps were generated for rat, guinea pig, and swine hearts during both sinus rhythm and in response to dynamic epicardial pacing (Figure 6). As expected, activation time measurements were faster during sinus rhythm (often occurring in the right ventricle before the left ventricle) and slower in response to external pacing. In the rat heart, activation time was 8.9 ± 1.9 ms (V_m) and 10.2 ± 2.1 ms (Ca²⁺) during sinus rhythm (Figure 6A), which increased to 27.7 ± 3.5 ms (V_m) and 30.4 ± 5.2 ms (Ca²⁺) with pacing (200 ms PCL; Figures 6B,C). Similar observations were noted in larger species; in the guinea pig heart activation time was 18.8 ± 5 ms (V_m) and 19.9 ± 6.3 ms (Ca²⁺) during sinus rhythm (Figure 6A), which increased to 29.4 ± 3.4 ms (V_m) and 31.9 ± 4.8 ms (Ca²⁺) with pacing (200 ms PCL, Figures 6B,C). Piglet heart activation time was 32.7 ± 7.3 ms (V_m) and 37.8 ± 10.2 ms (Ca²⁺) during sinus rhythm (Figure 6A), which increased to 67.0 ± 10.5 (V_m) and 77.4 ± 12.2 ms (Ca²⁺) with pacing (350 ms PCL, Figures 6B,C).

Calcium flux plays a central role in excitation-contraction coupling, wherein perturbations in calcium handling (e.g., alternans, calcium leak from the sarcoplasmic reticulum) are associated with an array of cardiac dysfunctions, both electrical and mechanical (Weiss et al., 2011; Wang et al., 2014; Kanaporis and Blatter, 2015; Johnson et al., 2019; Sutanto et al., 2020). Calcium alternans are a potentially deleterious physiological oscillation, which are observable as alternating amplitude fluctuations in optical signals. The latter has been mechanistically linked to T-wave alternans that can segue into lethal ventricular arrhythmias (Weiss et al., 2011; Ng, 2017). Experimentally, cardiac alternans have been more commonly observed at faster heart rates and/or cooler temperatures (Egorov et al., 2012; Millet et al., 2021); accordingly, we utilized burst pacing to induce alternans in isolated, intact hearts (Figures 7A–C). Although not observed at slower pacing rates (200 ms PCL), intact guinea pig hearts displayed both APD alternans and calcium transient amplitude alternans at faster rates (120 ms PCL, Figures 7A,B). Further, calcium transient duration maps revealed spatially discordant alternans at faster rates, in which two regions of the tissue (right vs. left ventricle) were out of phase. When considering the occurrence of alternans between species, we observed variable susceptibility with increasing pacing frequency (relative occurrence at 100 ms PCL: 14% in mice, 42% in rats, and 57% in guinea pigs; Figure 7C).

Pharmacological agents are frequently employed to identify the mechanisms responsible for cardiac responses. As one example, isoproterenol is commonly used to induce heart failure through its positive chronotropic effects (Balakumar et al., 2007; Greer-Short and Poelzing, 2015; Chang et al., 2018). Using Langendorff-perfused rat hearts, we demonstrated that acute isoproterenol exposure (30 nM) results in a 40% increase in sinus rate (280.3 ± 32.4 baseline vs. 392.0 ± 37.7 BPM with isoproterenol), and 17.7% shortening of CaD₈₀ with dynamic pacing at 120 ms PCL (89.9 ± 5.9 baseline vs. 74 ± 14.4 ms with isoproterenol; Figures 8A,B). Further, ryanodine (5 μM) modulates calcium release from the sarcoplasmic reticulum, and is often used to demonstrate signal specificity in dual optical mapping studies (Choi and Salama, 2000; Jaimes et al., 2019a). Using KairoSight, we demonstrated a marked 92% reduction in the calcium transient amplitude with minimal effects on optical action potentials collected from intact rat hearts (Figure 8C). Optical mapping studies are often used to monitor spiral wave formation and re-entry arrhythmias (Moe and Jalife, 1977; Pumir et al., 2005), which may occur due to tissue heterogeneities from a myocardial infarction or incomplete cardiac ablation. To demonstrate the utility of our optical mapping approach and data analysis, an epicardial radiofrequency ablation was applied to mimic the physiological environment that can promote such arrhythmias (Mercader et al., 2012). Optical signals were collected before and after radiofrequency ablation, and the analyzed signals show a typical pattern of electrical activity and calcium fluctuations outside the ablated area (Figure 9). Within the ablated area, we observed irregular and diminished signals from both the voltage and calcium signals. Accordingly, optical signal analysis via KairoSight can be applied to other applications focused on pharmacological treatment, tissue injury, or pathophysiology.