CalTrack

Supplemental Digital Content is available in the text.


INTRODUCTION
Calcium is a universal signaling molecule that can evoke extensive changes in the proteome and transcriptome of all cells and has particular roles in excitable neural cells and cardiomyocytes. In addition to cell signaling, calcium transients in cardiomyocytes can directly influence contraction, relaxation, and arrhythmogenicity. As such, dynamic calcium measurements are key to understanding cardiomyocyte physiology, pathology, and the therapeutic efficacy and safety in delivered compounds 1 .
Investigations into cardiovascular biology employ fluorescent indicators of cellular calcium, including cell permeable fluorophores 2,3 , and viral delivery of genetically-encoded calcium sensors 4, 5 that may target sub-cellular structures such as sarcomeres 6,7 . Despite the successful application of these strategies to primary cells derived from experimental models or explanted human tissues 8 , there remains an unmet need for higher throughput calcium screening assays, in particular to address human immortalized embryonic stem cell and induced pluripotent stem cell (iPSC)-derived cardiomyocytes. With this goal, we developed CalTrack, a set of MatLab algorithms with the flexibility to analyze cellular calcium transients from many cellular formats in an un-biased, rapid, analysis platform. CalTrack uses high-throughput fluorescent cellular imaging to enable analyses of wildtype 9 and mutant iPSC derived cardiomyocytes (iPSC-CMs) derived from patients or mutated using clustered regularly interspaced short palindromic repeats (CRISPR)/Cas-9 methodologies 10 .
Phenotyping of iPSC-CMs has recently become easier with techniques that allow rapid automated assessments of contractile function 11,12 and that address challenges of cellular heterogeneity by increasing experimental throughput, as greater data volume enables statistical power to interrogate small effects. While the development of more physiologically relevant models remains an important endeavor to overcome cellular immaturity, rapid phenotyping of iPSC-CMs can be performed with CalTrack, identifying subtle changes that are critical to understanding pathogenic signals. Existing strategies with these capabilities typically require expensive hardware and skilled manual interventions (See Online Table I). Additionally, to allow automated, high throughput assessment of calcium transient changes for therapeutic screening, we aimed for efficient data acquisition and analyses. CalTrack is freely available using a MatLab pipeline, or provided as a compiled version for users without MatLab access.
CalTrack defines calcium transients in primary cardiomyocytes from animal models 6 and iPSC-CMs, including cells on patterned substrates 13,14 or in engineered three-dimensional tissue systems 15 (Figure 1 and Online Figure I & II). The algorithm automates background subtraction, performs photobleach correction, uses masks to identify individual cells in image stacks with multiple cells per field of view, averages all identified transients, and performs transient fitting with parametric output. This code relies only upon providing a directory of video files (.avi, .vsi, .mov, .czi, .m4v, .mp4 and others) that can be converted to .tif stacks, or previously extracted fluorescence data in an excel file (Figure 2). CalTrack is user friendly (See Online User Manual) and computationally inexpensive, so that it can be run on a local computer. On average, the algorithm requires 140 seconds to convert 10 video files (250 frames) to .tif stacks and 55 seconds to analyze and export computed parameters into excel files. In comparison, manual conversion of similar videos (120 seconds), extraction of calcium traces (200 seconds), and computation of the parameters is considerably slower. In addition, manual computation of average representative traces, with baseline correction can impair standardization. We show that CalTrack's rapid automated analysis and fitting of hundreds of cells on a local computer provide robust statistical power for analyzing changes in cellular calcium transients in comparison to a commercial software. As CalTrack only requires fluorescence microscopy, or extracted traces, to rapidly analyze large data sets, we suggest that this platform provides a critical resource for the cardiac biology community to accelerate experimentation in physiology, pathology, and therapeutic screening.

Data Availability.
The CalTrack code and trial data sets are openly available and maintained on GitHub at the following URL: https://github.com/ToepferLab/CalTrack.

Mathematical basis of CalTrack calcium transient analysis.
To extract calcium traces, CalTrack relies on the Bio-Formats toolbox 16 for MatLab (Mathworks Inc. Natwick, MA, USA). CalTrack initially reads all frames in the video stack and averages the value of all pixels to generate a mask. Where multiple cell segmentation is required the average image is subjected to a difference of Gaussians filter 17 to increase contrast in local boundaries. Contrast-limited adaptive histogram equalization, and erosion-dilation enhance this to produce a segmented cell mask. This is applied to extract the raw fluorescence within each cell per frame, as the sum of all pixels in each cell (n): where represents each pixel intensity.
Below we provide a simplified mathematical version (see Online Supplement for additional details) of the process of extraction and analysis of traces that have successfully passed quality control and have either been corrected for photobleach or have been deemed by the user not to require photobleach correction. Previously extracted traces can also be processed with CalTrack.
First, to select the beginning of each fluorescent transient and perform transient segmentation, the entire trace is converted to an array of the difference of each intensity value with its previous value: The onsets of fluorescence increase, which correspond to the start of calcium transients, match the peak values of D. Peaks in D (termed D p ) are selected as onsets of calcium-driven fluorescent transients by thresholding to a minimum prominence of 50%. The beginning of each transient is selected by subtracting an offset of 10% of the cycle length to D p (e.g. 100ms for 1 Hz pacing), in the original trace. Individual transients are cropped from the whole fluorescent trace by specifying their beginning and ending points at D p -offset and D p+1 -offset respectively. Next, the transients are averaged to a single fluorescence intensity transient unless the option of measuring parameters on each individual transient is selected by the user, in which case the following is performed on every transient. When measuring parameters on each individual transient, a measurement of the standard deviation of all individual transients' parameters is provided as a quality control metric.
As an alternative to automatic transient segmentation, CalTrack offers the user the option to perform this process by defining the beginning of the first calcium transient via event markers such as the time of electrical pacing, if known.
To characterize all temporal parameters of the obtained average transient, the baseline intensity must be determined. This is achieved by averaging the last points of the trace that correspond to a temporal window of 20% of the cycle length (e.g. 200ms for 1 Hz pacing). Subsequently, the peak of the trace is selected as the maximum intensity value and the transient magnitude is calculated as max -baseline. F max /F 0 is calculated as the peak value divided by the baseline. Next, since fluorescent traces may have a lower baseline at the end of the trace than at the beginning, a more robust value for the baseline is defined as baseline + (0.03 x magnitude). This enables automated definition of the beginning and end of each transient in the fluorescent trace, and is the smallest change that can support automation, while minimally altering the measured temporal parameters (Online Figure III). The redefined baseline value is used to calculate all temporal parameters via linear interpolation (i.e. interpolation of time at intensity values corresponding to 0%, 10%, 50% and 90% of the magnitude, which has been re-calculated with the re-defined baseline), with the exception of tau (and its fitting parameters), which is calculated by fitting the decaying arm of the fluorescent trace with an exponential decay curve with equation , and calculating tau as . The goodness-of-fit value is also calculated and reported for quality control. For traces without uniform electrical pacing (i.e. more fluorescent transients are identified than the manually-input pacing frequency justifies) or traces that display aberrations reminiscent of irregular behaviour, such as early or delayed after-depolarizations (Online Figure IV), additional information is reported as output. This includes the beat-to-beat time as the average time interval between peaks, the number of intervals used to compute the beat-to-beat time, and the cell and beat number where early or delayed after-depolarizations occurred. These traces may later undergo post hoc processing. Here an automated analysis of the regular events is carried out (i.e. excluding the irregular parts of the trace) and reported measurements include all of the above in addition to the number of total and irregular beats, as well as classifying irregular beats as an early or delayed after-depolarization.
Finally, for quality control, the signal to noise ratio is calculated for the extracted traces as the average of the transient values (above baseline) divided by the standard deviation of values below the baseline.

Production of synthetic calcium transients for CalTrack benchmarking.
Synthetic calcium transients were simulated using a biophysically detailed electro-mechanical model 18 of adult human ventricular myocyte obtained through the coupling of the electrophysiology ToR-ORd 19 and the contractility Land 20 models. Simulations were conducted in MatLab, using the numerical solver ode15s with solutions reported every 1ms, corresponding to an acquisition frequency of 1000 frames per second (FPS). The model was stimulated at 0.5, 1, and 2 Hz for 200 beats, delivering a stimulus current of -53 A/ F for 1ms. The last 5 beats of each simulation were saved. For each pacing frequency, 20 5-beat traces were generated by scaling either the L-type calcium or the rapid delayed rectifier potassium currents' conductance. Calcium traces were down sampled to match experimental acquisition frequencies. Noise equivalent to 2.5 times the standard deviation of the final 300ms of each calcium trace was added to each trace. Resemblance to commonly acquired experimental data was confirmed on all traces visually before analysing with CalTrack.

Generation of WT and TNNI3 R21C/+ missense iPSC-CMs for calcium transient analyses.
A heterozygous pathogenic missense variant TNNI3 R21C that causes hypertrophic cardiomyopathy (HCM) 21 was introduced using CRISPR/Cas9 technology as previously described and in the Supplemental Methods 10, 22-24 (Please see the Major Resources Table in the Supplemental Materials). Targeted iPSC subclones were sequenced to confirm the TNNI3 R21C/+ genotype (See Online Figure V), differentiated into iPSC-CMs via Wnt pathway modulation, and plated in a 2D assay plate as described in detail in the Supplemental Methods and identically as previously described 25 . Cells were treated with dimethylsulfoxide (DMSO) or Mavacamten 0.3-3 µM as described in detail in the Supplemental Methods.

Guinea Pig cardiomyocyte isolation and manipulation.
Guinea pig studies were performed with protocols that were reviewed and approved by the Animal Welfare and Ethical Review Board at the University of Oxford and conform to the UK Animals (Scientific Procedures) Act, 1986. Adult left ventricular cardiomyocytes (adolescent male, 10-15 weeks in age) were isolated and immediately processed for ratiometric calcium analyses as previously described 26 (see Supplemental Methods and please see the Major Resources Table in the Supplemental Materials.). Paced Fura2-loaded cells were studied using the Ionoptix (Waltham, MA) platform IonWizard, according to manufacturer's protocols, and by CalTrack (see Results). Alternatively, isolated cardiomyocytes were cotransduced for 48 hours with adenovirus carrying a red genetically encoded calcium indicator (RGECO) 6 and a cDNA encoding either WT TnnI3 or TnnI3 R145G, and cultured in plates for imaging 6 . Videos of 0.5 Hz electrically paced cardiomyocytes at 37°C were acquired at 25 FPS. Pharmacologic effects were assessed in guinea pig cardiomyocytes that has been pre-treated with either DMSO or 10 µM levosimendan for 15 minutes at 37°C prior to imaging.

Statistical analysis.
All individuals performing data analysis were blinded to the treatment group under study. Post-analysis processing was unblinded. Statistical analyses were performed using GraphPad Prism version 8.4.2 for macOS (GraphPad Software, San Diego California USA, www.graphpad.com). Normality in all data sets, defined as alpha < 0.05, was assessed by the D'Agostino-Pearson normality test. Single comparisons of data modelled by a normal distribution was assessed by a double tailed Student's t test; multiple comparisons use a one-way ANOVA with post-hoc correction for the number of comparisons. Data that was not modelled by a normal distribution, was assessed by a non-parametric Mann-Whitney test. When multiple comparisons were tested in un-paired data a Kruskal-Wallis was used with post-hoc Dunn's correction.  Figure VII) where the normality test was not satisfied, data was tested with a non-parametric Friedman's test with Dunn's multiple comparisons correction. In all instances a significance cut-off of p < 0.05 was used.

CalTrack pipeline overview and applications.
CalTrack analyzes image stacks/videos acquired with fluorescent calcium reporters in iPSC-CMs or isolated adult cardiomyocytes ( Figure 1). Alternatively, extracted traces (collected in excel files) that bypass the need to derive data from stacks/videos can also be analyzed. After incubation with chemical calcium dyes such as Fura-2, Fluo-4, 2 or following transduction with viruses carrying calcium probes (e.g., RGECOs/GCaMPs 27 ), cells are imaged using a fluorescent microscope at a range of magnifications (typically 100X-10X) to provide image stacks of either single or multiple cells per field of view ( Figure  2). Image acquisition rates should at minimum be 25 FPS, but ideally should be 40-100 FPS to allow the accurate capture of calcium transients. However, data acquired at up to 1000 data points per second can be analyzed (Online Figure VI). CalTrack applies a mask to these image stacks to identify either single or multiple cells in a field of view (Online Figure I). In the case of imaging single iPSC-CMs at 100X or equivalent optical magnification, CalTrack identifies the area that the cell occupies in the field of view and applies a fixed mask to the cell to measure fluorophore intensity in each frame of the acquisition ( Figure  2A-B). Fluorophore intensity is therefore measured within the cell's boundary at maximum relaxation throughout each frame of the acquisition ( Figure 2D), which is used to determine the parameters that define the calcium transient. These include the decay constant tau (Tau), determined by curve fitting to the mean transient, time to peak (T on ), time to 90% of peak (T90 on ), time to 50% of peak (T50 on ), time to 10% of peak (T10 on ), peak fluorescence over baseline fluorescence (F max /F 0 ), time for calcium transient decay (T off ), 10% decay (T10 off ), 50% decay (T50 off ), 90% decay (T90 off ), calcium transient duration (CD), and signal to noise ratio for the calcium trace. Examples of the CalTrack fitting and parametric determination are illustrated in Figure 2E and in the CalTrack user guide.

CalTrack robustly detects cardiomyocytes in low magnification image stacks.
Imaging calcium transients in multiple cells within a single field of view at low magnification can improve throughput. To enable the analysis of this data, CalTrack constructs a cell mask to identify individual cells, and excludes rounded dead cells by filtering out non-elongated cells. This strategy yielded a 1.36% false negative rate and 4.

CalTrack corrects for linear and exponential baseline drift in calcium transient acquisitions.
Baseline drift in fluorescent microscopy with calcium indicators can be caused by multiple events including photobleaching, loss of fluorophore from the cell, or cell damage. CalTrack enables the investigator to select whether elimination from processing or correction of baseline drift is desirable. To correct baseline drift, a second-degree polynomial, applicable for both linear and exponential drifts, is used to determine the baseline trend; this is subsequently subtracted from the original trace to detrend the data  Figure 3C). The same applies to an exponential decay in baseline, which can be successfully modelled and removed from datasets with this approach (Figure 3D-E).

CalTrack provides analysis of irregular calcium transients including early after-depolarizations (EADs) and delayed after depolarizations (DADs).
To account for irregular fluorescent transients, which cannot be analyzed as part of the main automated analysis, CalTrack has additional post hoc analysis capabilities. These analyses require high signal to noise ratio in the extracted traces to allow robust detection of irregular transients. CalTrack classifies irregular traces as non-adherence to pacing and/or calcium spikes reminiscent of early and delayed afterdepolarizations (Online Figure IV). CalTrack can exclude the irregular events from the multi-beat trace and measure parameters from the regular transients as it would usually in adherent transients ( Figure  2). The irregular transients are subsequently analyzed and CalTrack provides a count of the number of total transients, the number of excluded events, and the number of irregular events, including designation of events as early or delayed after-depolarizations.

Assessment of quantitative calcium concentrations using CalTrack.
We harnessed CalTrack to quantify intracellular calcium using the ratiometric indicator Fura2 in WT adult guinea pig cardiomyocytes (Online Figure VI). When running CalTrack the user can either provide a calibration curve for the ratiometric indicator or alternatively, a calibration curve is generated within CalTrack by inputting the fluorescence intensity values at known calcium concentrations for a specific ratiometric indicator. This calibration is then used with the fluorescence ratio of the ratiometric indicator to generate absolute raw and smoothed calcium concentration traces, demarcating T 0 , providing profiles with temporal parameters from baseline and peak calcium concentrations, and quantifying the baseline and peak concentration of intracellular calcium (Online Figure VI).

Benchmarking of CalTrack with synthetic simulated calcium transient data.
CalTrack has been designed to function for multiple pacing frequencies, allowing the user to define the experimental pacing frequency. To test the fidelity of the underlying CalTrack code, we used simulated calcium transients obtained from 20 virtually generated human cardiomyocytes and several pacing frequencies from 0.5 Hz to 2 Hz (Figure 4). Variation of electrical pacing frequency affects simulated calcium transient parameters as expected (Figure 4 A-I). When using CalTrack to analyze these data, the mean over 20 simulated calcium transients shows marked changes in calcium transient characteristics with pacing frequency (Figure 4B). Normalized calcium peak (equivalent to F max /F 0 in experiments) decreases with increasing pacing frequency ( Figure 4C). T on and T off are faster at higher pacing frequencies ( Figure  4D-E). The calcium transient decay constant, tau, decreased with increasing pacing frequencies ( Figure  4F). T50 on , T50 off and CD were accelerated as pacing frequencies increased (Figure 4G-I).
To determine the interference of noise with measurements, random noise was added to simulated calcium traces at 1 Hz pacing (Online Figure VII). Measurements of peak normalized calcium, tau, time to 50%, peak calcium, and time to 50% and complete decay were unchanged at signal to noise ratio 15-70. However, the time to 50% peak calcium was increased (Online Figure VII H) and time to complete decay was decreased (Online Figure VII G) at low signal to noise ratio of 10. Therefore data acquired below signal to noise ratio of 15 may be unreliable and should be closely scrutinized by the end user.

Quantitative calcium analyses in adult TnnI3 R145G/+ cardiomyocytes with benchmarking comparison between automated CalTrack analysis and manual, user-driven IonWizard analysis.
We compared ratiometric calcium data from adult guinea pig cardiomyocytes, which had either been transfected with human TnnI3 +/+ 9 or with the R145G variant troponin I (TnnI3 R145G/+ ) 6, 28 , which had been loaded with Fura2 ( Figure 5). Both algorithms detected significantly increased baseline and peak calcium concentrations without altering calcium amplitudes (Figure 5 A-C ). The times to 10%, 50% and 90% of peak calcium were not significantly different, although CalTrack detected a modest increase in the time to peak calcium ( Figure 5D-G). Both software detected abnormalities in calcium decay times, although not identically and prolonged tau in the TnnI3 R145G/+ vs. WT cardiomyocytes ( Figure 5H-K). While these analyses demonstrated comparable data acquisition and interpretation by both software platforms, the overall operator times for using IonWizard was 20-fold longer on average per cell analyzed (CalTrack ~1 minute, IonWizard ~20 minutes).

CalTrack detects calcium transient responses to pharmacological agents, levosimendan and isoproterenol, in adult cardiomyocytes.
We studied WT adult guinea pig cardiomyocytes at baseline (n = 79) and after treatment (n = 96) with 10 M of the calcium sensitizer levosimendan 29 . Normalized mean calcium transients from 0.5 Hz paced cells ( Figure 6A) and calcium decays ( Figure 6B) that were obtained after levosimendan treatment increased F max /F 0 (p = 0.0022) (Figure 6C), and reduced tau (p = 1.9 x 10 -14 ) ( Figure 6D). T50 on was not statistically significantly altered by levosimendan ( Figure 6E), but T on was shortened (p = 0.026) ( Figure  6F). Levosimendan accelerated T50 off (p = 2.4x10 -9 ; Figure 6G) and T off (p = 0.023; Figure 6H) and reduced CD (p = 0.0034; Figure 6I). Drug-induced changes in fluorescence identified by CalTrack are consistent with prior studies in guinea pig cardiomyocytes 6 and pharmacologic activities of levosimendan. Levosimendan binds troponin C and increases calcium affinity, which underlies its positive inotropic effects, 30,31 and also potently inhibits phosphodiesterase activity 32 , thereby augmenting protein kinase A (PKA) and cAMP effects that increase cellular relaxation rates 33 and promote lusitropy.
CalTrack also detected the effects of the β-adrenergic agonist isoproterenol on 1 Hz paced WT iPSC-CMs (Online Figure VIII). In comparison to untreated cells, isoproterenol increased calcium kinetics, showing both a more rapid increase in normalized calcium fluorescence and significantly reduced tau, T on , and T50 off , consistent with previously reported data 34 .

CalTrack detects basal and drug-induced calcium changes in iPSC-CMs with an endogenous TNNI3 R21C/+ HCM variant.
We used CalTrack to assess the calcium transients from isogenic iPSC-CMs with and without an endogenous heterozygous pathogenic troponin I variant (TNNI3 R21C/+ ) that causes HCM 21 (see Supplemental Methods). A founder TNNI3 R21C/+ mutation in South Lebanon causes malignant HCM with sudden cardiac death, which often precedes hypertrophy 35 , likely due to abnormal calcium handling from disruption of PKA-mediated phosphorylation of its N-terminal molecular switch. 21,36 We studied 1 Hz paced cells treated with and without mavacamten 37 an allosteric myosin ATPase inhibitor 38 , which improves relaxation deficits associated with HCM thick filament variants [39][40][41][42] and also influences calcium via unknown mechanisms. 6 Using CalTrack outputs for the mean calcium transient in each cell analyzed (excel file), we defined an average transient at baseline and for each perturbation ( Figure  8). Mutant iPSC-CMs had increased T 50off , T off , and tau at baseline, which is consistent with previous work in a knock-in mouse model carrying the R21C variant 43 that exhibits marked diastolic insufficiency and slowed myofilament relaxation.

DISCUSSION
We demonstrate that CalTrack, a high-throughput automated analysis pipeline that capitalizes on advanced imaging capacities and acquisition of high frame rates by fluorescent microscopy, provides robust assessment of calcium transient in cultured or patterned cardiomyocytes, and cardiomyocytes within engineered tissues. This extends to quantitative ratiometric analysis of calcium concentrations within cells when data is acquired at high temporal resolution (>1000 data points per second). Harnessing rapid screening technologies to visualize and image multiple cells in single fields of view, CalTrack acquires and automatedly and rapidly analyzes data in high throughput.
Quantitative assessment of calcium transients by CalTrack can be used in computational investigations into the pathogenic mechanisms underlying cardiovascular conditions and their modulation under therapeutic interventions 44,45 . Computational modelling and simulation of cardiac pathophysiology can then be used to augment experimental datasets provided by CalTrack to enable a mechanism-driven understanding of cellular phenotypes 44,45 , thereby providing a platform for testing hypotheses in human cardiomyocytes as alternatives to animal models 46 so as to characterize physiologic and pathologic processes at baseline and in response to interventions.
Contemporary open source software for analyzing calcium transients focus on neuronal cell types [47][48][49][50][51] and characterize irregular traces generated by neurons. While these platforms have some utility in cardiomyocytes, they are cumbersome for use in this system, and cannot run automatedly. These issues limit high volume acquisition of cellular calcium transients and require significantly longer analysis times. Some laboratories overcome these issue through custom scripts, which typically lack automation, have variable parameter definitions, or limited parameter extraction precluding cross-study comparisons. [52][53][54][55][56] By contrast, software automation eliminates manual user input and associated user error/bias, so as to provide calcium transient data in a defined, standardized way that can be compared between cells and experimental conditions. While CalTrack is easily deployable for analysis across a variety of cardiac cells within a range of experimental systems, the platform has limitations. Analytical accuracy is compromised at low acquisition rates because of the rapid kinetics of the calcium transients generated in cardiomyocytes. For example, acquisition at 25 FPS will sample data every 40 ms, resulting in a theoretical temporal resolution of 92 ms (based on the Nyquist theorem, whereby resolution is equal to 1/2.3 of the sampling frequency). Hence the starting point of the calcium transient may be resolved as much as 40 ms further from whence it commences. This effect increases at lower sampling rates and is independent of the software used for analysis. Thus, although CalTrack uses interpolation to measure parameters smaller than the acquisition window, the acquisition frequency dictates accuracy, especially in parameters related to the start and fast phase of the transient. We suggest that this factor underlies the measurement difference of T50 on in WT human iPSC-CMs (~50 ms, acquired at 50 FPS) compared with WT guinea pig cardiomyocytes (~100 ms, acquired at 25 FPS). Nevertheless, CalTrack enables comparative analyses for lower frame rate data, where our findings using CalTrack are complimentary to previous findings generated using the same experimental data set that had been analyzed with manual user-driven analysis and previously reported 6 . Additionally, we note that individual transient registration may be perturbed with extremely high frame rate data, as oversampling reduces the prominence of point-to-point differences, from which the start of a calcium transient is defined. However CalTrack data acquired at 1000 FPS is down-sampled to 100 FPS solely to identify the start of the calcium transients; the raw 1000 FPS data is then used for parametric analyses. Therefore, high frequency data acquired by line scanning or by the use of a photo multiplier tube can also be robustly and rapidly analyzed by CalTrack. As with many platforms, high data noise may limit the accuracy of the algorithm in fitting key parameters, but this is largely overcome by the trace averaging used in CalTrack. Indeed, almost identical analysis fidelity of values were obtained with CalTrack and the IonOptix IonWizard software ( Figure 5). When needed, CalTrack also provides the end user the ability to check fitting fidelity for all steps of the analysis process, so that data quality control standards can be identified. Notably, CalTrack can also be used to automatedly analyze data obtained using a ratiometric calcium indicator such as Fura 2.
We demonstrate the utility of CalTrack to define perturbations in components of calcium homeostasis in adult guinea pig cardiomyocytes and human iPSC-CMs. We show that levosimendan, an agent with ionotropic effects by sensitizing troponin C 57 and a potent phosphodiesterase inhibitor 32 that stabilizes PKA, increased F max /F 0 , accelerated calcium decay and accelerated time to peak calcium, which manifested as shorter CDs in guinea pig cardiomyocytes similar to human myocardium 33 . iPSC-CMs treated with the β-adrenergic agonist isoproterenol activated calcium transients, due to PKA-dependent increases in L-type calcium channel activity 58 , release of sarcoplasmic reticulum calcium by phosphorylation of phospholamban 59 , and ryanodine receptor activity 60 .
CalTrack robustly detected differences in mutated sarcomeric thin filament proteins. Two pathogenic HCM variants in TNNI3 both had prolonged tau and longer T off times. CalTrack also identified an 8-fold higher frequency of delayed after depolarizations in TNNI3 R21C/+ iPSC-CMs when compared to WT iPSC-CMs. These findings result in decreased cellular relaxation, as assessed by SarcTrack in TNNI3 R21C/+ iPSC-CMs, which often manifest as diastolic abnormalities in HCM hearts 6,26 , and may promote after depolarizations and arrhythmias that contribute to sudden cardiac death 35 . The TnnI3 R145G/+ variant had a longer T on while the TNNI3 R21C/+ variant had a shorter T on when compared to respective WT cells. These findings are significant and are mirrored in the respective contractility profile of each variant 26 (Online Figure X). Notably, these data also show that the allosteric myosin ATPase modulator mavacamten, which corrects the contractile abnormalities in HCM thick filament variants, normalized tau and T off in TNNI3 R21C/+ iPSC-CMs albeit with a dose-dependent reduced F max /F 0 , time to peak, and CD (Online Figure IX). As such, mavacamten treatment may salvage some of the diastolic insufficiency caused by HCM mutations that are concomitant with increased lengths of calcium decays. These findings support recent studies 61 suggesting that mavacamten influences calcium cycling with potential therapeutic efficacy in addition to its established effects on myosin ATPase 38,61 .
Finally, we note that the combined use of endogenous genetically encoded green fluorescent protein (GFP) tag on titin 62 concurrent with the red spectrum RGECO calcium fluorophore allows concurrent imaging of both sarcomere contraction and calcium transients in cardiomyocytes (Online Figure X). Dual functional assessments with a cell permeant fluorophore represent the potential for future development and combination of SarcTrack with CalTrack, so as to enable wide-spread analyses using calcium-sarcomeric shortening loops (Online Figure XI B).
In summation, CalTrack has demonstrated applicability as a rapid automated calcium transient analysis platform for a wide variety of cardiomyocyte constructs and sources. CalTrack provides calcium amplitude and temporal parameters from single or multiple adult or iPSC-derived cardiomyocytes within a field of view and can analyze previously extracted data from proprietary equipment. The algorithm provides high-fidelity measurements, limited by the quality of data input (e.g., acquisition rate of image stacks and noise) and is easily adopted and applied on local computers with rapid automated data processing rates. We expect this open source tool will advance assessments of calcium transients across multiple experimental cardiovascular fields.

ACKNOWLEDGMENTS
We thank the members of the Image and Data Analysis Core at Harvard Medical School (HMS) for their discussions and technical assistance with programming CalTrack.

DISCLOSURES
None to disclose. Online Figures I -XI  Online Tables I -III Online Videos I -III Figure 1: The variety of cardiovascular model systems and constructs that can be automatedly analyzed using CalTrack. CalTrack can be applied to multiple cellular systems, both organized and disorganized primary cardiomyocytes from animal models and immortalized cell types (iPSC-CMs and ESC-CMs), using formats that include 3D tissue systems, organoids, and 2D cell culture systems.   . C) Increases in pacing frequency decrease normalized peak calcium concentration (equivalent to F max /F 0 ). D-I) Increasing pacing frequency significantly accelerates all calcium transient properties assessed by CalTrack (T on , T off , Tau, T50 on , T50 off and CD). Data presented as mean and standard deviation, colors of data points represent 0.5 Hz pacing (blue), 1 Hz pacing (green), and 2 Hz pacing (red). For Ca max /Ca 0 and Tau statistical analysis is performed by one-way ANOVA with a post-hoc Sidak correction . T on , T off , T50 on , T50 off and CD were tested with a paired non-parametric test Friedman test with Dunn's multiple comparisons correction, three comparisons. In both instances a p < 0.05 was defined as the significance cut-off. T90off is assessed to be longer in TnnI3 R145G/+ cardiomyocytes by CalTrack analysis. K) Tau is significantly longer in TnnI3 R145G/+ cardiomyocytes. Statistical analysis is performed by a two-tailed Student's t-test (Baseline, T on , T90 on , and T90 off ) or by nonparametric Mann-Whitney test (CaAmplitude, T10 on , T50 on , T10 off , T50 off , tau) with a significance cut-off of p < 0.05. No statistical comparisons are made between IonOptix and CalTrack. Levo treatment shortened T on (F), T50 off (G), T off (H). and CD (I). Data presented as mean and standard deviation. Statistical analysis is performed by a two-tailed Student's t-test (CD, T off , T50 off ) or by nonparametric Mann-Whitney test (F max /F 0 , Tau, T50 on ,T on ) with a significance cut-off of p < 0.05.  47) or TnnI3 R145G/+ (n = 88) guinea pig cardiomyocytes. The TnnI3 R145G/+ variant does not affect F max /F 0 (C), and Tau (D). T50 on is accelerated (E), but T on (F), T50 off (G), T off (H), and CD (I) are all slowed in comparison to WT. Data presented as mean and standard deviation. Statistical analysis is performed by a two-tailed Student's t-test (T on , T off , and CD) or by nonparametric Mann-Whitney test (F max /F 0 , Tau ,T50 on ,and T50 off ) with a significance cut-off of p < 0.05. Significances are denoted on each panel where * p < 0.05, ** p < 0.01, *** p < 0.005, **** p < 0.0001. Color of significance denotation defines the treatment group compared. Please see Online Table III for exact p values.

What Is Known?
 Measurements of Calcium transients in cardiomyocytes provide insights into homeostasis and pathophysiologic processes and define drug effects.  Tools that automatedly segment cells, extract calcium transients and output data in high throughput are not available.

What New Information Does This Article Contribute?
 CalTrack provides an automated, high-throughput calcium transient analysis pipeline for studying primary, iPSC-derived cardiomyocytes, and engineered cardiac tissues.  Analyses of cardiomyocytes with pathogenic thin filament variants reveals abnormal calcium transients in a cell model of hypertrophic cardiomyopathy (HCM).  Aberrant calcium transient observed in iPSC-derived cardiomyocytes with the HCM mutation TNNI3 R21C/+ are not fully normalized by application of Mavacamten.
CalTrack provides an automated and high-throughput platform for extracting and analyzing calcium transient data in cardiomyocytes. CalTrack can be used to analyze data from line-scans, photomultiplier tubes, fluorescent image stacks and movies. Cells can be automatically segmented to extract multiple calcium transients per field of view, from a wide variety of preparations, including single cells, organoids, and 3D tissues.
The CalTrack pipeline detected changes in calcium transients in response to pharmacological agents Leovsimendan and Isoproterenol. CalTrack also identified abnormal cellular calcium transients in guinea pig cardiomyocytes carrying the HCM variant TnnI3 R145G/+ and in human iPSC-CMs carrying the HCM variantTNNI3 R21C/+ . The TNNI3 R21C/+ variant caused a significant acceleration of T on , and longer duration of T off and calcium transient duration, abnormalities that were only partially corrected by Mavacamten, an allosteric modulator of myosin. We infer from this data that Mavacamten may have therapeutic utility in thin filament HCM, but that compounds with more specific calcium effects may be of additional benefit to these patients.