Fig 1.
Schematics of fluorescence recording apparatus and data processing.
(A) Optical apparatus to record action potential (AP) from hiPSC-derived cardiac microtissues. (B) Schematic of the data processing pipeline. A series of data processing steps were taken to measure AP metrics: ① AP recording. ② segmentation for identification of individual microtissues, ③ baseline correction to remove fluorescence changes due to dye bleaching or fluctuation of perfusion solution, ④ filtering to increase signal-to-noise ratio while preserving sharpness of AP upstroke, ⑤ calculation of first derivative and moving average subtraction to detect AP upstroke and repolarization, and ⑥ AP metric measurement.
Fig 2.
Selection of action potential metrics to detect ion channel block through computer modeling.
(A) Human action potential model by O’Hara et al. [52] and the underlying major ionic currents: Na+ (INa), transient outward K+ (Ito), Ca2+ (ICa), rapidly (IKr) and slowly (IKs) activating delayed rectifier K+ currents, and inward rectifier K+ current (IK1). (B) Illustration of AP metrics to detect AP shape changes by ion channel block; AP upstroke rise time, plateau phase (APD30 and APD50), and phase 3 repolarization (APD80, APDmxr, and APDtri = APDmxr−APD50). (C) Modeling of AP changes in response to specific block of indicated ion channels to 50% conductance (gX 50%). INa block exclusively alters AP upstroke; ICa block shortens APD by depressing the plateau Vm which accelerates repolarization; IKr block slows the phase 3 repolarization; and IKs block slightly prolongs APD. (D-F) Simulated changes in AP metrics following specific blocking of three major ionic currents (INa, ICa, and IKr) at 25%, 50% and 75%. Details of the modeling and parameter sets are described in Methods. (G) PCA plot of INa, ICa, IKr block. Feature extraction was done using principal component analysis (PCA). Distributions of AP metric changes are shown for INa (purple), ICa (orange), and IKr (blue) at 75% block. PCA indicates two major features that best distinguish AP metric changes between the five simulated blocks (INa, ICa, IKr, Ito, and IKs; see Tables 1 and 2).
Table 1.
AP metric changes under specific ion channel block in computer simulations.
Results are reported as mean +/- standard deviation (percent change of mean from cell mean without channel block). 40,000 single cells with channel conductances varied to reflect population-level heterogeneity were paced at 0.5 Hz as described in Methods. AP metrics were determined for the penultimate beat.
Table 2.
Principal component analysis of specific ion channel block in computer modeling.
PCA1 is a major axis that underlies overall APD prolongation and explains 79% of the variance in the simulation data. PCA2 mainly detects changes in the rise time of the AP upstroke with a smaller contribution due to shortening of APD30 and prolongation of APD triangulation. In computer simulation, the delay between stimulation and AP upstroke was small compared to the sampling rate and the stimulation delay parameter was not incorporated in calculating PCA axes.
Fig 3.
Effects of human cardiac fibroblast (hCF) percentage on microtissue compaction and characterization of AP.
(A) Phase-contrast image of 2D culture of purified hiPSC-CMs for microtissue generation. (B) Microtissue compaction assessed by reduction in spheroidal microtissue diameter over 7 days of culture. (C) Sample images illustrating microtissue compaction over indicated time. (D) Sample action potential traces with indicated hCF content. (E) AP characteristics of microtissues with indicated hCF content recorded after 6–8 days in culture. Each dot represents averaged AP metrics from each mold (n = 3 molds, 32–35 microtissues per mold).
Fig 4.
Segmentation of individual microtissues from fluorescence voltage images.
(A) Sample fluorescence image (left) and a histogram of the sample image (right). Traces from each pixel were transformed to frequency domain using FFT, and the maximum FFT amplitude within pacing cycle length were calculated to reconstruct a FFTmax image. (B) FFTmax image that highlights regions showing full AP amplitude. The histogram of FFTmax image (right panel) shows the best contrast image compared to fluorescence intensity in panel A. (C) Otsu’s thresholding and corresponding histogram in log scale showed detailed distribution of pixels with lower intensity. Since Otsu’s method selects a more conservative threshold value (red arrow), the segmentation often results in a smaller or distorted selection of microtissues (red circles). (D) Cross entropy thresholding via a minimum cross entropy algorithm improves the selection of pixels with AP signals (see the red arrow, mean threshold using cross entropy = 30 ± 17 vs. Otsu’s method = 72 ± 16, n = 30 microtissues). After segmentation, a region-labeling algorithm identified individual microtissues in a solid circular shape compared to Otsu’s method in panel C (mean area of microtissue detected using cross entropy = 160 ± 31 pixels vs. Otsu’s method = 99 ± 11 pixels, n = 30 microtissues). (E) Sample selected microtissue marked with a white box in panel D. (F) Superimposed traces from all the pixels within microtissue #6 in panel E. (G) Averaged trace from all the pixels in panel F.
Fig 5.
(A) Sample raw trace of AP from single pixel with baseline drift before correction. Dye bleaching and fluctuation from perfusion flow may cause slow drift in baseline. (B-I) Iterative method of polynomial fitting for baseline estimation using asymmetric least squares method. Step 1: Resample the original signal to lower time resolution to accelerate calculation (here, 128 points are sampled from 8192 points, panel B). Step 2: Fit the signal with least squares polynomial fitting (here 3rd order, panel C). Step 3: Apply reduced weight when the signal is above the polynomial fitting by multiplying asymmetric parameter p (here 0.001, panel D). Step 4: Apply polynomial fitting again and repeat adjusting weight (panel E). Step 5: repeat iteration until the baseline estimation converges (panel F-I). (J) Sample trace after 5 iterations of baseline correction.
Fig 6.
(A) Non-linear bilateral filter to preserve the sharpness of AP upstroke. Sample traces of AP before (black) and after applying Gaussian (top panel traces, blue to red) and bilateral (middle panel traces, blue to red) filters. AP upstroke is preserved after bilateral filtering even with a significantly larger window size (middle panel) compared to typical Gaussian filtering (top panel). The slope of AP upstroke was steeper with bilateral filter vs. gaussian filter (triangle for bilateral vs. black square plots for gaussian filter in the bottom panel), showing the advantage of edge-preserving bilateral filter to reduce noise in AP traces. (B) Upstroke detection using a moving average subtraction algorithm to detect the takeoff and peak time of AP upstroke. The moving average was calculated (red, window size = 45 points) and the original trace (black) was subtracted (blue), which detects a rapid change in AP upstroke. The middle traces show that the maximum of moving average subtraction (MAS) occurs at the initial rise of AP upstroke and the minimum of MAS occurs when the rapid rise of the AP upstroke is finished. The bottom panel compares the rise time of AP upstroke measured by time-to-max vs. MAS method. After sufficient filtering with 15-point bilateral filter, the standard deviation of the rise time measurements using MAS is significantly lower, increasing sensitivity to detect potential changes in AP upstroke by Na+ channel blockers. (C) AP upstroke analysis by MAS compared to Gaussian 2nd derivatives. The rise time measurement using MAS is less sensitive to the choice of window size (top), while the rise time measurement using Gaussian 2nd derivatives (middle) is highly sensitive to the choice of window size (bottom).
Fig 7.
Detection of AP repolarization.
(A) Repolarization time using moving average subtraction (MAS). MAS shows a secondary peak during the late phase of repolarization when the rapid repolarization slows down to reach the resting membrane potential. This time point is termed the maximum rate change of repolarization (APDmxr). (B) Automatic detection of APD30, APD50, APD80, APDmxr from a sample AP recorded from a single microtissue. (C) Comparison of APD measurements using MAS vs. 2nd derivative methods. A longer window size was required to measure APDs (>45) compared to AP upstroke analysis (~13, see Fig 6C). The mean value of APD using the MAS (red) method is less sensitive to the smoothing window size using the Gaussian 2nd derivative method (black). (D) Increasing the smoothing window size reduces the standard deviation of APD measurements. N = 30 microtissues.
Fig 8.
Experimental validation of AP metric changes in response to known ion channel blockers.
Representative AP traces before and 15 to 30 mins after drug exposure at indicated concentration (top) and automatically quantified AP metrics from microtissues containing 5% hCF after 6~8 days in culture (bottom). (A) Tetrodotoxin (TTX, 1 μM), a selective Na+ channel blocker, increased stimulation delay and rise time and moderately increased APD80 and APDmxr (n = 32 microtissues). (B) Flecainide (20 μM), a INa and IKr blocker, increased stimulation delay and rise time as well as APD30, APD50, APD80, APDmxr, and APDtri (n = 35 microtissues). The increase ion APDtri is caused by larger prolongation of APDmxr compared APD30 and APD50. (C) Nifedipine (2 μM), a selective L-type Ca2+ channel blocker, shortened APD30, APD50, APD80, APDmxr, and APDtri (n = 30 microtissues). (D) E4031 (2 μM), a selective IKr blocker, increased APD30, APD50, APD80, APDmxr, and APDtri (n = 32 microtissues). Due to pronounced APD prolongation, the time scale is shown over 2 sec for E4031 traces. Additionally, two sample action potential traces are shown for E4301, one without EAD and another with EAD (n = 7/32 microtissues). (E) Discriminating ion channel blocks using AP metrics and PCA. Distribution of AP metric changes are shown for TTX (purple), flecainide (green), nifedipine (orange), and E4031 (blue). Principal component axes are shown in Table 3.
Table 3.
Principal components analysis of experimental data using TTX, Flecainide, Nifedipine, and E4031 in Fig 8.
PCA1 is a major axis that underlies overall APD prolongation and explains 52% of the variance. PCA2 mainly detects changes in rise time of the AP upstroke with a smaller contribution of APDtri and explains 25% of the variance.
Table 4.
Accuracy of detecting ion channel blocks from AP shape changes from computer simulation data using multi-class logistic regression.
Results are reported as mean (standard deviation). Standard deviations were obtained by repeating the model fitting, hyperparameter tuning, and testing process on different training/test dataset splits as described in Methods.