Virtual blebbistatin: A robust and rapid software approach to motion artifact removal in optical mapping of cardiomyocytes

Figures

• 

  Fig. 1.. The drug blebbistatin disturbs iPSC-cardiomyocyte electrophysiology and sarcomere structure. (A and B) 2 µM blebbistatin, sufficient to eliminate cardiomyocyte monolayer motion, slightly increases Ca2+ upstroke duration (A) and action potential duration (APD; B). (C) Representative sarcomere structure (sarcomeric α-actinin staining, green, with Hoescht nuclear counterstain, blue) of a control iPSC-cardiomyocyte monolayer. (D) Representative sarcomere structure of iPSC-cardiomyocytes 4 h after 2 µM blebbistatin treatment. Error bars: SD. * indicates P < 0.05. n > 5. (Scale bar: 10 µm.)

• 

  Fig. 2.. Virtual blebbistatin stabilizes motion in images by mapping deformed regions back to their undeformed reference configurations. (A) A frame of a sequence is chosen to be a reference image. The image is divided into boxes which serve as templates [T(x)]. (B) A warping function W(x;p) is found for each region of each image that warps the coordinates x from the original template to the deformed region in the image by optimizing a set of parameters, p. (C, D, and E) Schematic illustration of the process. The software iteratively optimizes parameters to warp a region from the template (C) to a deformed region in a later frame (D and E). (F, G, and H) In the unwarped (stabilized) image series, the first frame (F) is an unchanged reference. Later frames (G and H) are “unwarped’ by interpolating (evaluating) pixel values at the warped coordinates in (D and E) and mapping those values to the material coordinates from the original frame.

• 

  Fig. 3.. The algorithms used for virtual blebbistatin are effective over a range of box sizes, and convergence is markedly improved by CLAHE. (A) CLAHE dramatically improved the percentage of regions that converged on a set of images of fluorescently labeled, contracting sheets of iPSC-CM. Three- and 5-point median temporal filtering of the individual pixel time series improved algorithm convergence to a lesser extent. The box size used for warping regions is shown on the x axis. (B) CLAHE substantially reduced the average number of iterations required for an individual region to converge. Temporal filtering was also effective in lowering the number of iterations for convergence but to a lesser degree. (C) rms image error (normalized to a range of 0 to 1) is a measure of the difference between the original template of a region and the warped region of the current frame. CLAHE reduced this error substantially for most of the box sizes tested, indicating an improvement in the performance of the algorithm. (D) Convergence metrics at 49-pixel box size using different preprocessing steps.

• 

  Fig. 4.. Virtual blebbistatin effectively removes motion from iPSC-CM monolayers and iPSC-micro-heart muscles (µHM). (A) Comparison of original (Top row) brightfield images to stabilized (Bottom row) images of a contracting iPSC-CM monolayer. The entire sequence represents approximately one beat cycle. (B) Subtracting the first frame (t = 0) from subsequent original (Top row) and stabilized (Bottom row) images in panel A produced heatmaps of the percent pixel intensity difference between each frame and the first frame. The difference in the stabilized images caused by noise and imperfect stabilization were due largely to the stationary particles within the frame. (C) iPSC-CM monolayer average speed, tracked by optical flow, before and after virtual blebbistatin. (D) iPSC-µHM net motion before and after virtual blebbistatin. (E) iPSC monolayer and (F) µHM motion reduction trials with different box sizes, indicating different optimal imaging processing box sizes for different applications. n = 5 per condition. [Scale bar: 200 µm (C and D), error bars: SD.]

• 

  Fig. 5.. Virtual blebbistatin improves accuracy in the optical analysis of action potentials and calcium transients in iPSC-CM monolayers. (A and B) Original action potential duration heatmap (A) and distribution (B). (C and D) Original heatmap of the time required for the calcium signal to decay 75% of its peak value (C) and distribution (D). (E and F) Virtual blebbistatin stabilized action potential duration (APD) heatmap (E) and distribution (F). (G and H) Virtual blebbistatin stabilized heatmap of the time required for the calcium signal to decay 75% of its peak value (G) and distribution (H). (I) APD 80 with virtual blebbistatin stabilization showed significantly higher kurtosis. (J) Similar centered SD for APD 80 was seen for original and virtual blebbistatin. (K) Calcium 75% decay with virtual blebbistatin showed significantly higher kurtosis. (L) Calcium 75% decay with virtual blebbistatin showed significantly less centered SD. *, ** and **** indicates P value < 0.05, 0.01, and 0.0001. Error bars: SD. n > 5.

• 

  Fig. 6.. The effect of motion artifact on calcium transient signals diminishes as the signal is averaged over a larger area. (A) Representative image of a contracting iPSC-CM sheet. Crosshairs indicate the location of the calcium signal probe, and circles indicate regions of radius 5, 15, and 25 pixels (6.6 μm, 19.8 μm, and 33 μm), respectively. (B) Normalized calcium signal intensity (0 to 1) at pixel location shown in crosshairs of (A). Original signal (blue online), and motion-corrected signal (orange online). The uncorrected signal exhibits artifacts such as deeper troughs that are not observed in the motion-corrected signal. (C) When averaging over a circular region with radius 6.6 μm, differences between the original and corrected signals are smaller than at the pixel level, but still visibly present. (D and E) As the radius of the circular region increased to 19.8 μm, and then to 33 μm, differences between the original and corrected signals diminished further until they were nearly imperceptible. (Scale bar 30 μm.)

• 

  Fig. 7.. Virtual blebbistatin allows direct analysis of calcium-contraction coupling. (A) Calcium-contraction phase plots for iPSC-CM monolayer on designated spots. (B) Calcium-contraction phase plots for iPSC-µHM on designated spots. (C) Calcium-contraction phase plots for hypertrophic cardiomyopathy prone, MYBPC3+/− iPSC-µHM on designated spots. (D) Normalized diseased (Left) and control (Right) tissue calcium-contraction coupling plots. Calcium signal intensity is shown on the x axis, contraction is shown on the y axis. Both calcium signal and contraction are normalized by setting the minimum value of each signal to zero, and the maximum values are scaled by the maximum value observed over all signals. Designated spots are shown in SI Appendix, Fig. S5.

• 

  Fig. 8.. A mathematical model of the interaction of actively contractile regions displays behavior like that observed in motion-corrected data from contracting tissues and cell sheets. (A) A mathematical model consisting of 4 motor-spring-damper elements. Each motor Mi is a contractile motor with displacement driven by a 1/Γ function repeated at 1 Hz, which resembles a cardiac action potential, in which Γ is the Gamma function. (B) Normalized contraction (0 to 1) vs. normalized motor signal (0 to 1) for the 4 elements. Direction of the phase trajectory in time is indicated by red arrows. The default nondimensional parameters were ki = 10, ci = 1, for all elements, and all motors had amplitude Mi = 4. The phase delay between adjacent regions was 0.1s. The different behaviors of the 4 regions are purely due to the phase delay. (C) If stiffness of the second (k2) is increased by a factor of 5 from the default value (leaving all other parameters unperturbed), the second element exhibits more slender hysteresis shape with a counterclockwise trajectory in the signal-contraction phase space. (D) If stiffness of the second element (k2) is decreased by a factor of 5 from the default value (leaving…