**1. Introduction**

The zebrafish has emerged to become a common and useful vertebrate animal model for cardiovascular research in recent years [1–4]. It is a powerful genetic system to study cardiac function due to several advantages. The ability to implement the main modern genome editing tools and techniques, including engineered nucleases (meganucleases, zinc-finger nucleases, transcription activator-like effector nucleases) and CRISPR gene editing, have been demonstrated for *Danio rerio* [5–8]. Various zebrafish transgenic lines can be used to enhance the visualization contrast of different transformation processes [9–11]. The use of the zebrafish as a model organism is of particular interest for the analysis of innate immunity [12] and for the study of viral [13–15], bacterial [16,17] and fungal infections [18,19].

Small size, ease of maintenance, short reproduction cycle with multiple (up to 2000) embryos and short period (3 days) of embryonic development make *Danio rerio* a convenient object for continuous observation and experimental research of the mechanisms by which genetic and disease-related modifications are being passed to the offspring.

Cardiovascular system functional characteristics of zebrafish are well studied in various conditions and in the whole life cycle. Their accurate analysis and patterning may indicate developmental disorders and presence of pathology at the very early stages. Detailed studies of cardiovascular system function are of grea<sup>t</sup> value for understanding cardiac failures and identifying novel therapies for human cardiovascular disease [20–22].

Zebrafish embryos develop in the external environment and are relatively small-sized and transparent in the optical wavelength range. These features allow one to easily visualize the circulatory system structure [23,24], observe heart and blood vessel formation and development, and simulate cardiovascular diseases [25–28]. While the zebrafish cardiovascular system functioning is well examined at larval stages [24,29–31], the mechanisms by which the vessel activity is initiated under natural conditions remain a subject of intense investigation. At the pharyngula stage (24–40 hpf) that precedes hatching [32], invasive procedures (excising, fixation, sectioning and staining) are not effective for real-time monitoring due to the fast-changing and vulnerable state of the emerging cardiovascular system.

Thus, biomedical optical imaging techniques are remaining promising in the field of in vivo zebrafish studies [2,33]. Though heart function analysis from bright field time-lapse image sequences [31,33–35] as well as quantitative research of zebrafish heartbeat initiation and stabilization [36] have been reported, blood vessel imaging during early embryonic development has not been carried out. Due to constant embryo motion within the shell, conventional processing algorithms based on image subtraction are inefficient and require precise initial image stabilization. In this research, we show that a combination of time-lapse bright-field microscopy and an advanced image processing technique represents an effective approach to in vivo stain-free cardiac activity mapping across the whole specimen and quantitative analysis of cardiovascular performance even at pre-larval developmental stages in the absence of any anesthesia.

#### **2. Materials and Methods**

#### *2.1. Experimental Animals*

Zebrafish embryos were from an existing stock at the Biological Faculty of Lomonosov Moscow State University. Before embryo collection, single species groups of males and females were kept by local aquarists in isolated 10 L glass aquaria at a temperature of 26 ◦C with the aquaria illumination turned on for 12 h daily. The selected groups were fed three times a day ad libitum on flake and *Artemia*. For embryo collection, males and females were placed together in a breeding tank and the embryos were collected immediately post-fertilization. A mixture of individuals from several different breeding groups was used during the experiment to maximize genetic variation among embryos. The two-cell development stage was used as initial time point during observation because the exact time of fertilization was difficult to determine. The development, survival rate and morphology of the collected embryos was not affected in comparison to the control group not subjected to the study.

#### *2.2. Experimental Setup*

The conventional off-the-shelf trinocular transmitted light bright-field microscope is the basis of the experimental setup (Figure 1). A Koehler system was implemented to achieve uniform illumination of the object. Images were formed by an optical system assembled of a flat field corrected apochromatic objective (10× NA 0.25) and a standard tube lens. To acquire the digital images, the complementary metal-oxide-semiconductor active-pixel image sensor (IDS uEye UI-3060CP-M-GL Rev.2, 1/1.2", 1936 × 1216 pixels) was installed onto the camera tube. The image sensor was connected to the PC for raw data acquisition and storage. The image processing and data analysis pipeline is described in Section 2.3.

**Figure 1.** Experimental protocol.

### *2.3. Experimental Protocol*

The single experimental dataset contained 3000 12-bit grayscale digital images with 1200 × 1200 pixel resolution obtained during a 60-s time series (50 fps). For preliminary geometrical calibration of the imaging setup, the images of the test chart were captured before the experimental dataset acquisition. The geometrical calibration procedure is an essential step for the magnification determining and estimation of the real image distortions. During image acquisition, the inspected embryo may move and rotate within the shell, and illumination conditions and background characteristics may also vary. Data analysis procedures require pre-processing of raw images to crop the region of interest, compensate for movement and eliminate the illumination non-uniformity (Figure 2). After these procedures, we obtained the well-matched and intensity-corrected spatiotemporal data cube *<sup>I</sup>*(*<sup>x</sup>*,*y*,*<sup>t</sup>*) including time-domain dependences *I*(*t*) for each image pixel with spatial coordinates *<sup>x</sup>*,*y*.

**Figure 2.** Image processing pipeline.

Figure 3 illustrates the main image pre-processing stages shown in pink color in Figure 2. First, the microscopic images of 26 hpf unhatched zebrafish embryo in a lateral position were to be cropped down to the embryo dimensions (Figure 3a). Second, to compensate for illumination non-uniformity and align the average intensity of all images in a series, we subtracted a smoothed image and constant term 127 (Figure 3b). Third, to ensure pixel-to-pixel matching of all images, we calculated local motion vectors and matched the images with respect to their directions and lengths. Figure 3c shows a locally matched image with crosses indicating the positions of each grid knot in the first (black) and

the last (white) image of the sequence. Shape and intensity of the temporal signal *I*(*t*) in each pixel of the obtained data cube *<sup>I</sup>*(*<sup>x</sup>*,*y*,*<sup>t</sup>*) characterizes morphogenetic events occurring during embryonic development. The blood flow changes may be detected by the periodic variations of the transmitted light intensity since blood cells absorb light more strongly than the surrounding tissues [37]. Plotting the spectrum of the temporal signal using Fourier transform allowed detection of the dominant frequencies, which corresponded to the cardiovascular activity. After obtaining the intensity deviation values in each image pixel, we could subtract the blood-free background and consider only the pixels with blood flow related to significant intensity oscillations. Thus, we obtained a series of well-matched blood flow images ready for cardiac activity analysis (Figure 3d). The blood circulation was clearly visible throughout the full cardiac cycle (Supplementary Video).

**Figure 3.** Stages of image pre-processing: (**a**) cropped image, (**b**) aligned image, (**c**) locally matched image and (**d**) blood flow image.
