2024 Volume 99 Article ID: 23-00239
The importance of the parent–progeny relationship tracking technique in single-cell analysis has grown with the passage of time. In this study, fundamental image-processing techniques were combined to develop software capable of inferring cell cycle alterations in fission yeast cells, which exhibit equipartition during division. These methods, exclusively relying on bright-field images as input, could track parent–progeny relationships after cell division by assessing the temporal morphological transformation of these cells. In the application of this technique, the software was employed for calculating intracellular fluorescent dots during every stage of the cell cycle, using a yeast strain expressing EGFP-fused Swi6, which binds to chromatin. The results obtained with this software were consistent with those of previous studies. This software facilitated single-cell-level tracking of parent–progeny relationships in cells exhibiting equipartition during division and enabled the monitoring of spatial fluctuations in a cell cycle-dependent protein. This method, expediting the analysis of extensive datasets, may also empower large-scale screening experiments that cannot be conducted manually.
Analyses of single cells, as opposed to cell populations, across diverse species have facilitated the elucidation of the characteristics of each individual cell. For instance, in studying the cell cycle, it is paramount to analyze the behavior of each cell in a specific phase, as opposed to aggregations of cells spanning multiple phases. One approach to single-cell analysis employs fluorescent proteins that are fused to cell cycle index proteins, and cells expressing such proteins are sorted by flow cytometry or observed directly by fluorescence microscopy. In single-cell tracking, suitable cells with fluorescent markers introduced for the purposes of the experiment must be observed over time under a fluorescence microscope. The creation of fluorescence-based cell cycle reporter strains conducive to practical application is complicated, and the types of available fluorescent proteins are constrained, thereby restricting the capacity to investigate these reporter cells (Sakaue-Sawano et al., 2008, 2017 ; Goto and Aoki, 2020). In the field of epigenetics, the imperative lies in elucidating the mechanisms through which hereditary traits are inherited by progeny cells. Thus, tracking parent–progeny relationships for each individual cell is more important than analyzing cell populations. To address this issue, fluorescence-based cell cycle reporter strains will be a potent tool to analyze both the cell cycle and the inheritance of epigenetic information, simultaneously, at single-cell resolution. These methods, however, require prolonged irradiation, yielding non-negligible cellular toxicity. Analysis of cell-tracking experiments further requires the processing of prodigious volumes of experimental data, which are too large to be evaluated manually.
The present study was conducted with the aim of developing software capable of tracking cells and estimating cell cycle phases at the single-cell level, without relying on fluorescent proteins, by combining fundamental image processing techniques. A tracking system capable of detecting generational alterations in epigenetic expression states at the single-cell level has previously been established (Mano et al., 2013), and software designed for the automated analysis of its results (Kanada et al., 2020). While this software significantly accelerated the analysis, thereby rendering genome-wide screening experiments viable, its applicability was confined to budding yeast, precluding its utilization in species characterized by symmetrical cell division and obscure parent–progeny relationships. This study describes the creation of an analogous software program tailored to fission yeast, Schizosaccharomyces pombe, which undergoes symmetrical cell division.
Novel tools have been created to adapt previously developed techniques in image processing and machine learning to biology (Chen et al., 2006; Li et al., 2008, 2009 ; Stringer et al., 2021). Each of these studies employed bespoke software programs. As an illustration, cellular segmentation was executed through image processing (Yang et al., 2005; Li et al., 2006; Zhou et al., 2008; Mashburn et al., 2012; Balomenos et al., 2015; Loewke et al., 2017; Al-Kofahi et al., 2018) or machine learning-based segmentation methods (He et al., 2007; Wang et al., 2007; Van Valen et al., 2016). Tracking segmented cells has been executed through an active contour model (Tarnawski et al., 2013), object tracking (Yang et al., 2005; Yan et al., 2009; Bise et al., 2011; Rapoport et al., 2011; Cornwell et al., 2020) and deep learning (Al-Kofahi et al., 2018; Cornwell et al., 2020), culminating in exceptionally precise tracking. Approaches involving pattern recognition (La Brocca et al., 2012) and machine learning (Nguyen and Kim, 2018; Ulicna et al., 2021) have also been used for tracking parent–progeny relationships. Moreover, machine learning techniques have been employed to automatically rectify errors caused by automated tracking (Löffler et al., 2021), with machine learning extending its utility well beyond the confines of image processing, in an automatic analysis algorithm. Given that these techniques constitute fundamental algorithms, their actual application in a comprehensive software solution for single-cell analysis necessitates a judicious amalgamation of these fundamental techniques. Furthermore, as each algorithm fundamentally presupposes its own specific species, the adaptation of these algorithms to alternative species presents a formidable challenge.
In this study, the utilization of OpenCV, an open-source computer vision library that incorporates essential image processing functions, is delineated. Our findings suggested that OpenCV could be effectively employed in the facile creation of software encompassing the requisite functions, even when the creator lacks expertise in the field of image processing. By amalgamating functions within OpenCV, we developed software encompassing four key features: cellular region segmentation, parent–progeny relationship tracking, cell cycle phase estimation solely leveraging bright field microscope imagery, and calculation of the number of fluorescent dots within individual cells. The first function employed the watershed method to segment regions of fission yeast cells undergoing equipartition during cell division. The second function facilitated the identification of parent–progeny relationships, both pre- and post-division, among numerous cells within a bright field microscope image. This was achieved by assessing the pixel count in the shared region between the nth and (n-1)th frames, serving as an index for distinguishing individual cells in the previous and next frames. The third function inferred transitions between the G1, S and G2 phases based on morphological alterations exhibited by fission yeast cells. These changes encompassed the observation of the septum, demarcated as the constriction occurring between parent and progeny cells, as well as the topological modifications between these cellular entities. The fourth function enabled use of this software for the quantitative measurement of biological activity through live imaging. Hence, this study describes the development of software predicated on the amalgamation of widely employed fundamental image processing algorithms, offering a valuable contribution to biological studies without the need for complicated algorithms like machine learning.
The software we developed was engineered to discern the division of fission yeast cells by evaluating morphological alterations within these cells. It also aimed to track parent–progeny relationships among individual cells both before and after cell division, while concurrently estimating cell cycle phases by evaluating morphological changes in single cells. This software required the input of a chronological series of bright field microscope images of fission yeast division, outputting cell cycle phases via the evaluation of morphological transformations within individual cells. The fundamental components of this program encompassed the segmentation of each cell within every image, the capacity to track cells consistently across all images, and the estimation of cell cycle phases relying exclusively on bright field images to detect morphological alternations. Furthermore, the quantitative measurement of biological activity necessitated a function to measure fluorescent dots within each cell.
Segmentation of individual cells within an imageCells from numerous species undergo symmetrical division. Given that our previous tracking software (Kanada et al., 2020) was specifically targeted for budding yeast, its applicability is ill-suited for species whose cells divide symmetrically. The development of analogous software targeted for cells that divide symmetrically necessitates the appropriate adaptation of each processing stage, encompassing segmentation, tracking and final analysis steps. Therefore, an initial effort was made to modify the segmentation stage to accommodate cells undergoing symmetrical division.
In our experimental setting, yeast cells are cultivated in a monolayer, enabling each cell to be captured in focus under the microscope, particularly in the case of fission yeast. In this ideal microscopic imaging, an accentuated boundary or contour is distinctly discernible around each cell (Fig. 1A; BF), rendering it a valuable feature for cell segmentation. This circumstance does not hold true for the diminutive “progeny bud” of budding yeast, because a parent cell frequently generates its progeny bud in a direction outside the microscope’s depth of field, leading to an out-of-focus image of the bud with an indistinct boundary or edge. Our previous software targeted for budding yeast therefore incorporated an edge enhancement mechanism with the aim of enhancing the weak boundary delineating progeny cells (Kanada et al., 2020). Conversely, fission yeast cells consistently remain within the depth of field even during the dividing phase. Consequently, we can inherently presume that all cells are capturable in focus, and that an accentuated boundary akin to that depicted in Fig. 1A BF is perpetually at our disposal for segmentation. This observation implies that cell segmentation can be achieved through a much more straightforward technique than that described by Kanada et al. (2020).
Our initial endeavor was to accomplish segmentation by discerning variations in brightness between the background and cellular regions through a straightforward binarization process grounded in a single threshold value. However, using this binarization approach alone could not achieve precise cell region segmentation, primarily because of noise, uneven lighting, individual differences, or other aberrations encountered during image capture (Fig. 1A; Binarization without median). To remove the noise prior to binarization, we next applied a median filter, which efficiently removes noise while concurrently preserving edges or boundaries of foreground objects. It was imperative to apply this filter with an appropriate kernel size, as it proves ineffective in noise reduction with a small kernel, while it obliterates even edges and boundaries when a larger size is used (Supplementary Fig. S1). Unfortunately, this approach failed to achieve the segmentation of cell regions, even when the threshold value was meticulously and manually calibrated (Fig. 1A; Binarization with median). To obtain cell regions, we opted to isolate connected components of the cell area using the binarized image subjected to median blur processing (Fig. 1A; Rough cell regions).
From Fig. 1A, it is evident that segmentation errors encompass the erroneous classification of authentic cell regions as the background and, conversely, the misidentification of background areas as cell regions, often leading to the grouping of multiple cells within a single region. Nevertheless, these results offer valuable information for precise cell segmentation: the central section of each cell region can be considered a “seed” corresponding to the respective cell, and a portion of the background region that maintains a substantial distance from each cell region is devoid of actual cells, thus allowing its designation as a “seed area” for the background. When median filtering is correctly administered and the binarization threshold value is aptly chosen, this observation remains valid with virtually no exceptions for images of fission yeast cells cultivated in a monolayer. The underlying concept is to exploit the information derived from the seeds extracted from the binarized images for the purpose of cell segmentation.
One of the techniques employed for image-based segmentation is the seed-based one. Within this category, the watershed algorithm stands out as a quintessential example. This algorithm interprets the brightness profile of an image as a topographic surface and expands the seeds through a flooding process to delineate watersheds, which demarcate region boundaries (Soille, 1999). Despite its relatively fundamental nature compared to contemporary segmentation techniques, the watershed technique remains a potent tool for cell segmentation, particularly within our application. As the watershed algorithm generates the same number of segmented regions as provided by the initial seeds, the suitable selection of the seeds is critical for the successful application of the watershed algorithm. The proposed software utilizes the seeds extracted from the median-filtered and binarized cell images for this purpose.
To extract the background seed, a morphological dilation operation is employed on the binarized image to shrink the background region. Specifically, a dilation with a 5×5 kernel is applied to the cell region within the binarized image for a specified number of iterations, and the residual background region is used as the seed for the background (Fig. 1B). This approach was expected to also be applicable for seed extraction within each cell region: by repeatedly dilating the background regions, the cell regions contract, leaving behind pixels that can serve as seeds for their respective cell regions. In practice, however, it failed to deliver precise seeds for cell regions, because a shallow dilation could not effectively separate adjacent cells in densely populated areas, while increasing the number of dilations eliminated the seed area for relatively small cells (Fig. 1C). To ameliorate this issue, our software incorporates the use of distance transformation, which converts the value of each pixel into the Manhattan distance from that pixel to the nearest background pixel. Initially, distance transformation is applied to each of the cell regions within the binarized cell image, followed by the extraction of pixels whose distance values fall below a predetermined threshold, designating them as the seed for each cell (Fig. 1C). These steps partition the entire set of pixels into three distinct categories: the background seed, the cell seed and other undefined regions. Starting from these seeds, the watershed algorithm is then applied to the undefined region, culminating in the completion of cell segmentation (Fig. 1D). Figure 1E shows the segmentation results achieved through our approach, demonstrating the attainability of precise segmentation with this simple method.
The actual implementation of the segmentation process requires binarization, median filtering, dilation, distance transformation and the watershed algorithm, all of which have been realized through the utilization of the OpenCV library in our software. The window size for the median filter was configured to 5×5. Given that the parameters employed in the proposed segmentation technique, such as the threshold values for binarization and distance transformation, as well as the number of dilations, are notably contingent on the average brightness and magnification factor of the target images, our software uses an empirical approach to determine these parameters. This is accomplished by adjusting corresponding sliders available on the GUI (graphical user interface). It is important to note that these parameters exhibit a high degree of stability and do not need frequent adjustments when a single time-lapse image sequence is being processed, provided that the magnification factor and image capture conditions remain unchanged.
Cell trackingIn cell division, a parent cell divides to generate two progeny cells. During this process, an unequal distribution of genetic information and cellular contents can introduce polarity within the progeny cells. Additionally, cells within a time-lapse experiment continuously change their locations due to growth, division and the flow of culture medium, among other factors. To effectively analyze biological properties of individual cells over time, it is imperative to maintain continuous tracking of the identity of each cell and the parent–progeny relationship of dividing cells across multiple frames. Consequently, our software incorporates two distinct types of cell tracking: interframe tracking of an identical cell and parent–progeny tracking for dividing cells. This is achieved by assigning a unique numerical identifier to each single cell. In the event of cell division, one of the two resultant cells inherits the parent cell’s identifier, while a new sequential number is assigned to the other progeny cell.
Bright-field imaging has no toxic effects on cells, permitting the selection of arbitrary small frame intervals in a time-lapse experiment. Our software operates under the assumption that the frame interval has been chosen to be sufficiently brief, rendering cell motion negligible both before and after cell division – for instance, with intervals of 10 min between frames. Our software exploits this assumption and the advantages of monolayer cultivation to realize the tracking process simply. As demonstrated below, our technique yields an accuracy that is adequate for subsequent analysis.
Let us assume that tracking has been successfully completed from the first to the (n-1)th frames. Figure 2A illustrates the relative positions of cells between the (n-1)th and nth frames, which were captured with a frame interval of 10 min. This depiction indicates that the cell with the greatest overlap corresponds to the same cell in the adjacent frames. Similarly, in the event of cell division, the two progeny cells share most of their regions with their parent cell, as illustrated in Fig. 2A (right column). These observations suggest that within our experimental environment, it suffices for tracking purposes to identify the cell with the greatest overlap from the preceding frame, even in instances of a division.
As shown in Fig. 2B, the target cell to be tracked in the nth frame exhibits overlap not only with itself but also with neighboring cells from the (n-1)th frame, due to interframe motion. Specifically, among these prospective candidate cells in the (n-1)th frame, the cell with the greatest overlap is identified by assessing the pixel count within their common region. Subsequently, the numerical identifier of this cell is inherited by the target cell, thus concluding the tracking process. Under this strategy, when two progeny cells emerge following cell division, they correspond to a single cell in the (n-1)th frame. In such instances, this single cell is recognized as their parent, and its numerical identifier is assigned to one of the progeny cells, while a new identifier is allocated to the other.
Figure 2C illustrates some examples of tracking results, with each identical cell represented in its unique color across two consecutive frames. The tracking success rates until a cell divides to eight cells have significantly improved, reaching 100% for cells during the migrating interphase and cytokinesis phase, when this algorithm is applied. Specifically, the improvement is from 67.7% to 100% during the migrating interphase cells and from 28.5 % to 100% during the cytokinesis phase (Supplementary Movie S1, S2). Because of the short frame interval and monolayer cultivation, our algorithm achieves precise tracking even during cell division. However, despite these advantages, occasional tracking errors may still occur. Therefore, a manual error correction interface has been integrated into the GUI of our software to address such situations.
Estimation of cell cycle phases using only bright field imagesThe fluorescence-based Fucci method is useful in assessing the cell cycle phase from an image (Sakaue-Sawano et al., 2008, 2017; Goto and Aoki, 2020). However, its application is constrained by the need to construct a cell strain and the requirement for numerous drug resistance markers, which in turn limits genetic manipulations of these cells. Consequently, we hypothesized that the cell cycle phase of any cell could be ascertained solely through the use of bright field images, obviating the need for complex cell strain construction. This could be achieved by integrating this technology with time-lapse analysis. First, we developed a technique for assessing the cell cycle using a time-lapse series of bright field images. In microscope images of fission yeast, cells featuring a septum were estimated as being in G1 phase, those with an indented center area were estimated as being in S phase, and cells immediately following division up to septum formation were estimated as being in G2 phase (https://dornsife.usc.edu/pombenet/fission-yeast-cell-cycle/) (Gould and Simanis, 1997; Meister et al., 2007; Garcia Cortes et al., 2016; G. Cortés et al., 2018) (Fig. 3A). Although distinguishing between G2 and M phases based solely on morphological characteristics within bright field images proved challenging, the presence of cells exhibiting a 1.7-fold increase in length compared to that immediately following cell division could be associated with the M phase, especially when contrasted with nuclear staining images. Consequently, when a cell’s length surpasses 1.7 times that of its post-division state, it could be roughly inferred to be in the M phase, while the remaining cells were categorized as being in the G2 phase. This approach enabled the differentiation of all cell cycle phases by amalgamating bright field images with fluorescence images obtained from nuclear staining (Fig. 3A; Nuclear fluorescence (reference)). Nonetheless, these estimations were approximate; the software was devised to estimate the cell cycle phase of individual cells in a manner that complemented the existing fluorescence method.
Given that the septum and indented area are situated along the central axis of the cell, these distinctive features were discriminated by examining the regions near the center of the cell (Fig. 3B). The major axis of the cell was determined by outlining a minimum bounding rectangle on the image using the watershed process (Fig. 3C). This bounding rectangle was considered the smallest area in which an object could be rotated, with its elongated side denoting the major axis of the cell. A new rectangle was then delineated on the central segment of the minimum bounding rectangle, encompassing either the septum or the indented area. The binarized image was cropped along the contours of this rectangle since the watershed-processed image lost information pertaining to the septum. Following this cropping, the resultant image was rotated into a landscape orientation via an affine transformation (Fig. 3D). The next steps were applied to the focused central region of the cell.
The G1 phase was identified by the presence of a septum within the central cropped area. The septum exhibited the same value as a cell wall in the binarized image, as both possessed well-defined edges. When representing the cell wall as white pixels and the remaining region as black, we systematically scanned the cropped center image starting from its top-left pixel, with the ratio of black to white pixels calculated for each row. A low ratio was indicative of the presence of a septum within the cell. Cells were classified as having a septum if the number of rows displaying a septum, namely rows with low ratios of black to white pixels, exceeded a certain threshold. Specifically, if these rows accounted for more than 20% of the length of the long side of a cropped center image, the cell was deemed to possess a septum, signifying that it was in the G1 phase (Fig. 3E).
Cells not in the G1 phase underwent an evaluation to determine if they were in the S phase. This involved examining whether an indented area existed within the central region of the cell, which was achieved by analyzing an image in which the binarized center part of the cell was cropped. In this context, an indented area on the cropped image was perceived as a concave polygon. Notably, the area of this polygon was smaller than that of the cropped image, and the polygon possessed more than four angles. These characteristics were discerned on the cropped image through the process of polygonal approximation, with both the area and the number of angles on the polygon computed. If the area of the polygon was smaller than that of the original cropped image and the polygon exhibited five or more angles, the cell was identified as having a concavity, thereby indicating that it was in the S phase (Fig. 3F).
Cells that did not fall within the G1 or S phases were categorized as being in the G2 phase. To measure their growth, the ratio of each cell’s length to that immediately following cell division was computed (Fig. 3G).
This method facilitated the tracking of alterations in the cell cycle of individual cells over time (Supplementary Movie S3).
Measuring intracellular fluorescent dots associated with the cell cycleCell imaging analysis has been amalgamated with the visualization of intracellular molecules, such as fluorescent proteins or fluorescence-labeled nucleic acids, to analyze their activities and intracellular localization. Our next objective was to develop a novel technique capable of simultaneously discerning fluorescent dots of intracellular molecules and cell cycle phases. We investigated the intranuclear presence of HP1, a central factor in heterochromatin regulation known for its role in repressing gene expression. Fission yeast Swi6, an ortholog of HP1, is a key factor in heterochromatin formation, with the ability to directly bind to chromatin to establish high-order chromatin structure (Nakayama et al., 2000, 2001). Furthermore, the localization of HP1/Swi6 varies throughout the cell cycle (Chen et al., 2008; Kloc et al., 2008): it is observed on a chromatin target sequence from the S to G2 phase, and then dissociates from chromatin, coinciding with increased H3S10 phosphorylation during the M phase (Fischle et al., 2005; Hirota et al., 2005; Chen et al., 2008; Kloc et al., 2008).
To concurrently evaluate intranuclear fluorescent dots of EGFP-fused Swi6 and the cell cycle phase, cell images underwent processing and segmentation to estimate the cell cycle stage of each individual cell. Simultaneously, a fluorescence image of Swi6 was taken. A bright field mask image for each cell was merged with the fluorescence image, and then fluorescent dots within each cell were measured (Fig. 4A, 4B). Regions of high brightness were interpreted as locations where Swi6 was bound. However, the pixel intensities varied between cells and frames. To address this, the maximum brightness for each image was ascertained, and all brightness values were normalized, with the maximum value set to 255. This procedure, while helpful, could amplify noise. To mitigate this effect, a median filter was applied to smooth the image, followed by binarization at an appropriate threshold. Throughout this process, regions with brightness values indicative of effective intensity were selected, and the connected components in the image were then labeled. If the area of Swi6 fluorescence exceeded a specific threshold, it signified the presence of Swi6 throughout the entire cell.
This software was used to measure the cell cycle and Swi6 protein while a single cell underwent multiple divisions, producing a total of 11 cells. The number of Swi6 dots within the cell or the frequency of their dispersion throughout the cell was determined in G2, S and G1 phases (C, Table 1). The binarization threshold for selecting fluorescent dots was set at 140, representing 55% of the maximum pixel brightness of 255. Values below 140 were removed.
| G2 | G2 (1.3-fold length ratio) | M | S | G1 | |
|---|---|---|---|---|---|
| No dots | 9.6% | 0% | 9.1% | 33.3% | 40.0% |
| 1 dot | 60.6% | 50.0% | 54.5% | 33.3% | 30.0% |
| 2 dots | 14.9% | 31.3% | 27.3% | 0% | 20.0% |
| 3 dots | 0% | 0% | 0% | 33.3% | 10.0% |
| spreading | 14.9% | 18.8% | 9.1% | 0% | 0% |
In the G2 phase, the fluorescent dots extracted by this method comprised single dots, accounting for 60.6%, followed by 14.9% with two dots and spreading, while none of these nuclei exhibited three or more dots. In the M phase, 54.5% displayed one dot, 27.3% exhibited two dots, while 9.1% revealed a spreading and no dots; again, none of the nuclei had more than three dots. Among the nuclei in the S phase, 33.3% each displayed three dots, one dot and no dots, with none presenting two dots or fluorescence spreading. In the G1 phase, distribution was as follows: 40% with no dots, 30% with one dot, 20% with two dots and 10% with three dots, and no fluorescence spreading was observed. The numbers of cells in the G1 and S phases were limited due to their brief durations, and spreading throughout the entire cell was not observed during these phases. Also, the presence of three dots was exclusively observed in the G1 and S phases. Within the G2 and M phases, we observed the disappearance of dots and a spreading of fluorescence. The disappearance of dots was not explored during the S and G1 phases, because it suggested de-clustering of HP1 from chromatin. These findings are consistent with observations indicating the disappearance of HP1 fluorescent dots during the M phase. The observation of three dots in the G1 and S phases may be attributed to the dual localization of Swi6p on the centromere and telomere of heterochromatin, as previously noted by Pidoux et al. (2000), who reported a typical range of 3–5 dots, with Swi6p dots persisting during mitosis. Nevertheless, they observed instances with 1–2 dots, spreading, and the disappearance of Swi6p. This inconsistency arose from the use of simple binarization to calculate dot numbers, which was not compatible with unprocessed fluorescence images. However, our algorithm demonstrated compatibility with this issue and provided a simplified analysis by applying an alternative algorithm such as signal amplification.
Our system facilitated the rapid establishment of a connection between cell function and the cell cycle by amalgamating cell tracking data derived from bright field images, cell cycle estimations, and fluorescence images.
The S. pombe strain TKYUF257 (h90, ade6-m216, his2-, leu1-32, swi6::Pswi6-EGFP-Swi6-TADH1-natMX6, ura4::Pura4-HTB1(Sc)-tdTomato-hphMX6) was used in the present study. Yeast cells were cultivated according to standard procedures (Moreno et al., 1991; Kajitani et al., 2017). C-terminal tagging (Bähler et al., 1998) was performed using PCR-based methods and the plasmid pFA6a-tdTomato-hphMX6 (Addgene).
Single-cell tracking experiments were performed as described (Mano et al., 2013; Kanada et al., 2020). Fission yeast cells were cultured on YES medium at 30 °C and collected in the logarithmic growth phase. Haploid fission yeast cells were trapped on CellASIC ONIX plates (4 chamber, 3.5–5 µm) and imaged using a CellASIC ONIX2 Microfluidic System (Merck) while running YES medium at a flow velocity of 2.0 psi. Images were taken with an Axis Observer Z1 (Carl Zeiss) microscope and a 40× Plan-Neofluar objective lens (NA = 1.3). Data for analysis were created from these images using ZEN 2.3 (blue edition) (Carl Zeiss).
Files necessary for analysisData for analysis were stored in three distinct folders: a BF folder for bright field images, an EGFP folder for EGFP fluorescence images and an mCherry folder for mCherry fluorescence images. Within each of these folders, files were systematically labeled with consecutive four-digit numbers, ensuring that files capturing frames simultaneously possessed identical file names.
C Sharp version and librariesC# 7.3 was employed within Visual Studio 2019 for our development. We utilized OpenCVSharp, a library based on the open-source OpenCV library, which had been converted to the C/C++ language to make it compatible with C#.
We would like to thank Dr. Jun-ichi Nakayama for providing the yeast strain.
