A Microimage-Processing-Based Technique for Detecting Qualitative and Quantitative Characteristics of Plant Cells

Abstract

When plants encounter external environmental stimuli, they can adapt to environmental changes through a complex network of metabolism–gene expression–metabolism within the plant cell. In this process, changes in the characteristics of plant cells are a phenotype that is responsive and directly linked to this network. Accurate identification of large numbers of plant cells and quantitative analysis of their cellular characteristics is a much-needed experiment for in-depth analysis of plant metabolism and gene expression. This study aimed to develop an automated, accurate, high-throughput quantitative analysis method, ACFVA, for single-plant-cell identification. ACFVA can quantitatively address a variety of biological questions for a large number of plant cells automatically, including standard assays (for example, cell localization, count, and size) and complex morphological assays (for example, different fluorescence in cells). Using ACFVA, phenomics studies can be carried out at the plant cellular level and then combined with ever-changing sequencing technologies to address plant molecular biology and synthetic biology from another direction.

1. Introduction

With the development of modern technology, increasingly more problems, including cleaning hazardous waste in inaccessible places [1], sensing chemicals dynamicallyand responding accordingly [2,3], and producing clean fuel [4], need novel and convenient approachesthat can rectify them. Synthetic biology has the potential to combine the investigative nature of biology with the constructive nature of engineering, thereby creating a novel intervention [4,5]. In recent years, there has been an increasing abundance of data on plant genomes. This has causedsingle-cell sequencing technologies to continue to mature, therebyenabling the satisfactory resolution of plant molecular mechanisms at the cellular level. For example, single-cell sequencing analysis of Arabidopsis protoplasts (root tips/leaves) tapped into the signaling and transcriptional regulatory networks among plant cells, revealing the responses of different plant cell types to changes in environmental signals. This also helped to provide a more complete spatio-temporal perspective for understanding plant cell differentiation, individual development, and physiological responses [6,7].

In addition to the classical terrestrial model plant (Arabidopsis thaliana), the comparative genomics and transcriptomics analysis of multiple algal species (unicellular/multicellular) has revealed a new perspective with regard to the importance of multicellular organisms in eukaryotic evolution [8]. These new theories have enhanced application of plant cells, such as Arabidopsis and Dunaliella cells, in creating and perfecting genetic devices and small modules. This will bring significant advances in synthetic biology [9,10]. Thus, research on the characterization of the cellular architectures of large numbers of plant cells (cellular phenomics) is essential for developing synthetic biology and fabricating practical organisms in a bid to solve problems. Examining plant cells by microscopy has long been a primary method for studying plants’ cellular function. Visual analysis can reveal biological mechanisms with the help of fluorescent light [11,12]. Currently, advanced microscopes can now easily collect thousands of high-resolution images of plant cells in a single day [13], which provides the necessary equipment base for the development of plant cellular phenomics. However, a bottleneck exists at the image analysis stage. Expert biologists have scored several pioneering large screens through visual inspection [14,15], whose interpretive ability will not soon be replicated by a computer. Still, for most applications, cell image analysis is strongly preferable to an analysis by eye. The latter is more labor-intensive and has lower throughput, possibly requiring months of tedious visual inspection. Therefore, there has been a pressing need for the identification and/or development of appropriate systems and software to produce reliable results from a large-scale microscopy picture in hours. Today, there are several commercial (for example, Imaris 9 [16], Volocity 6 [17], and Amira^TM [18]) and open-source (for instance, ImageJ 1.4 [19], CellProfiler 4.2 [20], Vaa3D 4.0.2 [21], BioImageXD X64 [22], Icy 4 [23], and Konstanz Information Miner (KNIME 4 [24])) software platforms designed for the analysis of biological microscopy images. Commercial platforms/software often focus on ease of use and broad coverage of image-processing tasks, and they target relatively unprofessional assignments.

It is important to note that the principal details of these image-processing algorithms are hidden, and this is an undesirable attribute in biological research. Conversely, these details are transparent in open-source platforms such as ImageJ, whose long existence, broad adoption, and extensible plug-in architecture have made it a tool of choice for scientists from various disciplines. Nevertheless, ImageJ was primarilydeveloped by biologists for biologists, and its architecture does not follow modern software-engineering principles. As a result, ImageJ is unable to engage in automatic processing [19]. To fully use modern technology in computers, several open-source image analysis platforms have been developed. These include the following: Vaa3D21, which can enhance visualization in three dimensions [21]; CellProfiler, which can achieve trainable segmentation of animal cells [20]; and Fiji, which can assemble various image-processing steps into complex pipelines using scripting languages [25]. However, none of the aforementioned software systems is designed for plant cell characteristics that are different from those of animal cells [26]). Moreover, results from the image analysis of plant cells using this software were inaccurate. This implies that, in practice, researchers still need to identify the characteristics of plant cells through human-scored analysis. However, human-scored image analysis is qualitative, usually categorizing samples as ‘hits’ (where normal physiology is grossly disturbed) or ‘non-hits, which would cause subtle samples to be missed [13,14]. By contrast, automated analysis rapidly produces consistent, quantitative measures for every image [27]. In addition, automated analysis of many cell micrographs allows useless or incorrect information to be filtered, thereby ensuringaccurate and reliable analysis results.

To integrate plant cytomics with other high-throughput sequencing data, the system for plant cytomics analysis should have the ability to run on a high-performance computing platform such as the Lunix system. The system parameters should also be flexible and easily adjustable, considering that the code is compiled in an open-source language such as Python. The accuracy of phenotyping and sequencing data determines the degree of refinement of the molecular mechanisms and regulatory networks revealed by the results [28]. Accuracy and refinement are essential in synthetic biology, where the emphasis is on engineering and conformity. This helps researchers to obtain systems-level conclusions from quantitative measures of a large number of features, even those that are undetectable by the eye, for every image [4,12].

In summary, while existing software and systems enable particular assays for certain cell types, high-throughput image analysis for plant cells has, up to this point, been impractical. The case is only slightly different where commercial packages are used with built-in algorithms for a limited set of cellular features and plant cell types, such as whole-stain leaves or roots [29,30]. There is a clear need for a robust, flexible, small-size (computational unit) platform for high-throughput plant cell image analysis. This will help to meet the need for high-throughput sequencing-dependent plant molecular biology studies.

This study describes a new method for developing a system for analyzing plant cell microscopy images. This method has been called the automated cell fluorescence values analysis method (ACFVA). ACFVA can simultaneously measure the localization, size, shape, and texture of various plant cell types in a high-throughput manner to collect raw data. Next, the raw data is filtered according to an internal reference to remove non-compliant values and retain clean data. Finally, the clean data is directly used in biostatistical analyses and visualization. Alternatively, it can be used as whole-phenome data that can be combined with other sequencing data to carry out more complex studies, such as GWAS. Therefore, ACVFA is well-suited to modern biology research as it allows the phenotypic characteristics of a large number of individual plant cells to be homogenized using a uniform standard before being presented as quantitative data. Please note that this paper does not focus on the technical details of the programs that have been developed in this system, nor on computational validation of the main published algorithms. It is also not based on a mechanistic study of any particular biological finding. Instead, the current study describes the system, validates the new method for a variety of real-world biological problems, and demonstrates the breadth of its utility (including for various plant cell types and assays). This study also hopes to stimulate ideas within the biological community for future applications of this new method, especially in molecular biology, epigenetics, and synthetic biology, where enormous data resources are involved.

2. Materials and Methods

2.1. Plant Materials and Vector Construction

The Arabidopsis ecotype Colombia (Col-0) was used as the plant material for protoplast isolation. After surface sterilization in 30% bleach, Arabidopsis seeds were kept at 4 °C for 48 h before sowing. The seeds were grown in vitro on half-strength Murashige Skoog (Sigma Co. Ltd., Shanghai, China) with 0.5% sucrose media (pH 5.7, 1.2% agar) in a growth chamber (20 °C) under white light (120 μmol m⁻² s⁻¹ photons) in 16 h light/day photoperiods for 10 days and then collected for protoplast isolation. The full-length cDNA of AtVDAC3 was amplified and cloned into a p1301-35S-EGFP vector (F: 5’ ATGGTTAAAGGTCCAGGACTCT 3’; R:5’ GGGCTTGAGAGCGAGAGCAATC3’) as well as approximately 1 kb rd29A promoter (F:5’CATTTAGACCTTATCGGAATT3’; R:5’ TTTCCAAAGATTTTTTTCTTT 3’) linked with the full-length cDNA of AtRD29A (F: 5’ ATGGATCAAACAGAGGAACC 3’; R:5’AAGCTCCTTCTGCACCGG3’) was amplified and cloned into a p1301-EGFP vector. Dunaliella cells were cultivated according to the protocol as previously described [31]. All plant material was kept in the laboratory of Jianghan University (Wuhan, China).

2.2. Arabidopsis Protoplast Isolation and Transfection Assays

Arabidopsis young leaves were cut into small pieces (width: 0.5 cm; length: 2 cm) and then completely submerged into digestion enzyme buffer (1.5% Cellulase R10, 0.4% Macerozyme R10, 0.4 M Mannitol, 20 mM KCl, 20 mM MES, pH 5.7, 10 mM CaCl2, and 0.1% BSA) for 6 h without light. After the protoplasts were released, enough W5 solution (154 mM NaCl, 125 mM CaCl2, 5 mM KCl, 2 mM MES, pH 5.7) was added to stop the digestion reaction. The protoplasts were centrifuged at 100× g for 10 min, washed twice with 25 mL pre-chilled W5 solution, and incubated on ice for 30 min. The protoplasts were centrifuged and resuspended in MMg solution (0.2 M mannitol, 15 mM MgCl2, and 4 mM MES, pH 5.7) to a final concentration of 2 to 5 × 105 cells/mL.

Protoplasts were transfected by a modified method from Yoo’s report [32]. Approximately 5 × 104 protoplasts (2 × 104 to 1 × 105) in 0.1 mL of MMg solution were mixed with approximately 10 (10 to 20) μg of plasmid DNA at room temperature. A quantity of 0.11 mL of a freshly prepared solution of 40% (v/v) PEG (MW 4000; FlukaCo. Ltd., Shanghai, China) with 0.1 M CaCl₂, enough carrier DNA, and 0.2 M mannitol was added, and the mixture was incubated at room temperature for 5–15 min. After incubation, 0.8 mL of W5 solution was added slowly, the solution was mixed, and protoplasts were pelleted by centrifugation at 100× g for 10 min. The protoplasts were resuspended gently in 1 mL of W5 and were incubated in 6-well plates coated at room temperature for 12–16 h in light. Flow cytometry and cell staining assays were used to detect the status of protoplasts before further analysis.

2.3. Confocal Laser Scanning Microscopy

Protoplasts were observed with a Leica TCS SP8 laser scanning confocal microscope using HC 10×/0.4 CS Plan-Apochromat, HC 20×/0.7 CS Plan-Apochromat, 40×/0.85 CS Plan-Apochromat or 63×/1.4 Oil Plan-Apochromat in multi-track channel mode. Excitation wavelengths and emission filters were 488 nm/band-pass 505–530 nm for GFP and 488 nm/band-pass 650–710 nm for chloroplast auto-fluorescence. In order to avoid errors caused by high laser energy intensities, the excitation wavelength energy for 488 nm was 0.5 mW, which was measured by a Gigahertz-optik PT9610 photometer at the end of the objective lens.

2.4. ACFVA Analysis Pipeline

All of the program was written in Python to enable integration with other existing data analysis methods and to facilitate the subsequent addition of other functionality. The user does not need to download any other attachments.

The system is divided into three parts: the first part utilizes the binary method and contour algorithm for image recognition and collects raw data. These algorithms are not limited to identifying cells based on fluorescence. They are applicable even if the image contains non-fluorescent cells such as paraffin slides. In most biological images, cells touch each other, causing the simple, fast algorithms used in some commercial software packages to fail. The Python-integrated OpenCV2 module can apply “Erosion” and “Dilation” to the image to separate contact cells easily. When the pixel standard of the image is sufficient, this software has no requirements for the dispersion of the main object. The only limitation of using this method is that the pixels of the images must achieve the requirements, which can be solved by turning up the camera resolution. After necessary preprocessing, such as Gaussian denoising, flood filling, contrast enhancement, impurity filling, etc., images can undergo gray processing according to the equation: $Gray=\sqrt[2.2]{\frac{R^{2.2}+{\left(1.5G\right)}^{2.2}+{\left(0.6B\right)}^{2.2}}{1+{1.5}^{2.2}+{0.6}^{2.2}}}$ [33]. Preprocessing of images includes, but is not limited to, the above steps. Grayscale is a conventional method in image processing. Compared with color images, grayscale images occupy less RAM and compute faster. Moreover, the grayscale of the image can increase visual contrast and highlight the target area. There are many grayscale methods, such as the maximum method, average method, weighted average method, etc. These algorithms all have errors in their calculation values, because these methods are not suitable for gamma-corrected images (the commonly seen 24-bit true-color images are all gamma-corrected images [34]). Therefore, we adopted the gamma correction algorithm to obtain the grayscale image. The second step is to binarize the preprocessed grayscale image. Through the adaptive threshold method, the grayscale image will be converted into a binary image. The binary image contains only two pixel values, 0 and 1, and is a pure black-and-white two-color image. The target of interest in the image will be white in this article. As a final step in cell identification, the contour recognition module of OpenCV2 will detect the cell contours in the binary image based on the contour algorithm [34], which can further reduce the amount of computational data compared with the edge detection algorithm based on the canny operator used by ImageJ. The conversion of grayscale and fluorescence values is part of the core technology of one of the patents in another project of the Smart Manufacturing team collaborating on this project (which is not readily available for publication), and the general explanation is that both grayscale and fluorescence within the contour range are converted to an intermediate value, which is used for subsequent pixel calculations. A simple scheme using one Arabidopsis protoplast image’s step-by-step results is also provided in Figure S1). A series of eigenvalues for the cells can be correspondingly captured in the original image according to the identified contours, such as fluorescence value, cell area, cell relative position, etc. Fluorescence is obtained by calculating the average gray level of all the pixels in the contour area; cell area is equal to the total area of the contour surrounding area; cell relative position marks the position of the first pixel in the upper left corner of the contour; etc. Although both contouring algorithms are used (and are also used by ImageJ for image recognition), the specific steps, step sequences, and step parameters in image recognition differ from one image to another. The ACFVA system has been accurately optimized for the recognition of non-fused single-cell images (protoplasts) of plants in the above respects, including, but not limited to, adjustment of the relevant parameters, and optimization for the recognition of images of fused cells (paraffin slices) is also in progress.

The second part of ACFVA is to select suitable internal references from the various features identified in the raw data and determine the internal reference thresholds, filter the raw data according to the internal reference thresholds, and obtain clean data for subsequent analysis. Through data filtering, we can ensure that the further analyzed cell phenotype data are all biologically meaningful, avoiding the technical errors caused by the intersection of disciplines. The choice of internal reference can be determined according to the purpose of the experiment, following the principle of relatively stable features in cell micrographs under different conditions, such as cell diameter, chloroplast fluorescence, and so on. In order to achieve intelligent analysis of a large number of plant cell images, the internal reference indicators, in addition to accurately evaluating the state of the plant cells, need to be easily accessible and not be dependent on the equipment and image resolution. For example, in this study, the internal references for Arabidopsis protoplast cells include cell diameter, ellipticity, and chloroplast fluorescence intensity.

There can be one or more internal reference indicators, and the weights of these indicators need to be analyzed statistically if more than one internal reference exists. In order to correlate with high-throughput sequencing data, the entropy weighting method (a common statistical method used in high-throughput sequencing data) was chosen to calculate three crucial internal reference indexes, and the training data for the formula calculations were provided from other related projects in this group. The third part of ACFVA is to perform the necessary biological statistical analysis and visualization of the clean data. Given that the initial goal of ACFVA development was to obtain a large amount of plant cell phenotypic data matching high-throughput sequencing data, statistical analysis and visualization procedures were interfaced with the existing high-throughput sequencing analysis system of the group. Of course, biostatistical analysis and visualization programs based only on cellular phenomes are also under development.

In summary, the pipeline of the ACVFA system is shown in Figure 1. With the ACVFA system, intelligent analysis of cellular phenomics based on large numbers of cell microscopy or tissue sections (including image recognition, data filtering, and data analysis) can be realized. In order to make this system more widely used, the three frameworks of the whole system can be adjusted and optimized appropriately (e.g., adding new steps or adjusting the parameters of the existing steps, etc.) because of the differences in the recognized images.

2.5. Image Recognition and Data Processing

Besides ACFVA, two commonly used biological image analysis software packages, ImageJ [19] and CellProfiler [20], are being studied for cell phenome analysis. Cell phenotypes were analyzed according to the official instructions for both programs. Although batching is possible in the recently developed plug-in for ImageJ through the writing of relevant scripts [25], this feature was not applicable in this study, considering that this requires a certain level of competence on the part of the user, which is not in line with the purpose of this study. The details of usage are listed in Table S2.

3. Results and Discussion

3.1. Suitable Prepossessing Is Necessary for ACVFA to Recognize Different Plant Cells

This study first demonstrated that ACFVA could accurately measure different biologically essential features of cell suspensions. This was done using several plant cell types, including live Arabidopsis protoplasts and Dunaliella cells, because they are both used as model organisms in plant molecular biology and metabolism studies. These cells are preferable because of their small genomes and detailed genomic information, which can provide an optimal research system for the study of plant growth and development, energy substance metabolism, and biosynthesis [4,26]. Nevertheless, identifying these cells using automated image analysis is challenging [30,35], and this has hindered the development of plant cell phenomics and the application of plant cells in synthetic biology, among other areas. Using the basic cell-culture methods described previously [31,36], Arabidopsis protoplasts and Dunaliella cells were prepared for experiments that are shown in this study. Direct comparison of image analysis methods is difficult because results from image analysis can be heavily skewed by how the software or method is tuned. On the other hand, commercial software packages are numerous and expensive. Furthermore, commercial software algorithms are proprietary and cannot be directly compared apart from the entire software package, including image preprocessing methods. Therefore, the best practical comparison is for the image analysis strategy to release results of these methods for the same image. For plant live-cell imaging analysis, in which living cells are imaged using phase contrast and/or fluorescence microscopy, image segmentation of single cells is necessary, as it gives crucial results for microscopy images, including cell counts, cell localization, and cell size. This also works in animal-cell microscopy image analysis [27,37].

Cell count, which is a straightforward phenotype, is used to probe cell proliferation/apoptosis/death in cytological research. As Arabidopsis protoplasts are a novel and convenient chassis cell in synthetic biology [10], three concentrations (low/medium/high) of Arabidopsis protoplasts were chosen as living plant cells for imaging using confocal microscopes. Microscopy images under brightfield were used for microscopy image analysis comparisons because this is the most convenient way to determine the state of the cells. For methods used to analyze the comparison of cell microscopy images with ACFVA, two classic image analysis programs were chosen: one was imageJ, a classic image recognition system that was developed based on Java and requires manual operation (two testers in this study); the other was CellProfiler, an image recognition software package explicitly developed for batch processing of medical cell pictures. For low concentrations of cells, there is no noticeable difference in cell count (30, 29, 45, 29 cells per image) using the various methods. However, the cell numbers counted by CellProfiler were much higher (>two-fold with medium and >20-fold with high cell concentration) when the number of cells was relatively larger. With both medium and high concentrations of cells, ACFVA and ImageJ exhibited similar results (Figure 2A). This indicates that picture segmentation and object recognition are the most challenging steps in automatic microscopy image analysis. The accuracy of these two steps determines that of the resulting cell measurements. This has proved to be challenging for many existing programs due to their poor ability to separate diverse plant cells with huge morphological and structural variations among different plant species, tissues, and cell types [38,39]. Considering how critical and challenging image segmentation and object recognition are, proper preprocessing is required when the pictures are very informative (medium- and high-concentration Arabidopsis protoplast maps) [40].

In characterizing cell morphology as a measure of cell health, cell size is also a fundamental quantitative metric. The ability to measure cell size is a direct indicator of the accuracy of an image recognition system [41,42]. For all cell concentrations, there was no difference in the cell size of Arabidopsis protoplasts, which ranged between 20–40 nm, as determined by ACFVA and ImageJ (Figure 2B). The findings were both consistent with those from previous reports [36]. Only at a low concentration was the identification of plant cells by CellProfiler accurate. It is only applied in measuring cell size at low concentrations. Unlike ACFVA, which gives information including cell localization and cell diameter for a single cell, CellProfiler just gave us three results: 10th-percentile diameter (the average diameter of the top 10% of small cells) was 13.6 μm; medium diameter (the average diameter of all the cells) was 20.1 μm; and 90th-percentile diameter (the average diameter of the top 10% of large cells) was 31.3 μm. To further compare these two automatic methods, the same results, 22.92 μm, 28.66 μm, and 34.48 μm, were determined with ACFVA. Although there was no significant difference (<2-fold) among the diameters measured by the two methods, the average diameter of the top 10% of small cells from CellProfiler was smaller than the standard diameter of the Arabidopsis protoplasts. Moreover, the correlation coefficient R² also proved that the diameters obtained by our new method were more stable and accurate than with CellProfiler (Figure 2C). Although Arabidopsis protoplasts were round or oval in shape, they also varied greatly in size, but when the number of cells is too large and overlapping, image recognition algorithms suitable for animal cells, such as CellProfiler, are unable to accurately segment and extract features from the image (Figure 2). Considering that the identification of plant cells by CellProfiler was only accurate at low concentrations, it is only applied in measuring cell size at low cell concentrations. Unlike ACFVA, which gives information that includes cell localization and cell diameter for a single cell, CellProfiler only yielded three results: the 10th-percentile diameter (the average diameter of the top 10% of small cells) was 13.6 μm; the medium diameter (average diameter of all the cells) was 20.1 μm; and the 90th-percentile diameter (the average diameter of the top 10% of large cells) was 31.3 μm. To further compare these two automatic methods, the same results, which were 22.92 μm, 28.66 μm, and 34.48 μm, were determined by ACFVA. Although there was no significant difference (<2-fold) among the diameters measured by the two methods, the 10th-percentile diameter provided by CellProfiler was smaller than with ACFVA. Moreover, the correlation coefficient R² also proved that the diameters obtained by ACFVA were more stable and accurate than those obtained by CellProfiler (Figure 2C). Although Arabidopsis protoplasts were round or oval in shape, they also greatly varied in size. However, when the number of cells is too large and overlapping, image recognition algorithms that are suitable for animal cells, which include CellProfiler, are unable to accurately segment and extract features from the image (Figure 2).

To test whether ACFVA could automatically rapidly identify a wide range of single plant cells, a microscopy image of Dunaliella suspension cells, a micrograph of paraffin sections of cells from cowpea pods, and a low-resolution microscopy image of plant single cells randomly downloaded from the Internet [37] were also analyzed using different methods. The cell numbers that were automatically counted by ACFVA in the three types of plant cells were almost identical to those from ImageJ (<1.1-fold). Due to the fact that the cell numbers in the two pictures were not small enough (>50), the cell numbers from CellProfiler were much higher (>100-fold for Dunaliella cells and >17-fold for other plant cells) compared with other methods (Figure 3). The significant errors in the Dunaliella and cowpea-pod cell numbers that were counted by CellProfiler may be due to their irregular cell shapes and contaminants, which are similar to the target cells in some ways. The cell diameters of cells in this study ranged from 10.26 to 11.48 μm, 11.13 to 23.69 μm, and 117.53 to 1141.70 μm, all of which were in line with previous reports. Consistent with the aforementioned results, the R² of the cell diameters of these three types of cells automatically measured by CellProfiler was lower than with ACFVA (Figure 3A–C).

In addition to comparisons of the accuracy of the different methods for image recognition, those for the time taken to analyze the images in batches by different methods were also made. To avoid technical errors, all the comparisons were done on the same computer with a Windows system. The same 10 images of plant cells, a set of five images of Arabidopsis protoplasts and another five of cowpea pods, were analyzed comparing between ACFVA and CellProfiler in this study, considering the user’s ability to write scripts. Disregarding the parameter-setting time, the time to process images for a single plant cell image is essentially the same for both software programs, and both are less than 30S (Figure 3D). CellProfiler ‘s image processing steps and parameter settings give the user a great deal of freedom, but also require that the user has a certain degree of knowledge of image recognition algorithms. Moreover, the whole process needs 5 to 10 min to set up the settings for each item. Unlike ACFVA (written in Python), which has no system dependency and can be run on PCs (Windows and MacOS systems) and servers (Lunix system, Table S2), so far, CellProfiler has not been able to run on Lunix-based servers, which makes it unsuitable for cellular phenotyping.

Compared with current biological image-processing systems, with suitable preprocessing for plant cells, ACFVA can automatically (no complex parameter settings), rapidly (the analysis was run on a desktop computer at a rate of >1 image/minute), and precisely (including cell localization, cell size, and cell shape) identify various single plant cells.

3.2. Rational Data Filtering Allows the Broad Applicability of ACFVA with Plant Cells

Plant cells such as Arabidopsis protoplasts are ideal as chassis cells in synthetic biology, mainly because they are eukaryotic, allowing exogenous DNA molecules to enter the chassis cell and be encoded correctly. They are also cellularly totipotent, with complete energy and metabolic pathways. Arabidopsis protoplasts are simple and economical as far as preparation is concerned, compared with animal cells. Therefore, in this study, ACFVA was applied to perform transfection efficiency calculations and detect environmental change (high-salinity stress). These are two common but useful analyses in molecular and synthetic biology using Arabidopsis protoplasts and they were done after demonstrating the ability of ACFVA to accurately identify plant cells and measure a large number of relevant phenotypes.

Gene transfection is a widely used technique in molecular studies and could considerably impact subsequent experiments. Accurately calculating the transfection efficiency of cells is a necessary and essential pre-requisite for most biological research. Counting the ratio of cells with positive fluorescent light to all cells is a direct method for calculating transfection efficiency. In this study, ACFVA was applied to compare the transfection efficiency of two systems using Arabidopsis protoplasts. This study compared the transfection efficiency of ‘with-carrier DNA’ and ‘without-carrier DNA’ systems using the same vector, 35S:AtVDA3-EGFP, in Arabidopsis protoplasts. AtVDA3 was reportedly involved in metabolite exchange between the organelle and the cytosol, prominently localized in the outer mitochondrial membrane, chloroplast, and nucleolus. After data filtering by ACFVA, 54 cells in ‘with-carrier DNA’ and 84 cells in ‘without-carrier DNA’ systems were identified as qualified Arabidopsis protoplasts for further analysis. Unlike animal cells, healthy plant cells have strong and stable chloroplast fluorescence, induced by a specific wavelength of the right energy. This can be easily recorded by a microscope. Like animal cells, plant cells’ cellular states (redox systems) can also be compared by fluorescent staining, though the states of plant protoplasts in different systems can easily be compared using the intensity of chloroplast fluorescence without any additional treatment (staining). The consistency of cell states among different systems (background consistency) is critical for subsequent experiments, which helps make plant protoplasts versatile chassis cells in synthetic biology. There were no significant differences in fluorescent light in the chloroplasts measured in the two systems (Figure 4A), suggesting that the cells in these two systems were in almost the same healthy state. Moreover, both are appropriate for gene transfection. Therefore, the ratio of EGFP-qualified protoplasts (positive EGFP and chloroplasts are shown in Figure 4A) to total qualified protoplasts (BF and chloroplasts are shown in Figure 4A) in each system can represent its transfection efficiency, respectively. After rational data filtering with ACFVA, there were 49 positive cells in ‘with-carrier DNA’ and 12 positive cells in ‘without-carrier DNA’ systems. The EGFP fluorescent-light-positive cell rates in the two systems were almost identical (Figure 4A), indicating that the host cell activity and positive cell activity of these two Arabidopsis protoplast transfection systems were consistent and that transfection efficiency was the main difference. Based on the ACFVA analysis, the transfection efficiencies were 90.74% and 14.29% in ‘with-carrier DNA’ and ‘without-carrier DNA’ systems, respectively.

The findings were consistent with previous reports suggesting that carrier DNA can improve the efficiency of gene transfection [43]. It should be noted that, without the data filtering process of ACFVA, the numbers of cells in the brightfield images of the two systems were 57 (with carrier DNA) and 89 (without carrier DNA), respectively. However, the numbers of cells in the EGFP images were 61 (with carrier DNA) and 37 (without carrier DNA), respectively, and the transformation efficiencies were 107.02% (with carrier DNA) and 41.57% (without carrier DNA), respectively. All these data suggest that data filtering was necessary for ACFVA with Arabidopsis protoplasts to analyze gene transfection efficiency in a more rapid and accurate manner, further enhancing its application in studying the effects of external chemicals on plant development and metabolism. Moreover, synthetic biology research requires biological models that can rapidly and accurately sense changes such as chemical stress in the external environment. The development of modern gene editing techniques and fluorescent tags enables fluorescent signals from biological models to be used in detecting changes in the external environment [3]. Various external stresses, such as NaCl, dehydration, ABA, and cold treatment can lead to the synergistic activation of Responsive-to-Dehydration 29A (RD29A), which encodes a hydrophilic protein [44] whose function is unknown [45].

In this study, we constructed a biological model using Arabidopsis protoplasts with high transfection efficiency as chassis cells, Arabidopsis RD29A as the sensor, and fluorescent tagging with EGFP as a reporter to detect the exit of NaCl in an external environment (250 μm NaCl). No significant differences were observed between the two treatments with regard to chloroplast fluorescence, which was also detected to balance the state of chassis cells (Figure 4B). These findings prove that the biological model was strong enough to detect external NaCl signals. It is important to note that the changes in reporter fluorescence were caused by NaCl, not cell activity. The reporter fluorescence (EGFP) significantly increased when NaCl was present in the external environment (Figure 4B). This confirms the accurate and rapid response to NaCl of the biological model combined with ACFVA, as presented in the current study.

After omitting data filtering, the results using ACFVA showed that the fluorescence values for EGFP in NaCl ranged from 10.82 to 98.32, with a mean value of 48.33. Contrastingly, the fluorescence values of EGFP in DMSO ranged between 7.34 and 72.41, with a mean value of 20.71. An analysis of significance showed that there was only a slight difference between the two sets of results (0.01 < p < 0.05). The R² values of the NaCl concentration calibration curves (R² = 0.9817 and R² = 0.7358) drawn by different vectors combined with ACFVA (R² = 0.9817 of AtRD29A, R² = 0.7358 of 35S) also demonstrate the accuracy of ACFVA combined with suitable components (RD29A promoter) for the detection of NaCl (Figure 4C, up). The NaCl concentration measured using the RD29A-calibration curve in combination with the fluorescence detected by ACFVA was almost in agreement with the results as measured by the salinometer (Sodium Analyzer Easy, Figure 4C, down). The reliability of the significance analysis is vital for determining the sensitivity of plant cells with regard to responding to external environmental stimuli. This analysis is also necessary for successfully applying the synthetic biology module. For example, it can be applied in monitoring salt concentration levels in wastewater [46,47]. High-salinity water poses hazards to the environment and affects agriculture, infrastructure, and communication under seawater. Salinity is one of the most critical variables in ocean monitoring, the marine environment, seasonal weather forecasting, aquaculture, and solar engineering. Therefore, an effective method for sensing salt solutions has been a much sought-for application in many fields, such as in agriculture [48], public health [49], and environmental management [50]. Many techniques have been proposed to measure salt concentrations, and these include optical techniques [51], infrared attenuated total reflection spectroscopy [52], microwave sensing [53], and bio-chemical sensing [54]. This study demonstrated that ACFVA, in combination with Arabidopsis protoplast cells, can quickly, intuitively, and accurately find liquids with NaCl. This means that monitoring chemicals in the environment using plant cells can be realized with this system. Compared with conventional chassis organisms, plant protoplasts are economical and easily accessible [10], On the other hand, ACFVA can quickly analyze large amounts of information from plant protoplasts, thereby improving accuracy through extensive data analysis [55,56].

4. Conclusions

In this study, we presented a new, automated, and quantitative image analysis method for plant cells. Automated image analysis for plant cells via ACFVA is faster when compared with previously described methods that rely on manual calibration. It is also more accurate than computational methods that are only suitable for animal cells. The methods developed in this work should facilitate quantitative analysis of a large number of plant cell microscope images (cellular phenomics) that can be combined with a variety of high-throughput sequencing genomics data. The results that are obtained using this ACFVA method can be applied for answering various synthetic biological questions [57].