MitoCellPhe reveals mitochondrial morphologies in single fibroblasts and clustered stem cells

Abstract

Mitochondria are dynamic organelles that differ significantly in their morphologies across cell types, reflecting specific cellular needs and stages in development. Despite the wide biological significance in disease and in health, delineating mitochondrial morphologies in complex systems remains challenging. Here, we present the Mitochondrial Cellular Phenotype (MitoCellPhe) tool developed for quantifying mitochondrial morphologies and demonstrate its utility in delineating differences in mitochondrial morphologies in a human fibroblast and human induced pluripotent stem cell (hiPSC) line. MitoCellPhe generates 24 parameters, allowing for a comprehensive analysis of mitochondrial structures and importantly allows for quantification to be performed on mitochondria in images containing single cells or clusters of cells. With this tool, we were able to validate previous findings that show networks of mitochondria in healthy fibroblast cell lines and a more fragmented morphology in hiPSCs. Using images generated from control and diseased fibroblasts and hiPSCs, we also demonstrate the efficacy of the toolset in delineating differences in morphologies between healthy and the diseased state in both stem cell (hiPSC) and differentiated fibroblast cells. Our results demonstrate that MitoCellPhe enables high-throughput, sensitive, detailed, and quantitative mitochondrial morphological assessment and thus enables better biological insights into mitochondrial dynamics in health and disease.

INTRODUCTION

Mitochondria are essential organelles responsible for a host of cellular functions, with a primary objective of meeting the cellular bioenergetics demands via the production of adenosine triphosphate (ATP) (1). The functioning of the mitochondria is intimately linked to dramatic modifications to its structure and remodeling that occurs within (2–5). The general morphology of the mitochondria is organized into an elaborate network that dynamically interconverts through frequent rounds of fission and fusion events that are tightly regulated to control overall morphology (4–8). The dynamic nature of the mitochondria varies with cellular needs, developmental stage, and tissue type, with the maintenance of structural integrity a necessary feature for mitochondria to function effectively. Perturbations in mitochondrial morphology occur in many human diseases including those associated with myopathies (9, 10), neurodegeneration (11–19), cancer (20), diabetes (21–23), and a host of other disorders. However, it is important to note that it is not clear if the mitochondrial morphological changes are correlative, rather than causative. Owing to the involvement of mitochondrial dynamics in regulating cellular functions, studies have focused on understanding the relationship between the morphology and function of the mitochondria in health and different disease processes.

Quantifying mitochondrial morphologies is challenging given the significant heterogeneity in mitochondrial length and degree of branching. The heterogeneity has thus resulted in diverse nomenclature and classification (e.g., fragmented, short, elongated, tubular, long, reticular, interconnected, fused, hyperfused, aggregated) and used to define the mitochondrial network. It is thus important to take into consideration the complexity that exists in the mitochondrial network within a single cell and accordingly assign correct nomenclature and classification to ensure that we capture the range of potential morphologies. Over the past decade, various image-processing algorithms have been developed to quantify mitochondrial morphology. Specific attempts have focused on applying morphometric image analysis tools to mitochondria using Image J or Image Pro Plus 5.1 (Media Cybernetics) to measure numerous mitochondrial parameters such as volume, length, number of nodes, etc., using fluorescent images (24–36). To fully realize the potential of image analysis platforms to generate large data sets that require high-throughput, automated, mitochondria-specific skeletonization, and bulk analysis of fluorescent images, we sought to develop a tool that would analyze fluorescently labeled mitochondria in images containing single cells or clusters of cells.

In this article, we introduce the mitochondrial-cellular phenotype (MitoCellPhe) tool, a software with a graphical user interface that makes it easier to quantify the morphology of the mitochondria. Specifically, MitoCellPhe does not require prior programming knowledge, can be easily adapted by any laboratory, and is flexible in allowing users to analyze batch or single images. Furthermore, this software provides a host of parameters that can provide better information to quantify and analyze morphological differences between mitochondria in different types of cells. Here, we use two different cell types to compare mitochondrial morphological characteristics generated by MitoCellPhe. We used the tool to characterize mitochondrial morphological differences in a sample of human induced pluripotent stem cells (hiPSC) and human fibroblast cell lines, which contain immature and mature mitochondria, respectively. We also tested the utility of the tool to delineate differences in mitochondrial morphologies between normal and diseased fibroblasts and hiPSCs. Our results show that MitoCellPhe can generate accurate skeletons and provide comprehensive information that allows users to characterize morphological differences on a single-cell level and at a cluster level. Furthermore, MitoCellPhe allows for fast and efficient analysis of high-throughput data and demonstrates the potential for identifying complex morphologies within the cell, and provides a powerful resource for identifying aberrant structures that might be involved in disease development.

MATERIALS AND METHODS

Fibroblast Cell Culture

Cultures of human BJ (ATCC CRL-2522) fibroblasts were obtained from the American Type Culture Collection (ATCC, Manassas, VA). The SBG1 (T8993G) diseased fibroblast was obtained from the Medical University of Salzburg, Austria. These cells were maintained in a fibroblast expansion medium that consisted of minimal essential medium (MEM) (Thermo Fisher Scientific, Waltham, MA), 10% fetal bovine serum (FBS) (GE Healthcare – HyClone, Chicago, IL), and 2 mM l-glutamine. Fibroblasts were enzymatically passaged in 0.05% trypsin-EDTA (Thermo Fisher Scientific). All cell cultures were maintained without the use of antibiotics, handled in Biosafety Type II sterile hoods regularly cleaned with UV irradiation and 70% ethanol, and grown in 37°C incubators at 5% CO₂ and 95% humidity. The culture medium was replenished every 2 days until cells became 80% confluent. Before use in experimentation at passage 8, cells were dissociated using 0.05% trypsin-EDTA (Thermo Fisher Scientific) and 20,000 cells were seeded into 35 mm dishes for fluorescence labeling and image analysis.

hiPSC Culture

Once reprogrammed from the BJ and SBG1-(T8993G) fibroblasts, hiPSCs were maintained in NutriStem hPSC xeno-free (XF) medium (Biological Industries, Cromwell, CT) with Stemolecule Y27632 Dihydrochloride Hydrate (Reprocell, Beltsville, MD) on a highly purified and refined laminin-511 E8 fragment matrix, iMatrix 511(Reprocell) on a 24-h feeding schedule. hiPSCs were enzymatically passaged once reaching 70%–80% confluency at a split ratio of 1:3 using StemPro Accutase (Thermo Fisher Scientific). All cell cultures were maintained without the use of antibiotics, handled in Biosafety Type II sterile hoods regularly cleaned with UV irradiation and 70% ethanol, and grown in 37°C incubators at 5% CO₂ and 95% humidity. When the cells were 70% confluent, they were passaged for fluorescence labeling and image analysis, as detailed in Fluorescence Labeling of Mitochondria. The fluorescence labeling and image analysis were performed on hiPSCs at passage 11 in culture.

Fluorescence Labeling of Mitochondria

To label the mitochondria, fibroblast cells were incubated with MEM NEAA basal medium (Thermo Fisher Scientific) containing 150 nM Mitotracker Red CM-H₂Xros (Invitrogen) for 30 min. In the FCCP treatment group, fibroblast cells were incubated with 0.7 µM FCCP for 30 min before the addition of the Mitotracker Red CM-H2Xros. At the end of the incubation period, the cells were washed three times with prewarmed Dulbecco’s phosphate-buffered saline (dPBS). The nucleus was stained by further incubating cells with basal medium containing Nucblue Hoechst (Invitrogen) for 15 min. Following this incubation, cells were washed several times with prewarmed dPBS to remove excess dye. At the end of the wash, MEM NEAA basal medium was added to each dish before image acquisition.

To visualize mitochondrial morphology in the hiPSCs, 10⁵ cells were seeded in a 35 mm precoated culture dish. The cells were incubated for at least 48 h before staining. Subsequently, a 100 nM solution of Mitotracker Red CMXRos (Invitrogen) prepared in serum-free culture medium was added to stain the cells in a 37°C incubator at 5% CO₂ and 95% humidity. At the end of the incubation period, the cells were washed 3× times with prewarmed Dulbecco’s phosphate-buffered saline (dPBS). The nucleus was stained by further incubating cells with basal medium containing Nucblue Hoechst (Invitrogen) for 15 min. Following this incubation, cells were washed several times with prewarmed dPBS to remove excess dye. At the end of the wash, MEM NEAA basal medium was added to each dish before image acquisition.

Live-Cell Fluorescence Microscopy

Fluorescence images of live cells were acquired using an EVOS FL inverted light/epifluorescence microscope with a 40×/0.65 objective and a Sony ICX445 monochrome CCD digital camera. Red fluorescence from Mitotracker Red CM-H₂Xros was measured using a 530 nM excitation and a 593 nM emission filter set. Blue fluorescence from Nucblue Hoechst was measured using a 360 nM excitation and a 447 nM emission filter set. All live cells were imaged on 35 mm dishes containing phenol-red free basal medium. Image acquisition was performed one dish at a time with a maximum time of 30 min between dishes. All dishes were stored in a humidified 37°C, 5% CO₂ incubator until image acquisition. All images of live cells were taken on the same day as the labeling of mitochondria. All live-cell images were exported as TIFF files for further analysis. Five to ten images were acquired per dish and three dishes were stained per trial. At least three independent trials were performed for both the fibroblast and hiPSC lines.

Mitochondrial Morphology Analysis

In this section, we discuss the underlying algorithms for the MitoCellPhe Skeletonizer and Analyzer. MitoCellPhe Skeletonizer skeletonizes a cell slide image and optionally segments it into individual cells (for noncluster analysis). It is implemented as a pipeline for the CellProfiler program. MitoCellPhe Analyzer analyzes skeletonized images by calculating 24 parameters that describe the structure of the mitochondrial networks in the skeleton and is implemented in ImageJ.

Cell skeletonization and segmentation (MitoCellPhe skeletonizer).

To analyze a raw cell image, the image must first be skeletonized. Skeletonization allows binary images to be reduced to one-pixel width. In addition, if the goal of the analysis is to evaluate individual cells, the image must then be segmented into regions that define individual cells. The methods are discussed in subsequent sections.

Image preprocessing.

Before skeletonizing a cell image, the image must be preprocessed to improve the quality of the skeleton. The image preprocessing consists of converting the stained sample image to skeletonized images for each cell detected. The input image is a three-color channel image displaying the applied red mitochondrial stain. To perform the morphology analysis, a binary mitochondrion skeleton image representing the stained sample image is required. The input image X = (x_R, x_G, x_B) is combined to form a grayscale image using Y = norm (x_R + x_G + x_B, 0,1).

The grayscale mitochondria images contain significant variation in pixel intensity across regions of the image. Therefore, to prepare the stained image for skeletonization, we used contrast limited adaptive histogram equalization (CLAHE) illumination correction (37) with a 10×10 kernel size, a clip limit of 0.02, and 256 bins. The small kernel size helps to even the brightness of the mitochondria. To generate the binary image for skeletonization, the CLAHE output is thresholded using a 2-class adaptive Otsu thresholding method. This method works well on stained images that have been CLAHE normalized (38). Afterward, the morphological skeleton can be generated.

Skeletonization.

The Otsu-thresholded image is then used as the start image for the skeletonization process. The skeletonization of the image is accomplished with Zhang’s method (39), which performs successive passes over the thresholded image to detect boundary pixels and whittle them down to one-pixel-wide branches. The segmentation process, if desired, can split the skeletonized image into separate regions for each cell in the image.

Segmentation.

If the stained sample image contains multiple cells, and they are not clustered close together (i.e., hiPSC lines), then MitoCellPhe can additionally segment the sample image into multiple images of each of its cells. Such cell types are called differentiated cell types, and those that are clustered are called undifferentiated cell types. For undifferentiated cell types, an aggregate morphology analysis can be performed without using segmentation. When run in MitoCellPhe Analyzer, the tool will perform morphology analysis on the entire image. For differentiated cell types, the images can be used directly to create object segments. Using the grayscale nonskeletonized image Y, the image is denoised by suppressing small features and subsequently blurred using a Gaussian filter to smooth the pixel intensity in the stain region. This helps the object detection recognize individual cells and reduce oversegmentation. The object detection algorithm uses the CellProfiler IdentifyPrimaryObjects module. This method does not require nuclei stain images or phase contrast cell images.

Mitochondrial morphology analysis (MitoCellPhe analyzer).

The morphology analysis can be computed on object morphological skeletons or a single aggregate morphological skeleton representing the stained sample. The morphology analysis can be run on the batches of morphological skeletons by specifying the root directory. The output contains the file names along with each of the computed parameters and is given in Table 1.

Table 1.

MitoCellPhe parameter table

Value	Definition	Parameter Evaluation
Skeleton area	The area of the skeleton within the image (i.e., the number of pixels the skeleton takes, converted to micrometers)	${\displaystyle \sum X}$
Punctate count	Number of skeletal components with zero branches (i.e., single points)	P = |{s ∈ S: |Es| = 0}|
Rod count	Number of skeletal components with one branch (i.e., straight lines)	R = |{s ∈ S: |Es| = 1}|
Network count	Number of skeletal segments with two or more branches	N = |{s ∈ S: |Es| ≥ 2}|
Punctate percentage	The ratio of punctate components to the total number of components	$\frac{100P}{P+R+N}$
Rod percentage	The ratio of rod components to the total number of components	$\frac{100R}{P+R+N}$
Network percentage	The ratio of network components to the total number of components	$\frac{100N}{P+R+N}$
Punctate length	Always zero	Wp = 0
Rod length	The length of the branch of a rod	wr
Network length	The sum of all branch lengths contained within a network	$w_n={\displaystyle \sum _{En}{w}_e}$
Mean rod length	The average length of one-branch skeletal segments	${\overline{w}}_r=\frac{1}{\left|R\right|}{\displaystyle \sum _r{w}_r}$
Median rod length	The median length of one-branch skeletal segments	$w_r^{\mathit{med}}={\mathrm{med}}_r(w_r{)}_0$
SD rod length	Population standard deviation of one-branch skeletal segments	$w_r^{\sigma }={\left(\frac{{\displaystyle {\sum }_r{\left(w_r-\overline{r}\right)}^2}}{\left|R\right|}\right)}^{1/2}$
Total network branch count	Number of branches contained in all networks	${\displaystyle \sum _n\left|E_n\right|}$
Mean network branch count	The average number of branches for each network	$\frac{1}{N}{\displaystyle \sum _n\left|E_n\right|}$
Mean network branch length	Average branch length for each branch in every network	${\overline{w}}_{en}=\frac{1}{{\displaystyle \sum _n\left|E_n\right|}}{\displaystyle \sum _n{\displaystyle \sum _{e\in E_n}{w}_e}}$
Mean network length	Average of every network’s length in an image	${\overline{w}}_n=\frac{1}{\left|N\right|}{\displaystyle \sum _n{\displaystyle \sum _{e\in E_n}{w}_e}}$
Median network length	Median of every network’s length in an image	$w_n^{\mathit{med}}={\mathrm{med}}_n\left(w_n\right)$
SD network length	Population standard deviation of every network’s length in an image	$w_n^{\sigma }={\left(\frac{{\displaystyle {\sum }_n{\left(w_n-{\overline{w}}_n\right)}^2}}{\left|N\right|}\right)}^{1/2}$
All branch count	Number of branches contained in networks or rods	${\displaystyle \sum _n\left|E_n\right|}+\left|R\right|$
Mean length of all branches	Average branch length for each branch in every network and rod	${\overline{w}}_z=\frac{{\displaystyle \sum _z{\displaystyle \sum _{e\in E_z}{w}_e}}}{{\displaystyle \sum _z\left|E_z\right|}}$
Median length of all branches	Median branch length for every branch in every network and rod	$w_z^{\mathit{med}}={\text{med}}_z\left(w_z\right)$
SD length of all branches	Population standard deviation of the branch length for every branch in every network and rod	$w_z^{\sigma }={\left(\frac{{\displaystyle \sum _z{\left(w_z-{\overline{w}}_z\right)}^2}}{{\displaystyle \sum _z\left|E_z\right|}}\right)}^{1/2}$
Mean network and rod length	The average length of all networks and rods	${\overline{w}}_{nr}=\frac{{\displaystyle \sum _z{\displaystyle \sum _{e\in E_z}{w}_e}}}{\left|N\right|+\left|R\right|}$
Median network and rod length	The median length of all networks and rods	$w_{\mathrm{nr}}^{\mathit{med}}={\text{med}}_{\mathrm{nr}}\left({\displaystyle \sum _{e\in E_z}{w}_e}\right)$
SD network and rod length	Population standard deviation of the length of all networks and rods	$w_z^{\sigma }={\left(\frac{{\displaystyle \sum _z{\left({\displaystyle \sum _{e\in E_z}{w}_e}-w_{nr}\right)}^2}}{\left|N\right|+\left|R\right|}\right)}^{1/2}$

The definition and calculations for the different parameters are generated by the MitoCellPhe analyzer.

A mitochondrial skeleton consists of one-pixel-wide segments, and can be modeled as a graph where each line segment is an edge and endpoints are vertices. The skeleton graph S contains disconnected components s which represents the topology of the mitochondrial networks. A component s has edges E_s, edge weights E_s, and vertices Vs. The edge weight w_e of an edge e ∈ E_s, is equal to the pixel length of edge e in the skeleton. For each component s ∈ S, we differentiate between three different topologies: punctuates, rods, and networks.

A component s ∈ S is a punctate if |E_s| = 0, a rod if |E_s| = 1, and a network if it |E_s| > 1, where |E_s| is the cardinality of the set. In Table 1, to represent the network component of the skeleton graph S and make the notation simpler, we partition S into three sets, punctates p ∈ P, rods r ∈ R, and networks n ∈ N. Then, we make a fourth set of network and rod components, Z = N ∪ R to compute statistics on all edges of the graph.

Statistical Analysis

To ensure scientific rigor, a total of 79 cells across four biological replicates were used for the analysis related to human fibroblasts. Since fibroblast cell lines have different cellular morphologies, the parameters generated were normalized by the cell surface area. Unlike the fibroblast cell lines, the hiPSC lines were not segmented. One hundred images across five biological replicates were used for the analyses related to hiPSCs. Since hiPSCs prefer to grow in colonies, normalization was applied to the resulting data according to the surface area measured and the number of nuclei in these areas. The data for the different parameters are presented as the means ± standard deviation (SD) and were generated using MS Excel 2016. Post hoc Tukey’s HSD test was used to identify differences among specific groups. Data are presented as the means ± standard deviation (SD) and were analyzed using the GraphPad Prism 8 program (GraphPad Software, San Diego, CA). A P < 0.05 was considered significant.

Data and Tool Availability

The data supporting the findings of this study are available within the article and from the authors on request. All codes, manuals, and related files associated with the operation of MitoCellPhe will be released through the website (bioenergetics.uark.edu).

RESULTS

Descriptors of Mitochondrial Morphology

Understanding mitochondrial morphology is important in determining the condition of the cell with descriptors such as tubular fragmented and hyperfused (40) often used to characterize mitochondrial morphology. Other descriptors such as individuals and networks are also used to identify networks of mitochondria (34, 35) and are important in helping to identify mitochondrial morphology in different cell types. In general, these descriptors are invaluable in making predictions on mitochondrial dynamics as it relates to cellular health. Fragmented mitochondria seem to be the predominant morphology observed in mitochondrial dysfunctions (41, 42), whereas fused mitochondria are associated with cell survival mechanisms (43, 44). It is therefore imperative that we can identify these different morphologies to further delineate their contributions to cellular health and disease. MitoCellPhe provides us with the ability to capture these morphological differences with descriptors (Table 1 and Fig. 1) that are more comprehensive. MitoCellPhe allows us to differentiate between rods, punctate, and branched structures; and recognizes three distinct object types: punctate (no branches), rods (one branch), network (two or more branches) (Fig. 1). MitoCellPhe computes 24 different values that can be used in evaluating mitochondrial dynamics and thus provides investigators with more information to make predictions on mitochondrial dynamics. The more comprehensive nature of MitoCellPhe thus allows us to dig deeper into the morphological differences of the mitochondria between different cell types.

Figure 1.

Mitochondrial morphology descriptors. The mitochondrial structures are classified into three groups: punctates have no branches, rods contain one branch, and networks consist of structures with two or more branches that meet at a junction pixel. A branch is defined as a path from one pixel to another (i.e., Pixel A to Pixel B). Scale bar = 100 µM.

MitoCellPhe Generates Detailed Skeletons

In addition to having more detailed descriptors for mitochondrial morphology, the steps used in MitoCellPhe allow for preprocessing of images, to generate more details in the skeletons (Fig. 2). We adapted previously described steps (45) to generate the skeletons. These steps use an efficient three-dimensional (3-D) parallel thinning algorithm to extract both the medial surfaces and medial axes of a 3-D object and preserve the topological and geometric conditions of the object. The product is a desirable skeleton (Fig. 3) that is more sensitive to noise and increases speed. The accuracy of the skeletons is important for generating the parameters described in Table 1–as skeletons that are not representative of the original image can skew the results obtained.

Figure 2.

MitoCellPhe flowchart. Flowchart showing the steps from image input to generation of skeletons and the parameter tables for differentiated (fibroblast; A) and undifferentiated (hiPSC) cells (B). The images for the fibroblast cell lines had to be processed to generate segmented skeletons, whereas the images for the hiPSCs did not undergo any segmentation. hiPSC, human induced pluripotent stem cell.

Figure 3.

Skeletonization of differentiated and undifferentiated cell lines. Representative images of differentiated (fibroblast) and undifferentiated (hiPSC) cell lines showing the steps involved in skeletonization. The differentiated cell lines were stained with 150 nM Mitotracker red CM-H₂Xros, which stains actively respiring mitochondria, whereas the undifferentiated cells were stained with 100 nM Mitotracker red CMXros dye. Scale bar = 100 µM. hiPSC, human induced pluripotent stem cell.

Human Fibroblasts Display Extensive Networks

To determine the sensitivity of MitoCellPhe, we used a healthy human fibroblast cell line (Fig. 4), as they are differentiated somatic cell lines with mature mitochondria, usually characterized by elongated and networks of mitochondria (43). Therefore, we expect to observe parameters consistent with this morphology. Indeed, we observed higher rod and network counts relative to punctate (Table 2). On average, 81% of branched structures contribute to the mitochondrial morphology in the healthy fibroblast cell lines. Even when we analyzed individual cells, ∼85%–96% of the mitochondrial structure are branched compared with the ∼4%–15% with punctate structures. Furthermore, we observe that rods are the most predominant branched structure, contributing ∼60% to the structure of the mitochondria in healthy fibroblast cell lines. Healthy mitochondria should exhibit a balance between fission and fusion, a balance observed in the punctate and network counts (Table 2). Our observations are consistent with what has been predicted for fibroblast cell lines with healthy mitochondria (40). However, it is important to note that there are slight differences when we make comparisons between segmented cells. For example, the punctate counts in cell2 and cell3 are almost three times that of cell1 (Table 2). This highlights the power of MitoCellPhe, as one can use this tool to identify differences in the mitochondrial structure at a single cell and population level.

Figure 4.

Mitochondrial dynamics of healthy differentiated cell lines. Representative image of differentiated fibroblast cell line showing the steps involved in skeletonization and segmentation. The segmented skeletons’ analysis and results are reported in Table 2. Scale bar = 100 µM.

Table 2.

Mitochondrial morphology data of differentiated cell lines

Value	Combined	Segments
Counts		Cell1	Cell2	Cell3
Punctate count (P)	0.00469 ± 0.00301	0.00149	0.00444	0.00445
Rod count (R)	0.0142 ± 0.00452	0.0254	0.0210	0.0190
Network count (N)	0.00507 ± 0.00197	0.00721	0.00755	0.00632
Mean network branch count (MNB)	0.00321 ± 0.00203	0.00242	0.00147	0.00108
Total network branch count (TNB)	0.0764 ± 0.0188	0.0702	0.0749	0.0585
All branch count (R+TNB)	0.0906 ± 0.0198	0.0955	0.959	0.0775
Count percentages
Percentage P, %	19.05 ± 9.08	4.38	13.45	14.96
Percentage R, %	59.96 ± 9.54	74.45	63.68	63.78
Percentage N, %	20.98 ± 4.58	21.17	22.87	21.26
Size of the skeleton
Skeleton area, µM2	0.0346 ± 0.00656	0.0334	0.0336	0.0255
Lengths
Punctate length, µM	0.00 ± 0.00	0.00	0.00	0.00
Rods
Mean rod length, µM	0.00167 ± 0.000781	0.00186	0.00119	0.000722
SD rod length, µM	0.00190 ± 0.00109	0.001737	0.001056	0.000676
Median rod length, µM	0.00104 ± 0.000484	0.00135	0.000863	0.000507
Networks
Mean network length, µM	0.00683 ± 0.00531	0.00443	0.00266	0.00193
SD network length, µM	0.0148 ± 0.0101	0.00580	0.00478	0.00503
Median network length, µM	0.00193 ± 0.00131	0.00267	0.00107	0.000968
Network branches
Mean network branch length, µM	0.000379 ± 0.000135	0.000456	0.000268	0.000209
Mean length of all branches, µM	0.000375 ± 0.000135	0.000442	0.000266	0.000196
SD length of all branches, µM	0.000341 ± 0.000146	0.000379	0.000203	0.000148
Median length of all branches, µM	0.000276 ± 0.0000949	0.000332	0.000203	0.000158
Branches (rods + networks)
Mean network and rod length, µM	0.00197 ± 0.00125	0.00130	0.000894	0.000601
Median network and rod length, µM	0.000366 ± 0.000167	0.000445	0.000278	0.000163
SD network and rod length µM	0.00793 ± 0.00514	0.00322	0.00269	0.00264

The values for different mitochondria parameters are reported by MitoCellPhe. On the left, the means ± SD values are reported for all cells, whereas the right shows values for the segmented skeletons ad shown in Fig. 4. All data have been normalized to the surface area of their respective cells. The combined data is an average of 79-segmented cells.

Undifferentiated Human Induced Pluripotent Stem Cells Have Smaller Networks of Mitochondria

We next examined mitochondrial morphology in a healthy undifferentiated human induced pluripotent stem cell (hiPSC) line derived from the human fibroblast cell line described above. Previous studies have reported that reprogramming results in mitochondrial remodeling with more fragmentation observed in the hiPSCs relative to differentiated cell lines (46, 47). We chose both of these cell lines (human fibroblasts and hiPSCs) to highlight changes in mitochondrial morphology observed in cells during different stages of development. Unlike human fibroblasts, hiPSCs grow in colonies, making it particularly challenging to segment the cells for the same comparisons as the human fibroblast cell lines. Instead, batch analysis of images of hiPSCs was conducted (Fig. 5) and a comparative analysis with human fibroblasts was done.

Figure 5.

Mitochondrial dynamics of healthy undifferentiated hiPSC lines. A representative image of undifferentiated hiPSC lines showing the skeleton derived from the MTR fluorescent image. The clustered skeletons’ analysis and results are reported in Table 3. Scale bar = 100 µM. hiPSC, human induced pluripotent stem cell.

Similar to what was observed in the human fibroblast cell line, an increase in the number of rods relative to punctate and networks was observed in the hiPSCs (Table 3). This result suggests that the healthy undifferentiated hiPSCs consist of both networks and fragments of mitochondria. When compared with the differentiated fibroblast, however, the hiPSCs have more fragmented mitochondria. In the human fibroblast cell line, punctate comprise only 19% of the overall mitochondrial structure, whereas it contributes to 23% in the hiPSC. Furthermore, the human fibroblast cell line has 81% of branched structures contributing to the overall mitochondrial morphology, compared with the 76% we observe in the hiPSCs (Tables 2 and 3). Our result is consistent with previous reports that suggest that during reprogramming to hiPSCs, mitochondria undergo remodeling to adopt a more fragmented morphology (46–48). The predominant morphologies are long rods, which could play a role in biogenesis and maintain cell survival (43, 44). This could also imply that even in undifferentiated cell lines like the hiPSCs, a healthy balance between fission and fusion is necessary.

Table 3.

Mitochondrial morphology data of undifferentiated BJ-hiPSCs

Value	Data Average of 100 images
Counts
Punctate count (P)	7.04 ± 3.26
Rod count (R)	16.68 ± 5.53
Network count (N)	6.94 ± 1.85
Mean network branch count (MNB)	0.53 ± 0.25
Total network branch count (TNB)	93.70 ± 25.95
All branch count (R+TNB)	110.16 ± 29.97
Count percentages
Percentage P, %	23.10 ± 8.25
Percentage R, %	53.33 ± 8.31
Percentage N, %	22.56 ± 6.02
Size of the skeleton
Skeleton area, µM2	35.70 ± 9.37
Lengths
Punctate length, µM	0.00 ± 0.00
Rods
Mean rod length, µM	0.22 ± 0.11
SD rod length, µM	0.26 ± 0.13
Median rod length, µM	0.13 ± 0.065
Networks
Mean network length, µM	0.96 ± 0.46
SD network length, µM	1.54 ± 0.81
Median network length, µM	0.38 ± 0.18
Network branches
Mean network branch length, µM	0.070 ± 0.032
Mean length of all branches, µM	0.067 ± 0.030
SD length of all branches, µM	0.053 ± 0.030
Median length of all branches, µM	0.052 ± 0.023
Branches (rods + networks)
Mean network and rod length, µM	0.32 ± 0.15
SD network and rod length, µM	0.96 ± 0.49
Median network and rod length, µM	0.054 ± 0.030

The values for different mitochondria parameters are reported by MitoCellPhe. The means ± SD values are reported for 100 images of healthy control BJ-hiPSCs. All data have been normalized to the surface area of their respective images and cells in each image. The number of cells was determined from the counts of Nucblue-stained nucleus. hiPSCs, human induced pluripotent stem cell.

MitoCellPhe Reveals Fragmented Mitochondrial Phenotype in Human Fibroblasts with Pathogenic mtDNA Mutation (T8993G) Modeling Leigh Syndrome

Having confirmed that MitoCellPhe can generate detailed skeletons, which allows for quantification of the mitochondrial morphology in healthy differentiated and undifferentiated cell lines, we proceeded to analyze differences between healthy control and one diseased fibroblast cell line carrying mtDNA mutation impacting subunit of the Complex V-ATP synthase. We analyzed the healthy BJ control and one patient fibroblast cell line (SBG1-FB: T8993G) carrying one of the most prevalent pathogenic mtDNA mutation in patients with Leigh syndrome (LS) (49, 50). We analyzed mitochondrial morphology in the diseased SBG1-(T8993G)-FB under basal conditions and after FCCP treatment. We have shown previously that the mutation in SBG1-(T8993G)-FB results in a fragmented mitochondrial morphology (51). Using MitoCellPhe, we were able to confirm that SBG1-(T8993G)-FB exhibit trends toward slightly fragmented mitochondria relative to the healthy control (BJ-FB) (Fig. 6). This is evident by the significant decrease (P < 0.05) in mean rod length (by 17.55%) (Fig. 6F) and mean network length (by 11.41%) (Fig. 6G) in the SBG1-(T8993G)-FB relative to the healthy control BJ-FB. Furthermore, the mean length of all branches (Fig. 6H) was lower by 13.06% in the SBG1-(T8993G)-FB relative to the healthy control BJ-FB. There is also a slight increase in punctate and network count (smaller networks) (Fig. 6, B and D) in SBG1-(T8993G)-FB compared with the healthy control BJ-FB, further supporting the presence of fragmentation in this cell line. In addition to having fragmented mitochondria, the SBG1-(T8993G)-FB appeared to have fewer mitochondria overall relative to the healthy control. This is supported by the trend towards lower skeleton area (by 7.45%) (Fig. 6A), rod count (by 3.08%) (Fig. 6C), and mean network branch count (by 16.30%) (Fig. 6E) in the SBG1–(T8993G)-FB relative to the healthy control BJ-FB.

Figure 6.

Mitochondrial morphology of healthy control BJ-FB and diseased SBG1-(T8993G) FB in the absence and presence of FCCP. The mitochondria in the SBG1-(T8993G) FB lines are smaller than those of the control fibroblast. This is evident in the decrease in skeleton area (A), slight increase in punctate count (B), decrease in rod count (C), decrease in network count (D), decrease in mean network branch count (E), and decrease in mean rod length, mean network length, and mean lengths of all branches relative to control under basal conditions, without FCCP treatment (F–H). Treatment with FCCP resulted in mitochondrial fission, with SBG1-FB (T8993G) showing similar response to FCCP like the healthy control. All data are representative of 10–14 images taken from three independent dishes/treatment group. The bars represent minimum and maximal values, and each black dot represents different data points. The dark and light gray bars represent the control fibroblast without and with FCCP treatment (−FCCP vs. +FCCP). The red and pink bars represent the SBG1-FB (T8993G) without and with FCCP treatment (−FCCP vs. +FCCP). ****P < 0.0001, **P < 0.01, *P < 0.05.

Next, we analyzed mitochondrial morphologies in both BJ-FB and SBG1-(T8993G)-FB cell lines under coupled and uncoupled respiration conditions, wherein trifluoromethoxy carbonyl cyanide phenylhydrazone (FCCP), a mitochondrial uncoupler, was used to induce maximal respiration. FCCP treatment promotes mitochondrial depolarization, subsequently resulting in the fragmentation of mitochondrial networks (52). We predicted that healthy BJ-FB cells should be able to rapidly remodel their mitochondria to maintain energy homeostasis. This is evident by an increase in the punctate, rod, network, and all branch counts (Fig. 6, B–D, Supplemental Fig. S1B; see https://doi.org/10.6084/m9.figshare.15060813), and a subsequent decrease in the mean network branch counts, mean rod length, mean network length, mean length of all branches, mean network branch length, and mean network and rod length (Fig. 6, E–H, Supplemental Fig. S1, F and G). Our results support this prediction, as we recorded a significant (P < 0.05) trends toward changes in these parameters in both the healthy control BJ-FB and diseased SBG1-(T8993G)-FB after treatment with FCCP. FCCP treatment resulted in fragmentation, as is evident by significantly (P < 0.05) higher counts of punctate, rods, and networks (Fig. 6, B–D), suggesting that depolarization by FCCP is causing remodeling of the mitochondria in both the diseased SBG1-(T8993G)-FB and the healthy control BJ-FB. Furthermore, the decrease in mean network branch count, mean rod length, mean network length, and mean length of all branches (Fig. 6, E–H) suggests that the rods and networks are fragmenting into smaller structures as the mitochondria adjust to the stress induced by FCCP in both cell lines. It is worth pointing out that the overall mass of the mitochondria does not change after FCCP treatment, as the skeleton area in both the control and the SBG1- (T8993G)-FB remained indistinguishable when comparing between the FCCP treated and untreated groups (Fig. 6A).

MitoCellPhe provides us with other very important details as well. For instance, we are able to identify based on the percentages (Supplemental Fig. S1, C–E) of each structures (punctate, rods, networks) that the rod structure (Supplemental Fig. S1D) is the most predominant morphology in both the healthy control BJ-FB and diseased SBG1-(T8993G)-FB. The rod comprises 60.30% and 59.30% of the mitochondrial structure in the healthy control BJ-FB and diseased SBG1-(T8993G)-FB cell lines, respectively. This perhaps ensures that during remodeling the rods can either be fragmented into smaller rods and punctates or fuse to form large networks. MitoCellPhe also provides more details on the branches within a network, as we observe that although FCCP results in decrease to the mean network branch count (Fig. 6E), it does not affect the aggregate amount of branches in the network (Supplemental Fig. S1A). Taken together, the comprehensive output derived from MitoCellPhe helped us identify fragmentation as the predominant morphology in the SBG1-(T8993G)-FB. Furthermore, we were able to show that treatment with FCCP resulted in mitochondrial remodeling in both the healthy control BJ-FB and diseased SBG1-(T8993G)-FB cell lines, similar to what we have reported previously (51).

MitoCellPhe Reveals Increase in Networks and Decrease in Punctates in Human iPSCs with Pathogenic mtDNA Mutation (T8993G) Modeling Leigh Syndrome

Having determined the utility of the MitoCellPhe toolset to detect different mitochondrial morphologies in healthy control BJ-hiPSCs (Table 3), we decided to analyze differences between healthy control BJ-hiPSC and diseased SBG1-(T8993G)-hiPSCs. As was done with the control BJ-hiPSCs, we stained the diseased SBG1-(T8993G)-hiPSCs, with the mitochondrial specific stain (MitoTracker Red), obtained fluorescent images, skeletonized, and analyzed the images. Our analysis demonstrated that the SBG1-(T8993G)-hiPSCs exhibit a statistically significant increased skeletal area (denoting a higher mitochondrial occupying area) (18.96%, P < 0.0001) when compared with control BJ-hiPSCs (Fig. 7A). Based on our analysis, we also determined that mitochondrial morphologies in BJ-hiPSCs contained 21.75% punctates (Fig. 7B, Supplemental Fig. S2C; see https://doi.org/10.6084/m9.figshare.15060918), 55.37% rods (Fig. 7C, Supplemental Fig. S2D), 22.89% networks (Fig. 7D, Supplemental Fig. S2E), whereas SBG1-(T8993G)-hiPSCs contained 18.89% punctates (Fig. 7B, Supplemental Fig. S2C), 56.30% rods (Fig. 7C, Supplemental Fig. S2D), and 24.81% networks (Fig. 7D, Supplemental Fig. S2E).

Figure 7.

Mitochondrial morphology of SBG1-(T8993G) hiPSCs. Different mitochondrial morphological parameters were determined and analyzed in comparison with the BJ-hiPSC (control cell line) to quantify mitochondrial skeletal area occupied by the mitochondria (A), punctate count (B), rod count (C), network count (D), mean network branch count (E), mean rod length (F), mean network length (G), and mean length of all branches (H). All data are representative of five to seven analyzed images obtained from seven to nine independent dishes from three independent experiments. The bars represent minimum and maximal values including all points, and each black dot represents different data points. The gray bars represent the BJ-control hiPSC, whereas the pink bars represent the SBG1-(T8993G) hiPSC. ****P < 0.0001. hiPSC, human induced pluripotent stem cell.

In SBG1-(T8993G)-hiPSCs, we observed a statistically significant increase in the network percentage (8.39%, P < 0.001) (Supplemental Fig. S2E), mean network branch count (68.3%, P < 0.0001) (Fig. 7E), mean rod length (63.22%, P < 0.0001) (Fig. 7F), mean network length (76.22%, P < 0.0001) (Fig. 7G), mean length of all branches (75.04%, P < 0.0001) (Fig. 7H), mean network branch length (75.42%, P < 0.0001) (Supplemental Fig. S2F), and mean network and rod length (88.5%, P < 0.0001) (Supplemental Fig. S2G) when compared with control BJ-hiPSCs. Although not statistically significant, we observed a 3.12% increase in network count (Fig. 7D), 5.32% increase in total network branch count (Supplemental Fig. S2A), 4.74% increase in all branches (networks + rods) count (Supplemental Fig. S2B), a 1.69% increase in rod percentage (Supplemental Fig. S2D) in SBG1-(T8993G)-hiPSCs when compared with control BJ-hiPSCs. We also observed a statistically significant decrease in punctate count (17.37%, P < 0.0001) (Fig. 7B), punctate percentage (13.14%, P < 0.0001) (Supplemental Fig. S2C), in SBG1-(T8993G)-hiPSCs when compared with control BJ-hiPSCs. Although not statistically significant, we observed a 3.25% decrease in rod counts (Fig. 7C) in SBG1-(T8993G)-hiPSCs when compared with control BJ-hiPSCs. Overall, these observations related to the presence of more mitochondria (defined as sum of punctates, networks, rods) with significant decrease in punctates and significant increase in networks. Decrease in punctates implies reduced fragmentation in the diseased SBG1-(T8993G)-hiPSCs when compared with control BJ-hiPSCs. Although we did not observe a significant change in rod counts and rod percentage, we observed a significant increase in mean rod length (63.22%) and combined mean network and rod length (88.5%), which indicates that the diseased SBG1-(T8993G)-hiPSCs were tending towards fused mitochondria. To further understand the complexities of the structure of the mitochondria in the diseased state, we estimated the nature of the branches. The significant increase in mean network branch count (68.3%) in conjunction with significant increase in mean length of all branches (75.04%) and the significant increase in mean length of branches in the network (75.42%) indicates that mitochondria have coalesced into a complex network structure. The presence of the hyperfused state is an indication of abnormal mitochondrial phenotypes (53), which could thus indicate that the diseased SBG1-(T8993G)-hiPSCs exhibit bimodal dependence on OxPhos and glycolysis that may affect their proliferation and self-renewal potential(54). Taken together, the comprehensive output provided by MitoCellPhe assisted in identifying the subtle differences that exist in the pluripotent state and more importantly between healthy control BJ-hiPSCs and diseased SBG1-(T8993G)-hiPSCs.

DISCUSSION

Mitochondria are dynamic organelles that undergo various remodeling processes to meet specific cellular needs (2). Understanding the differences between the structures of the mitochondria in healthy and diseased cells can enable us to target specific therapies for various disorders that result from mitochondrial dysfunction. Equally important is the ability to delineate differences in mitochondrial morphologies during the continuum of human development from an early stage embryo to specialized cells/tissues during the differentiation process. In this study, we demonstrate that MitoCellPhe is a useful tool for quantifying mitochondrial dynamics in a single or cluster of cells. The automated segmentation allows for further analysis of important mitochondrial structures on a single-cell level. This feature provides the user the advantage of making a comparative analysis of mitochondrial structures within a cell population and among a population of cells. Furthermore, MitoCellPhe allows for high-throughput batch processing and generates 24 features that can be useful for characterizing mitochondrial morphology. The comprehensive nature of the parameters computed for each sample can be useful for disease classification, drug safety testing, or intercellular communication. This simple graphical user interface is user friendly and does not require the user to have prior programming knowledge or strong technical background in computer science.

In this study, we were able to show that MitoCellPhe can generate skeletons that recapitulate the structures present in the fluorescent images (Figs. 4 and 5). The skeletonization of the fluorescent image must not result in the loss of important details, as this can affect the output. As is noted in the methods section, we have developed the skeletonizer and analyzer as two different applications. In the context of usage, having two separate applications provides users options. Specifically, if users already have other tools for generating skeletons from their fluorescent images, the analyzer platform can easily incorporate the skeletonized images to generate an output. In the future, we will combine these two applications to streamline the process.

As shown in our study, MitoCellPhe can identify mitochondrial structures and provide an accurate snapshot of the mitochondrial morphology in differentiated and undifferentiated cell lines (Tables 2 and 3). It was able to identify the predominant tubular and fragmented structure characteristics of differentiated and undifferentiated cell lines, respectively. During the process of reprogramming from fibroblasts, studies have shown that hiPSCs undergo a metabolic shift from oxidative phosphorylation to glycolysis (55, 56). This metabolic shift leads to physiological changes that correspond with the fragmentation of mitochondria to adapt to the new metabolic system. Therefore, the mitochondrial morphology converts from a branched connected elongated network into clustered, fragmented mitochondria (46–48, 55). In our analysis, MitoCellPhe was able to capture this dynamic change in mitochondrial structure between the differentiated fibroblast and undifferentiated naïve hiPSCs. Our results showed a higher punctate percentage (23% hiPSCs vs. 19% fibroblasts) and lower network percentages (53% hiPSCs vs. 60% fibroblasts) in our study (Tables 2 and 3). In addition, testing the efficacy of the tool in two different experimental scenarios demonstrated the utility of MitoCellPhe in deciphering differences between healthy control BJ-fibroblasts and diseased SBG1-(T8993G)-fibroblast (Fig. 6, Supplemental Fig. S1) and between healthy control BJ-hiPSC and diseased SBG1-(T8993G)-hiPSC (Fig. 7, Supplemental Fig. S2). Results demonstrate the significant potential for ascertaining differences in mitochondrial morphologies and the ability to correlate with biological differences. In future studies, when hiPSCs will be subjected to specific directed differentiation strategies (e.g., neural, cardiac, immune), MitoCellPhe can be employed to study mitochondrial remodeling that promotes differentiation into various cell lineages. In addition, we can also use MitoCellPhe to study mitochondrial dynamics in various patient- and disease-specific hiPSCs and their differentiated derivatives that correspond to mitochondrial disorders created in our laboratory (57).

Although MitoCellPhe was developed to measure mitochondrial dynamics, there are other potential applications to monitor other organelles within the cell. We expect that it could be used to study other organelles and dynamic structures that contribute to cellular health and integrity. For example, the lysosomes are dynamic organelles that also undergo remodeling into a tubular network when macrophages and dendritic cells become activated (58, 59). Lysosomes and mitochondria also work together to induce autophagy and mitophagy—processes that help with the clearance of cellular waste (60). By staining cells with dyes specific to both mitochondria and lysosomes, MitoCellPhe can provide us with further details on the structural remodeling that occurs in both mitochondria and lysosomes to promote autophagy and mitophagy. For example, the information gained from evaluating the structural remodeling because of changes in both mitochondria and lysosomes could contribute to the development of precision therapies to target cancer cells and promote the clearance of tumor cells. Aside from the lysosome, the mitochondria also interact with the endoplasmic reticulum. Mitochondria-endoplasmic reticulum contacts (MERCs) have been implicated in aging and aging-related diseases (61), cancer (62), neurodegenerative disease (63), and cardiovascular diseases (64). These studies have shown that MERCs regulate mitochondrial dynamics and contribute to cellular health. In the future, we envision expanding the use of MitoCellPhe to studying other subcellular structures such as the MERCs and lysosomes to identify how changes in these structures contribute to health and disease. Future work will include combining machine learning with the MitoCellPhe application to make predictions about mitochondrial diseases based on mitochondrial morphology and function. We would also make the process more automated and expand it to multiple cell types so that other research laboratories can use this tool with very little technical expertise. Finally, we note that the current implantation only works with 2-D cultures, as both the skeletonization and analysis algorithms assume a 2-D image. Specifically, for generating 3-D skeletons, either some additional preprocessing or the use of a new algorithm is required, as Lee’s algorithm requires a 2-D image as input. One could hypothetically separate the 3-D image as a stack of 2-D images and perform skeletonization on each layer, but this is not natively supported in the current version. For the analysis, since we are using Fiji/ImageJ as our backend, and Fiji/ImageJ can handle 3-D imagery, it is possible to adapt the analysis script to natively handle 3-D imagery. We will thus expand the use of MitoCellPhe to three-dimensional images in the future to test its efficacy on three-dimensional skeletonization. MitoCellPhe thus holds the potential as a less invasive diagnostic tool for identifying various metabolic disorders.

SUPPLEMENTAL DATA

Supplemental Fig. S1: https://doi.org/10.6084/m9.figshare.15060813.

Supplemental Fig. S2: https://doi.org/10.6084/m9.figshare.15060918.

GRANTS

This research was supported in part by funding from DoD W81XWH-16-1-0181 (to S. Iyer).

DISCLAIMERS

The funders had no role in study design, data collection, and analysis, decision to publish, or preparation of the article.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.B.B. and S.I. conceived and designed research; A.B.B., F.M., B.L., C.M., R.R.R., and S.I. performed experiments; A.B.B., F.M., B.L., C.M., R.R.R., J.Z., and S.I. analyzed data; A.B.B., F.M., B.L., C.M., R.R.R., J.Z., and S.I. interpreted results of experiments; A.B.B., F.M., B.L., C.M., R.R.R., and S.I. prepared figures; A.B.B., F.M., B.L., C.M., R.R.R., J.Z., and S.I. drafted manuscript; A.B.B., F.M., B.L., C.M., R.R.R., J.Z., and S.I. edited and revised manuscript; A.B.B., F.M., B.L., C.M., R.R.R., J.Z., and S.I. approved final version of manuscript.