Here, we present a MATLAB implementation of automated detection and quantitative description of lipid droplets in fluorescence microscopy images of fission and budding yeast cells.
Lipid metabolism and its regulation are of interest to both basic and applied life sciences and biotechnology. In this regard, various yeast species are used as models in lipid metabolic research or for industrial lipid production. Lipid droplets are highly dynamic storage bodies and their cellular content represents a convenient readout of the lipid metabolic state. Fluorescence microscopy is a method of choice for quantitative analysis of cellular lipid droplets, as it relies on widely available equipment and allows analysis of individual lipid droplets. Furthermore, microscopic image analysis can be automated, greatly increasing overall analysis throughput. Here, we describe an experimental and analytical workflow for automated detection and quantitative description of individual lipid droplets in three different model yeast species: the fission yeasts Schizosaccharomyces pombe and Schizosaccharomyces japonicus, and the budding yeast Saccharomyces cerevisiae. Lipid droplets are visualized with BODIPY 493/503, and cell-impermeable fluorescent dextran is added to the culture media to help identify cell boundaries. Cells are subjected to 3D epifluorescence microscopy in green and blue channels and the resulting z-stack images are processed automatically by a MATLAB pipeline. The procedure outputs rich quantitative data on cellular lipid droplet content and individual lipid droplet characteristics in a tabular format suitable for downstream analyses in major spreadsheet or statistical packages. We provide example analyses of lipid droplet content under various conditions that affect cellular lipid metabolism.
Lipids play crucial roles in cellular energy and carbon metabolism, synthesis of membrane components, and production of bioactive substances. Lipid metabolism is fine-tuned according to environmental conditions, nutrient availability and cell-cycle phase1. In humans, lipid metabolism has been connected to diseases, such as obesity, type II diabetes and cancer2. In industry, lipids produced by microorganisms, such as yeasts, represent a promising source of renewable diesel fuels3. Cells store neutral lipids in so-called lipid droplets (LDs). These evolutionarily conserved bodies are composed of triacylglycerols, steryl esters, an outer phospholipid monolayer and associated proteins1. LDs originate in the endoplasmic reticulum, exert cell-cycle or growth-phase dynamics, and are important for cellular lipid homeostasis1. LD number and morphology can be used as a convenient proxy when assaying lipid metabolism under various growth conditions or when screening a panel of mutants. Given their dynamic nature, techniques capable of analyzing the properties of individual LDs are of particular interest in studies of lipid metabolism.
Various yeast species have been used to describe lipid-related metabolic pathways and their regulation, or used in biotechnology to produce interesting compounds or fuels1. Furthermore, for model yeasts, such as the budding yeast Saccharomyces cerevisiae or the distantly related fission yeast Schizosaccharomyces pombe, genome-wide deletion strain libraries are available that can be used for high-throughput screens4,5. Recently LD composition and dynamics have been described in S. pombe6,7,8,9, and mutants related to lipid metabolism have been isolated in the emerging model yeast Schizosaccharomyces japonicus10.
Numerous techniques are available to study LD content and dynamics. Most employ some kind of staining of LDs with lipophilic dyes such as Nile Red or BODIPY 493/503. The latter shows more narrow excitation and emission spectra, and increased specificity towards neutral lipids (LDs) as opposed to phospholipids (membranes)11. Fluorimetric and flow-cytometry methods have been used successfully in various fungal species to uncover genes and growth conditions that affect storage lipid content12,13,14,15. While these methods are suitable for high-throughput applications, they cannot measure the numbers and morphology of individual LDs in cells, which can differ dramatically between growth conditions and genotypes. Coherent Raman scattering or digital holographic microscopy are label-free methods that yield LD-level data, but require specialized expensive equipment16,17,18. Fluorescence microscopy, on the other hand, can provide detailed data on LD content, while utilizing commonly available instruments and image analysis software tools. Several analysis workflows exist that feature various degrees of sophistication and automation in cell/LD detection from image data, and are optimized for different cell types, such as metazoan cells with large LDs19,20,21, or budding yeasts17,22,23. Some of these approaches only work in 2D (e.g., on maximum projection images), which may fail to reliably describe the cellular LD content. To our knowledge, no tools exist for determination of LD content and morphology from fission yeast microscopic data. Development of automated and robust LD-level analyses would bring increased sensitivity and enhanced statistical power, and provide rich information on neutral lipid content, ideally in multiple yeast species.
We have developed a workflow for LD content analysis from 3D fluorescence microscopy images of yeast cells. Live cells are stained with BODIPY 493/503 and Cascade Blue dextran to visualize LDs and determine cell boundaries, respectively. Cells are immobilized on glass slides and subjected to z-stack imaging using a standard epifluorescence microscope. Images are then processed by an automated pipeline implemented in MATLAB, a widely used (commercial) package for statistical analyses. The pipeline performs image preprocessing, segmentation (cells vs. background, removal of dead cells), and LD identification. Rich LD-level data, such as LD size and fluorescence intensity, are then provided in a tabular format compatible with major spreadsheet software tools. The workflow was used successfully to determine the impact of nitrogen source availability on lipid metabolism in S. pombe24. We now demonstrate the functionality of the workflow in S. pombe, S. japonicus and S. cerevisiae, using growth conditions or mutants that affect cellular LD content.
1. Preparation of Solutions and Media
2. Cell Cultivation
3. Lipid Droplet Staining
4. Setting up the Microscope and Imaging
5. Image Analysis
The whole procedure is summarized in Figure 1 for the fission yeasts (the budding yeast workflow is analogous), and below we provide examples of how the workflow can be used to study LD content in three different yeast species under various conditions known to affect cellular LD content. Each example represents a single biological experiment.
Figure 1: Schematic diagram of the experimental and analytical workflow. The workflow for fission yeasts is shown as an example. Please click here to view a larger version of this figure.
First, we analyzed S. pombe cells (Figure 2). Wild-type (WT; h+s) cells were grown to exponential phase in either the complex YES medium or defined EMM medium. Compared to YES, fewer LDs and higher LD staining intensity per unit of cell volume were detected in EMM (Figure 2A-C). Moreover, individual LDs formed in EMM medium were larger and displayed increased total staining intensity (Figure 2D, E). This is in agreement with previous findings of increased storage lipid content in cells grown in EMM24. The ppc1 gene encodes a phosphopantothenate-cysteine ligase required for coenzyme A synthesis. The temperature-sensitive ppc1-88 mutant shows a marked decrease in LD content when grown at the restrictive temperature31, providing an example of cells with low BODIPY 493/503 signal (Figure 2A). Accordingly, compared to wild type (grown at 32°C), smaller LDs with lower total staining intensity were detected in ppc1-88 cells grown in YES following a shift to 36°C (Figure 2D, E), without any apparent change in LD number per unit of cell volume (Figure 2B).
Figure 2: Impact of growth media and lipid metabolism mutation on LD content in S. pombe. Wild type (WT) and ppc1-88 cells were grown to exponential phase in the complex YES or defined EMM medium, as indicated. WT cells were grown at 32 °C. The temperature-sensitive ppc1-88 cells were grown at 25°C and shifted to 36°C for 2 hours prior to analysis. (A) Representative unprocessed microscopic images of LDs stained with BODIPY 493/503. A single optical slice is shown for each condition; 10% overlay with inverted blue channel was added to better visualize cell boundaries. Scale bar = 10 µm. (B) Number of identified LDs per unit of cell volume. (C) Fluorescence intensity of identified LDs per unit of cell volume. (D) Distributions of total fluorescence intensities of all identified LDs. ***, ### unpaired Wilcoxon test p = 1.7 x 10-107, p = 3.7 x 10-132, respectively. (E) Distributions of volumes of all identified LDs. ***, ### unpaired Wilcoxon test p = 6.8 x 10-71, p = 1 x 10-64, respectively. Data in panels B-E were derived from 242, 124 and 191 cell objects for the WT YES, WT EMM and ppc1-88 samples, respectively. Please click here to view a larger version of this figure.
Next, we quantified LD content in S. japonicus cells (h+ matsj-2017)32 from exponential and early-stationary cultures grown in YES (Figure 3A). Cells entering stationary phase showed markedly decreased number of LDs per unit of cell volume compared to exponentially growing cells (Figure 3B), while volume-normalized LD fluorescence intensity decreased slightly between the two conditions (Figure 3C). The early stationary-phase LDs were typically moderately larger in size and had moderately higher total fluorescence intensity compared to LDs from exponentially growing cells (Figure 3D, E).
Figure 3: LD content in S. japonicus cells changes with growth phase. Exponentially growing (LOG) and early stationary phase (STAT) cells were analyzed. (A) Representative unprocessed microscopic images of LDs stained with BODIPY 493/503. A single optical slice is shown for each condition; 10% overlay with inverted blue channel was added to better visualize cell boundaries. Scale bar represents 10 µm. (B) Number of identified LDs per unit of cell volume. (C) Fluorescence intensity of identified LDs per unit of cell volume. (D) Distributions of total fluorescence intensities of all identified LDs. *** unpaired Wilcoxon test p = 1.3 x 10-114. (E) Distributions of volumes of all identified LDs. *** unpaired Wilcoxon test p = 2.4 x 10-85. Data in panels B-E were derived from 274 and 187 cell objects for the LOG and STAT samples, respectively. Please click here to view a larger version of this figure.
Finally, we analyzed S. cerevisiae cells of the widely used BY4741 laboratory strain (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) grown to exponential and stationary phase, respectively, in the complex YPAD medium. Budding yeast cells typically accumulate storage lipids upon entry into stationary phase1, and we were able to recapitulate these findings (Figure 4). Stationary cells contained somewhat fewer LDs per unit of volume compared to exponentially growing cells (Figure 4B), but their volume-normalized LD fluorescence intensity almost doubled (Figure 4C). This sharp increase in overall LD content was due to the much higher fluorescence intensity and volume of individual LDs in stationary phase (Figure 4D, E).
Figure 4: LD content in S. cerevisiae cells changes with growth phase. Exponentially growing (LOG) and stationary phase (STAT) cells were analyzed. (A) Representative unprocessed microscopic images of LDs stained with BODIPY 493/503. A single optical slice is shown for each condition; 10% overlay with inverted blue channel was added to better visualize cell boundaries. Scale bar represents 10 µm. (B) Number of identified LDs per unit of cell volume. (C) Fluorescence intensity of identified LDs per unit of cell volume. (D) Distributions of total fluorescence intensities of all identified LDs. *** unpaired Wilcoxon test p = 4.6 x 10-78. (E) Distributions of volumes of all identified LDs. *** unpaired Wilcoxon test p = 3.7 x 10-63. Data in panels B-E were derived from 430 and 441 cell objects for the LOG and STAT samples, respectively. Please click here to view a larger version of this figure.
Thus, our analysis workflow can detect changes in LD number, size and lipid content in three different and morphologically distinct yeast species under various conditions that positively or negatively affect cellular LD content.
The understanding of lipid metabolism and its regulation is important for both basic biology, and clinical and biotechnological applications. LD content represents a convenient readout of lipid metabolism state of the cell, with fluorescence microscopy being one of the major methods used for LD content determination. The presented protocol allows automated detection and quantitative description of individual LDs in three different and morphologically distinct yeast species. To our knowledge, no similar tools exist for the fission yeasts. The MATLAB scripts required for image processing are included as Supplementary files, and are also available from the Figshare repository (DOI 10.6084/m9.figshare.7745738) together with all raw and processed image and tabular data from this manuscript, detailed descriptions of the CSV output files, and R scripts for downstream data analysis and visualization. Also, the latest version of the MATLAB scripts is available from GitHub (https://github.com/MartinSchatzCZ/LipidDots-analysis).
Successful LD analysis is largely dependent on the quality of the raw fluorescence images obtained. For optimal performance of the segmentation algorithms, clean glass slides devoid of dust particles should be used for microscopy, the cells should form a monolayer (the actual number of cells per field of view is not a critical parameter), and should not contain a large proportion of dead cells. Also, the z-stack imaging should start slightly below and end slightly above the cells. Depending on the particular microscopic setup, users may need to adjust some of the parameters in the image processing scripts (such as “th” for image background intensity threshold). While the current method is able to detect and describe individual LDs in the segmented cell objects, the workflow does not produce truly single-cell data due to difficulties with automated separation of all individual cells. Instead, LD content per unit of cell volume generalized for the whole sample is reported. This limitation may hamper data interpretation in analyses of heterogeneous cell populations. Also, care should be taken when working with cells with altered transport of small molecules (e.g., efflux pump mutants), as this might affect the intracellular BODIPY 493/503 concentration and LD staining, as observed for the Nile Red lipophilic dye33,34.
Staining the medium with the cell-impermeable Cascade Blue fluorescent dextran is a convenient way of distinguishing cells from the background35, which can be applied to many (if not all) yeast species. It also helps with automated removal of dead cells from the analysis as these will turn blue upon staining. Any dying or sick (and thus partially permeable for dextran) cells detected as alive can be removed during data analysis steps based on the “IntensityMedianBlue” value of the detected cell objects. In principle, the whole workflow can be used to detect various other cellular structures, such as DNA repair foci, provided the structures can be labelled with suitable fluorophores. The workflow should also be applicable to cells of other (yeast) species, further broadening its utility.
The authors have nothing to disclose.
This work was supported by Charles University grants PRIMUS/MED/26, GAUK 1308217 and SVV 260310. We thank Ondřej Šebesta for help with microscopy and development of the image analysis pipeline. We thank the ReGenEx lab for S. cerevisiae strains, and JapoNet and Hironori Niki’s lab for S. japonicus strains. The ppc1-88 strain was provided by The Yeast Genetic Resource Center Japan. Microscopy was performed in the Laboratory of Confocal and Fluorescence Microscopy co-financed by the European Regional Development Fund and the state budget of the Czech Republic (Project no. CZ.1.05/4.1.00/16.0347 and CZ.2.16/3.1.00/21515).
12-bit monochromatic CCD camera Hamamatsu ORCA C4742-80-12AG | Hamamatsu | or equivalent | |
Adenine hemisulfate salt, ≥99% | Merck | A9126-25G | |
BODIPY 493/503 (4,4-Difluoro-1,3,5,7,8-Pentamethyl-4-Bora-3a,4a-Diaza-s-Indacene) | Thermo Fisher Scientific | D3922 | for neutral lipid staining |
D-(+) – Glucose, ≥99.5% | Merck | G7021 | |
Dextran, Cascade Blue, 10,000 MW, Anionic, Lysine Fixable | Thermo Fisher Scientific | D1976 | for negative staining of cells |
Dimethyl sulfoxide, ≥99.5% | Merck | D4540 | or higher purity, keep anhydrous on molecular sieves |
EMM broth without dextrose | Formedium | PMD0405 | medium may also be prepared from individual components |
Fiji/ImageJ software | NIH | or equivalent; for visual inspection of microscopic data | |
High precision cover glasses, 22×22 mm, No 1.5 | VWR | 630-2186 | use any # 1.5 cover glass |
Image Processing Toolbox for MATLAB, version 10.0 | Mathworks | ||
Lectin from Glycine max (soybean) | Merck | L1395 | for cell immobilization on slides |
MATLAB software, version 9.2 | Mathworks | ||
Microscope slide, 26 x 76 mm, 1 mm thickness | Knittel Glass | L762601.2 | use any microscope slide fitting your microscope stage, clean thoroughly before loading cells |
Olympus CellR microscope with automatic z-axis objective movement | Olympus | or equivalent | |
pentaband filter set | Semrock | F66-985 | brightfield, green and blue channels are sufficient |
Signal Processing Toolbox for MATLAB, version 7.4 | Mathworks | ||
SP supplements | Formedium | PSU0101 | |
standard office computer capable of running MATLAB | |||
Statistics and Machine Learning Toolbox for MATLAB, version 11.1 | Mathworks | ||
Universal peptone M66 for microbiology | Merck | 1070431000 | |
UPLSAPO 60XO objective | Olympus | or equivalent | |
Yeast extract | Formedium | YEA03 | |
Yeast nitrogen base without amino acids | Formedium | CYN0405 |