Quantification of lipid area within thermogenic mouse perivascular adipose tissue using standardized image analysis in FIJI

Abstract

Quantification of adipocyte size and number is routinely performed for white adipose tissues using existing image analysis software. However, thermogenic adipose tissue has multilocular adipocytes, making it difficult to distinguish adipocyte cell borders and to analyze lipid proportion using existing methods. We developed a simple, standardized method to quantify lipid content of mouse thermogenic adipose tissue. This method, using FIJI analysis of hematoxylin/eosin stained sections, was highly objective and highly reproducible, with ~99% inter-rater reliability. The method was compared to direct lipid staining of adipose tissue, with comparable results. We used our method to analyze perivascular adipose tissue (PVAT) from C57BL/6 mice on a normal chow diet, compared to calorie restriction or a high fat diet, where lipid storage phenotypes are known. Results indicate that lipid content can be estimated within mouse PVAT in a quantitative and reproducible manner, and shows correlation with previously studied molecular and physiological measures.

Introduction

Established methodologies are useful for estimation of lipid within white adipose tissue (WAT), which is the most abundant adipose tissue in the body (1). Adipocytes in WAT have a single lipid droplet that fills the cytoplasm (unilocular), making it easy to identify cell boundaries and quantify cell size and cell number, using tools such as Adiposoft (2) and Adipocount (3). Both programs use threshold algorithms to select lipid space from surrounding tissue. Human peri-aortic adipose tissue has some thermogenic properties, yet morphologically appears as unilocular adipocytes when tissue is derived from adult donors (4). Thus Adipocount and Adiposoft can be used for analysis of human perivascular adipose tissue (PVAT).

In contrast to the unilocular adipocytes used for lipid storage in WAT, brown adipose tissue (BAT) and mouse PVAT are filled with multilocular adipocytes, where numerous lipid droplets of varying size occupy the cytoplasm, making cell borders difficult to distinguish. Often these droplets are quite small, and cannot be easily measured. Multilocular thermogenic adipocytes are known for their fat-burning phenotype, where mitochondrial uncoupling from ATP generation results in lipid breakdown to release heat. BAT and PVAT in the mouse are thermogenic adipose tissues. PVAT also releases factors, which can differ under varying phenotypes, that have local effects on the vasculature (6). PVAT is found adjacent to large vessels in the body, specifically around the aorta where it can be collected with relative ease. In the mouse, BAT is found as interscapular lobes underneath subcutaneous WAT located on the back. The multilocular phenotype of thermogenic adipose tissue pose a challenge for morphometric quantification using current methods. Staining methods include oil red O (5) or lipid probes such as BODIPY (4,4-difluoro-3a,4a-diaza-s-indacene) or LipidTOX (6). These methods are more time consuming than routine histology, and require cryosectioning, expense of the reagent, and fluorescence microscopy. We validated a standardized method to estimate lipid content within thermogenic adipose tissue (BAT and PVAT) that can be performed on routine histological slides stained with hematoxylin/eosin (H&E). The protocol is low cost, has high inter-rater reliability, and results are comparable to quantification using lipid stains. This method allows for reproducible and comparative lipid assessment in complex adipose tissues with multilocular properties.

Methods/Design

Sample Collection and Processing.

All protocols involving mice were approved by the Institutional Animal Care and Use Committee of the Maine Medical Center. The overall phenotype of the mice have been previously characterized (7). Thoracic aortas were collected from C57BL/6J mice. PVAT naturally surrounding the aorta was left intact and attached to the vessel. A small piece (roughly ~2mm) of the aorta with PVAT was placed into 10% formalin, and rocked overnight at 4°C. Similarly, BAT was taken from the interscapular fat pad on the back of the mouse, separated from the subcutaneous white adipose tissue, and was subsequently treated the same as the aorta/PVAT. The following day, samples were washed twice in PBS and moved to 70% ethanol prior to staining.

Sample Staining and Image Acquisition.

Formalin-fixed samples were processed for paraffin embedding and sectioning (5μm sections), and stained with H&E. Images were acquired at original 400x magnification using a Zeiss Axioscope 40 with attached Canon EOS 60D 18MP DSLR. Images were saved as .JPG files (2592×1728 px). Images originally taken at 200x magnification were also tested; however it was found that ImageJ/FIJI was better equipped to pick up details in the adipose tissue at 400x original magnification, due to small lipid droplets in mouse thermogenic adipocytes. Using the 40x objective or higher is recommended for obtaining images for use in this protocol. Example images from WAT, BAT, and PVAT highlighting phenotypic differences are shown in Figure 1. The imaging method utilized the image processing software ImageJ and the extension FIJI (8).

Figure 1. Phenotypic diversity of mouse adipose tissue depots.

Adipose tissue was collected from adult C57BL/6 male mice and processing for histology. Shown are hematocxylin/eosin-stained adipose tissue from gonadal white adipose tissue (A), interscapular brown adipose tissue (B), and perivascular adipose tissue surrounding the thoracic aorta (C). Scale bar = 25μm.

Frozen Tissue Sectioning and LipidTox Staining.

Thoracic aortas with attached PVAT were collected and embedded in OCT prior to sectioning. Tissues were sectioned at −30°C onto glass slides and set overnight at −80°C. Slides were warmed then incubated at room temp in a coplin jar with a paper towel soaked in 10% formalin at the base for 10 minutes. Slides were then washed in water for 40 min, and permeabilized in PBS + 0.1% Tween-20. Sections were stained with LipidTox (ThermoFischer Scientific cat# H34475 lot 1936383) dissolved in DMSO, and DMSO was used as a negative control. Stain was applied for 1h at room temperature. Slides were washed 3×10 min in PBS and then mounted in Vectashield Antifade Mounting Medium with DAPI (Vector Laboratories cat# H-1200–10). Slides were imaged using a Leica SP8 confocal microscope and analyzed in ImageJ.

Protocol Overview

Segmentation.

In FIJI the image is converted to 8-bit grayscale, and follows two pathways as indicated in Figure 2. The first is segmentation based on light/dark areas, using “Find Maxima” (9). This generates a “map” of the tissue differentiated from background. The flood fill tool is used to convert background areas to black.

Figure 2. Process for image assessment and quantification of multilocular adipose tissue.

Input images of hematoxylin/eosin stained sections are converted to grayscale (a), and then follow two pathways: identification of the region of interest (b-c) and thresholding of the image (d). The results from both pathways are combined using the «AND» function in the «Image Calculator» tool in ImageJ to produce the final image (e) (this corresponds to step 7 in the protocol). This image is then measured as the final lipid volume, and compared to the entire selected region (c).

Thresholding.

The second pathway applies an Otsu threshold to the image (10). This thresholding algorithm assumes a bimodal distribution of foreground and background in an image to convert the grayscale into a binary result (15). With the grayscale H&E images, there is a clear distinction between background (white) and foreground (pink) seen in the presence of two peaks on the histogram of each image, which allows the threshold to easily separate the two rapidly. Comparing the binary, Otsu processed images to the 8-bit, original H&E stain, there is a clear visual match between what is seen as white and black space, and what is stained as tissue and nuclei. The white space is either background or space previously occupied by lipid. The lipids are removed during tissue processing, however their absence is seen as intracellular white space, and is the basis of the estimation of lipid content. When thresholding and segmentation are complete, the result is two separate images.

Final Image Calculation and Measurements.

The final output is a single image that contains lipid space in white, and all non-lipid features and background in black. The last step involves selecting and measuring the white area using ImageJ/FIJI tools, to provide a lipid value, which can then be expressed as a percentage of total area.

Protocol

1. Orient ImageJ/FIJI to read the images with the desired binary settings.

   1. “Process” -> “Binary” -> “Options”
   2. Turn “Black Background” ON. This will tell ImageJ/FIJI to treat black space as back, and white space as front, and measure the white space.
2. Drag the desired image into the ImageJ/FIJI program.

   1. File types used were .JPG files. Any file type that is accepted by ImageJ/FIJI will work (including .TIFF or .PNG files). Due to amount of images being processed, files were saved from the camera as .JPG to minimize storage space needed.

3. “Image” -> “type” -> “8-bit”

4. “Process” -> “Find Maxima”

   1. Set prominence to a value that shows selection in all regions of the tissue, including white areas (typically a value between 5–30 is appropriate, there is no right value for this, just as long as it selects areas to omit).
   2. Change output type to “Segmented Particles”
   3. Select “Ok”
   4. The output image should be similar to Figure 2b. Once set, the settings for prominence value and output type will hold for each image.

5. Use the Flood Fill tool -> to fill in any non-tissue segments black. The original image can be used as a guide. Use the zoom tool to increase magnification for accuracy. To set a pure black, use the colorpicker tool () to carefully select one of the segment lines which should appear black, or double click the colorpicker tool, double click on one of the black/white overlapping boxes at the bottom of the window (there should be two), and change the sliders all to 0, and it should say black. If this value is not pure black, the selection/measurement tool will not work. This setting will hold until modified. The result will be similar to Figure 2c.

6. Select the other window (the 8-bit image) and use “Image” -> “Adjust” -> “Threshold”. Change the default threshold algorithm to “Otsu”, make sure that “Dark Background” is checked, then hit apply. These settings will maintain between images. The white area inside the tissue will be the lipid space, similar to the image shown in Figure 2d.

7. Remove the background of the image and constrain the white space to being entirely within the tissue sample, such as in Figure 2e.

   1. “Process” -> “Image Calculator”
   2. Set Image 1 to the Threshold image
   3. Set Operation to “AND” (depending on the version of ImageJ, this may need to be performed with the “Add” function).
   4. Set Image 2 to the Segmented image
   5. Select “Create new window” and hit “OK”
8. Select the combined image (from step 7), use “Edit” -> “Selection” -> “Create Selection”, which will select all the white areas. Go to “Analyze” -> “Measure” to measure this area (in pixels).

   1. To set the program to measure area, go to “Analyze” -> “Set Measurements”.
   2. Select “Area” and deselect everything else. Now, upon performing the Measure function, the measurement will be the area selected in pixels.
9. Perform this same operation on the segmented image (from step 7) to get the area that the selection was taken from.

   1. Performing step 8–9 will give you two values, the area of the lipid space and the total region area, allowing calculation of lipid space area.

Protocol Comparison to Lipid Staining Method

LipidTox was used to quantify lipids in frozen sections of mouse BAT and PVAT from C57BL/6 male mice, with tissues from the same mice used for paraffin embedding and processing to use with our protocol. Multiple images were taken at the same focal length and analyzed independently. The H&E images were analyzed using the newly established protocol and LipidTox images were manually processed using ImageJ.

The processing of LipidTox images was determined using visual threshold levels, as different sections had variable intensities of fluorescence. As seen in Figure 3, brightfield and DAPI images were taken to identify total tissue. A tissue mask was created, and a threshold was applied to separate lipid from background similarly to the thresholding used for the H&E images. A difference, however, is that the Otsu threshold was not used, because lipid was clearly omitted when using this algorithm. Instead, a manual threshold was determined by the rater to represent all fluorescent stain, and background areas were omitted. No specific algorithm was used, as all algorithms available in ImageJ/FIJI displayed varying levels of lipid being excluded, or background being included. Therefore, the rater would select a threshold value (between 0–255) based on the histogram of each LipidTox image. This value would be specific to each image, and the rater would attempt to pick a value that included all stained lipid, and ommitted all black background. However, this was a subjective method that did not follow to a strict algorithm. Percentage of tissue with fluorescence was quantified.

Figure 3. Comparison of lipid quantification methods.

A) Mouse perivascular adipose tissue was obtained from Immunofluorescence images of a frozen PVAT section from a C57BL/6J mouse. Each channel is labeled with its corresponding stain/label. Red arrows on the Brightfield and LipidTox channels indicate lipid that is seen outside of the tissue, that may have been mechanically removed from the tissue during the staining process. Images taken at 100x magnification and digitally zoomed 4x to result in effective 400x magnification, scale bar 50μm. B) Brightfield image of an H&E-stained section of PVAT from the same mouse that was treated with the IF stains. Image taken at 400x magnification, scale bar 25μm. C) Comparison graph of LipidTox and H&E-stained images, each compared based on the percent lipid that was identified based on the images. Comparison occurred between a sample of PVAT and BAT from the same mouse.

Comparison between these two methods is in Figure 3. There was no significant difference in calculated lipid area in PVAT between the two methods. In the BAT, our new protocol yielded a slightly higher lipid content compared to the LipidTox quantification. These differences may be due to the greater subjectivity of the analysis process for the LipidTox stain, or may be artifacts of the sectioning and staining process. We observed that lipid would often be detected outside of the tissue proper (red arrows in A), suggesting lipid release from droplets during the procedure. The processing for LipidTox analysis was substantially more time consuming, and due to the lipids within the tissues, these sections did not adhere well to slides, leading to detachment during the staining process.

Assessment of Changes in Lipid Content in Mouse PVAT and Protocol Reproducibility

Lipid content in mouse PVAT changes based on diet and genetics. We previously characterized PVAT from C57BL/6 mice fed a high fat diet, standard control chow diet, or with calorie restriction (7). We applied the current lipid quantification method to PVAT from that study to consider our results with the molecular and morphological features of PVAT. Detailed assessment of body weight, adiposity, and PVAT phenotype were previously reported (7). Briefly, high fat diet (~60% fat) was given to mice at eight weeks of age for 12 weeks, while controls were fed standard chow (~6% fat) during that time. For calorie restriction, normal daily food intake was monitored for two weeks, and mice were given 70% of normal daily intake for five weeks. Aortic PVAT was collected, processed for paraffin-embedding, sectioned and H&E stained for all groups.

We utilized our protocol to estimate lipid for each group, and tested interrater reliability with multiple independent raters. All raters had highly consistent values, with samples from each diet group (n=10 per group) assessed by three independent raters (Table 1). Results from a two-way ANOVA indicated there was no significant differences between raters for any of the diet groups. Due to the strong correlation seen from this assessment, two raters analyzed the entire dataset for each of the three diets (n=38 per group), since we validateed high reproducibility between raters. Morphology of PVAT demonstrated a visual difference in lipid between groups, and corresponded to the quantified lipid estimate, as seen in Figure 4A. We present data from two raters in Figure 4B. There was a highly statistically significant difference between lipid percentage of all groups (adjusted p value of <0.0001 for each). This level of significance was seen across both raters of the data. Additionally, statistical analysis indicated that again there was no significance between raters within a diet group as indicated in Figure 4B.

Table 1. Interrater reliability.

A of sample C57BL/6J PVAT images across calorie restricted, control, and high fat diets (n=10 each) were assessed independently, and a two-way ANOVA was performed to compare values between three raters. There were no statistically significant differences (ns) between raters for any group.

Group	Rater 1 Mean ± SEM	Rater 2 Mean ± SEM	Rater 3 Mean ± SEM	Greatest difference between raters	Statistical difference
Calorie Restricted	25.2 ± 2.4%	25.2 ± 2.5%	24.3 ± 2.1%	0.95%	ns
Control	50.5 ± 3.4%	50.6 ± 3.2%	50 ± 3.5%	0.52%	ns
High Fat	66.5 ± 3%	66 ± 3%	66 ± 3%	0.52%	ns

Figure 4. C57BL/6J PVAT phenotype shift under varying diet conditions as quantified using our ImageJ protocol.

A) Images of representative H&E-stained sections for each diet analyzed in this case study. Each image is labeled with the corresponding diet. Scale bar = 25μm for all images. B) Percent lipid measurements of samples images from all studied diets across two raters. Each rater independently quantified each sample image from all diet types, with the results being plotted. Each n=39 per diet. A two-way ANOVA was performed using GraphPad Prism to determine significance between diet types, and multiple t-tests were used for each pair of raters within a diet to determine significance between raters.

Discussion/Conclusion

The ability to assess adipocyte size and number in WAT has been critical in understanding mechanisms of adipocyte hypertrophy related to obesity and metabolic diseases. PVAT is a unique adipose depot that has significant local effects on the vasculature (11). In the mouse, PVAT, particularly around the thoracic aorta, is multilocular and has similar phenotypic and molecular features as interscapular BAT, as seen in Figure 1 (12). Current methods that exist to quantify WAT prove ineffective in providing accurate measurements for multilocular thermogenic tissues. The method presented here is highly reproducible, low cost, and easy to perform on ImageJ/FIJI, an open source software. This method allows for manual definition of desired tissue region, which is useful in structurally diverse tissues. This method was validated using PVAT from a previously characterized mouse dietary experiment, and was rigorously tested using multiple raters to assess reliability.

Several factors are important to consider. The first involves identifying the prominence value for the “Find Maxima” function. This function was able to be manipulated by changing the numerical value that was defaulted by FIJI to 10.00. Upon selecting the «Find Maxima» feature, a window opens that allows manipulation of the prominence value, as well as output type which should be changed to «Segmented Particles» (see protocol step 4). The prominence value that will appear upon opening the window is 10.00, however this value can be selected and changed. The default value was applicable to most images, however at times the image would be over or undersegmented. The former resulted in the tissue borders being indistinguishable, and the latter resulted in it being included with some regions of the background. It was determined that this occurred due to the light levels of the image, which would vary depending on the conditions in which the image was taken. Fortunately, this did not alter the result of the threshold due to the same level of contrast being seen between the stained and unstained regions no matter the light level of the image. To address this issue, different values can be used to improve the quality of the segmentation. If the resulting image was not segmented enough, a smaller value can be used to increase sensitivity of the thresholding. If the image was too segmented, a larger value can be used. We observed that within a set of images taken at the same time, the same value typically worked for all images.

Another consideration for analysis of mouse PVAT images is the vascularity of the tissue, which can be seen histologically. We found that it was beneficial to place the image windows in FIJI side by side, and ensure the grayscale image was visible while manually selecting the adipose region. In general, the selection was conducted to be more stringent, and opt for removing areas of PVAT rather than including areas of non-adipocytes. Likewise, since mouse interscapular BAT anatomically lies deep to subcutaneous WAT, it may occur that WAT is captured within samples, depending on the method of tissue collection. In those instances, it may also be necessary to omit those regions. Overall, our protocol proved easy to master and perform, and highly reproducible between raters. In addition to thermogenic adipose tissue such as mouse PVAT and BAT, we believe that the protocol can be adapted to quantify the level of steatosis in other tissues, such as liver, heart, or muscle.