- Top of page
- Research Methods and Procedures
Objective: Fat cell size is a fundamental parameter in the study of adipose tissue metabolism, because it markedly influences the cellular rates of metabolism. Previous techniques for the sizing of adipocytes are often complicated or time-consuming. The aim of this study was to develop a new, computerized method for rapid and accurate determination of adipocyte size in a cell suspension obtained by incubating human or rat adipose tissue biopsies with collagenase.
Research Methods and Procedures: The cell suspension was placed between a siliconized glass slide and a cover slip. Using the reference method [designated as (R)], the cell diameters were determined manually using a microscope with a calibrated ocular. The new method presented here [designated as (C)] was based on computerized image analysis.
Results: After two well-defined corrective adjustments, measurements with (R) and (C) agreed very well. The small remaining differences seemed, in fact, to depend on inconsistencies in (R).
Discussion: We propose that (C) constitutes a valuable tool to study fat cell size, because this method is fast and allows the assessment of a sufficient number of cells to get reliable data on size distribution. Furthermore, images of cell preparations may be stored for future reference.
- Top of page
- Research Methods and Procedures
The major function of the fat cell is to store and release energy. Depending on the amount of lipid stored, the diameter and the volume of the adipocyte can change ∼20-fold and several thousand fold, respectively. In clinical and metabolic studies of adipose tissue, an accurate method for the determination of fat cell size is of importance, because adipocyte size markedly influences the cellular rates of metabolism. Both glucose metabolism and lipid mobilization are increased in enlarged fat cells (1, 2, 3, 4). In contrast, the stimulating effects of insulin on the rates of glucose oxidation and incorporation into the triglycerides are inversely related to the size of the fat cells (5, 6). Factors regulating cytokine release within adipose tissue also seem to include the size of the fat cells (7, 8), and differences in adipose tissue cellularity have been implicated as a possible link between obesity and diabetes (9). Moreover, not only adipocyte size but also adipocyte number may be affected by metabolic, endocrine, and pharmacological influences through, for example, differentiation of pre-adipocytes (i.e., recruitment of small fat cells) (10).
There are several different techniques for counting and sizing adipose cells in small samples of human or animal adipose tissue. Microscopic measurements on conventional histological preparations or frozen-cut adipose tissue (11, 12), determination of DNA content (13), and automatic counting or sizing of osmium tetroxide-fixed fat cells or unfixed adipocytes in a suspension using a Coulter electronic counter (12, 14, 15) are some of the methods that have been described. In our laboratory, fat cell isolation with collagenase, followed by microscopic determination of adipocyte diameter (Ref. 16, slightly modified), has been used for several years, and a similar technique has been used by others (17). However, after the collagenase procedure, cell diameter must be determined immediately, because the suspension cannot be stored without affecting cell morphology. This is a considerable drawback when several samples must be analyzed. In a study by Lavau et al. (18), the time spent handling the isolated fat cells was minimized by measuring adipocyte diameter, not under the microscope, but on printed and enlarged photomicrographs. Still, this method and the others that have been described are often both tedious and time-consuming. Recently, a new method measuring the cross-sectional area of adipocytes in histology sections with computer image analysis has been published (19). This method allows a large number of adipocytes to be measured rapidly. However, when isolated adipocytes are studied, as in many ex vivo and in vitro experimental set-ups, fast and accurate determination of the size of adipocytes in a suspension is important.
In this paper, we describe a new computerized technique [designated as (C)1] for determination of the size of adipocytes isolated by collagenase digestion. The main advantage with this technique is that a large amount of visual data may be stored and processed in a short period of time.
- Top of page
- Research Methods and Procedures
Two of the most evaluated methods for determining fat cell size are 1) automatic determination after osmium fixation (12, 14) and 2) microscopic measurements with an eyepiece micrometer after freeze cutting or collagenase digestion (12, 16, 17). The former method is applicable to cells of all sizes, a large number of cells can be counted, and the reproducibility is good. Still, the technique requires osmium fixation for 24 to 72 hours, depending on the amount of tissue, and the fixation may be accompanied by cell swelling (14). Moreover, it is not possible to store information on what has actually been sized. In this paper, the second method has been used as the reference (R). This method requires the assessment of diameters of single cells, one by one, by direct observation. In practice, this means that the total number of measured cells will be restricted and possible conclusions will be curtailed.
Here, we present a computer-based method for determination of fat cell size after collagenase digestion. With this method, a large number of cells may be measured in a short period of time. It is also possible to save images of the measured cells for future reference. In addition, observer bias can be minimized by such a method. The procedure is carried out in two steps (i.e., registration of the images and subsequent analysis of these images).
Registration of Images
One advantage of direct eyepiece measurement of fat cell size is that it gives a very precise measure, because focus is adjusted for each cell. At the same time, however, the need to assess single cells constitutes a disadvantage because it makes the technique slow and tiresome. To capitalize on the speed of the computer-based method, the focus is set only once for each visual field. However, this inevitably means that a number of cells will be somewhat out of focus. Increasing the depth of focus of the microscope reduces this effect. Furthermore, in a separate control experiment, we showed that this effect did not appreciably influence the reliability of the obtained result (Figures 3 and 4).
The main advantage of computer-based image analysis is that a large amount of visual data may be processed in a short period of time. To obtain reliable results, however, it is essential that the analyzed images are well standardized. This is achieved through standardization of the microscope set-up and through automatic contrast enhancement at the image processing level. Also, the necessary observer inputs during the calculations should be few; in our method, they were restricted to the setting of a single segmentation parameter. Furthermore, selection of the inclusion criteria that determine what should be included as relevant image structures is critical.
For all practical purposes, day-to-day differences in light intensity of the microscope and in the amplification of the electronic equipment result in a variation of contrast and brightness in the registered image. To compensate for this variation, all images were normalized through automatic contrast enhancement, which expands the intensity levels to the full 8-bit range of the software.
In the process of segmentation, the intensity level that differentiates relevant and irrelevant structures must be decided. In our technique, this amounted to the setting of one parameter at each measurement occasion, i.e., once for 20 to 40 visual fields, depending on the requirements of the experiment. Although this choice implies a moment of subjective judgment, control experiments indicated that the final results of the measurements were very insensitive to this setting (Figure 5).
In the final processing of the images, all areas bordered by fat cell contours would be measured by the software, including deformed cells, cells cut along the image borders, and areas between cells. For this reason, two inclusion criteria were defined, i.e., maximum degree of oblongness (Feret ratio, the ratio between the longest and shortest extension of the area > 0.8) and minimum degree of contour smoothness (ratio between true surface area and perimeter-derived surface area > 0.8). A lower limit of the cell size was also set, corresponding to a cell diameter of 22 μm at 8× magnification, reflecting the useful resolution of the image. DeMartinis and Francendese have described a population of very small fat cells of ∼8 to 35 μm in some mammalian species (20). The smallest of those cells, 8 to 21 μm, were not measured using the present technique. However, as reported by DeMartinis and Francendese, the very small fat cells below ∼20 μm are not spherical, but rather oval, and were thus excluded irrespective of size by the computer program. Thus, if these “pre-adipocytes” of ∼8 to 20 μm are to be studied, other methods must be used.
Breakage of large fat cells during collagenase treatment, resulting in nonrepresentative samples containing lipid droplets, has been reported in the literature (14). However, with the precautions used, i.e., a gently shaking water bath during collagenase treatment, freshly siliconized glassware, and no centrifugation of the cells, there is no indication of an increased rupture of adipocytes during preparation of the cell suspension (16). Thus, the proportion of lipid droplets in the cell suspensions is usually very low. One considerable advantage with the new computerized method is that the images can be stored and later studied regarding what has actually been measured. If separate spherical formations, identified as lipid droplets, are present on the images, these formations can easily be excluded from the calculations. The adipocytes can be stained with crystal violet to facilitate adipocyte identification. However, it has not been found necessary to stain the samples routinely.
In the original description (12, 16) of the sampling technique adopted in (R), the representation of the measured cells was ensured by inclusion of all cells that fell along a line, scanning perpendicularly across the visual field. In this study, cells that were adjacent to a vertical line, positioned arbitrarily across the cell preparation, were measured. Such a procedure will inevitably exaggerate the number of big cells, because they will be included even if their centers lie far away. It is also possible that subjective judgment might influence the selection of measured cells, which means that the sampling error in (R) may not easily be corrected by mathematical means.
In (C), a sufficient number of arbitrarily selected visual fields are registered and analyzed. Also in this case, selection is influenced by the procedure; more large cells than small cells along the borders of the visual field become transected and disregarded. However, this error may be corrected fully by mathematical means.
The coefficient of variation of cell size in the cell samples was between 29% and 38%, indicating that a sample size of 100 cells, as in (R), could reveal differences in mean cell diameter of around 7% at p < 0.05 between cell preparations. Considering that the size distribution seems somewhat skewed (Figure 2), the precision is likely to be less. When cells within a defined size range are of interest or when cell distribution pattern is under study, the sample size must be increased in an inverse proportion to the size of the subsample to achieve the same degree of precision, pointing to the need for the presented methodology.
Comparison between (C) and (R)
When the experimental conditions had been established, fat cell size was measured on 39 biopsies (23 human and 16 rat) both with (C) (around 11, 300 and 8700 cells, respectively) and with (R) (around 11, 700 and 6900 cells, respectively). In Figure 6, the obtained size distribution is shown for human cells. The general pattern of the distribution is similar for the methods, with a dominating peak and a less prominent hump for smaller cell diameters. However, two discrepancies were observed: 1) (C) gave a smaller mean cell diameter than (R): 78.2 ± 23.1 and 89.4 ± 31.1 μm, respectively; and 2) it seemed that there were fewer small cells measured with (C).
A difference in average size could be caused either by a systematic difference in size estimation by the two methods or by a systematic difference in sampling.
In fact, when cells were measured both interactively on the registered image and with the automatic procedure, it was obvious that (C) underestimated cell size by a factor of 1.06 (Figure 7). This difference depended on the image analysis processing, in which the cell diameter is calculated from the surface area inside the cell contours. Because the degree of underestimation was linear over the whole size range, this error could be corrected by including the factor 1.06 in the computer program. After such a correction (Figure 8), the distributions agreed better, and the difference in average cell size was smaller: 82.9 ± 24.5 and 89.4 ± 31.1 μm for (C) and (R), respectively.
Individual cells were also measured with both (C) and (R) (Figure 9). In these measurements, the same proportionate underestimation of cell diameters by (C) (x1/1.07) was obtained, reflecting the computerized processing of the images. There was, however, also a negative y intercept of the linear regression curve of (C) on (R). Because no significant y intercept was obtained when the very same cells were measured interactively on the registered image, the only possible explanation is that there was, in fact, a systematic bias in (R). Such a bias might reflect the way the observer delimited cell contours. If this bias is taken into account, the discrepancy between mean diameters obtained with (C) and (R) would be further reduced: 82.9 ± 24.5 and 84.7 ± 31.1 μm (mean ± SD), respectively.
Another explanation contributing to the observed shift in the size distribution (Figure 8) could be that big cells were more likely to be selected in (R). This effect is illustrated in Figure 12.
Figure 12. This composite figure is based on the same data as in Figure 6. The distribution of cell diameters is illustrated, determined by the reference method (line curves), with the distribution according to the computerized method given in the background for comparison (gray area). Only the main peaks are shown (compare Figure 6). On the left, noncorrected reference data are shown, whereas the other two diagrams illustrate the distributions after corrections, considering sampling along a 30-μm-broad band (middle) or along a line (right). It is obvious that compensation for this putative sampling error improves the agreement between the distribution profiles obtained by the reference method and the computerized method.
Download figure to PowerPoint
In the lowest size range, (C) seemed to pick up fewer cells than (R). However, when cells <44 μm were scrutinized in the registered video images, all cells above the defined limit (>22 μm) were picked up. Another 10% were cells smaller than 22 μm. This little difference, however, could not explain the difference in cell number. One possibility is that the discrepancy reflects a subjective bias to include too many cells in the smallest range with (R).
Mean Cell Diameter
The agreement between the two methods in the determination of mean cell diameter was good, with the data points scattered around the line of identity when (C) was plotted against (R) (k = 0.97 ± 0.01). The size of the coefficient of variation around the regression line, 6.5%, reflects the variation in the individual methods (Figure 1), together with the systematic sampling error and the suspected subjective bias in (R) (step II in Figure 1). Thus, as applied here, the coefficient of variation of determinations on single cell samples (steps II to IV in Figure 1) was 3.6% and 1.2% in (R) and (C), respectively. The dispersion between determinations of mean cell diameter of cells from the same fat biopsy but from different cell samples (steps I to IV in Figure 1) could be bigger, again highlighting the need for large samples and the importance of homogeneous preparations.
When measurements made with (C) and (R) were compared, certain differences were observed regarding cell size distribution and mean cell diameter. The differences depended to a certain extent on the computerized processing of the image data. These differences were defined in detail and could be compensated for. The remaining differences seemed to depend on inconsistencies in (R) as applied. They should not be compensated for but should merely highlight the need for caution when data obtained by different methods are compared.
We have presented a computer-based method to measure cell size on adipocyte suspensions obtained from collagenase-treated adipose tissue biopsies. The technique is fast enough to easily allow the assessment of 10-fold more cells from fresh preparations than conventional methods. We propose that the method constitutes a powerful tool to study not only average cell size, but also more subtle variations in the size distribution of fat cells from tissue biopsies. At the same time, images of the measured cells may be stored for future reference.
[(1.) For cells of point size, the effective area will be equal to the true area of the visual field = width (w) × height (h). For cells with Ø = d, cells with the center closer than d/2 from the periphery will be excluded, which means that the effective area for these cells will be (w − d) × (h − d). Thus, if the apparent relative frequency of cells with Ø = d is f, then the true relative frequency (F) will be F = f × [w × h]/[(w − d) × (h − d)]. (2.) Strictly, only those cells that are grazed or intersected by the vertical line are sampled. For cells with Ø = d, only those cells with the center closer to the vertical line than d/2 will be included, which means that the effective area for these cells is d × l (l = the traveling distance along the line). Thus, if the apparent relative frequency of cells with Ø = d is f, then the true relative frequency (F) will be F = f × k/d, where k is an arbitrary constant. In order to simulate the practical procedure, one might assume that all cells with the borders within a certain distance (a/2) from the vertical line are included. Then the true relative frequency will be F = f × k/(d + a).]