Quantification of islet size and architecture

Abstract

Human islets exhibit distinct islet architecture particularly in large islets that comprise of a relatively abundant fraction of α-cells intermingled with β-cells, whereas mouse islets show largely similar architecture of a β-cell core with α-cells in the periphery. In humans, islet architecture is islet-size dependent. Changes in endocrine cell mass preferentially occurred in large islets as demonstrated in our recent study on pathological changes of the pancreas in patients with type 2 diabetes.¹ The size dependency of human islets in morphological changes prompted us to develop a method to capture the representative islet distribution in the whole pancreas section combined with a semi-automated analysis to quantify changes in islet architecture. The computer-assisted quantification allows detailed examination of endocrine cell composition in individual islets and minimizes sampling bias. The standard immunohistochemistry based method is widely applicable to various specimens, which is particularly useful for large animal studies but is also applied to a large-scale analysis of the whole organ section from mice. In this article, we describe the method of image capture, parameters measured, data analysis and interpretation of the data.

The entire tissue section is captured by a modified method of “virtual slice image capture”^2-6 using a microscope with a 10x objective. One virtual slice image is typically composed of several hundreds of optical panels. Each virtual slice taken at multiple fluorescent channels is merged into one composite (shown as insulin in green, glucagon in red, somatostatin in white and nuclei in blue in Fig. 1A). An example of a detailed regional view that contains several islets including a small cluster of β-cells is shown on the right. Note that there is no overlap among the endocrine cell populations. A 20x objective (or higher) may be used depending on the size of tissue sections and each application. However, we have found that the image resolution using a 10x objective is sufficient for the subsequent computer-assisted image analysis at a single cell level. An important consideration may be the capacity of computer-assisted data processing in the standard laboratory setting, since the size of image files is typically in several gigabytes.

Figure 1. Large-scale capture and computer-assisted semi-automated analysis of the whole tissue section. (A) Virtual slice view of a human pancreatic section (female, 66-y old) immunostained for insulin (green), glucagon (red), somatostatin (white) and nuclei (blue). A series of contiguous images of a specimen is collected (illustrated as boxed panels) and merged into a single image montage (i.e., virtual slice; arrowed). A composite is made by merging four overlapping virtual slice images. Shown on the right is an example of a detailed regional view that contains several islets including a small cluster of β-cells. (B) Views of each channel showing cellular composition: a. β-cells; b. α-cells; c. δ-cells; d. nuclei; and e. a composite of all three endocrine cells and nuclei. Note that there is no overlap among the endocrine cell fractions. (B, f) Reconstructed endocrine cell distribution within each islet based on the captured center coordinates of each cell type within the given islet, which parameter can be used to count the number of each endocrine cell type and analyze cellular composition and geographic islet architecture. (B, g) Total endocrine cell area shown as a converted 8-bit mask after automatic thresholding. (B, h) Total islet area that includes unstained fractions such as intraislet capillary. (C) Table summarizing data obtained through the computer-assisted large-scale analysis. Note that each islet including a small cluster is designated with an identification number so that specific information on a given structure can be obtained.

Quantification of cellular composition (i.e., β-, α-, and δ-cell populations and nuclei; Fig. 1B, a-d) is performed using a macro written for ImageJ (free software that can be downloaded at http://rsbweb.nih.gov/ij/). A macro is a custom-written script that provides instructions for the quantification of interest in each application. A composite of all three endocrine cells and nuclei is shown in Fig. 1B, e. To analyze islet architecture, the center coordinates of each endocrine cell type within a given islet are measured. First, the DAPI fluorescent signals are converted to 8-bit masks and watershed to obtain masks of individual nuclei. The pixels surrounding nuclei masks are quantified with respect to each endocrine hormone staining (i.e., insulin, glucagon, or somatostatin) to identify which hormone is most prevalent around each nucleus and each coordinate is recorded. DAPI signals outside of islets are not included. The coordinates are used to count the number of each endocrine cell type and analyze cellular composition and geographic islet architecture. Reconstructed endocrine cell distribution within each islet is shown in Figure 1B, f. Total endocrine cell area is measured using a converted 8-bit mask after automatic thresholding (Fig. 1B, g). Total islet area that includes unstained fractions such as intraislet capillaries is measured by automated contouring of each islet structure (Fig. 1B, h). Each islet including small clusters is designated with an identification number so that specific information on a specific structure can be obtained as summarized in the table. A video capturing the computer-assisted analysis is provided online (Video 1).

Figure 2 illustrates islet size dependent changes of endocrine cellular composition in humans compared with mice. Islet sizes are in a wide range from a single endocrine cell to a large islet consisting of several thousand cells. In addition, smaller islets are more frequent. Therefore, a logarithmic size scale is appropriate for the histogram of islet sizes because it gives fine bins for the high number of small islets and large bins for the low number of large islets. This provides not only sharp size categories of small islets but also statistically adequate islet samples at large size bins. Here islet areas are divided by the single-cell area, 170 μm²,⁷ to make them dimensionless values representing the number of cells in a given islet area. We have also provided a conversion meter between an effective islet diameter (i.e., a diameter of a normalized area of a perfect circle) and the logarithmic value of dimensionless islet area so that readers can intuitively perceive the size scale chosen. In mouse islets, the β-cell is the major component of islets throughout the size distribution (Fig. 2A, left). However in human islets, while small islets show similar cellular composition with mice, in larger islets (> 60 µm in diameter; 1) the fraction of α- and δ-cells increases (Fig. 2A, right). These large islets typically exhibit relatively intermingled architecture of β- and non-β-cells.^8,9 Note that such changes in large islets are not an intrinsic characteristic of human islets, but are also observed in mice under insulin resistance such as pregnancy, obesity, diabetes and inflammation.⁹ Interestingly, as we reported, a preferential loss of β- and α-cells was observed in large islets of over 60 µm in diameter in patients with T2D.¹ Currently, little is known about the relationship between such changes in islet structures and the functional properties of islets. As we have shown, the regulation of islet size (limited to ~500 µm in diameter) throughout mammals that we have studied,^9,10 the increased ratio of non-β cells and intermingled architecture suggests an important role for intraislet paracrine effects of the endocrine cell network. Figure 2B shows the relative contribution of each bin of small to large islets to total endocrine cell area (left, mouse and right, human corresponding to Fig. 2A). The greater number of endocrine clusters and small islets (gray bars) does not markedly share the total area (plotted in a red line), but the fewer number of large islets mainly comprises the islet mass. The contribution of large islets to the total endocrine cell area is highlighted in Figure 2C with the cutoff points in islet size of > 60, 100 and 150 μm in diameter (left, mouse; right, human corresponding to Fig. 2B).

Figure 2. Islet size dependent changes of endocrine cellular composition in human islets. (A) Left, mouse pancreas (a CD-1 female mouse at 3-mo). Relative frequency of islet size (gray bar) and ratios of α (red), β (green), and δ (blue) cells within islets are plotted against islet size; means ± SEM. Note that islet size is presented as a logarithmic scale considering the high number of small islets and the low number of large islets. In addition, islet area is divided by the single-cell area (170 μm² ; 7) to make them as dimensionless values representing the number of cells in a given islet area. See the conversion between logarithmic islet area (logarithmic) and effective diameter (μm). Right, human pancreas (same as in Figure 1). While small islets show the similar endocrine cell composition of the dominant fraction of β-cells in mouse islets, α- and δ-cell fractions increase in large islets in humans. (B) Fraction of islet size distribution (gray bar) and fraction of total islet area (red line). (C) The contribution of large islets to the total endocrine cell area is plotted with the cutoff points in islet size of > 60, 100 and 150 μm in diameter (left, mouse; right, human corresponding to Figure 2B).

Additional parameters we measure are circularity and Feret’s diameter of each islet structure.² Circularity reports the degree of roundness of a structure, where 1.0 corresponds to a perfect circle. Feret’s diameter is the longest distance within a structure. With a value of area, each and every islet in the whole specimen can be visualized in a 3D scatter plot (Fig. 3). The color-coding of the density of islets provides a visual assessment of the size and shape distribution. Figure 3A shows an example of non-diabetic (ND) young lean and old overweight persons; left, a 15-y old male with BMI = 16 and right, a 51-y old female with BMI = 29, where the latter exhibits larger and more spherical shaped islets. Another example of age-matched female subjects with non-diabetic (left, 63-y) and T2D (right, 66-y) is shown in Figure 3B. Preferential loss of large islets is noted in the patient with T2D. These plots are generated using Mathematica (Wolfram Research).

Figure 3. Islet size and shape distribution. Three-dimensional scatter plots. Each dot represents a single islet/cluster with reference to size (area) and shape (circularity and Feret’s diameter). The density of islets is color-coded from sparse to dense. (A) Right, a 15-y old male with BMI = 16. Left, a 51-y old female with BMI = 29. (B) Left, a non-diabetic 63-y old female. Right, a 66-y old female with T2D.

A simple quantification of islet architecture is to calculate cell-cell distances between every pair of two cells within an islet using a center coordinate of each cell. The conceptual basis of the analysis is illustrated in Figure 4A with two model islets: (Model 1) the core of β-cells accompanied by α-cells in the periphery (which is typically observed in wild-type mouse islets); and (Model 2) β- and α-cells intermingled (which is often observed in large islets in humans). When the cell-cell distances (between β-β, α-α and β-α cells) in Model 1 are plotted against relative frequency, the core-forming β-cells fit into a bell-shaped curve, whereas the α-cells in the periphery lead to a rightward shift in the perfect spherical islet. In Model 2 with intermingled β- and α-cells, however, all three cell-cell distance curves are overlapped in a bell-shape. We then validated our models using isolated mouse and human islets. Shown in Figure 4A is a pair of representative results from over 100 islets examined (Hara et al., unpublished). Although all three cell-cell distance curves are more similar in human islets compared with mouse islets, they are not as perfectly coincident as Model 2 in which α- and β-cells are randomly intermingled. To examine islet architectures in details, we have introduced cell-cell contact probabilities.¹ As schematically described in Figure 4B, we identified the neighborhood of each cell based on center coordinates of individual cells within an islet, and calculated the cell-cell contact probabilities. For example, the probability, Pαβ, quantifies how many α-cells contact with β-cells in a given islet. We calculated the cell-cell contact probabilities of Models 1 and 2 for mouse and human islets (Fig. 4C). For a given composition of α- and β-cells, Pα and Pβ, the cell-cell contact probabilities could be easily predicted in the case of a random mixture of them: Pαα = PαPα; Pββ = PβPβ; and Pαβ = PαPβ+PβPα = 2PαPβ. The predictions are based on the probabilities that two neighboring sites are independently occupied by two α-cells; two β-cells; and one α-cell and one β-cell or vice versa. As expected, Model 1 showed distinctive cell-cell contact probabilities compared with the random mixture where β-cells prefer to contact with β-cells, while Model 2 showed the same probabilities with the random mixture. Mouse and human islets also showed the preferential attachment between β-cells. We have used this method in two-dimensional pancreatic sections, and obtained the same conclusion in human islets.¹ The characterization of human islets thus concluded that although more abundant α-cells are intermingled with β-cells in the islet core as well as periphery, they are clearly not a random mixture of α- and β-cells. Note that the cell-cell distance distributions and contact probabilities could be generalized with the inclusion of δ-cells, the third endocrine cell type.

Figure 4. Characterization of islet architecture. (A) Cell-cell distance distributions between α-cells (+; red), α- and β-cells ( × ; blue), and β-cells (*; green) in Model 1, Model 2, and representative mouse and human islets. Three-dimensional positions of α- (red) and β-cells (green) in isolated islets were determined by confocal microscopy. (B) A schematic diagram to show how to calculate cell-cell contact probabilities between α-cells (Pαα), β-cells (Pββ), and α- and β-cells (Pαβ) in an islet. (C) Cellular compositions of α-cells (Pα) and β-cells (Pβ); and cell-cell contact probabilities were calculated for the above Model 1, Model 2, mouse, and human islets. Note that PαPα, PβPβ, and 2PαPβ in the parentheses correspond to the cell-cell contact probabilities in the case of a random mixture of α- and β-cells for the given Pα and Pβ.

In summary, the large-scale computer assisted image capture and analysis minimizes sampling bias and allows us to quantitate islet size distribution, cellular composition and architecture. As we have shown in our study,¹ the representative data obtained from each human pancreas tissue enabled us to reveal the pathogenesis of T2D within a limited sample size, where the variability of pancreas pathology in humans compared with that of inbred mice is noted. Ongoing studies include regional differences (i.e., head, body and tail regions) of the human adult pancreas, fetal and neonatal development, and pathogenesis associated with aging. Our aim is to provide a baseline of information on the pathophysiology of human pancreas and further fill a gap in our understanding of islet biology, particularly in terms of translational studies of animal models. Shifting research from experimental animals to humans is not only impractical but also limits our investigations in a major way, not to mention disregarding a wealth of knowledge accumulated in the field based on animal studies. It is rather important to know the similarities and differences between preclinical models and human pathophysiology so that bi-directional studies should be more advanced and become practical. With recent major advances in genetics and epigenetics, future studies should further integrate pathophysiology with genomic and epigenomic data.

Kilimnik G, Zhao B, Jo J, Periwal V, Witkowski P, Misawa R, et al. Altered islet composition and disproportionate loss of large islets in patients with type 2 diabetes. PLoS One. 2011;6:e27445. doi: 10.1371/journal.pone.0027445.

Note

Supplemental material can be found at: www.landesbioscience.com/journals/islets/article/19256