Heterogeneity of the Human Pancreatic Islet

Functional heterogeneity of β-cells in regard to glucose-responsive insulin secretion was reported decades ago (1–3). With the recent advancement of technologies, specialized pacemaker cells, termed “hubs,” were identified as 1–10% of β-cells with a high degree of connectivity, high glucokinase, and low insulin, Pdx1, and Nkx6.1 expression (4). Similarly, molecular signatures of subpopulations of β-cells have been documented in maturation (Fltp [5], Srf [6], and Ucn3 [7]), proliferation (C1 [8]), and aging (IGF-IR [9]). Several other markers differentially expressed in subgroups of β-cells have also been reported, including PSA-NCAM (10), ST8SIA1 and CD9 (11), and MST1 (12). To date, all of these markers are solely claimed by each group of authors and not validated by other investigators. Therefore, relationships among these markers are still unknown (13). This is an exciting era in which new molecular insights of β-cell heterogeneity have begun to be revealed, which is thus currently one of the most focused areas in diabetes research.

Importantly, such heterogeneity extends to the pancreatic islet in its cytoarchitecture, particularly in humans, unlike in rodent islets (14–19). This heterogeneity of the human islet has not been systematically studied to date and thus may not have been properly recognized. It is well known that islets differ markedly in size throughout species from small to large islets consisting of thousands of endocrine cells. We have shown islet size–dependent variability in humans, including islet architecture, cellular composition, and intraislet capillary density (20). Humans can have 1 million islets (21). Many studies in the past were undertaken with only a few islets per donor. In some studies at single-cell levels, for example, ∼50–70 cells were considered to represent one donor’s endocrine pancreas, where a single islet with a 150-μm diameter can have ∼1,000 cells.

The heterogeneity of the human pancreas is observed in four levels, in an ascending order: 1) β-cell heterogeneity, 2) islet heterogeneity, 3) intraindividual regional variability of islet distribution, and 4) interindividual heterogeneity. In this study, we aimed to show the islet heterogeneity in humans to demonstrate the need for baseline work, before exploring β-cell heterogeneity, by using a small number of so-called representative islets that have been somewhat allowed in the field. A study design inattentive to the aforementioned four levels of the heterogeneity of the human pancreas could result in misleading conclusions.

Human pancreata were provided by the Gift of Hope Organ & Tissue Donor Network Organ Procurement Organization in Chicago. Written informed consent from a donor or the next of kin was obtained for use of a sample in research. The use of deidentified human tissues in the study was approved by The University of Chicago Institutional Review Board. Over the course of developing a three-dimensional (3D) imaging method, several specimens were randomly used unless otherwise specified.

Pancreas tissues from 8-week-old C57BL/6J male mice were used. All of the procedures involving animals were approved by The University of Chicago Institutional Animal Care and Use Committee.

The following primary antibodies were used: mouse monoclonal anti-insulin (cat. no. ab8304, RRID:AB_2126399; Abcam, Cambridge, MA), mouse monoclonal anti-human glucagon (cat. no. WH0002641M1, RRID:AB_1841774; Sigma-Aldrich, St. Louis, MO), mouse monoclonal anti-somatostatin (cat. no. ab30788, RRID:AB_778010; Abcam), mouse monoclonal anti–pan-endocrine (HPi1; cat. no. MA5-16126, RRID:AB_11157008; Thermo Fisher Scientific, Waltham, MA), rabbit polyclonal anti-CD31 (cat. no. ab28364, RRID:AB_726362; Abcam), rabbit anti–α-smooth muscle actin (α-SMA; cat. no. ab5694, RRID:AB_2223021; Abcam), and DAPI (Invitrogen, Carlsbad, CA). The primary antibodies were conjugated with a combination of amine-reactive fluorophores (N-hydroxysuccinimide-esters; Thermo Fisher Scientific, Kalamazoo, MI).

A frozen human pancreas tissue block (∼5 mm thick) was fixed in 4% paraformaldehyde, embedded in 2% agarose gel, and mounted on a vibrating microtome. Sections (600–800 µm thick) were collected in cold PBS. These macrosections were then immunohistochemically stained overnight. Optical clearing was performed by sequential incubation with 20, 50, 80, and 100% (w/v) solutions of d-fructose and 0.3% (v/v) α-thioglycerol (Sigma-Aldrich) for 1–2 h each at 35°C with gentle agitation. A Leica SP8 laser scanning confocal microscope was used to image tissue slices mounted between coverslips. 3D reconstruction and analysis were performed using Fiji and Imaris software (Bitplane, Concord, MA).

Human pancreatic islets were captured in situ in thick tissue slices (600–800 µm in depth). Shown in Fig. 1Aa is a large-scale capture of a pancreatic tissue slice from a 13-year-old girl, immunostained for insulin (green), glucagon (cyan), CD31 (red), and α-SMA (yellow). Here, large blood vessels (i.e., arteries and veins) and arterioles are labeled with α-SMA (22). Note that most of the islets appear to have one feeding arteriole. This stack of multifluorescent images was surface rendered in 3D (Fig. 1Ab), which provides solid views of the morphology of structures and the basis for volume quantification. Various sizes of islets were located along with arterioles (Fig. 1B). Also observed was heterogeneity in intraislet capillary density, which generally tended to be higher in larger islets (Fig. 1C), as we have previously reported (20).

The proportion of three major islet cells was quantified in 3D within intact pancreatic tissues. The total volume of individual islets was assessed by surface rendering of HPi1 staining [23]). A pancreatic tissue slice from a 24-year-old man was immunostained for insulin (green), glucagon (cyan), somatostatin (blue), and HPi1 (red) (Fig. 2A). Our 3D quantification method is shown step by step in Fig. 2B–D using a cluster of three islets boxed in Fig. 2A. First, fluorescent signals from the z-stack of all four channels were processed by binary conversion using Fiji (Fig. 2B). Next, the individual islet boundary was identified by 3D surface rendering of pan-endocrine fluorescent binary signal using Imaris (Fig. 2Ca). Then, the volume of each islet and three types of endocrine cells was measured in 3D using fluorescent signals (Fig. 2Cb–e). An example of z-stack views from top to bottom of islets is shown in Fig. 2D. Here, to recapitulate the common practice of selecting a small number of islets based on the idea of “the prototypical islet,” five islets of different sizes were selected for each donor (Fig. 2E). The percentage of islet volume constituted by β-cells (insulin, green), α-cells (glucagon, cyan), and δ-cells (somatostatin, blue) was quantified. Marked intradonor islet variability was observed across a wide range of islets of varying volume.

To recapitulate the common practice of selecting a small number of islets based on two-dimensional (2D) imaging and further demonstrate the resulting variance, we performed simulation by randomly selecting islets (as 1, 2, 3… up to 2,500 without replacement) from the whole human pancreas. Here we used previously characterized 10 age-matched male subjects without diabetes (21). The simulation workflow is shown in Fig. 3A. A subset of islets >40 μm in diameter was sampled from each of 10 subjects, and the average β-cell area was measured (Fig. 3Aa). This sampling was repeated 1,000 times, and the mean β-cell area was plotted (Fig. 3Ab). The total number of islets examined and subject information obtained from our previous study are provided. The distribution of β-cell composition by islet size was examined in three groups of islets according to a range of diameters (Fig. 3Ac). Figure 3B depicts the simulation performed in the 10 human pancreatic samples. Mean ± SD and median β-cell area values are displayed for each donor. The distribution of β-cell composition sorted into three diameter ranges (100–200, 200–300, and 300–400 μm) revealed the extent of the islet heterogeneity within a similar size range in all individuals examined (Fig. 3C). Mean ± SD and median β-cell proportion for islets 100–200 μm in diameter are shown for each subject. A summary plot of the simulation in Fig. 3D provides the margin of error from the mean β-cell ratio of all islets in the pancreas depending on the number of islets sampled for each of 10 samples. Each point represents the difference between the mean β-cell ratio from a single sampling simulation for each sample size and the mean β-cell ratio of all islets from that subject. CIs for β-cell proportion are summarized in Table 1. Accordingly, to achieve a 95% CI of width of 5% around the mean β-cell ratio of the whole pancreas, a minimum of 400 islets would need to be examined to take the islet heterogeneity in humans into consideration. Importantly, note that the CIs are parameter specific. To assess the number of islets required for different parameters, large-scale data collection of individual measurements will be needed.

In 2D immunostaining, as shown in Fig. 4A (glucagon in red, insulin in green, and somatostatin in gray), mouse islets exhibit well-known mantle-core structure in that non–β-cells were mostly found in the periphery, although they do not form a complete mantle (Fig. 4Aa). In human islets, non–β-cells are found in the core of islets, from small to large ones (Fig. 4Ab–d), as has been previously documented (14–19). Note that α-cells were not randomly dispersed or scattered throughout the islets, as previously reported (14,15). Oftentimes, there appears to be a lining of α-cells surrounding a cluster of β-cells when observed in 2D slices, which has been referred to as “a subunit” resembling rodent islets (24–27). The authors noted that islets with higher proportions of β-cells (i.e., fewer α-cells) showed a clear pattern of such subunits of the rodent mantle-core organization, whereas some large islets with a lower proportion of β-cells (i.e., more α-cells) gave “the suggestion of random organization” with less clear subunit formation. Nonetheless, it was still considered that there was clear clustering of β-cells, “just smaller clusters” (27).

We hypothesized that depending on how a given islet was cut and imaged in 2D, some regions would show such structures. Therefore, we tried to reproduce similar 2D views by optically slicing a 3D z-stack image of solely the glucagon channel to highlight the lining of α-cells (Fig. 4B). Shown here are some optically sliced views from top to bottom (Fig. 4Ba–f), 3D surface-rendered views of islets (Fig. 4Bg), and the side views of g (Fig. 4Bh). The first column depicts a mouse islet with a typical mantle-like pattern of α-cell distribution. However, no clear mantle-like pattern was observed in human islets. Here, we selected the above-mentioned two types of islets from three individuals. The amount of α-cells (i.e., larger or smaller proportion of an islet) was judged by 3D reconstructed views. The left column of each islet pair shows an islet with relatively fewer α-cells, and the right column shows an islet with more α-cells: I, a 16-year-old girl; II, a 59-year-old man; and III, a 62-year-old woman. In I, optical slicing of 3D z-stack images of a small islet with fewer α-cells may appear to be a rodent-like mantle pattern in selected 2D slices (Fig. 4Bd–f). However, we failed to observe the same structure when this islet was reconstructed in 3D (Fig. 4Bg and h). Interestingly, a larger islet with more α-cells showed subunit-like lining of α-cells (suggested by the aforementioned authors) in some of its 2D optical slices (Fig. 4Bb–e). However, no rodent-like subunits were observed in 3D (Fig. 4Bg and h). In the Supplementary Data, a montage of the entirety of optical slices of this islet (n = 90, 1 μm increment) is shown in Supplementary Fig. 1, together with a movie of the top-to-bottom view (Supplementary Movie 1). A 3D view of the islet is shown in Supplementary Movie 4. Similarly, as demonstrated in II and III, regardless of the proportion of α-cells, a certain lining of α-cells that may be suggestive of “a mantle-core pattern” can be observed depending on which optical plane was chosen in 3D as well as how a given islet was sectioned in 2D. We further captured a large islet (e.g., Feret diameter = 335 μm) with abundant α-cells in which a mantle-like structure was composed of a single layer of α-cells, which resembles the islet shown in Fig. 4Ad. Note that the structure of α-cells found in these islets is rare. Montage and movies for II, III, and Fig. 4C are provided in the Supplementary Data (Supplementary Figs. 2–4 and Supplementary Movies 2, 3, and 5–11).

The 3D analysis also revealed an architectural relationship between different types of endocrine cells and the islet capillary bed. The limitation of 2D analysis is particularly evident for the network structure such as vasculature, neurons, and the extracellular matrix. Markers for all of these networks only appear as “dots” or “short lines” in 2D images of thinly cut sections, as we recapitulated such 2D views out of 3D imaging shown in Fig. 5Aa–d. However, when these optical 2D slices were reconstructed in 3D, vascular markers appearing once as dots or short lines resolved to a complex network of intraislet capillaries (Fig. 5Ae and f). Clustering of the same type of cells was observed, naturally to a higher extent for β-cells. Islet architecture varied throughout the size distribution, and all types of endocrine cells were capable of direct contact with islet capillaries (Fig. 5B–E). This is opposed to a previous report that α-cells were exclusively aligned along blood vessels (28).

The current study has demonstrated considerable islet heterogeneity and that there is no “stereotypic islet” in humans. Decades of research in rodent islets have concluded a somewhat stereotypical islet structure of a β-cell core with α- and δ-cells in the periphery. However, we have shown the atypical nature of the human islet, even within the same individual with regard to cellular composition as well as islet architecture. Endocrine cells are not dispersed or scattered throughout the human islets. Indeed, the predominant population of β-cells formed clusters within an islet, as did α-cells as well, to a lesser extent. We previously examined the relative attractions between islet endocrine cell types computationally and found that the attractions between homotypic cells were slightly, but significantly, stronger than the attractions between heterotypic cells, commonly in mouse, pig, and human islets (29). Interestingly, the difference between β-cell–to–α-cell attraction and β-cell–to–β-cell attraction was minimal in human islets, maximizing the plasticity of islet structures. Small islets usually have a low proportion of non–β-cells and thus tend to resemble the rodent mantle-core pattern; however, in this study, we also found some small islets with non–β-cells located in the center of islets. In addition, contrary to a report by Bosco et al. (28), vasculature penetrated small islets as well as the clusters of β-cells in larger islets. The authors also proposed “trilaminar epithelial plates” of islet organization, comprising one layer of β-cells sandwiched between two α-cell–enriched layers, and this structure should have a folded pattern and vessels circulate along both of its sides (28). We failed to observe such an organization in 3D.

An arrangement of rodent-islet-like subunit formation in the human islet has long been proposed, based on immunohistochemical analysis of thinly cut pancreatic tissue sections (24–27). Indeed, it had once been quite convincing, before the recent technological advancement of 3D analysis including microscopes, tissue clearing methods, and image analysis software. It may demonstrate our inability to reconstruct 2D phenomenon in 3D in the human brain in general. For example, we may tend to fill apparent gaps of the lining of α-cells as a continuous “mantle” and further extend it to an imaginative spherical structure in 3D (16). Such a phenomenon, in which the human brain perceives patterns as organized beyond their real presence, was recognized since as early as 1923, and described as the Gestalt Principles of Grouping (30). In the 3D analysis of the human pancreatic tissues, we did not find any rodent-islet-like subunit formation.

Unfortunately, heavy workload would not justify studies with an extremely small number of islets per donor. Such studies could lead to biased interpretations of experimental results and, therefore, would not ultimately benefit investigators themselves. Reasonable experimental design is the most important for the time and efforts undertaken for any studies. Then the question is how many islets out of a potential 1 million would represent the endocrine pancreas of one human. A range of “stereologically” appropriated numbers of islets to examine may depend on types of studies. In our study of β-cell/islet mass and islet number in the whole human pancreas, we first verified the frequency of sampling tissues (31). We found that with the unbiased examination of an entire tissue section, one section can sufficiently represent a pancreatic tissue block of ∼1 cm in thickness from head to body to tail regions. Therefore, we divide a pancreas into consecutive blocks (∼5 mm) with alternating collection between paraffin embedding and snap freezing. Then paraffin-embedded sections (5 μm) are used for computer-assisted quantification of β-cell/islet mass. While singly scattered endocrine cells and small clusters are always included, in this study, only islets >40 μm in diameter were examined for meaningful sizes of islets. The total number of islets examined in 10 cases was in a range of ∼3,500–14,000 (average ± SD of 8,774 ± 3,078), reflecting the interindividual islet heterogeneity (32).

Our rationale for the whole pancreas analysis, despite its laborious effort, is based on the striking intraindividual regional variability of islet distribution (21,31,33,34). Studies including only a limited sample of pancreas may miss important variance both within the same pancreas and across individuals. Therefore, similarly, regarding islet heterogeneity, we propose that sufficient numbers of islets should be first explored in each experiment. Such pilot studies, while requiring certain significant commitment, should strengthen individual work in the long run. Studies using the state-of-art new technologies may be considerably expensive, labor intensive, and far from high throughput. On such occasions, rather than compromising with a small number of specimens, we should proactively seek for collaboration and further a nationwide and international consortium.

Computer modeling, including in silico screening, is suggested as one of the useful tools for the interrogation of β-cell heterogeneity by Benninger and Hodson (13). In their Perspective, the authors discussed the importance of using intact islets compared with dissociated endocrine cells, since interactions between cells within the islet are critical and tissue architecture influences gene expression. Furthermore, while enzymatic dissociation usually reduces cell viability with altered transcriptomic/protein features, the authors warn that some subpopulations with immature phenotypes, including hubs, appear more susceptible to endoplasmic reticulum stress and apoptosis/cell death. In fact, in the groundbreaking hubs study by Hodson and colleagues (4), a potential limitation of the study of recording Ca²⁺ oscillations in β-cells within a ∼20-µm confocal plane, not in the entire islet, was compensated by mathematical modeling approaches such as Monte Carlo–based correlation analyses and power law link-probability distribution. The existing gaps in experiments between dissociated single cells and intact islets, as well as between isolated islets and islets in vivo, may be filled with computer modeling, at least to some extent.

In the very near future, with the anticipated rapid advancement of artificial intelligence (AI) and accompanied machine learning technology, many experimental approaches and techniques that have been believed to deserve “expertness and dedication of human beings” would be widely replaced with the AI, sooner or later. Importantly, biased data fed to the AI to teach a machine learning algorithm have been shown to significantly influence its behavior (the world’s first psychopath AI [http://norman-ai.mit.edu/]). We propose that during this transitional era, it is critical to perform arduous but important baseline work in order to build the foundation for future studies.

Acknowledgments. The authors acknowledge Dr. Piotr Witkowski and Dr. J. Michael Millis at The University of Chicago, and Dr. Martin Jendrisak and the entire team of the Gift of Hope Organ & Tissue Donor Network in Chicago for providing the human pancreas tissues used in the current study.

Funding. The study was supported by National Institutes of Health, DK-117192 and DK-020595, to the University of Chicago Diabetes Research and Training Center (Physiology Core), DK-072473, and AG-042151, and a gift from the Kovler Family Foundation to M.H. Imaging was performed at The University of Chicago Integrated Light Microscopy Facility.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.P.D. and M.H. performed tissue preparation, data collection and analyses, and manuscript preparation. M.H. designed the study. M.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Data and Resource Availability. All data generated or analyzed during this study are included in the published article (and its online Supplementary Data). No applicable resources were generated or analyzed during the current study.