figure b

Introduction

Type 1 diabetes is an autoimmune disease in which insulin-producing beta cells are damaged by the immune system leading to loss of beta cell function and, ultimately, irreversible insulin deficiency [1]. However, the immunological mechanisms that lead to recognition of beta cells by the immune system are not fully understood. Genetic susceptibility predominantly mapping to specific MHC class II alleles confers an increased risk for some patients. Population studies have indicated that HLA class II (HLA-II) variants of HLA-DRB1, -DQA1 and -DQB1 genetically predispose patients to type 1 diabetes [2, 3].

Classical HLA molecules are cell surface glycoproteins involved in antigen presentation, typically categorised as HLA class I (HLA-I) and HLA-II [3]. HLA-I molecules are expressed on all nucleated cells, while HLA-II molecules are known to be constitutively expressed by immune antigen-presenting cells (monocytes, macrophages and dendritic cells), B lymphocytes and activated T lymphocytes [4]. In other cell types (e.g. some epithelial cells), HLA-II expression can be induced under certain conditions (e.g. by exposure to IFN-γ) [5]. In the healthy pancreas, beta cells do not express HLA-II [6], but they do express low levels of HLA-I, similar to any other nucleated cell. HLA-I hyperexpression is a hallmark of type 1 diabetes [7] that can also be present in autoantibody positive (Aab+) donors [8] and thus might be essential in sensitising beta cells for immune recognition. However, the expression of HLA-II by islet beta cells of patients with type 1 diabetes has been historically controversial. In the 80s, Bottazzo et al described HLA-II expression by some beta cells in one individual with type 1 diabetes [9]. Foulis et al reported ‘aberrant’ HLA-II expression by many beta cells in islets expressing both insulin and HLA-I. This HLA-II expression in the beta cells was present in 27 of the 35 individuals with type 1 diabetes examined, suggesting it was quite a frequent phenomenon unique to type 1 diabetes (i.e. absent in type 2 diabetes and chronic pancreatitis) [10, 11]. However, other groups could not replicate these results, and several reports indicated that HLA-II expression was not observed in beta cells [12, 13] or it was present in the cells surrounding some islets [12, 14]. Others reported a low HLA-II expression in a few beta cells in just some donors with type 1 diabetes (e.g. 1/18 patients with recent onset type 1 diabetes [14]), or in a limited number of islets (<1%) [14,15,16]. A recent study by Russell et al re-invigorated the old debate when they reported an upregulation of HLA-II-associated genes in bulk-sorted and single beta cells isolated from donors with type 1 diabetes [17]. They also found the presence of HLA-II protein and its transcriptional regulator, HLA-II MHC transactivator protein (CIITA), expressed by a subset of insulin+CD68 beta cells in patients with type 1 diabetes. Other studies had previously demonstrated that HLA-II expression may be triggered in vitro on beta cells exposed to combinations of proinflammatory cytokines [18,19,20,21], even though they failed to induce it with IFN-γ or TNF-α alone [18,19,20].

The aim of our project was to precisely quantify the expression of HLA-II in the beta cells of human pancreatic tissue sections obtained from the Network of Pancreatic Organ Donors (nPOD), to investigate the cellular sources of HLA-II within the islets and to analyse whether HLA-II expression could be induced in isolated human islets or in reaggregated human islets.

Material and methods

Pancreatic organ donors and islet donors

Formalin-fixed paraffin-embedded (FFPE) tissue sections of human pancreas were obtained through nPOD from 15 brain dead organ donors (non-diabetic n = 5, Aab+ n = 5, and type 1 diabetic donors n = 5) whose age, sex and BMI had no statistically significant differences between disease groups (electronic supplementary material [ESM] Fig. 1). The demographic and clinical information of donors are detailed in Table 1.

Table 1 Demographic and clinical characteristics of nPOD donors

Isolated human islets were obtained through the Integrated Islet Distribution Program (IIDP, City of Hope, USA; http://iidp.coh.org/), and 3D InSight Human Islet Microtissues (reaggregated islets) were obtained through InSphero (InSphero AG, Schlieren, Switzerland). The islets used for the production of InSphero 3D reaggregated islets were obtained through Prodo Laboratories (Irvine, CA, USA). All the islet donors were non-diabetic; details are summarised in the human islet checklist (ESM Table 1). All samples were provided anonymously, collected with consent from next of kin, and exempt from institutional review board oversight.

Human isolated islets and reaggregated human islets culture

3D InSight Human Islet Microtissues were produced and supplied by InSphero according to their protocols [22]. Islet microtissues were cultured in 3D InSight Human Islet Maintenance Medium (InSphero, CS-07-005-01) for 24 h upon arrival. Isolated human islets from IIDP were incubated in CMRL medium (Thermo Fisher Scientific, USA, catalogue number 11530037) supplemented with 10% FBS, 2 mmol GlutaMAX (Gibco, Thermo Fisher Scientific), 100 U/ml penicillin, and 100 μg/ml streptomycin upon arrival. After 48 h, islets were handpicked and transferred to a U-bottom 96-well spheroid carrier plate (Perkin Elmer, USA, catalogue number 6055330), seeding five islets per well.

Prior to functional and microscopic analysis, isolated and reaggregated human islets were cultured for 4 days with varying concentrations of proinflammatory cytokines (IFN-γ [Sigma-Aldrich, Germany, #I17001], TNF-α [Sigma-Aldrich, #H8916], IL-1β [Sigma-Aldrich, #H6291]) or with vehicle control (medium) (ESM Table 2). Every condition was replicated five times, and redosings with cytokines and media exchange were performed every 1–2 days.

Prior to RNA sequencing, islet microtissues were treated for 7 days with a cytokine cocktail containing PBS supplemented with 0.1% BSA (Sigma-Aldrich, #A7888), TNF-α 10 ng/ml (Thermo Fisher Scientific, #PHC3016), IFN-γ 10 ng/ml (Sigma-Aldrich, #I3265), and IL-1β 2 ng/ml (Sigma-Aldrich, #I17001). In this experiment, a total of 60 islet microtissues from two donors were used: two different conditions were tested in each donor, and five technical replicates were used per condition (with three islets per technical replicate). Redosings with cytokines and media exchange were performed every 1–2 days.

Glucose-stimulated insulin secretion test

The functionality of islets was assessed by glucose-stimulated insulin secretion (GSIS) assay. Low glucose (2.8 mmol) and high glucose (16.6 mmol) solutions were prepared in KRB solution supplemented with 0.5% BSA. First, supernatants were collected, and islets were equilibrated to low glucose for 1 h. Next, islets were incubated at 37°C with low glucose for 2 h, followed by high glucose for 2 h, and then low glucose again for 2 h. Supernatants were collected after every incubation step and stored at −20°C. ELISA was then performed to detect insulin concentrations in the supernatants, according to the manufacturer’s instructions (ALPCO, USA, catalogue number 80-INSHU-E01.1).

RNA sequencing, library creation and data analysis

RNA extraction and processing were performed at InSphero (Switzerland). At the end of the cytokine treatment, three islet microtissues were pooled, washed once in 70 μl of PBS without Ca2+/Mg2+, and then lysed in 15 μl of 1x Enhanced Lysis Buffer (BioSpyder Technologies, USA). Crude islet lysates were used to generate sequencing libraries relying on the standard TempO-Seq chemistry. Generation of sequencing libraries and RNA sequencing was performed by BioSpyder using TempO-Seq technology (https://www.biospyder.com/). A Human Whole Transcriptome 2.0 assay, consisting of 22,537 probes targeting 19,701 genes, was used. Purified libraries were run on the Illumina sequencing platform (Illumina, USA). After sample demultiplexing and obtaining single FASTQ files, reads alignment and reads counting were done by BioSpyder using the TempO-SeqR program (BioSpyder Technologies). Further TempO-Seq data analysis was done by InSphero via proprietary R-based (https://www.R-project.org/ [23]) analytical pipeline. The resulting probe-wise raw count table was collapsed towards gene-wise count table by summing counts for probes associated with the same gene. Additionally, genes for which there were zero counts across all samples were removed from further analysis. Data normalisation and differential expression analysis (DEA) was performed as implemented in the DESeq2 R package [24]. Furthermore, surrogate variable analysis (SVA) was applied to remove unintended batch effects in the data by using SVA R package version 3.36.0 [25]. Benjamini–Hochberg procedure of false discovery rate (FDR) was used to correct for inflated p values across the three contrasts defined in the study [26]. FDR cut-off was set to 0.001 to distinguish significant differential expression.

Immunofluorescence staining

FFPE pancreatic nPOD sections

Slide-mounted tissue sections were de-paraffinised and rehydrated in descending concentrations of alcohol. Antigen retrieval was performed by incubating slides in citrate buffer pH 6 at 95°C for 20 min. Slides were positioned in a 3D printed immunostaining slide rack (https://3dprint.nih.gov/discover/3dpx-012172) for the remainder of the staining procedure. Tissues were blocked with 10% normal goat serum for 1 h at room temperature and then stained with antibodies detailed in ESM Table 3. Finally, slides were counterstained with Hoechst 33342 (Thermo Fisher Scientific, catalogue number H3570) and mounted using ProLong Gold anti-fade (Thermo Fisher Scientific).

Isolated islets and reaggregated islets

Islets were fixed using 4% paraformaldehyde for 2 h at 4°C and then permeabilised with 0.5% Triton-X in goat serum dilution buffer for 2 h at room temperature while shaking. Islets were blocked using Human FcR block (BD Biosciences, BD, USA) for 20 min, then stained with antibodies diluted in 0.3% Triton-X in goat serum dilution buffer (ESM Table 3), and counterstained with Hoechst 33342. During all the incubation times, the islets were gently shaking in the plate. Finally, islets were transferred to Superfrost slides and mounted using ProLong Gold anti-fade.

Image acquisition

Widefield images of whole pancreatic tissue sections were obtained using Zeiss AxioScan.Z1 slide scanner (Zeiss, Germany) with a ×20 (0.8 numerical aperture [NA]) air objective. Z-stacks were collected with a 0.5 μm step size, and the best focused slice was chosen for each region and channel using a custom macro (https://github.com/saramcardle/Image-Analysis-Scripts, folder ‘Zen OAD Macros’). Additionally, high resolution images were obtained from randomly selected islets across each case and were acquired with the Zeiss laser scanning confocal microscope LSM880 with a ×63 (1.4 NA) oil objective using a 0.3 μm step size. High resolution images from isolated native and reaggregated islets were acquired using Zeiss laser scanning confocal microscope LSM780 with a ×40 (1.3 NA) oil objective (for InSphero islets) or a ×20 (0.8 NA) objective (for IIDP islets) using a 0.7 μm step size (obtaining about 40 z-slices per imaging field). From the tissue sections, we also obtained enhanced high resolution images from several beta cells and macrophages using confocal microscope LSM880 Airyscan, with a ×63 oil (1.4 NA) objective in order to identify the cellular location of the markers. All 8-bit images were acquired using the full dynamic intensity range and the same strategy was used to acquire all cases from the same experiment.

Image analysis

Widefield whole slide image analysis with QuPath v0.2.2 [27]

Minimum intensity thresholds for each channel were calculated as described previously [28], and 15 exocrine regions were selected across the tissue (ESM Fig. 2a). We trained a machine learning pixel classifier (random trees) to identify the islet regions automatically (ESM Fig. 2b). Next, we programmed a Groovy script to automatically expand the islet boundaries by 10 μm to define the peri-islet regions (ESM Fig. 2c). This script calculated the areas of the annotations (islet, peri-islet and exocrine), the mean intensity of the fluorescent signal and the area occupied by each marker alone or in combination in every annotation. Next, we optimised the built-in automatic cell detection based on the Hoechst signal and created insulin+ and HLA-II+ cell classifiers. We also trained a machine learning based object classifier to classify the cells that were CD68+. We programmed another Groovy script to retrieve the total number of cells and how many express each marker inside the annotations (ESM Fig. 2d). The density of beta cells expressing HLA-II was measured as the number of insulin+ HLA-II+ cells/islet area (cells/mm2). The exact same process and parameters were applied to all cases.

Confocal image analysis with Zen

Orthogonal views and maximum intensity projections (MIP) of z-stacks were generated using Zen Blue (v2.5, Zen blue edition, Zeiss, Germany). To analyse the cellular detail of the Airyscan images, central and peripheral (top and bottom) regions were defined for each beta cell based on the nucleus position, and an MIP was created for each region (ESM Fig. 3). Background thresholds for each channel were determined using Fiji ImageJ [29]. Islet boundaries were set based on insulin staining and then analysed using the co-localisation module in Zen Blue using the same defined thresholds for all the cases. Within each islet, percentages of positive area and mean intensity for every channel were calculated. Weighted Mander’s correlation coefficients were calculated for two channels at a time, representing the weighted percentage of overlapping pixels to total pixels.

Image analysis with Imaris

Enhanced high resolution confocal Airyscan images were used to create 3D representations of beta cells (v9.6 Imaris x64, Oxford Instruments, UK). Thresholds were set above background and the co-localisation module was used to create a new channel to visualise the overlap between two markers. Iso-surfaces were created to visualise the fluorescent signal of HLA-II and Hoechst within 3D space. A 3D approximation of the cell outline was created based on the insulin signal.

Statistical analysis

Prism 9 software (GraphPad Prism version 9.1.0 for macOS, GraphPad Software, USA, www.graphpad.com) was used for statistical analysis. First, we performed a descriptive statistical analysis, plotted the data to identify outliers and created box-plot graphs. We used the Shapiro–Wilk test for normality and determined that the data was not normally distributed. We used the non-parametric Kruskal–Wallis test to calculate pvalues and identify significant differences between more than two groups, and the Dunn’s multiple comparisons test to assess specific differences between groups. We used the non-parametric Mann–Whitney U test to compare the means of two groups. Two-way ANOVA was used to analyse differences in the GSIS response between islets receiving different concentrations of cytokines, with the Tukey’s test to assess specific differences between multiple pairwise comparisons. p values <0.05 were considered significant and adjusted when applying multiple comparisons.

Results

Insulin-containing islets from patients with type 1 diabetes express more HLA-II than islets from non-diabetic or Aab+ donors

Immunostaining of pancreatic sections revealed that some insulin+ cells within islets of individuals with type 1 diabetes expressed HLA-II, but this HLA-II expression was rare in non-diabetic or Aab+ donors (representative images in Fig. 1). We performed a comprehensive analysis of HLA-II expression within the islets using different approaches.

Fig. 1
figure 1

Representative images of islets from non-diabetic (non-DM), autoantibody positive (Aab+), and donors with type 1 diabetes (T1D). Shown are representative widefield images of islets from non-diabetic (nPOD #6098), single Aab+ (nPOD #6123), double Aab+ (nPOD #6197), and T1D (nPOD #6399) donors. nPOD pancreas sections were stained for antibodies against insulin (green), CD68 (white), HLA-II (red), and were counterstained with Hoechst (blue). Scale bar, 50 μm

From the entire tissue sections of all nPOD donors, we retrieved data from 7415 islets using QuPath. We observed a higher percentage of HLA-II-positive area in the insulin-containing islets (ICIs) of individuals with type 1 diabetes (24.31%) compared with non-diabetic (3.80%), Aab+ (2.31%), and type 1 diabetic insulin-deficient islets (IDIs) (0.67%) (Fig. 2a). The mean intensity of the HLA-II fluorescent signal within the type 1 diabetic ICIs was also significantly higher than in type 1 diabetic IDIs and that of non-diabetic and Aab+ islets (Fig. 2b). These differences were statistically significant (all p < 0.0001), despite HLA-II varying between individuals and between islets of the same individual (ESM Fig. 4).

Fig. 2
figure 2

Characterisation of HLA-II expression within islets. Quantification in QuPath of widefield tissue scans. (a) HLA-II positive area as a percentage of the total islet area. Islets from individuals with type 1 diabetes were split into insulin-containing islets (T1D ICIs) and insulin-deficient islets (T1D IDIs). (b) Mean intensity of HLA-II fluorescent signal within islet areas. (c) Co-localisation of HLA-II and insulin signals, represented as the percentage of HLA-II+ pixels that overlap with the totality of the insulin+ pixels. Each data point represents one islet. The number of islets analysed per group and the mean ± SD are shown in the tables below. Statistical significance was calculated using Kruskal–Wallis test and all p values were <0.0001. Dunn’s test was used for multiple comparisons (****adjusted p < 0.0001)

In islets from non-diabetic donors, the majority of the HLA-II signal could be attributed to immune cells that were either circulating in the islet capillaries or located very close to the islet (Fig. 1, Fig. 3a). However, in donors with type 1 diabetes, there was variable expression of HLA-II on the beta cells that could not be mapped to nearby immune cells (Fig. 1, Fig. 3b).

Fig. 3
figure 3

HLA-II expression pattern in beta cells and macrophages. nPOD pancreas sections were stained with antibodies against insulin (green), CD68 (white), HLA-II (red) and counterstained with Hoechst (blue). Representative confocal images of macrophage and beta cell morphology from (a) non-diabetic nPOD #6271 donor and (b) type 1 diabetic nPOD #6399 donor. Arrows point to beta cells and macrophages. (c) Quantification from QuPath analysis of the mean intensity of HLA-II fluorescent signal in beta cells and macrophages in donors with type 1 diabetes (T1D) and non-diabetic donors. Subsequent comparison of the mean intensity of HLA-II in (d) beta cells (insulin+) and (e) macrophages (CD68+HLA-II+) by disease status. (f) Quantification of the mean intensity of HLA-II signal in macrophages and beta cells expressing HLA-II in donors with type 1 diabetes. Each data point represents one cell. For each graph the mean ± SD is shown. Statistical significance was calculated using Mann–Whitney U (**** p < 0.0001). Scale bar, 5 μm (a) or 10 μm (b)

Additionally, the confocal data obtained from 301 islets confirmed a statistically significant higher expression of HLA-II (both percentage of positive area and mean intensity) in the ICIs of individuals with type 1 diabetes when compared with the other groups (ESM Fig. 5a, b).

HLA-II is expressed in pancreatic beta cells of patients with type 1 diabetes

We first calculated the percentage of the insulin signal that was co-expressed with HLA-II in order to confirm that beta cells were able to express HLA-II. 45.89% of the total insulin signal co-localised with HLA-II in type 1 diabetic ICIs, while in the non-diabetic and Aab+ islets, most of the insulin signal did not co-localise with HLA-II (6.07% and 4.14%, respectively) (Fig. 2c). Consistent with this finding, confocal data analysis demonstrated that the highest co-localisation of HLA-II within the total insulin signal was in type 1 diabetic ICIs (ESM Fig. 5c).

Next, we identified the number and percentages of cells inside islet areas that expressed HLA-II, insulin, CD68, or any combination of these markers (ESM Fig. 2d). We obtained data from 338,480 cells. In patients with type 1 diabetes, 27.65% of the cells inside the ICIs expressed both insulin and HLA-II, while in non-diabetic and Aab+ donors 0.96% and 0.59% of the islet cells, respectively, expressed both markers. We observed differences between islets of the same individual and between individuals (Fig. 4a,b). The mean density of cells expressing both insulin and HLA-II in islets of type 1 diabetic individuals was 960.4 ± 559.7 cells/mm2, which was significantly higher than observed in non-diabetic and Aab+ individuals (66.9 ± 118.4 cells/mm2 and 40.27 ± 42.28 cells/mm2, respectively).

Fig. 4
figure 4

Analysis of cells expressing both HLA-II and insulin within the islets. Quantification in QuPath of widefield pancreatic tissue scans. (a) Percentage of cells in the islets positive for both insulin and HLA-II. (b) Number of cells in the islets positive for both insulin and HLA-II. Mean ± SD is shown. Variation between individuals and among diagnosis groups can be noted. T1D ICIs, insulin-containing islets from patients with type 1 diabetes

Insulin+ cells expressing HLA-II in donors with type 1 diabetes are not macrophages

In individuals with type 1 diabetes, 2333 cells expressed both insulin and HLA-II, and the majority of them (85.34%) did not express CD68. By visual observation, beta cells and macrophages have clear differences in their morphology, and each present a distinct pattern of HLA-II expression (Fig. 3a,b). We quantified the intensity of the HLA-II signal in all beta cells and macrophages throughout the five type 1 diabetic and five non-diabetic cases. In total, 132,295 beta cells (insulin+) and 7949 macrophages (CD68+HLA-II+insulin) were analysed. The mean fluorescent intensity of HLA-II was higher in macrophages than in beta cells in both groups (Fig. 3c). Beta cells in individuals with type 1 diabetes presented significantly higher HLA-II fluorescent intensity when compared with the beta cells of non-diabetic donors (Fig. 3d). Macrophages presented similar intensity in the HLA-II signal regardless of disease status (Fig. 3e). In patients with type 1 diabetes, when both cell types expressed HLA-II, the mean fluorescent intensity of HLA-II was higher in macrophages (CD68+HLA-II+insulin) than in beta cells (insulin+HLA-II+CD68) (Fig. 3f).

Beta cells express most of the HLA-II in their cytoplasm

In order to identify the location of HLA-II within beta cells, we acquired enhanced resolution images of beta cells from two type 1 diabetic donors (nPOD #6399 and #6198). In these beta cells, we quantified the expression of HLA-II in the central and peripheral (top and bottom) parts of the cells (ESM Fig. 3). In the central regions (representative of the cytoplasm), 37.4% of the area expressed HLA-II and most of it (89.8%) was co-localised with the insulin signal (Fig. 5). HLA-II expression accounted for 12.1% and 12.4% of the area in the top and bottom peripheral surfaces, respectively (which we defined to be representative of the cell membrane compartments), and between 0.1% and 11.9% of their area expressed HLA-II alone. In summary, most of the HLA-II signal was cytoplasmic, accumulating in vesicular clusters that localised within insulin rich vesicles, and was sparsely expressed in the peripheral membrane compartments.

Fig. 5
figure 5

Expression of HLA-II in a beta cell. (a) Representative confocal MIP of a beta cell expressing HLA-II (red) and insulin (green) from one donor with type 1 diabetes (nPOD #6399). The MIP was visualised in 3D space with Imaris software, and the signal from each marker was progressively 3D iso-surfaced beginning with (b) Hoechst (blue), then (c) an approximate cell outline based on the insulin signal (green), and finally (d) HLA-II (red). Co-localisation between the HLA-II and insulin signal was calculated in Imaris, and (e) the co-localised signal (white) was subsequently 3D iso-surfaced. All 3D processed images show a top-down view of the cell. Scale bar, 5 μm

Isolated human islets and reaggregated human islets from non-diabetic donors can express HLA-II under inflammatory stress

Isolated human islets treated with high concentrations of IFN-γ and TNF-α +/− IL-1β induced an increase in HLA-I (Fig. 6b,c) and HLA-II expression (Fig. 6d,e). Representative images shown in Fig. 6a. HLA-II signal largely co-localised with insulin in these stressed islets (Fig. 6f). However, these differences were not statistically significant.

Fig. 6
figure 6

Expression of HLA-I and HLA-II in isolated human islets. Isolated islets from a non-diabetic donor were stained with antibodies against insulin (green), HLA-I (orange), HLA-II (red), and counterstained with Hoechst (blue). (a) Representative MIPs after treatment with media or proinflammatory cytokines (scale bar, 100 μm). Quantification of the percentage of positive area of (b) HLA-I and (d) HLA-II, mean fluorescent intensity of (c) HLA-I and (e) HLA-II signal, and (f) the percentage of co-localisation between insulin and HLA-II (total of insulin+ pixels which overlap with HLA-II+ pixels within the islets). Statistical significance was calculated using Kruskal–Wallis and no significant differences were found. Each data point represents one islet and bars represent the mean ± SD

InSphero reaggregated islets were treated with IFN-γ and/or TNF-α, alone or in combination with IL-1β. GSIS demonstrated statistically significant differences in the islet function depending on the proinflammatory cytokines received (p < 0.0001) (ESM Fig. 6b). Reaggregated islets receiving different concentrations of IFN-γ, alone or in combination with TNF-α, had a higher secretion of insulin in response to both low and high glucose than the control islets receiving media (ESM Fig. 6b). Control islets had a low expression of HLA-I and almost no HLA-II (representative images ESM Fig. 6a). We observed a statistically significant increase in both HLA-I (mean intensity p = 0.0015; percentage of positive area p = 0.024) and HLA-II expression (mean intensity p = 0.0090; percentage of positive area p = 0.0011) depending on the proinflammatory cytokines received (Fig. 7a–d). There was a statistically significant increase in the co-localisation of HLA-II within the insulin signal (p = 0.0183) in the islets receiving cytokines (Fig. 7e). Reaggregated islets receiving IFN-γ in combination with TNF-α +/− IL-1β experienced the highest increase in HLA-II expression. Different concentrations of TNF-α alone did not impair the islet function nor did it induce HLA-I or HLA-II expression.

Fig. 7
figure 7

Expression of HLA-I and HLA-II in reaggregated human islet microtissues. Quantification from Zen software of the percentage of positive area of (a) HLA-I and (c) HLA-II; mean fluorescent intensity of (b) HLA-I and (d) HLA-II, and (e) the percentage of co-localisation between insulin and HLA-II (total of insulin+ pixels which overlap with HLA-II+ pixels within the islets). Each data point represents one islet, and for each group bars show the mean ± SD. Statistical significance was calculated using Kruskal–Wallis test and all the differences were statistically significant (p < 0.05): HLA-I mean intensity p = 0.0015; HLA-I percentage of positive area p = 0.024; HLA-II mean intensity p = 0.0090; HLA-II percentage of positive area p = 0.0011; co-localisation of HLA-II within the insulin signal p = 0.0183. Dunn’s test was used to compare multiple groups and statistically significant differences between groups are indicated in the graphs (*adjusted p < 0.05)

HLA-I and HLA-II mRNA expression are upregulated in islets from non-diabetic donors upon culture with proinflammatory cytokines

Prior studies have shown that reaggregated human islets using InSphero’s hanging drop system have an average endocrine cell purity of 97.3% [22]. As expected, islet markers were highly expressed in the reaggregated microtissues from our two healthy donors (data not shown). Proinflammatory cytokines did not induce an upregulation of immune-specific genes (ESM Fig. 7); however, it did induce a statistically significant upregulation of several HLA-I and HLA-II genes (Fig. 8). HLA-I genes HLA-A, -B, -C and -E, and HLA-II genes HLA-DOA, -DQA1, -DQB1, -DRA, -DRB1 and -DRB5 were significantly upregulated in the beta cells of both donors (Fig. 8 and ESM Fig. 7). HLA-DMB was upregulated in just one donor (ESM Fig. 7).

Fig. 8
figure 8

HLA-II and HLA-I mRNA expression in reaggregated human islet microtissues cultured with vehicle control (media) or proinflammatory cytokines. (a) Combined Log2 fold change of HLA-I and HLA-II mRNA transcripts between cytokine and vehicle treated islet microtissues from two healthy donors. Significantly upregulated transcripts are shown in red. Benjamini–Hochberg procedure of FDR was used to correct for inflated p values across the three contrasts defined in the study. FDR cut-off was set to 0.001 to distinguish significant differential expression. Data are presented as mean with associated 95% confidence interval. Normalised counts of (b) HLA-I and (c) HLA-II mRNA transcripts in response to vehicle and cytokine culture. There are five technical replicates per condition from each donor, and each replicate is the result of three islets that were pooled together. Each data point represents three islet microtissues

Discussion

In the 1980s, Bottazzo et al [9] and Foulis et al [10] reported HLA-II expression in beta cells of patients with type 1 diabetes. However, whether pancreatic beta cells are able to express HLA-II has remained controversial. This was in part due to the lack of reproducibility by other groups [12,13,14,15,16] and the theory that these cells could be macrophages engulfing beta cells [12, 30]. HLA-II expression in beta cells is subtle and heterogeneous, so the lack of sensitivity of the previous imaging systems may explain why it was not detected. With the current high/enhanced resolution and automated microscopy systems, along with the reliable reagents we have at our disposal, what was once perhaps considered a daunting venture is now achievable.

In the current study, we provide evidence with immunofluorescence and mRNA expression assays that beta cells can express HLA-II in patients with type 1 diabetes and in healthy isolated islets that are exposed to inflammatory stress. This should put the older debate to rest, allowing future efforts to be focused on discovering a potential mechanism for direct recognition of beta cells by autoreactive CD4+ lymphocytes.

Our results show that type 1 diabetic ICIs express more HLA-II than IDIs or islets from donors in the other groups. Aab+ donors did not exhibit higher levels of HLA-II than the non-diabetic donors. Most of the cells which express both insulin and HLA-II in the pancreas of patients with type 1 diabetes do not express CD68. The semi-automated analysis with the cell classifier is very consistent and prevents selection bias, but at the same time poses a challenge when a cell’s shape is elongated, as can occur with macrophages.

We observed at a cellular level that most of the HLA-II signal was present in the cytoplasm, but some HLA-II could be localised to the membrane. Our method is limited because we did not have a membrane marker and we analysed tissue sections instead of dispersed beta cells. By using FACS, it was reported that HLA-II can be expressed on the surface of islet beta cells treated with proinflammatory cytokines [21], and HLA-II was present on the membrane of about 15% of the beta cells from patients with type 1 diabetes [17].

With the IIDP islets, we observed that high concentrations of IFN-γ and TNF-α were able to induce HLA-II expression in most of the beta cells. With InSphero islet microtissues, lower doses of these cytokines induced a significant increase in HLA-II expression, but in a lower proportion of beta cells. The different concentrations of cytokines used and the fact that the isolated and reaggregated islets originated from different donors and are inherently different in vitro model systems may explain the differences. Transcriptomic analysis demonstrated that beta cells display increased levels of HLA-II and HLA-I genes after being cultured with proinflammatory cytokines. This further supports the findings and suggests that HLA-II can be induced as a response to inflammatory stress. In the past, it was reported that IFN-γ alone was insufficient to induce HLA-II expression in beta cells and required a combination of at least two proinflammatory cytokines to obtain such an effect [18,19,20,21, 31]. The more sensitive technology used in the present study may explain our finding that IFN-γ alone induces a moderate increase in the HLA-II signal in reaggregated islets, and an even greater level of HLA-II is induced in response to the combination of cytokines. Only one previous study that analysed native human islets from two non-diabetic donors reported an increase in HLA-II expression after stimulation with IFN-γ, as measured by FACS [32].

In the past it was observed that HLA-II could be present only in islets that contain insulin and hyper-express HLA-I [10]. This is consistent with our findings, where we only observed HLA-II in the ICIs, but not in the IDIs, of donors with type 1 diabetes. Also, islets receiving proinflammatory cytokines upregulated both HLA-I and HLA-II.

Whether beta cells can present autoantigens to CD4+ T cells, and what effect this might potentially cause, is not clear yet. Direct antigen presentation by beta cells may result in the activation of autoreactive T cells and lead to autoimmunity, or contrarily, it may induce a regulatory response in the CD4+ T cells.

Other authors could not find the co-stimulatory molecules CD80 or CD86 in the beta cells of mice [32] or humans [17, 33]. However, in a study of NOD mice islets, beta cells from infiltrated islets expressed I-Ag7 (MHC-II), and they were able to stimulate CD4+ T cell proliferation [32]. This suggests that HLA-II on beta cells may render them capable of presenting antigens to CD4+ T cells. Autoreactive CD4+ T cells recognising islet beta cell peptides are present in peripheral blood of both healthy donors and individuals recently diagnosed with type 1 diabetes [34]. However, these autoreactive cells show a polarisation towards IFN-γ production in individuals with type 1 diabetes, whereas in non-diabetic individuals, those cells are polarised to secrete IL-10. Hence, it is possible that expression of HLA-II in individuals with type 1 diabetes could be pathogenic, because their autoreactive CD4+ T cells exhibit polarisation towards a proinflammatory phenotype in response to islet autoantigens, supporting a later role for HLA-II in beta cell destruction [34]. It was recently reported that the blood leucocyte peptidome includes insulin peptides and echoes that of the pancreatic islets in NOD mice [35]. This may translate to humans, and we may be able to identify specific immunogenic peptides that could be presented by the beta cell.

Conversely, HLA-II may be recognised by Tregs [36], and if co-expressed with molecules such as programmed death-ligand 1 (PD-L1), it may induce a regulatory T cell response [37]. Recent studies have shown that PD-L1 is expressed on beta cells in ICIs of donors with type 1 diabetes, but not on those of non-diabetic individuals, and have demonstrated that IFN-γ and IFN-α can induce PD-L1 expression on isolated human islets in vitro [38, 39]. These findings taken together with our observations suggest that HLA-II and PD-L1 expression can occur in parallel. In the context of autoimmune thyroid diseases, thyrocytes can also co-express PD-L1 and HLA-II in situ, and cultured thyrocytes can express PD-L1 and HLA-II in response to IFN-γ stimulation [40, 41]. However, concurrent HLA-II and PD-L1 expression in beta cells and its implications for type 1 diabetes pathogenesis has not yet been studied.

Overall, our findings support a potential role for HLA-II in type 1 diabetes and provide evidence that insulin-producing beta cells are able to express HLA-II. Further studies are needed to determine the mechanism by which HLA-II expression is achieved by beta cells and how it impacts type 1 diabetes pathogenesis.