Comprehensive Characterization of Islet Remodeling in Development and in Diabetes Using Mass Cytometry

Abstract

The pancreatic islet is the functional and structural unit of the pancreatic endocrine portion. Islet remodeling occurs in both normal development and pathogenesis of type 1 (T1D) and type 2 diabetes (T2D). However, accurately quantifying changes in islet cellular makeup and hormone expressions poses significant challenges due to large intra- and inter-donor heterogeneity and the limited scalability of traditional methods such as immunostaining. The cytometry by time-of-flight (CyTOF) technology enables simultaneous quantification of more than 30 protein markers at single-cell resolution in a high-throughput fashion. Moreover, with distinct DNA and viability markers, single live cells can be explicitly selected in CyTOF. Here, leveraging the CyTOF data generated by the Human Pancreas Analysis Program, we characterized more than 12 million islet cells from 71 donors. Our data revealed continued age-related changes in islet endocrine cell compositions, but the maturity of endocrine cells is reached by 3 years of age. We also observed significant changes in beta cell numbers and key protein expressions, along with a significant increase in bihormonal cells in T1D donors. In contrast, T2D donors exhibited minimal islet remodeling events. Our data shine a light on the islet dynamics during development and diabetes pathogenesis and suggest divergent pathogenesis processes of T1D and T2D. Our comprehensive approach not only elucidates islet plasticity but also establishes a foundation for integrated CyTOF analysis in islet biology and beyond.

mass cytometry, CyTOF, islet remodeling, diabetes, development

Determining the alterations in cellular makeup and the degree of phenotypic changes in human islet cells during development, aging and in diabetes can help us understand the plasticity of endocrine cells and the pathophysiological mechanisms of diabetes. However, accurately quantifying changes in islet cellular composition and hormone expression is challenging due to significant intra- and inter-donor variability and the limited scalability of traditional methods like immunostaining. We previously pioneered the cytometry by time-of-flight (CyTOF) (1, 2) workflow for multiplexed protein level analysis at single-cell resolution in a high-throughput fashion in human islets (3). Here, leveraging the large CyTOF dataset generated by the Human Pancreas Analysis Program (HPAP) (4), we annotated more than 12 million cells from 71 donors with 33 protein makers per cell. We comprehensively characterized the endocrine composition and hormone expression levels and enumerated bihormonal cells in islets from adult controls, child controls, and donors with type 1 diabetes (T1D) or type 2 diabetes (T2D). The large sample size, combined with a semisupervised machine learning workflow, facilitates a robust and thorough depiction of the islet remodeling process. Our work also establishes a foundation for future integrated analysis of CyTOF datasets in islets and other organ systems.

Material and Methods

CyTOF Sample Preparation

Organ procurement and processing were carried out by HPAP (4). CyTOF samples were prepared as previously described (3). Briefly, islets were dissociated with 0.05% trypsin, then washed and filtered with a 40 μm cell strainer. Cells were stained with cisplatin for live/dead discrimination (5), then fixed and permeabilized. After barcoding with donor identity labels, cells from different donors were pooled and labeled with an antibody cocktail. Postlabeling, cells were washed and stained with DNA intercalator Iridium. Right before sample acquisition, cells were washed with MilliQ H₂O. Multielement normalization EQ beads were used for signal normalization.

CyTOF Data Preprocessing

Raw CyTOF FCS files were downloaded from PANC-DB (https://hpap.pmacs.upenn.edu/) (4). Only data from HPAP040 or later were used due to the shifts in the antibody panel and the differences in signal intensity in the early samples. FCS files were sequentially gated to select EQ beads negative and single live cells (Fig. 1A) (3). To ensure even sampling, donors with 30-fold fewer cells than the average cell numbers from all donors were excluded (excluded donors were HPAP079, HPAP095, HPAP096, HPAP103, HPAP110, HPAP121, HPAP124, and HPAP125). A merged Seurat object was created containing raw count data of 33 antibody channels per cell [CD3, CD4, CD8, CD9, CD11c, CD14, CD16, CD19, CD27, CD44, CD45, CD45RA, CD45RO, CD49F, CD56, CD68, CTLA4, Eomes, PD-1, HLA-ABC, HLA-DR, Ki67, GLUCAGON (GCG), C-PEPTIDE (C-PEP), SOMATOSTATIN (SST), GHRELIN (GHRL), POLYPEPTIDE (PPY), PDX1, ST8SIA1, PDGFRA, pS6, EpCAM, and pan-cytokeratin]. A hyperbolic arcsine with a cofactor of 5 was used to transform the raw count data following the CyTOF convention.

Figure 1.

CyTOF facilitates single-cell protein level analysis in human islets. (A) Sequential gating strategy to select single live cells for downstream analysis. Cells were first gated as events that were negative for beads (left). Single cells were then identified and selected based on DNA content and event length (middle). Finally, live cells were distinguished from dead cells based on cisplatin exclusion (right). Percentages of cells in each gate in 1 representative sample are shown. (B) Integrated UMAP visualization by cell type (top) and donor (bottom). The data harmonization process effectively controls for donor effect. (C) Violin plots illustrating the marker protein expressions in each annotated cell type. (D) Endocrine cell composition in different donors. Each dot represents 1 donor. The error bars indicate SEM. * Indicates significant differences compared to adult controls with adjusted P-value <.1 using 1-way ANOVA and Dunnett post hoc test. (E) Changes of endocrine cell frequencies with age. The regression line for each cell type was fitted with loess smoothing.

Abbreviations: CyTOF, cytometry by time-of-flight; UMAPs, Uniform Manifold Approximation and Projection for Dimension Reduction.

Cell Type Annotation and Protein Expression Analysis

We applied the atomic sketch integration method from the Seurat V5 pipeline (6) to perform integration, dimension reduction, and cell type annotation. This workflow contains the following major steps: (1) A representative subset of cells (a “sketch”) was sampled from each dataset; (2) the sampled cells were integrated and clustered to define cell types; (3) the full datasets were reconstructed based on the structure of the subset of cells; (4) the cells from the full datasets were annotated based on the labels from the sketch. Two thousand representative cells from each donor were sampled and integrated using the IntegrateLayers function with FastMNN method, and clustered using FindNeighbors and FindClusters. Cell types were identified based on marker protein expression: GCG for alpha cells, C-PEP for beta cells, SST for delta cells, GHRL for epsilon cells, PPY for pancreatic polypeptide (PP) cells, pan-cytokeratin and CD49F for exocrine (ductal/acinar) cells, EpCAM negative for mesenchymal cells, CD45 for immune cells and within immune cells, CD3 for T cells, and CD68 for macrophages. All cells from the full datasets were embedded and annotated using ProjectIntegration and ProjectData. Endocrine cell type proportions were calculated relative to the total endocrine cells per sample and validated against HPAP annotation using the concordance correlation coefficient (ccc) from the epi.ccc function of epiR package. All functions were implemented with default parameters.

To examine the endocrine cell compositional changes with age, regression was performed in ggplot2 in R with the argument stat_smooth(method = “loess”). Loess fitting was chosen because it does not impose a structure but performs a local weighted regression to create a smooth curve representing the conditional means.

To compare the protein expression levels of cells from different donor groups, the pseudobulk approach was used. Briefly, the AggregateExpression function was applied to sum the protein expression of all cells from the same donor for each cell type. Differential expression analysis was performed in each cell type to identify proteins with different levels between different donor types using FindMarkers with default parameters. A Bonferroni corrected P-value < .10 was considered significant.

To estimate frequencies of bihormonal cells, biaxial plots with C-PEP vs 1 of the 4 hormones (GCG, SST, GHRL, and PPY) were used for manual gating, with the number of beta cells as the denominator for percentage.

Results

An Integrated Reference Map of Single-cell Marker Protein Expression in Human Islets

To systematically evaluate the protein levels of pancreatic islet endocrine cells during development and in T1D and T2D, we leveraged the CyTOF data generated by the HPAP (4). This dataset consists of 33 multiplexed protein measurements of single islet cells from 38 adult controls, 6 child controls, 9 T1D donors, and 18 T2D donors (Table 1). Based on DNA staining (Ir191 and Ir193) and live/dead labeling (cisplatin), we selected single live cells for downstream analysis (Fig. 1A). Postfiltering, 12 059 196 cells were analyzed, with an average of 169 848 cells per donor (Table 1).

Table 1.

Donor characteristics

	Adult control	Child control	T1D	T2D
Number of donors (F/M)	38 (19/19)	6 (2/4)	9 (2/7)	18 (11/7)
Number of cells/donor (mean (range))	125 329 (7560-357 574)	174 958 (49 975-369 201)	268 659 (70 839-465 679)	212 723 (36 644-626 246)
Age (years) (mean (range))	36.8 (19-64)	8.8 (3-15)	20.1 (9-32)	49.9 (34-63)
BMI (kg/m2) (mean (range))	30.1 (18.0-38.7)	17.9 (12.0-24.1)	22.1 (15.9-32.7)	33.9 (24.3-45.5)
HbA1c (%) (mean (range))	5.4 (4.7-6.7)	5.5 (5.3-5.9)	11.2 (7.8-16.1)	7.8 (5.0-12.0)
% of alpha cells (mean (range))	48.7 (19.1-81.9)	29.3 (10.8-43.4)	79.5 (50.1-93.8)	50.6 (9.2-92.3)
% of beta cells (mean (range))	43.1 (14.9-68.4)	53.7 (38.0-67.2)	9.1 (0.3-37.6)	39.3 (3.7-75.0)
% of delta cells (mean (range))	5.4 (1.7-13.1)	7.7 (3.4-15.3)	7.7 (0.4-14.7)	5.7 (0.4-15.0)
% of epsilon cells (mean (range))	0.1 (0-0.4)	1.4 (0.3-3.8)	0.3 (0-1.2)	0.05 (0-0.1)
% of PP cells (mean (range))	2.7 (0.5-9.5)	7.9 (3.4-16.7)	3.4 (0.2-15.8)	4.3 (0.3-17.8)

Abbreviations: BMI, body mass index; HbA1c, hemoglobin A1c; PP, pancreatic polypeptide; T1D, type 1 diabetes; T2D, type 2 diabetes.

We harmonized the 12 million cells from all the donors using Seurat V5 (see method) (6). Postintegration, cells clustered based on their cell types but not donor origins (Fig. 1B). Identified cell types included all 5 pancreatic endocrine cell types (alpha, beta, delta, epsilon, and PP), exocrine cells (ductal and acinar), mesenchymal cells, macrophages, and T cells (Fig. 1B). Each cell type exhibited its canonical maker expression pattern (Fig. 1C).

To validate the cell type calling procedure, we compared the cell type frequency from our pipeline with the HPAP annotation, observing high concordance for alpha (ccc = 0.91), beta (ccc = 0.89), delta (ccc = 0.81), and PP cells (ccc = 0.68) (Supplementary Fig. S1) (7). Discrepancies arose with epsilon cells in 2 samples (HPAP052 and HPAP092). For these 2 samples, the HPAP method, using tSNE graphs of individual donors, did not separate epsilon populations, whereas cell types identified with our integrated method displayed expected marker protein levels (Supplementary Fig. S2) (7). After removing the 2 outliers, the ccc between the 2 annotation methods in epsilon cells increased from 0.09 to 0.67 (Supplementary Fig. S1D) (7).

Changes in Endocrine Cell Compositions across Age and Diabetes

We next examined the differences in cell type proportions in different types of donors. In adult controls, beta cells comprised 15% to 68% of all islet endocrine cell types. The percentage of beta cells in adult controls was not significantly associated with donors’ hemoglobin A1c levels (linear regression r² = 0.042, P = .22). Consistent with the literature (8-10), T1D donors showed a significant decrease in beta cell proportion and an increase in alpha cell proportion compared with adult controls (Fig. 1D). Of note, some residue beta cells were detected in all of the T1D donors analyzed, with a frequency between 0.3% and 37.6% (Table 1), in line with histological studies (11-14). Again, the percentage of beta cells was not significantly associated with T1D donors’ hemoglobin A1c levels (linear regression r² = 0.12, P = .36).

Child controls showed a significant decrease in alpha cell proportions and significant increases in epsilon and PP cell proportions compared with adult controls (Fig. 1D). To better assess the relationship between the endocrine cell proportions and age, we performed a local weighted regression analysis with loess fitting, restricting our analysis to adult controls and child controls (Fig. 1E). We observed that the proportion of alpha cells increased with age and stabilized at ∼30 years old, followed by a decrease at ∼50 years old. In contrast, the beta cell proportion decreased with age, stabilized at ∼30 years old, and increased at ∼50 years old. Proportions of delta, epsilon, and PP cells decreased with age followed by plateaus at ∼ 20 to 30 years old (Fig. 1E). To be noted, since we analyzed the proportion of different endocrine cell types rather than their absolute mass, changes in the composition of 1 cell type affect the proportions of the others.

Because of the prominent sex differences in diabetes risk (15, 16), we investigated whether these differences could be associated with variations in endocrine cell composition between sexes. PP and epsilon cell proportions differ significantly between the 2 sexes across different donor types (2-way ANOVA, P < .05). Pairwise comparison between the 2 sexes further showed that child female controls had higher epsilon cell proportions compared to child male controls (multiple comparisons with Sidak's test, adjusted P < .1). None of the other comparison groups between the 2 sexes showed statistical significance (Supplementary Fig. S3A) (7). Both females and males show similar patterns of age-related endocrine-cell compositional change (Supplementary Fig. S3B) (7).

Protein Level Changes in Islet Endocrine Cells across Age and Diabetes

We next tested for differential expression of protein markers in each endocrine cell type across different ages and disease conditions. Within all the endocrine cell types, only beta cells displayed significant protein level changes (Fig. 2). Specifically, T1D beta cells had significantly reduced C-PEPTIDE (C-PEP) expression compared with beta cells from all of the other donor types (Fig. 2B). Beta cells from adult controls had significantly lower HLA-ABC levels compared with beta cells from other donor types, whereas beta cells from T1D had significantly higher HLA-ABC levels compared with beta cells from other donor types (Fig. 2F). Among markers that did not show any changes in beta cells in different types of donors were the proliferation marker Ki67 and PDGFRA (17) and beta cell subtype markers CD9 and ST8SIA1 (18). None of the proteins showed significant sex-specific differences (2-way ANOVA, adjusted P > .1) (Supplementary Fig. S4) (7).

Figure 2.

The distributions of protein expressions across cell types and donor types. Ridge plots showing the levels of (A) GLUCAGON, (B) C-PEPTIDE, (C) SOMATOSTATIN, (D) GHRELIN, (E) POLYPEPTIDE, and (F) HLA-ABC. The curves associated with proteins that display significant differences between different types of donors (adjusted P-value <.1 with Wilcoxon test) are outlined in red. All protein markers show similar expression levels across different donor types except C-PEPTIDE, which is significantly reduced in beta cells of T1D donors, and HLA-ABC, which is significantly increased in beta cells of T1D donors and decreased in beta cells of adult controls.

Bihormonal Cells

Increased bihormonal cells have been postulated as a mechanism of compensation in diabetes (19, 20). Focusing on beta cells, we quantified the frequencies of each type of bihormonal cells (including C-PEP+ GCG+, C-PEP+ SST+, C-PEP+ GHRL+, and C-PEP+ PPY+ cells) (Fig. 3A). We observed that each of the 4 types of bihormonal cells had significantly increased frequencies in T1D donors compared with adult controls (Fig. 3B). In contrast, child controls and T2D donors did not show increased frequencies of bihormonal cells compared with adult controls (Fig. 3B). The proportions of different bihormonal cell types did not significantly differ between the 2 sexes across different donor types (2-way ANOVA, P > .05) (Supplementary Fig. S5) (7).

Figure 3.

Type 1 diabetes donors have increased frequencies of bihormonal cells. (A) Gating strategy for each type of bihormonal cells. (B) Percentage of bihormonal cells in different donors. Each dot represents 1 donor. The error bars show SEM. * indicates significant differences compared to adult controls with adjusted P-value <.1 using 1-way ANOVA and Dunnett post hoc test.

Discussion

Here, using the rich CyTOF data generated by HPAP (4), we systematically annotated and evaluated more than 12 million islet cells from 71 donors and 33 protein levels per cell. Our study provides an integrated map of marker protein profiles in islet cell types and captures islet remodeling during development and in T1D and T2D.

We observed age-related changes in endocrine cell compositions. In control donors, between 25 and 60 years old, alpha cells showed the highest abundance among all islet endocrine cell types. The continued increase of alpha cell percentage before 30 years old could be explained by the highest basal replication rate of alpha cells among all endocrine cells (3, 21, 22). On the other hand, the increase in beta cell proportions beyond 50 years old could be associated with the long lifespan and low turnover of beta cells (23-25). The higher frequency of epsilon cells in younger donors implies a higher abundance of endocrine progenitor cells in this age group, as the epsilon cell marker GHRL overlaps with the endocrine progenitor marker NEUROG3 (26-28). We did not find more bihormonal cells in children. This suggests that by age 3 (our youngest donor), islet endocrine cells are fully mature, as high levels of bihormonal cells are typically seen in the fetal pancreata (29-32).

In T2D donors, there were no changes in islet cell type proportions, hormone protein levels, cell proliferation, subtypes, and bihormonal cells. This result suggests that the underlying processes associated with these parameters (beta cell transdifferentiation, proliferation, and subtype transitions) may not be a significant factor involved in T2D pathogenesis and favor alternative mechanisms of beta cell dysfunction including dedifferentiation, as suggested by recent literature (33-35). We could not identify dedifferentiated cells that have lost their hormone markers but retain other endocrine markers such as chromogranin A, because the CyTOF setup did not include the necessary markers to detect this change. In the same vein, we might have missed some abnormalities in the T2D islets because of the limitations of the number of markers included in the CyTOF workflow. Moreover, the cross-sectional nature of the human study precludes longitudinal tracking of cellular changes, potentially overlooking some dynamic events.

In contrast to T2D, we observed numerous significant changes in the beta cells of T1D donors including reductions in beta cell proportion and C-PEP protein levels, increases in the frequencies of bihormonal cells, and increases in the HLA-ABC levels. The combination of cell number reduction and C-PEP level reduction indicates that both the mass and function of beta cells are impaired in T1D, consistent with current understanding (36). The increases in frequencies of bihormonal cells in T1D donors suggest a large degree of pancreatic cell plasticity in the context of severe beta cell loss, as has been shown in mice (37, 38). The hyperexpression of HLA class I molecules in T1D beta cells agrees with previous findings and is thought to be involved in beta-cell recognition by CD8+ T cells (10, 11, 39-41). Of special note, HLA and insulin constitute the 2 major genetic loci associated with T1D risk (42, 43). Hence the observed changes in HLA-ABC and C-PEP protein levels in T1D beta cells offer a direct mechanism of T1D etiology. Apart from beta cell intrinsic defects, the increase in bihormonal cells suggests that active islet remodeling contributes to or results from T1D. The differences in T1D and T2D islet changes support the view that these 2 types of diabetes have different pathogenic processes (44).

Overall, we observed minimal differences between females and males in terms of islet endocrine cell composition, frequencies of bihormonal cells, and marker protein levels. This suggests that the differential diabetes risks between these 2 sexes are likely due to more subtle differences that were not captured in the CyTOF system. Of note, large transcriptomic differences in islet cells between females and males are observed using single-cell RNA sequencing (scRNA-seq) (16, 45). The different results between CyTOF and scRNA-seq might originate from the differential sensitivities, specificities, and scales of the 2 technologies.

In summary, using the large CyTOF dataset, we comprehensively characterized islet cell changes across development and disease. Our study provides insights into islet cell plasticity in (patho)physiological conditions. Our analytic pipeline provides a framework for future integrated analysis of large-scale CyTOF data and facilitates research endeavors in islet biology and other biological fields.

Acknowledgments

This manuscript used data acquired from the HPAP (HPAP-RRID:SCR_016202) database (https://hpap.pmacs.upenn.edu), a Human Islet Research Network (RRID:SCR_014393) consortium (UC4-DK-112217, U01-DK-123594, UC4-DK-112232, and U01-DK-123716). The authors thank Dr. Terra Bradley for the careful editing of the manuscript.

Disclosures

The authors have nothing to disclose.

Data Availability

Original data generated and analyzed during this study are included in this published article or in the data repositories listed in the References.

Author notes

Maria Pilar Toledo and Gengqiang Xie contributed equally to the work.