Spatial Environment Affects HNF4A Mutation-Specific Proteome Signatures and Cellular Morphology in hiPSC-Derived β-Like Cells Free

Heterozygous mutations in the HNF4A gene are associated with maturity-onset diabetes of the young 1 (MODY1) (1), a monogenic autosomal-dominant type of diabetes. Given the inaccessibility of human pancreatic tissue and insufficient recapitulation of the human phenotype of MODY1 in current in vivo models (2), human-induced pluripotent stem cells (hiPSCs) provide an excellent alternative source to study patient-derived models of MODY1. Previous studies of hiPSC-based models of MODY1 (3–5) have not demonstrated mutation-specific cellular phenotypes in end-stage pancreatic β-like cells. Notably, protocol-optimizing experiments to improve directed pancreatic β-cell differentiation have used three-dimensional (3D) environment strategies (6) where cell aggregation (7,8) and alginate encapsulation (9) have detected better morphological and functional outcomes. In this study, we assessed the properties of mutated and corrected HNF4A hiPSC-derived pancreatic β-like cells in two different 3D environments (AggreWell aggregate condition and alginate encapsulation) to demonstrate mutation-specific phenotypes in β-like cells in MODY1.

We reprogrammed fibroblast cells as previously described (4,10) from a donor from the Norwegian MODY Registry after written informed consent and with experiments approved by the Regional Committee on Medical Ethics of Western Norway (REK 2010/2295). We used isolated human pancreatic islets from three deceased donors without diabetes (REK 2011/782) and the human embryonic stem cell line H1 as control.

Alt-R CRISPR/Cas9 reagents were provided by Integrated DNA Technologies (Coralville, IA). Briefly, 10⁵ iPSCs were mixed with 12.5 μL ribonucleoprotein, 5 μL electroporation enhancer, and 2.5 μL homology-directed repair (HDR) template (stock 3 nmol/L) in 40 μL Nucleofector solution 2 (Lonza); electroporated; and subsequently incubated for 48 h. Dissociated single cells were grown until colonies emerged. Ninety-six colonies were picked and screened. Positive clones from restriction fragment–length polymorphism assay, using BccI enzyme on exon 7–amplified segments, were validated by Sanger sequencing. Seven colonies, including D13, D94, E75, and E87, were found to be corrected by the HDR template.

We characterized iPSCs as previously described (10), including assessment by flow cytometry and immunofluorescence as described below. The chromosome analysis was done at the Cell Guidance Systems Genetics Service, Cytogenetic Laboratory, Babraham Research Campus, Cambridge, U.K. We further evaluated pluripotency potential in mutated and corrected hiPSC lines using an hPSC functional identification kit (SC027B; Bio-Techne Ltd.). Specifically, three uniquely formulated media were used to differentiate hPSCs into the three germ layers endoderm, ectoderm, and mesoderm. The three germ layers were detected by immunocytochemistry using the antibodies for the transcription factors SOX17 (endoderm), Otx2 (ectoderm), and brachyury (mesoderm) (SC027B; Bio-Techne).

The directed differentiation of hiPSCs to β-like cells was performed as previously described (4,11). To reduce variability, we performed the different directed differentiations in parallel in the same multiwell plate for all conditions with the same batch of differentiation cocktail. At the end of stage 5 (S5), 5.0 million cells were embedded in alginate beads, as previously described (12), 1.2 million cells were used to form cell aggregates (8), and ∼5 million cells were kept in adherent culture. The beads, aggregates, and planar cultures were differentiated until stage 7 (S7), as previously described (4,9).

As described elsewhere, we performed, Western blotting (13), flow cytometry (14), immunofluorescence (4), and confocal imaging (9). For Western blotting, we used the primary antibodies HNF4A (1:1,000; Cell Signaling Technology), PDX1 (1:1,000, R&D Systems), or GAPDH (1:1,000, Santa Cruz Biotechnology); otherwise, we used the following: mouse anti-Oct-3/4 (sc-5279; Santa Cruz Biotechnology), rabbit anti-NANOG (ab21624; Abcam), goat anti-SOX17 (AF1924; R&D Systems), goat anti-PDX1 (AF2419; R&D Systems), mouse anti-HNF4A (PP-H1415-00; R&D Systems), mouse anti-Nkx-6.1 (F55A12-S; Developmental Studies Hybridoma Bank, University of Iowa), and guinea pig anti-insulin (A0564; Dako). The samples were mounted in Prolong Diamond Antifade Mounting Media (P36970; Life Technologies).

We used AggreWell 400 (34411; STEMCELL Technologies) as recommended by the manufacturer. Briefly, 1.2 × 10⁶ cells were seeded per AggreWell in stage 6 (S6) medium, which generated aggregates of ∼1,000 cells. After 48 h incubation, the aggregates were transferred to low-adherence plates and kept on an orbital shaker (100 rpm). We encapsulated cells with alginate and processed the capsules as described elsewhere (9).

RNA was extracted from 2D Matrigel-differentiating cells, aggregates, and encapsulated cells using an RNeasy Micro Kit (QIAGEN) and processed and sequenced (mRNA sequencing for AggreWell samples and 3′ RNA transcriptomics [QIAseq UPX 3′ Transcriptome Kit; QIAGEN] for alginate samples; 2.5 and 10 ng purified RNA for the alginate and 2D samples, respectively) as previously described (9).

The lysed cells and human islets (positive control) were processed and analyzed with mass spectrometry as previously described (4,15), using 25 μg of each sample and three TMT11plexes, with a reference mix of all samples in 2 of the 11 samples in each plex to allow crossplex analysis. Pathway analysis was performed as previously described (4).

As previously described, we assessed S7 cells by assays of static glucose-stimulated insulin secretion (GSIS) (4), dynamic GSIS (16), and oxygen consumption rate (OCR) (17) using 60 handpicked S7 cell clusters in dynamic GSIS (perfused with high glucose [20 mmol/L] from 42 to 84 min, whereas other fractions were exposed to low glucose [1.67 mmol/L]) and 40–60 S7 cell clusters per 24 wells during OCR. The final concentrations of 20 mmol/L glucose, 5 μmol/L oligomycin (Cell Signaling Technology), 5 μmol/L carbonyl cyanide 3-chlorophenylhydrazone (Sigma-Aldrich), and 5 μmol/L rotenone (Sigma-Aldrich) were added in order. The OCR values were normalized to the average baseline values measured in assay media containing 1.67 mmol/L glucose.

Statistical analysis was performed in Excel version 14.7.7 and GraphPad Prism 7.0.0 software. A two-sided t test was used, and P < 0.05 was considered significant. Supplementary Fig. 11 was built in R 3.6.1 using the packages tidyr version 1.1.3, dplyr version 1.0.0, scico version 1.2.0, and ggplot2 version 3.3.5.

The RNA-sequenced data sets have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (accession no. GSE188827). The proteome data sets have been deposited in ProteomeXchange through the Proteomics Identification Database (accession no. PXD025054).

We generated several mutated (HNF4A^+/c811dupA) hiPSC lines (the hiPSC lines 6A, 6D, and 6F in Supplementary Fig. 1) by reprogramming fibroblasts from a patient with MODY1 (N904-1 in Fig. 1A and Supplementary Table 1) with a c.811dupA/p.Ile271fs mutation (Fig. 1B). Using these lines, we created corrected (HNF4A^+/corrected) mutation-free isogenic hiPSC lines (the E75, D13, D94, E87 hiPSC lines in Supplementary Fig. 1) by CRISPR/Cas9 HDR gene editing (Fig. 1C–E and Supplementary Figs. 2–3). We subsequently assessed the integrity of the mutated and corrected hiPSC lines (10) (Fig. 1F and Supplementary Figs. 1 and 4–7).

The mutated and corrected hiPSCs were subjected to directed differentiation (Supplementary Fig. 8) toward pancreatic β-like cells, as done previously (4), but by adding two different 3D environments (aggregate formation in AggreWell and alginate encapsulation) to the standard 2D format (Fig. 2A). Using global proteomics analysis of end-stage pancreatic β-like cells (S7), we found that the cell lines cultured in AggreWell or in alginate capsules yielded distinctively different proteomics signatures (disregarding mutational status) that were also distinctively different from the lines cultured in 2D and from islet controls (Fig. 2B). The 2D and 3D environments did not differentially impact the single-protein levels (measured by proteomics) of key β-cell markers (Fig. 2C and Supplementary Fig. 9A), which were also expressed at a similar level by immunocytochemistry (Fig. 2D). Although proteomics analysis of S7 cells did not detect the HNF4A protein, we confirmed HNF4A expression at intermediary stages by other methods (Supplementary Figs. 8 and 10).

Given that the alginate 3D environment had indeed produced a unique proteomics signature of S7 cells, we next assessed whether the mutational status affected the alginate-encapsulated S7 proteome (Fig. 3A). Focusing first on the mutated alginate-encapsulated S7 cells (class a in Fig. 3A), we identified glycolysis and gluconeogenesis (predicted as inhibited) as the top canonical pathway, consistent with the known role of HNF4A to regulate metabolic glucose sensing in pancreatic β-cells (18), followed by fatty acid β-oxidation and stearate biosynthesis (predicted as activated) (Fig. 3B and Supplementary Fig. 9B). In contrast, when extending the pathway analyses to the proteome of corrected cells (class b in Fig. 3A), we were not able to predict significant differences in glycolytic and glyconeogenetic pathways or in stearate biosynthesis and fatty acid β-oxidation pathways (Fig. 3B), supporting the notion that gene correction rescued the phenotype. Analogously, from the analysis of the proteome data set at the single-protein level (Fig. 3C and D and Supplementary Fig. 9C and D), four proteins among 10 enzymes in the glycolytic pathway displayed lower levels in the alginate-encapsulated 3D environment compared with 2D for mutated lines but not in corrected lines, although these were not reproduced in the transcriptome (Supplementary Fig. 11).

Since we found that also the aggregated 3D environment had produced a unique proteomics signature of S7 cells, we next assessed the mutated AggreWell-cultivated S7 lines. Strikingly, we observed that the mutated lines formed irregular spheroids that clustered together (Fig. 4A and Supplementary Fig. 12). In contrast, the corrected lines formed regular round shapes, which did not cluster (Fig. 4A and Supplementary Fig. 12). Next, we assessed whether the mutational status affected the alginate-encapsulated S7 proteome and transcriptome (Fig. 4B and Supplementary Fig. 11), first focusing on mutated AggreWell-encapsulated S7 cells (class a in Fig. 4B). We identified GP6 signaling and apoptosis (predicted inhibition) and EIF2 signaling and transfer RNA (predicted activation) as top canonical pathways (Fig. 4C). Since the GP6 signaling pathway plays a role in collagen-induced activation and aggregation (19), we assessed the proteome at the single-protein level and found that indeed, reduced levels of collagen and extracellular matrix (ECM)–related proteins (Fig. 4D), validated at the mRNA level (Supplementary Fig. 11). Finally, we assessed and identified mutation-specific functional readouts uniquely present in the aggregated 3D context (Supplementary Fig. 13).

In this study, we were able to identify for the first time a mutation-specific phenotype in MODY1 β-like cells facilitated by growing the cells in two different 3D environments and comparing the cellular and proteomic readouts to the 2D background. We were also able to suggest two specific molecular mechanisms whereby the two different 3D culture effects could be mediated. Each of these specific molecular mechanisms showed a dependency on the type of 3D context chosen. The cellular confinement in the alginate context helped to identify a metabolic phenotype with lower levels of glycolytic proteins, potentially affecting glucose sensing. The structural scaffolding in the AggreWell 3D context helped to identify a structural collagen-associated phenotype with irregular clusters, a unique proteome with lower levels of structural ECM proteins (laminins, collagens) as well as a functional readout.

The two chosen 3D environments challenge the differentiating pancreatic β-like cells in unique ways, including the levels of oxygen delivery, nutrient supply, and cell-to-cell contact. The 3D cell aggregation/AggreWell context is preferred by several investigators in generating pancreatic β-like cells (7,8,20) because it mimics the in vivo pancreatic islets in size, level of cell-to-cell contact, and the self-organized compact spheroid arrangement (21). In the 3D alginate encapsulation context, the cells are immobilized and lack cell-to-cell contact; however, the alginate gel-pore network still allows transport of oxygen and diffusion of nutrients and waste products (22). Furthermore, alginate encapsulation is suitable for future therapeutic transplantation (23) and can also protect the cells in bioreactor culture systems (24).

In conclusion, our findings warrants a careful consideration and selection of an appropriate 3D context when studying stem cell models. Our study also provides a patient-specific diabetes model that can be further explored for potential new drug targets that can be used in precision diabetes medicine.

Acknowledgments. The authors thank the Islet Distribution Program at the University of Oslo for providing human islets samples. They also thank Dr. Steven P. Gygi and the Taplin Mass Spectrometry facility at Harvard Medical School for use of mass spectrometers, A.H. Knudsen for technical help, and Lars A. Akslen at the Center for Cancer Biomarkers, Department of Clinical Medicine, University of Bergen, for strategic support. The confocal imaging was performed at the Molecular Imaging Center, Department of Biomedicine, University of Bergen.

Funding. This work was funded by Bergen Forskningsstiftelse (grant BFS2014REK02 to H.R.), Diabetesforbundet (to H.R.), Johan Selmer Kvanes legat (to H.R.), Novo Nordisk Foundation - Komite for Endokrinologi og Metabolisme - Norden (ENDO) (grant NNF17OC0027258 to H.R.), Novo Nordisk Foundation (grant NNF15OC0015054 to S.C.), the Research Council of Norway through its Centres of Excellence funding scheme (project #262613 and project #301178 to M.V.) and Norwegian Research Foundation UiO:LifeScience Convergence Environments II (grant 247577 to H.S.), Western Norway Regional Health Authority (grant 911985 to H.R.), and National Institutes of Health/National Institute of General Medical Sciences (grant R01 GM132129 to J.A.P.).

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

Author Contributions. M.C. performed immunofluorescence imaging and analyses. M.C., C.W., A.M.S., Y.B., S.C., L.M.G., H.S., and H.R. interpreted the observations. M.C., C.W., Y.B., and J.G. performed the differentiation, sample preparation for proteomic and RNA sequencing analyses, and immunofluorescence staining, and analyzed data. M.C., A.M.S., Y.B., V.L., H.S, and H.R. conceived the experiments. C.W. performed the FACS analyses. A.M.S. performed the CRISPR/Cas9 editing and iPSC integrity studies. Y.B. and S.C. analyzed the proteomics data. Y.B., H.S., and H.R. wrote the manuscript. S.A. performed the immunoblotting. S.A. and H.S. generated human islet preparations. J.A.P. performed the tandem mass tag–labeling experiment and mass spectrometry analysis. E.T. provided the skin biopsy samples. E.T., P.N., I.N., and H.R. provided clinical data. M.V. performed the RNA sequencing bioinformatics analyses. All the authors reviewed, commented on, and edited the manuscript. H.R. 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.