Reprogramming human gallbladder cells into insulin-producing β-like cells

The gallbladder and cystic duct (GBCs) are parts of the extrahepatic biliary tree and share a common developmental origin with the ventral pancreas. Here, we report on the very first genetic reprogramming of patient-derived human GBCs to β-like cells for potential autologous cell replacement therapy for type 1 diabetes. We developed a robust method for large-scale expansion of human GBCs ex vivo. GBCs were reprogrammed into insulin-producing pancreatic β-like cells by a combined adenoviral-mediated expression of hallmark pancreatic endocrine transcription factors PDX1, MAFA, NEUROG3, and PAX6 and differentiation culture in vitro. The reprogrammed GBCs (rGBCs) strongly induced the production of insulin and pancreatic endocrine genes and these responded to glucose stimulation in vitro. rGBCs also expressed an islet-specific surface marker, which was used to enrich for the most highly reprogrammed cells. More importantly, global mRNA and microRNA expression profiles and protein immunostaining indicated that rGBCs adopted an overall β-like state and these rGBCs engrafted in immunodeficient mice. Furthermore, comparative global expression analyses identified putative regulators of human biliary to β cell fate conversion. In summary, we have developed, for the first time, a reliable and robust genetic reprogramming and culture expansion of primary human GBCs—derived from multiple unrelated donors—into pancreatic β-like cells ex vivo, thus showing that human gallbladder is a potentially rich source of reprogrammable cells for autologous cell therapy in diabetes.


Introduction
Type 1 Diabetes Mellitus (T1DM) is a metabolic disorder characterized by hyperglycemia caused by autoimmune-mediated destruction of pancreatic β cells leading to a complete or near total loss of the hormone insulin [1,2].Daily administration of exogenous insulin is still the mainstay of treatment, but this is not always sufficient to prevent chronic complications of the retina, kidneys, peripheral nerves, and blood vessels.Replacement of β cells (either by whole pancreas or islet cell transplantation) is the best available method for restoring a physiologic glycemic control and improvement of diabetes complications [3,4].As with other types of organ transplantation, shortcomings related to islet or whole pancreas transplantation include the high medical and surgical costs, insufficient supply of available donor pancreata, and lifetime requirement for immunosuppression therapy-that is variably associated with adverse side-effects [3,[5][6][7].Therefore, alternative sources of abundant transplantable β cells are needed to fill the supply gap [8][9][10][11][12][13][14][15].
Based on the common developmental origin of the ventral pancreas, the liver and its associated biliary tree from the posterior ventral foregut [34] and from reports of ectopic pancreatic tissues found in extrahepatic biliary tree [35][36][37], our group previously showed that murine gallbladder can be dependably reprogrammed into insulin-producing islet-like cells after forced expression of Pdx1, MafA, and Neurog3 ex vivo [10,38].Here, we embarked on the very first reprogramming, from multiple donors, of human primary gallbladder and cystic duct cells.The overarching aim of this study was to genetically reprogram highly expandable patient-derived human gallbladder cells (GBCs) to produce insulin for potential autologous transplant in T1DM patients in the future.The gallbladder and cystic duct are attractive sources of potentially renewable β-like cells not only for their close developmental origin to the pancreas, but also for their amenability for surgical extirpation without serious adverse health effects [10].In this study, we developed a reliable method for the large-scale expansion of patient-derived human GBCs ex vivo for the purpose of genetic reprogramming to β cell fate.We reprogrammed GBCs into insulin-producing pancreatic β-like cells by combining adenoviral mediated transduction of pancreatic endocrine transcription factors PDX1, MAFA, NEUROG3, and PAX6 and small molecule-dependent transdifferentiation. Global mRNA and microRNA expression profiles and protein immunostaining experiments indicate that reprogrammed GBCs (rGBCs) adopted an overall β-like state and could transiently engraft in immunodeficient recipients.

Extrahepatic biliary tissue specimens
Normal human gallbladder and cystic duct specimens (collectively referred to as extrahepatic biliary tissues or EHBT) were obtained from the Department of Surgical Pathology at the Oregon Health & Science University (OHSU).The pathologist/research pathology technician-onduty selected and excised grossly normal EHBT fragments (1 cm 3 to 3 cm 3 ) and kept them on ice in DMEM/F12 supplemented with 100 units/mL Penicillin and 100 μg/mL Streptomycin (Mediatech, Manassas, VA, USA).Tissue samples were anonymized as outlined in the IRB study exemption approved by the OHSU Institutional Review Board.Single cell suspensions of gallbladder and cystic duct (collectively referred to as gallbladder and cystic duct cells or GBCs) were prepared by protease digestion.First the tissue (EHBT) was incubated in 1X EBSS (Gibco, Grand Island, NY, USA) supplemented with 10 mM EGTA (Sigma, St. Louis, MO, USA) and 1% v/v HEPES buffer (Mediatech, Manassas, VA, USA) for 5 minutes at room temperature.The EHBT fragments were then digested in 5 mg/mL collagenase II (Gibco, Grand Island, NY, USA) in 1X HBSS (HyClone Laboratories, South Logan, UT, USA), 20 μg/mL DNaseI (Roche, Laval, QC, Canada) and 1% v/v HEPES (Mediatech, Manassas, VA, USA) for a total of 2-hours with intermittent agitation in a 37 ˚C water bath.EHBT tissue chunks were subsequently strained through a 100 μm nylon mesh (BD Falcon, Bedford, MA, USA) and the resulting suspension of GBCs was washed and treated with ACK lysis solution (0.15M NH 4 Cl, 10 mM KHCO 3 , 0.1 mM EDTA, pH 7.4, 0.22 μm filter-sterilized) for 2 minutes at room temperature to remove red blood cells.Finally, GBCs were washed and strained using a 40 μm nylon mesh (BD Falcon, Bedford, MA, USA) and suspended in DMEM/F12 supplemented with 0.5% fetal bovine serum (Sigma-Aldrich, St. Louis, MO, USA) and kept on ice.To archive tissues, EHBT samples were cut into less than 1 cm 3 fragments and embedded in Tissue-tek cryomatrix (Sakura, Tokyo, Japan) and stored at -81 ˚C.

RNA isolation, cDNA synthesis and quantitative PCR
Reprogrammed GBCs (rGBCs) were washed with 1X PBS once and Trizol reagent was directly added to the cells.RNA was isolated following the Trizol standard protocol.For cDNA synthesis, 1-5 μg of RNA was reverse transcribed using M-MLV reverse transcriptase (Invitrogen, Carlsbad, CA, USA) following the manufacturer's recommended protocol in the presence of 25 μg/mL oligo(dT)18, 0.5 mM dNTP, 10 mM DTT, 40 units RNaseOUT recombinant ribonuclease inhibitor (Invitrogen, Carlsbad, CA, USA).SYBR-based quantitative PCR was performed by utilizing either the iQ TM 5 or CFX96 Touch TM Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA) to measure gene expression.The qPCR reactions were comprised of Platinum 1 Taq polymerase, 2.5 mM MgCl 2 , 10 μM DNA primers, 10 mM dNTPs, and 0.5X SYBR green (Invitrogen, Carlsbad, CA, USA).Thermocycling reactions were run as follows: 45 cycles of 15 s at 95 ˚C, 20 s at 68 ˚C, and 25 s at 72 ˚C.PCR primers are shown in S1 Table.

Measurement of intracellular and secreted c-peptide
For measurement of intracellular of C-peptide content, we harvested two-week old rGBC by rinsing with prewarmed PBS 3 times and lysing in 1X cell lysis buffer (Cell Signaling Technology, Danvers, MA, USA) with a complete mini-protease inhibitor cocktail (Roche, Mannheim, Germany).The cells were scraped off the dish, sonicated, and centrifuged at 14,000g at 4 ˚C, saving the supernatant for C-peptide ELISA (Mercodia, Uppsala, Sweden).The C-peptide secretion assay was done as described previously [17].Specifically, 2-week old rGBC in 12-well plates were incubated in fresh prewarmed day 14 reprogramming medium (MM3) for 2 hours at 37 ˚C, then followed by rinsing three times with plain Krebs-Ringer bicarbonate buffer (KRBB) 20 minutes each at 37 ˚C.Subsequently, rGBC were incubated with 150 μL KRBB with or without 1mM or 2.8 mM glucose for 2 hours in a humidified 37 ˚C incubator.Supernatants were collected, centrifuged to remove cell debris, and stored at -80 ˚C.Into the same wells, we added 150 μL KRBB with 25 mM glucose and incubated for 2 additional hours and harvested the supernatants.Total protein contents were measured using a BCA protein assay kit (Pierce Biotechnology, Rockford, IL, USA).Glucose-stimulated C-peptide was measured using Ultrasensitive Human C-peptide ELISA kit (Mercodia, Uppsala, Sweden).

Flow cytometry
Dissociated GBCs were incubated in Hpi1 or Hpi2 mAb (courtesy of Oregon Stem Cell Antibody Core) [40] for 30 min at 4 ˚C and washed with cold DMEM prior to labeling with APCconjugated anti-mouse IgG (Jackson ImmunoResearch, West Grove, PA, USA).Propidium iodide (10 μg/mL) (Molecular Probes, Eugene, OR, USA) was included to mark dead cells.Cells were analyzed with a FACSCalibur or sorted by Influx-GS (BD Biosciences, San Jose, CA, USA) at 15 psi using a 100 μm nozzle.FlowJo software (Treestar, Ashland, OR, USA) was used to analyze flow cytometric data.

RNA sequencing
FACS-sorted rGBC were subdivided based on positive staining with the Hpi1 mAb [40] and collected directly into a 1.5 mL microcentrifuge tube with 750 μL Trizol-LS.RNA was isolated using a modified Trizol-LS protocol and subsequent column-based method using Qiagen RNeasy mini kit (Qiagen GmbH, Hilden, Germany).RNAseq libraries were constructed and indexed using TruSeq RNA library prep kit v2 (Illumina Inc., San Diego, CA, USA).Short read sequencing assays were performed by the OHSU Massively Parallel Sequencing Shared Resource using Illumina HiSeq2000.RNA sequencing data are available from ArrayExpress database (accession number E-MTAB-5556).RNA library sequencing reads were aligned and annotated according to UCSC RefSeq (hg19 assembly).We used START Shiny App (https://kcvi.shinyapps.io/START)[41] for pairwise comparisons and statistical calculations of p-values and adjusted p-values.Absent or lowly expressed genes in all samples (RPKM<1) were not included.Differentially expressed genes were picked based on log 2 FC>1 and p<0.01.We further selected differentially expressed genes (log 2 FC>5 and p<0.01) for gene set enrichment analyses (GSEA) [42].Heat maps and hierarchical clustering (one-minus Pearson correlation or Euclidian distance) were generated using the GENE-E or Morpheus online tool (Broad Institute, MIT) and graphs were made using Prism 7 (GraphPad Software, Inc., La Jolla, CA, USA).Venn diagrams were generated using Pacific Northwest National Laboratory Venn Diagram Plotter.

microRNA sequencing
Total RNA from FACS-sorted and unsorted rGBC and untransduced GBC were isolated using Qiagen microRNeasy minikit (Qiagen GmbH, Hilden, Germany) following the manufacturer's protocol.microRNA libraries were generated and processed by the University of Pennsylvania's Next Generation Sequencing Core using Illumina truSeq small RNA library preparation kit (Illumina, San Diego, CA, USA) configured for 50 cycle and sequenced on Illumina HiSeq 2500 (Rapid-V2).MicroRNA sequencing data are available from ArrayExpress database (accession number E-MTAB-5554).Adapter sequence was trimmed from the 3'-end of the sequences allowing up to 1 mismatch every 4 bases.Trimmed reads were aligned against pre-miRNAs from mirBase18 using Bowtie.Only reads with a single best alignment were considered.Trimmed reads aligned to premiRNA were checked for overlap with annotated or inferred mature forms (miRbase v18).We compared rGBC and GBC with publicly accessed datasets on pancreatic β cells [43] (cell.betadataset GEO accession number GSM1262771).We calculated differential expression by using a simple fold change MvA comparison analysis (where M = log 2 rGBC/β fold change and A = log 2 rGBC average expression).We selected microRNAs with RPM>5 and FC>10 as differentially expressed.Heat maps, Venn diagrams, and graphs were created similar to RNA sequencing analyses.

Transplantation of reprogrammed gallbladder cells
All animal experiments were done at OHSU and approved by the OHSU Institutional Animal Care and Use Committee.Transplantation experiments involved the use of 8-12 week old NSG (NOD.Cg-Prkdc scid Il2rg tm1Wjl / SzJ) [44,45] or NSG-RIP-DTR [NOD.Cg-Prkdc scid Il2rg tm1Wjl / Tg(Ins2-HBEGF) 6832Ugfm/SzJ] mice [46].Mice were kept in conventional mouse rooms at constant temperature (70-72 ˚F) and 12h light/dark cycle with free access to food and water.Day 14 rGBCs were washed with 1X PBS and collected using rubber cell scrapers and transferred into 15-mL conical tubes.rGBC clusters were centrifuged at 1,200 rpm for 7 minutes at 4 ˚C and the supernatant was discarded.For transplantation under the kidney capsule, rGBC pellets (5x10 5 -2x10 6 cells) were mixed with 50 μL whole blood collected from a NSG mouse and allowed to clot.rGBC-clot mixtures were deposited under the left kidney capsule of an anesthetized (isoflurane vaporizer set at 2% with O 2 flow rate of 1 liter/min) NSG mouse (thirty-four animals received non-reprogrammed GBC and seventeen received rGBC).For transplantation of rGBCs into the mammary fat pad (n = 17 mice), epididymal fat pad (n = 17 mice), and dorsal cutaneous fat (n = 12 mice), the scraped rGBC pellets were suspended in ~40 μL hydrogel (Advanced BioMAtrix, Carlsbad, CA, USA) and injected directly into the fat depots.Between 3-5 independent in vitro reprogramming experiments were performed to generate transplantable rGBCs in 3-5 cohorts of mice per transplant site.Post-operative mice were assessed for overall health and wellbeing following standard operating procedures set by OHSU Department of Comparative Medicine.After indicated time points, xenografts from kidneys and fat pads were excised and fixed in 4% paraformaldehyde in 1X PBS for 6-10 hours at 4 ˚C.Tissues were washed in 1X PBS and immersed in 30% sucrose in 1X PBS for 18-24 hours at 4 ˚C.Fixed tissues were blotted dry, cut into smaller fragments (less than 1 cm 3 fragments), and embedded in Tissue-tek cryomatrix (Sakura, Tokyo, Japan), frozen in methylbutane/liquid nitrogen, and stored at -80 ˚C.

Immunofluorescence
OCT tissue blocks were cut into 5-8 μm thick cryosections using a Reichert 2800 Frigocut cryostat (Reichert Scientific Instruments, Buffalo, NY, USA) onto Superfrost Plus slides (Fisher Scientific, Pittsburgh, PA, USA) and allowed to dry for 10 minutes.Slides were rinsed three times with 1X PBS, permeabilized with 0.1% Triton X-100 in 1X PBS at room temperature for 10-15 minutes, and washed three times with 1X PBS.Slides were blocked with 5% normal goat serum/normal donkey serum in 1X PBS for at least 30 minutes at room temperature.Slides were incubated with primary antibodies (1:100 dilution v/v) (S2 Table ) overnight in a humidified chamber at 4 ˚C.Primary antibodies were removed and slides were rinsed three times with 1X PBS.Fluorochrome-conjugated secondary antibodies (goat and donkey; 1:500 dilution v/v) were incubated for 1-2 hours at room temperature.Slides were washed three times with 1X PBS prior to mounting with Fluoromount-G with DAPI (Southern Biotech, Birmingham, AL, USA).Slides were evaluated using a Zeiss Axioskop 2 plus microscope or LSM 700 confocal microscope (Carl Zeiss, Jenna, Germany).

Statistical analyses
All statistical calculations were performed using Prism 6.0/7.0 (GraphPad Software, Inc., La Jolla, CA, USA), unless otherwise stated.Unpaired t-tests were done to determine statistical significance (α = 0.05).Correlation between two datasets was measured by Pearson correlation coefficient (95% confidence interval).

Adult human gallbladder is a rich source of expandable autologous cells for pancreatic endocrine reprogramming
For human gallbladder to be a practical source for β cell replacement, a reliable culture and expansion method needs to be in place.We cultured human GBCs on a subconfluent feeder layer of irradiated LA7-RSPO1 cells and a culture medium modified from murine gallbladder medium [10,39].This enabled robust growth and reliable passaging regardless of donor source.Using this protocol, we successfully cultured and serially passaged GBC from over 100 gallbladder and cystic duct specimens ~80% of the time.The addition of cholera toxin (increased cAMP) [47], nicotinamide (increased nicotinamide adenine dinucleotide) [48], Nacetylcysteine (antioxidant) [49], non-essential amino acids, and R-Spondin (enhancement of canonical Wnt signaling) [50] significantly accelerated the expansion of human rGBCs comparable to its murine counterpart, which eventually allowed us to propagate millions of human rGBCs in vitro.Cultured GBCs had a high nucleocytoplasmic ratio with prominent nucleoli, a cellular diameter between 7 μm and 12 μm, and grew in tightly packed colonies flanked by the larger irradiated feeder cells (Fig 1A).GBCs could be successfully cultured up to 12 passages allowing for robust expansion in vitro.After initial seeding of 5x10 5 primary cells, GBCs could be expanded to 1x10 8  To guide our in vitro genetic reprogramming and differentiation protocols, we used RNAsequencing data from both human GBCs and FACS-sorted β cells.The comparative study of the transcriptomes of GBCs and human pancreatic β cells allowed us to ascertain candidate pancreatic endocrine factors for genetic reprogramming.Gene set enrichment analyses of the top 224 differentially expressed genes in human β cells relative to GBCs (log 2 FC>5 and p<0.01) indicated that these genes (missing in GBCs) are involved in regulation of β cell development, gene expression, peptidase activity, insulin secretion, synthesis, and processing (S3 Table ).The baseline gene expression profile indicated that primary GBCs did not produce insulin and were deficient or expressed very low levels of β cell-associated lineage, differentiation, and identity factors (PDX1, MAFA, NEUROG3, PAX6, NEUROD1, NKX2-2, and NKX6-1) (Fig 1C).Previous studies utilized some of the abovementioned transcription factors for the induction of a β-like state in different mouse and human tissues [10, 18-20, 22, 51].Similar to our previous work with mouse gallbladder [10,38], we used replication-defective adenovirus serotype 5 (RDAd5) encoding for multiple transcription factors in both monocistronic and multicistronic expression cassettes (Fig 1D

Transduction of M6PN transcription factors induced the pancreatic endocrine program in rGBC
GBCs transduced with reprogramming factors began to show morphological changes from a monolayer into three-dimensional clusters as early as two days after transduction and cell growth slowed markedly after one week.Cells that were not incorporated into the threedimensional clusters of rGBC eventually died within two weeks.Induction of insulin expression in reprogrammed GBCs (rGBCs) was evident on day four after viral transduction (S1G Fig) .We extended our observation period up to nineteen days to determine the temporal pattern of insulin expression in rGBCs.In M6PN-transduced GBCs, insulin mRNA peaked around nine days and began to decrease slightly after two weeks.Importantly, the combination of all four transcription factors resulted in the highest insulin expression.M6PN-rGBCs had 4 to 5 orders of magnitude greater insulin expression relative to 3-factor (M6P) or 1-factor (N) transduction groups, which also remained consistent over the entire period of observation (Fig 2A).The kinetics of GBC reprogramming was affected not only by the pancreatic endocrine factors introduced but also by the reprogramming media and is consistent with previous work with pancreatic ducts [17].Pre-treatment of GBCs with DMSO and the addition of retinoic acid, T3, Alk inhibitor (SB431542), and ISX-9 to the reprogramming medium further improved both INS and NKX6-1 levels (S2 Fig) .Based on these observations, we performed subsequent reprogramming experiments for a period of 14 days to measure expression levels of β cell-associated factors induced in rGBCs.Quantitative RT-PCR showed that rGBCs upregulated many important β cell-associated genes to levels similar to those seen in intact human islets.Whereas, INS and NKX6-1 were 10-100 times less in rGBCs, respectively (compared to islets), NEUROD1, NKX2-2, RFX6, PAX4, MAFB, and HOPX expression were 5-50 times higher in rGBC compared to pancreatic islets.Other genes related to insulin processing (PCSK1) and granule storage/release (CHGA, TMEM27, SYP, KCNJ11, ABCC8) were likewise strongly upregulated.Among other islet-produced hormones, GCG was barely induced while SST and GHRL were significantly expressed.PPY gene expression was around 50 times lower in rGBCs than human islets.While many pancreatic endocrine genes were highly induced, the reprogrammed cells did not fully silence GBC genes.For example, the GBC-related MUC5B gene was still highly expressed in rGBCs (Fig 2B).The temporal expression of pancreatic endocrine genes in rGBCs followed the same trajectory as INS-peaking between day 9 and 14 and declining afterwards (Fig 2C and 2D).With the exception of INS, NKX6-1, and PPY, all other pancreatic endocrine-associated genes (e.g.NEUROD1, NKX2-2, RFX6, PAX4, MAFB, HOPX, SST, GHRL, PCSK1, CHGA, TMEM27, SYP, KCNJ11, ABCC8) had greater expression levels than whole human islets (Fig 2B -2D).Overall, insulin-producing rGBCs showed a β-like expression profile with additional expression of δ, ε, and γ endocrine genes, but also retained a partial biliary character.To further evaluate the global effects of genetic reprogramming on human GBCs, we compared the complete transcriptomes of unreprogrammed GBCs and human β cells to rGBCs using RNA-sequencing.The heat map illustrated that rGBCs significantly upregulated genes that are associated with β cell state while downregulating genes that are more characteristic of parental GBCs (Fig 2E).The total number of expressed genes detectable in both human β cell and GBC was 15,249 (RPKM>1), with 9,254 (60.7%) genes not differentially expressed [ABS(log 2 FC)<1 and/or p>0.01)].There were 5,995 differentially expressed genes ([ABS(log 2 FC>1 and p<0.01), whereby 3,037 (19.9%) genes were more highly expressed in GBCs ("GBC genes") while 2,958 (19.4%) were upregulated in human β cells ("Beta genes") (Fig 2F).Of these 2,958 "Beta genes", 1,965 (66.43%) were induced in rGBCs (Fig 2G).Gene set enrichment analysis of the 151 most differentially expressed of these "Beta genes" in rGBCs (log 2 FC>5 and p<0.01) demonstrated that rGBCs upregulated the molecular signatures involved in the regulation of (a) β cell development, (b) gene expression in β cells, (c) MODY (Maturity onset diabetes of the young), (d) and Type 1 Diabetes Mellitus (S4 Table ).However, the analysis also revealed that 993 "Beta genes" remained uninduced in the rGBC population.Gene set analysis showed that these uninduced "Beta genes" in rGBC are involved in generic transcription activity, extracellular matrix proteins, plasma membrane proteins and nervous system development (S5 Table ).Notably, ZNF family of genes encoding for Krueppel C2H2-type zinc-finger family of proteins were well represented (35 out of 44 genes) among the transcription regulators (Reactome_Generic Transcription pathway) which were differentially expressed in human β cells but not in rGBCs.In addition, 1,809 out of 3,037  ).Interestingly, 2,610 genes (28.2%) were unique to rGBC and differentially expressed compared to both unreprogrammed GBC and human β cells (Fig 2H).
In addition to gene expression analysis, reprogramming was also assessed at the protein level.Consistent with the gene expression data, insulin protein production (as measured by Cpeptide) was found on day five (Fig 3A In aggregate, these data show that our reprogramming protocol induced not only a β-like state but also limited δ and ε-like states. To further determine functional characteristics of β cells, rGBCs were tested for their ability to release insulin in response to glucose.The bulk rGBC population had an intracellular Cpeptide content 2 orders of magnitude lower than intact human pancreatic islets (Fig 3H ), taking into account that approximately 10% of total rGBC were C-peptide positive (Fig 3F).The effective intracellular amount was calculated to be 10% of that produced in intact islets.Moreover, similar to intact human islets, rGBCs exhibited good response to glucose stimulation in vitro (Fig 3I).

Enrichment for insulin-producing rGBCs by FACS using a pan-islet endocrine cell surface-reactive monoclonal antibody
The C-peptide+ rGBCs generated by our protocol also included rGBCs positive for other islet hormones (Figs 2B-2D, 3D-3G and S3 Fig) .To separate and enrich for INS+ rGBC, we tested the human islets cell surface markers (Hpi1 and Hpi2) previously used to specifically isolate primary human islet cells by FACS [40,52].Using Hpi1, we were able to enrich insulin-positive rGBCs by flow cytometry.Genetic reprogramming of GBC using M6PN resulted in the upregulation of Hpi1-reactive surface antigen by a factor of 11.6 (Fig 4A).C-peptide-positive rGBCs were exclusively Hpi1+ by immunocytochemistry (Fig 4B).FACS separation based on Hpi1 staining also resulted in the enrichment β-cell specific genes.Hpi1+ rGBCs had marked increases in the mRNA expression of INS (4.8-fold increase) (p = 0.024), NKX6-1 (3.6-fold increase), NEUROD1 (2.12-fold increase), and PCSK1 (2.6-fold increase) over that of the unsorted bulk rGBC population.As expected the gene expression differences were markedly higher when Hpi1+ cells were compared to Hpi-cells.NKX6.1 was increased by 4.7x10 3 -fold, PCSK1 by 55-fold and NEUROD1 by 50-fold.Moreover, INS gene expression was >1,000 higher in Hpi1+ than Hpi1-rGBCs (p = 0.0023) (Fig 4C).In contrast, the pan-islet surface antibody Hpi2 did not enrich for INS or any other pancreatic endocrine genes (S4A and S4B Fig).
Given that Hpi1 is a pan-islet endocrine surface marker, Hpi1+ FACS-sorted rGBCs enriched not just for INS, but also for other islet endocrine genes such as SST, GHRL, and PPY with the exception of GCG (Fig 4G).This suggests that the antigen for Hpi1 does not exclusively associate with insulin-producing cells but with pancreatic endocrine cells in general.More importantly, the expression of hallmark β genes such as NEUROD1, NKX2-2, NKX6-1, RFX6, and MNX1 that are associated with β cell fate, identity, function, and maintenance were thoroughly enriched in Hpi1+ rGBCs (Fig 4H).Furthermore, Hpi1+ rGBCs had greater expression of genes related to insulin processing (PCSK1), packaging (CHGA, SYN), and response to glucose (KCNJ11, ABCC8) relative to Hpi1-rGBC, except for TMEM27 (Fig 4I).

microRNA profiling of reprogrammed human gallbladder cells
The development, cellular identity, function, and physiologic state of pancreatic β cells are governed by interactions or network of both coding and non-coding RNAs [53,54].Above, we demonstrated that rGBCs displayed a similar β cell mRNA expression profile.Given that the reprogramming strategy used was based on ectopic gene expression of pancreatic endocrine transcription factors, we sought to determine whether this reprogramming also resulted in a β cell non-coding RNAs profile.There were 1,896 high confidence (based on miRbase v21) [55] mature microRNAs enumerated in our small RNA library sequencing, 894 microRNAs present in both human β cells and primary GBCs, 63 exclusively β cells, 705 exclusively in primary GBCs, and 235 microRNAs were absent in both cell types.Of the 894 microRNAs detectable (RPM>1) in both GBC and β cells, 81 were more highly expressed in β cells (RPM>5 and FC>10), and 185 enriched in primary GBC (RPM>5 and FC>10).In addition, there were 422 that were not differentially expressed (FC<10 and/or RPM<5 in both cell types) (Fig 5A).Unbiased hierarchical clustering revealed that the global microRNA contents of rGBC had a closer relationship with β cells than with primary GBCs (Fig 5B).Of the 81 β cell-specific microRNAs, 54 (60.1%) were upregulated in rGBCs while only 14 (7.57%) out of the 185 GBC microRNAs remained expressed in rGBCs.Moreover, fifteen of the twenty most highly expressed β microRNAs showed robust expression in rGBC compared to GBC (Fig 5C).In contrast, the top twenty GBC microRNAs were poorly represented (four out of twenty) in rGBC (Fig 5D).Taking into account the overall microRNA expression profiles, there was even stronger correlation between rGBC and β cells (R 2 = 0.725) compared to rGBC and primary GBC (R 2 = 0.440) (Fig 5E).This finding provides further evidence that rGBCs had a microRNA expression levels highly similar to primary human β cells.
Although the overall microRNA expression profile of rGBC looked more similar to β cells, there were several GBC microRNAs that were still upregulated in rGBC (Table 1).Scatterplot correlation analysis revealed hsa-miR-31-5p as one of the most differentially expressed micro-RNA in rGBC and primary GBC that is low to absent in β cells (Fig 5E and 5G and Table 1).These microRNAs are possible targets for suppression to improve the β cell fate conversion of GBC in the future.
We have shown in this study that the surface antibody Hpi1 was effective in enriching for C-peptide+ rGBC.The transcriptome profile of Hpi1+ rGBC was more β-like than unsorted rGBC, while the overall microRNA expression profiles of Hpi1+rGBC and unsorted rGBC were equally very similar to β cells (R 2 = 0.767 versus R 2 = 0.766, respectively) (S5A Fig)  In order to promote long-term engraftment of rGBC, we tested the addition of endothelial (HUVEC) and human mesenchymal stem cells (MSC) into the in vitro reprogramming of human GBC starting on reprogramming day 10 (or 10 days post-transduction).Incorporating HUVEC and MSC with cells from various tissues to form organotypic structures in vitro had been utilized for successful engraftment [56][57][58].We mixed a ratio of 1x10 6 10-day old rGBC with 5x10 5 HUVEC and 2x10 5 MSC.Between 24 and 48 hours, this heterotypic monolayer culture started to condense into a tight structure which became more tissue-like in consistency after five days (S6D  staining revealed that rGBC (C-peptide) and HUVEC cells (huCD31) were rare to absent, while MSC (huCD44) was abundant at that time point (S3D-S3F Fig) .Hence, mesenchyme-driven condensation of rGBC in vitro did not enhance the engraftment of rGBC in immunodeficient mice compared to direct transplantation of rGBC clusters.

Discussion
Our study is the first report of a robust methodology for expansion culture and genetic reprogramming of primary human gallbladder cells from more than a hundred unrelated donors.Previous attempts at in vitro cultures of human gallbladder tissues were hampered by inefficient plating, poor passagability, limited cell yield, and fibroblast contamination [59][60][61][62].Our reliably scalable in vitro expansion of GBCs is a vast improvement from previous methods as evidenced by a greater than 80% plating and multiple passaging efficiency based on more than one hundred distinct gallbladder specimens, the majority of which were smaller than 5 cm 3 in dimension.This new method provided us with millions of primary human GBCs which we genetically reprogrammed with key pancreatic transcription factors.
Our rGBC highly recapitulated the β-like cell state not only in its strong insulin production, response to glucose challenge, and global gene expression signature but also in the overall microRNA expression profile.These are indications that ectopic expression of four transgenes (i.e.M6PN) can drastically remodel both the coding and non-coding genetic landscape of GBC to assume a different cellular state (i.e.β-like cell).As we and others had previously reported, the minimal requirement for PNM to reprogram mouse gallbladder [10], mouse pancreatic acinar/ductal cells [17,20,22], and mouse liver cells [19] parallels that of reprogramming human GBC, further suggesting that a ductal phenotype shared by GBC [63][64][65] with epithelial cells in the liver and pancreas is conducive to genetic reprogramming using PNM.The high reprogramming potential of endoderm-derived ductal cells as evidenced by this report and others [10,17,19,20,22] is correlated to their epigenetic proximity (biliary tree and ventral pancreas) [34].Majority of these studies using the same three reprogramming factors utilized rodent cells/organs as targets for reprogramming.Without taking into account for the inherent differences in species or cellular microenvironment (e.g. in vitro vs. in vivo, primary vs. transformed lines, transdifferentiation protocol for in vitro studies, timing of assays), the comparative reprogramming efficiencies of other adult endoderm derivatives like rat and mouse exocrine cells [20,22], gastric antral cells [33], and Sox9+ cells in the liver [18,19] were better than human rGBCs.However, contrary to liver, stomach, and pancreas reprogramming, the gallbladder has the advantages of easy access for surgical removal without any dire health sequelae and from a practical perspective-scalable ex vivo expansion [10].
Among major β cell transcription factors lacking in human GBC, only the addition of PAX6 to PNM-but not NEUROD1, NKX6-1, NKX2-2-enhanced reprogramming.PAX6 is one of the major regulator of pancreatic endocrine transcription and determinant of normal islet development [66] and was not induced in GBC after PNM transduction.Although PAX6 was not required for GBC reprogramming, its addition to PNM factors clearly improved the induction of insulin and key β-associated genes, especially NKX6-1, NEUROD1, NKX2-2, and RFX6.Interestingly, several of these key endocrine genes were induced at higher levels compared to whole human islets.Many of these activated genes are expressed downstream of the reprogramming factors [67].These adenovirally encoded reprogramming factors M6PN were constitutively overexpressed in rGBCs under the potent CMV promoter, thus, potentially induced downstream genes at significantly higher levels than found in human islets.Consequently, it is still not known the most optimal stoichiometry (or even factor combinations) and actual expression levels of reprogramming factors needed for human gallbladder to completely and efficiently convert to mature and long-lived β-like cells.Definitely, forced overexpression of master reprogramming factors will most likely lead to unregulated and supraphysiologic levels of target downstream genes as observed in this study.Unlike that seen in mouse gallbladder reprogramming [10], GCG was not upregulated in human rGBC.A previous study had implicated PAX6 with the activation of the GCG promoter [17], which we did not observe in this study, pointing to other binding partners neither available nor induced in human rGBC.It's been posited that the extrahepatic biliary tree (GBC included) and the ventral pancreas arose from the same SOX17+ progenitor cell [34].Interestingly, studies performed in rat islets showed that the ventral pancreas to be glucagon-poor while the dorsal pancreas to be glucagon-rich [68][69][70].Therefore, if rGBC reprogramming was biased towards a "ventral pancreatic endocrine" state, then this could point to why rGBCs had absent glucagon expression.
Compared to a previous study in primary human pancreatic CD133+ ductal cells (huP133+) [17] the transduction of the same four transcription factors M6PN in human GBCs induced greater insulin gene induction (0.01 rGBC vs.0.002 huP133+ relative to human islets).The insulin induction was even higher in Hpi1+ rGBC (>0.05).Both rGBCs and reprogrammed huP133+ had comparable insulin secretion response to high glucose.Overall expression of β cell-associated genes in rGBCs were either greater or comparable to reprogrammed huP133+ [17].In comparison to primary mouse rGBCs, human rGBCs yielded relatively better in vitro reprogramming efficiency after transduction with M6PN factors [38].However, both reprogrammed huP133+ and mouse rGBCs were able to mature and survive longer in vivo [17,38].
Given the highly similar transcriptomes and microRNAomes of rGBC and human β cells, we hypothesized that the surface antigens found in human islets may be induced in rGBC as well.The availability of a pan-islet surface antibody (Hpi1) enabled us to further enrich for the most reprogrammed rGBC, enabling us to subdivide rGBC into two subpopulations: (a) highly β-like Hpi1+ rGBC and (b) poorly reprogrammed Hpi1-rGBC.Of the two, only Hpi1 + rGBCs genetically and phenotypically resembled real β cells, therefore the use of this monoclonal antibody in rGBC (and possibly other reprogrammed cell types) will be useful in ensuring purity of rGBC in future experiments and eventually for transplantation purposes.
Although the rGBC we had so far generated was highly β-like, there are several issues-that remain to be addressed for rGBC to achieve a truly β-cell state and transplantation utilityincluding the suboptimal reprogramming efficiency (between 10-25%), transient engraftment in vivo, presence of residual biliary signature, and persistence of transgenes during reprogramming.The reprogramming rate of less than 25% may be attributed to gene and microRNA sets associated with β cell state that remained deficient in rGBCs.Notably, ZNF family of genes encoding for Krueppel C2H2-type zinc-finger family of proteins were well represented (35 out of 44 genes) among the transcription regulators that were highly expressed in human β cells but not in rGBCs.On the other hand, examples of genes implicated in the resistance to reprogramming/insulin induction are CDX2 [33], FOXO1 [33], HES1 [35], NR5A2 [33], and PTF1A [71].Based on our transcriptome analysis, primary human GBC expressed significant amounts of CDX2, FOXO1, HES1, and NR5A.Even after M6PN reprogramming the resulting rGBC still expressed high levels of these four factors, which may warrant silencing of these negative regulators in future reprogramming of GBC.Although microRNAs hsa-miR-3168 and hsa-miR-7-2-3p were upregulated in β cells but not in rGBC, none of the two have proven link to β cell program [72].Similarly, hsa-miR-31-5p was differentially expressed in rGBC and it remains to be tested whether this non-coding RNA targets critical genes involved in the β cell program.
Vector-encoded transgenes as measured by GFP (which is co-expressed with Ad5-M6P) in rGBC showed continued expression.Ideally, transgene expression like that of NEUROG3 should be short-lived, since it is not expressed in adult β cells but only transiently early in pancreatic development during endocrine specification [73].The continued expression of NEU-ROG3 may have contributed to the immature and polyhormonal phenotype [74], the finite duration of INS expression, and overall viability of rGBC [75].Given the limited lifespan of rGBCs, we were not able to determine how the temporal withdrawal of the exogenous reprogramming factors will affect expression of endogenous factors and overall stability of rGBC reprogramming.Indirect immunofluorescence staining patterns of NEUROD1, NKX2-2, GHRL, and PP in rGBC are indicative of how persistent transgene expression of NEUROG3 and heterogeneous viral transduction produce partial reprogramming.Moreover, the mostly cytoplasmic localization of NEUROD1, NKX2-2, and NKX6-1 proteins potentially contribute to the incomplete cell fate conversion of rGBCs into β cells.The cross-regulation of NKX2-2 and NEUROD1 are predictive of the type of islet hormones (GHRL, PP, GCG) expressed in NEUROG3+ cells [74].We have noticed the variable GFP intensity in transduced rGBC which suggests that transgene expression of M6PN may be different from cell to cell.Therefore, NEUROG3+ cells have variable transgene expression levels which suggest a heterogeneous population of NEUROG3+ rGBCs resulting to a proportion of immature β phenotype in rGBC.
Recent advances in the differentiation of pluripotent stem cells (PSCs) into functionally glucose-responsive insulin-producing β-like cells have pushed the promise of cell therapy for diabetes closer to reality [11,14,15].However, there are several obstacles that need to be overcome such as an effective and durable transplantation methodology that will protect these PSC-derived β-like cells from host immune response as the widely available stem cell lines are allogeneic.The potential for teratoma formation or uncontrolled proliferation is present and life-long immunosuppression is also assured unless patient-derived PSCs are generated.Even with patient-derived PSCs, some of the disadvantages include the long time required, the complicated and variable nature of reprogramming, expansion, and potential clonal variation, intrapatient variation, and epigenetic abnormalities, and heterogeneous differentiation process to β-like cells [76].The reprogramming of adult cells like the GBC (albeit not yet as highly efficient as PSC-derived β-like cells) still offers advantages over PSC differentiation: (a) autologous transplantation without the time-consuming and expensive reprogramming to pluripotency and differentiation, and (b) absent teratoma risk (based on greater than six months transplantation of GBC in NSG mice).
In summary, we had developed, for the first time, a reliable and robust culture expansion and genetic reprogramming of multiple unrelated patient-derived adult primary GBCs into pancreatic β-like cells ex vivo, thus showing that the human gallbladder is a potentially rich source of reprogrammable cells for autologous cell therapy for diabetes.