Cells with Surface Expression of CD133 high CD71 low are Enriched for Tripotent Colony-Forming Progenitor Cells in the Adult Murine Pancreas

Progenitor cells in the adult pancreas are potential sources of endocrine beta cells for treating type 1 diabetes. Previously, we identified tri-potent progenitor cells in the adult (2–4 month-old) murine pancreas that were capable of self-renewal and differentiation into duct, acinar, and endocrine cells in vitro . These progenitor cells were named pancreatic colony-forming units (PCFUs). However, because PCFUs are a minor population in the pancreas (~1%) they are


INTRODUCTION
The pancreas plays a critical role in regulating metabolism and is composed of three major cell lineages: acinar cells, which secrete digestive enzymes such as amylase; duct cells, which transport digestive enzymes into the gut and secrete mucin to fend off pathogens; and endocrine cells, which secrete hormones such as insulin and glucagon that are important for glucose homeostasis.
There has been an intense debate about whether the adult pancreas harbors stem and progenitor cells that can give rise to new insulin-producing beta cells. Because of their roles as progenitors in the early embryo, duct cells in the adult pancreas have long been the assumed source of new beta cells [6]. However, in mouse models using Cre-Lox lineagetracing techniques, contradictory results either positively [7,8] or negatively [3,[9][10][11][12][13] implicate ducts as the source of progenitor cells in the adult pancreas. In addition to ducts, cells expressing markers for acinar (Ptf1a [14,15], carboxypeptidase A [16], or elastase [15]) or endocrine cells (insulin [9] or glucagon [17]) have been shown to give rise to new beta cells in the adult murine pancreas. Together, these results suggest that there are multiple mechanisms by which beta cells can be replenished from different cellular sources in the adult pancreas. However, the aforementioned in vivo lineage-tracing studies did not directly address properties of multi-potentiality and self-renewal-two criteria necessary to define a stem cell.
To investigate progenitor cells in the adult pancreas, we and others have turned to in vitro single cell analysis. We designed unique 3D colony assays that allow quantitative analyses of single progenitor-like cells from adult murine pancreas. Using these assays, we found that the adult murine pancreas contains tripotent progenitor cells that are capable of multilineage differentiation and robust self-renewal in vitro [18][19][20]. These progenitor-like cells were named "pancreatic colony-forming units" (PCFUs). Consistent with our findings, other laboratories confirmed that dissociated single cells from the adult pancreas of mice [21,22] and humans [23] can be propagated in high concentrations (>33% vol/vol) of Matrigel in vitro where they generate cystic "organoid" colonies similar to what we observed [18][19][20]. Together, these results demonstrate that single cells with the capacities for self-renewal and multi-lineage differentiation in vitro are present in the adult pancreas. However, these PCFU progenitor cells remain poorly characterized.
One major roadblock to effective characterization of PCFUs is that they constitute a minor population in the adult pancreas (~1% in 2-4 month-old mice) [18,20]. Thus, purification of these progenitor cells is needed. Previously, a transgenic mouse model was used for enrichment of PCFUs [18]. In this mouse model, expression of enhanced green fluorescence protein (EGFP) reporter was driven by Sox9 regulatory loci (75 kb upstream and 150 kb downstream sequences) in a CD1 out-bred background [24]. However, this strategy prohibits enrichment of PCFUs from other mouse models. Therefore, identification of cell surface markers expressed by PCFUs could lead to an alternative enrichment method. CD133 (prominin 1) is a cell-surface marker commonly used to enrich various stem cells from adult tissues [25], and adult pancreatic ducts of humans and mice express CD133 [26][27][28][29]. Our prior work demonstrated that CD133 + , but not CD133 − pancreatic cells from adult C57Bl/6 (B6) mice, contained PCFUs [18,20]. However, only one in about twenty pancreatic CD133 + cells was a PCFU [20], consistent with a recent report [22].
The goal of this study was to identify an additional cell-surface marker that, when combined with CD133, could further distinguish and enrich PCFUs. Cell surface markers that were previously known to enrich non-pancreas progenitor cells, including CD71 (transferrin receptor), were screened. CD71 transports iron from the extracellular space into cells, and higher levels of CD71 expression are detected in erythroid progenitor cells [30] as well as in some cancer cells [31]. The adult pancreas expresses CD71 [32]; however, its role in the pancreas has not been characterized. Here, we report that CD71 expression status fractionates pancreatic CD133 + ductal cells in adult mice. Among the subpopulations, CD133 high CD71 low cells are the most enriched for PCFUs, and approximately one in three CD133 high CD71 low cells is a PCFU. This enriched population will enable further studies on putative pancreas stem and progenitor cells in vivo.

Expression of CD133 and CD71 in the adult murine pancreas
CD133 is expressed by the ductal cells of the adult pancreas [26][27][28][29]. We used flow cytometry analysis to screen other known cell-surface markers to determine their ability to fractionate adult CD133 + pancreatic cells. CD71 was found to divide CD133 + ductal cells into three major subpopulations: CD133 + CD71 − , CD133 high CD71 low , and CD133 low CD71 low cells, as indicated by regions (R) 1 to 3, respectively (Fig. 1A), suggesting heterogeneity of ductal cells.
Quantitative (q) RT-PCR analysis of freshly sorted cells indicated that cells in both R1 and R2 expressed higher levels of ductal genes (Krt19, Krt7 and Sox9) and Pdx1 compared to unsorted cells (Fig. 1B). Compared to R2, R1 cells expressed higher levels of endocrine markers, Insulin1 and Glucagon (Fig. 1B). Because the lower limit of the R1 gate was close to the double negative cells, some endocrine cells may have been contaminated during sorting. In the subsequent RNA-seq experiments (see below), the lower boundary of the sorting gate for R1 was moved upward, and as a result no difference in the expression of Insulin1 and Glucagon between R1 and R2 cells was observed (Supplementary Table 1). R3 cells did not express significant levels of the aforementioned pancreatic lineage markers (Fig. 1B).
To further confirm the expression of CD71 in the ducts, immunohistochemical staining on sections of adult murine pancreas was performed. As expected, CD133 was expressed in the ductal structures ( Fig. 1C; dotted line). CD71 was found to co-localize with CD133 in the ducts (Fig. 1C). For additional confirmation, ductal cells that expressed Sox9 also stained positive for CD71 (Fig. 1D), again demonstrating that CD71 was expressed in the ductal region.
Pdx1 protein was detected in the CD133 + ductal cells by double immunostaining analyses (Fig. 1E), although the overall intensities of Pdx1 was lower in ducts compared to that in the islet cells, which are known to express Pdx1 [34]. Interestingly, the staining intensities for Pdx1 or CD133 varied among individual ductal cells (Fig. 1E), again suggesting heterogeneity of ductal cells.
To further characterize R1 and R2 ductal cells, genome-wide gene expression analysis using RNA-seq was performed ( Supplementary Fig. 1A). Compared to R1, R2 cells expressed higher levels of several ductal and epithelial cell markers including Krt19 (Fig. 1F), consistent with the above qRT-PCR analysis (Fig. 1B). R2 cells expressed higher levels of cyclin A2, a positive regulator of cell cycle gene (Fig. 1F), suggesting that R2 cells may be ready for proliferation and forming colonies. Indeed, R2 but not R1 ductal cells gave rise to colonies in culture (see below). Pathway analysis of the RNA-seq data revealed that R2 and R1 cells have propensities for cell migration and immune response, respectively ( Supplementary Fig. 1B). Taken together, these results demonstrate that, unlike the Sox9/ EGFP mice used in our previous study [18], CD71 expression status fractionates pancreatic CD133 + ductal cells into subpopulations in adult mice and that ductal cells are heterogeneous.
Most CD133 − cells expressed CD71 at higher levels ( Fig. 1A; R4) or at undetectable levels (double negative cells). Approximately 80% of CD133 − CD71 high (R4) cells expressed TER119 ( Supplementary Fig. 2), an erythroid lineage-specific marker [33], suggesting that R4 contains mostly red blood cells or their immediate precursors. Red blood cells were lysed before antibody staining and flow cytometry analysis. This manipulation resulted in a reduced percentage of CD133 − CD71 high (R4) cells ( Supplementary Fig. 3), further confirming that R4 contains red blood cells.

Murine adult pancreatic CD133 high CD71 low (Region 2) cells are most enriched for colonyforming progenitors
Previously, we devised two pancreatic colony assays, one containing Matrigel and the other a laminin hydrogel, which support the formation of Ring or Endocrine/Acinar colonies, respectively (Supplementary Fig. 4) [18]. These colony types are morphologically distinct. Ring colonies are cystic and contain mostly ductal-like cells [18,20]; whereas, Endocrine/ Acinar colonies are small and compact, composed mostly of endocrine-and acinar-like cells and fewer ductal cells [18]. A single progenitor cell capable of giving rise to a Ring or an Endocrine/Acinar colony is designated as a PCFU-Ring or a PCFU-Endocrine/Acinar, respectively.

Ring colonies derived from the CD133 high CD71 low (R2) cells are ductal-like, express progenitor markers, and contain self-renewing colony-forming cells
To test the lineage compositions of individual Ring or 1° Endocrine/Acinar colonies, microfluidic qRT-PCR analyses were performed ( Fig. 3A & B). Microfluidic qRT-PCR has a reaction volume in the nano-liter range, permitting gene expression analysis from a single colony [18,35]. Colonies were individually handpicked under direct microscope visualization (n=21 per group). Expression of a panel of genes indicative of pancreatic progenitor, ductal, endocrine, and acinar lineages was analyzed.
Three-week-old Ring colonies continued to express CD133 and CD71 ( Fig. 3A & B), suggesting that the originating CD133 high CD71 low (R2) cells may have either given rise to ductal cells or self-renewed as progenitors. To test whether the freshly sorted CD133 high CD71 low (R2) cells could self-renew, 3-week-old primary Ring colonies (n=8) were individually handpicked, dissociated, and re-plated into Matrigel assays for a total of six passages (Fig. 5). All eight primary Ring colonies initiated secondary and subsequent colonies (Fig. 5), indicating that single PCFUs-Ring contained in the CD133 high CD71 low (R2) population were capable of self-renewal in vitro.
Primary "Endocrine/Acinar" colonies derived from the CD133 high CD71 low population express higher levels of endocrine and acinar markers, and lower levels of ductal markers compared to Ring colonies In contrast to Ring colonies, 1° Endocrine/Acinar colonies expressed higher levels of endocrine (Insulin1, Insulin2, Glucagon) and acinar (Amylase 2A) genes, and lower levels of ductal (Sox9, Mucin1) genes in single-colony RT-PCR analyses ( Fig. 3A & B). Protein expression of Amylase (an acinar marker) (Fig. 6A) in whole-mounted 1° Endocrine/Acinar colonies was detected but did not co-localize with C-peptide + cells (a surrogate marker for de novo synthesized insulin). Similarly, Krt19-expressing ductal cells did not co-stain with insulin (Fig. 6B). Surprisingly, double immunostaining of glucagon and C-peptide in 1° Endocrine/Acinar colonies revealed that cells expressing C-peptide were all positive for glucagon ( Fig. 6C; yellow color). This is in contrast to our prior results on the 2° Endocrine/ Acinar colonies that express single hormones [18]. Transmission electron microscopy showed both insulin-like and non-insulin granules within cells in CD133 high CD71 low (R2)derived 1° Endocrine/Acinar colonies (Fig. 6D), further confirming the double-hormonal phenotype. CD133 high CD71 low (R2)-derived 1° Endocrine/Acinar colonies expressed higher levels of several beta cell maturation markers, such as Glut2, Glucokinase, Pcsk1 and Pcsk2, compared to Ring colonies ( Fig. 3A & B). Therefore, we tested whether these 1° Endocrine/ Acinar colonies were glucose responsive in C-peptide secretion in vitro (Fig. 6E). Pooled (~1,000) 1° Endocrine/Acinar colonies released 1.43-fold more C-peptide into the media when presented with 16.7 mM compared to 2.5 mM D-glucose (Fig. 6E). When colonies were bathed with 16.7 mM D-glucose plus 10 mM theophylline (a cAMP potentiator), 2.13fold more C-peptide was released (Fig. 6E). Intracellular C-peptide levels in these colonies (0.65±0.32 μg/g proteins) were ~3,700 fold lower than that in adult murine islets (2,427±1,395 μg/g proteins). Thus, despite the simultaneous expression of insulin and glucagon as well as the low intracellular C-peptide levels, CD133 high CD71 low (R2)-derived 1° Endocrine/Acinar colonies are functional in C-peptide secretion in vitro.

Primary Endocrine/Acinar colonies give rise to ductal, acinar, C-Peptide + Glucagon − , and C-Peptide − Glucagon + cells in vivo
Attempts to dissociate and re-plate 1° Endocrine/Acinar colonies into 2° laminin hydrogel culture did not result in the formation of 2° Endocrine/Acinar colonies. In the positive control experiment, dissociated Ring colonies did form 2° Endocrine/Acinar colonies. These results suggest that the current laminin hydrogel culture condition may not support the development of the 2° Endocrine/Acinar colonies from the 1° Endocrine/Acinar colonies. To test whether an in vivo environment may allow further development of pancreatic lineages, 10-day-old 1° Endocrine/Acinar colonies were placed underneath the renal capsules of streptozotocin-induced diabetic mice (~12,000 colonies per kidney; n=2). The grafts were examined 5 weeks post-transplantation (Fig. 7A). By week 5, cells were either insulin + glucagon − or insulin − glucagon + as shown by immunohistochemical staining (Fig.  7B). In addition, Krt19 + ductal cells and Amylase + acinar cells were also detected in the grafts (Fig. 7B). The ductal-like cells were most likely originated from donor cells because they were located inside the grafts (Fig. 7A, H&E staining and Fig. 7B, blue box); however, further confirmation is required because kidney tubules are known to express KRT19 [36]. Taken together, these results demonstrate that 1° Endocrine/Acinar colonies derived from the CD133 high CD71 low population have tri-lineage differentiation potential in vivo and are capable of giving rise to insulin + glucagon − cells.
CD133 has been shown to positively enrich progenitors from embryos and neonates of murine [26,28,29,39] and human [29] pancreas, as well as from adult human pancreas [23]. We [20] and others [22] have previously shown that pancreatic progenitor cells from adult mice can be enriched by CD133 selection. However, the cell culture techniques reported by others were more difficult to enumerate colonies. Cells were mixed in high concentrations (>33% vol/vol) of Matrigel and placed only at the edge of the culture well. After solidification, the Matrigel was thicker at the periphery than towards the center of the culture dish. When viewed under a phase-contrast light microscope, the images of the resulting colonies were largely overlapping, preventing effective counting. Quantification of colony was achieved by sparse organoid plating or single cell deposition methods [22], which were labor-intensive and resource-consuming. In contrast, the methylcellulosecontaining semi-solid media that we used allowed us to evenly spread Matrigel and colonies across the culture well. In addition, a grid was attached underneath the culture well to further facilitate accurate numeration by avoiding double counting [40]. Using our quantitative assays, we find that approximately one in three adult pancreatic CD133 high CD71 low cells is a PCFU-Ring (Fig. 2C), compared to ~1 in 100 pre-sorted pancreatic cells [18,20] (Fig.  2C), which translates to a 33-fold enrichment. The efficiency of formation of Ring colonies from CD133 high CD71 low cells has now reached ~30% (Fig. 2C), compared to ~5% from CD133 + selection alone in B6 mice [20] and ~15% from CD133 + Sox9/EGFP + cells from CD1 mice [18]. Importantly, the use of commercially-available antibodies rather than transgenic mice will enable the characterization of adult PCFUs in many laboratories.
As mentioned, Ring colonies in our previous studies developed from sorted CD133 + [20] or CD133 + Sox9/EGFP + cells [18]. Consistent with the characteristics of those colonies, the Ring colonies derived from CD133 high CD71 low cells also express higher levels of ductal (Fig. 3B) and pancreatic progenitor cell markers, such as Pdx1 or Ngn3 (Fig. 3B), compared to 1° Endocrine/Acinar colonies. Similar to previous results [20], cells from a handpicked Ring colony derived from CD133 high CD71 low cells generate secondary and subsequent Ring colonies over six serial passages in Matrigel (Fig. 5), demonstrating the capacity for selfrenewal of the originating, single PCFUs-Ring.
A new observation from the current study is that cells in 10-day-old 1° Endocrine/Acinar colonies grown in laminin hydrogel express insulin and glucagon simultaneously (Fig.  6C&D). This is in contrast to the insulin + glucagon − or insulin − glucagon + cells in the 2° Endocrine/Acinar colonies that we identified in a previous report [18]. Previously, freshly sorted CD133 + Sox9/EGFP + cells were initially plated into the Matrigel assay, with or without exogenous R-spondin1, a Wnt signaling agonist, for 3 weeks to generate Ring colonies. When Ring colonies were collected and dissociated into single cell suspension and re-plated into laminin hydrogel for an additional 10 days, the 2° Endocrine/Acinar colonies that formed expressed either insulin or glucagon [18]. These results suggest that our former two-step culture method, in which the PCFUs were expanded prior to the induction of endocrine differentiation, permitted more time for the progenitor cells to develop and mature into single hormonal positive cells. The 1° Endocrine/Acinar colonies derived from freshly sorted CD133 + Sox9/EGFP + cells were not examined in the previous study [18] because of the low incidence of 1° Endocrine/Acinar colonies.
Five weeks post-transplantation into diabetic mice, CD133 high CD71 low cell-derived, 10-dayold 1° Endocrine/Acinar colonies gave rise to insulin + glucagon − and insulin − glucagon + cells as well as ductal and acinar cells (Fig. 7). These results demonstrate that CD133 high CD71 low cells are progenitor cells with tri-lineage differentiation capacity. An in vivo study reported by Thorel et al. [17] showed that severe beta-cell injury (>99% destruction) led to the formation of insulin + glucagon + cells, followed by the appearance of insulin + glucagon − cells. Because the regenerated beta cells were lineage-traced from glucagon-expressing cells, it was concluded that alpha cells trans-differentiated into beta cells [17]. It is possible that CD133 high CD71 low cells described in this study differentiated into insulin + glucagon − cells in vivo via an insulin + glucagon + intermediate cell stage in vitro. However, further lineage tracing experiments are needed to test this hypothesis.
In summary, we report that adult pancreatic progenitor cells from B6, non-transgenic mice can be highly enriched using two cell-surface markers and live-cell sorting. Approximately 30 and 10-fold enrichment of PCFUs-Ring and 1° PCFUs-Endocrine/Acinar, respectively, in CD133 high CD71 low cells were achieved compared to unsorted cells. It should be noted that only in vitro self-renewal and multi-lineage differentiation by the CD133 high CD71 low cells are demonstrated in the current study, and these results do not prove that progenitor cells exist in vivo in the adult pancreas. Further confirmation will require examining multi-lineage reconstitution and self-renewal of CD133 high CD71 low cells transplanted into an animal model with a severely injured pancreas. Finally, the results presented in this study have direct implications for in vitro generation of glucose-responsive beta-like cells to treat diabetes patients.

Mice
C57BL/6J (donors; 2-4 months old) (The Jackson Laboratory, Bar Harbor, ME, USA) and NOD-SCID (recipients; 2 months old) (Charles River Laboratory, Wilmington, MA, USA) were used in this study. All mice were maintained under specific pathogen-free conditions, and animal experiments were conducted according to the Institutional Animal Care and Use Committee at the City of Hope.

Preparation of Single Cell Suspensions
Murine pancreata were dissected, cleared of fat tissue under a dissecting microscope, and rinsed three times in cold DPBS containing 0.1% bovine serum albumin, penicillin, and streptomycin (PBS/BSA). Whole pancreata, in a dry petri dish placed on ice, were chopped using spring scissors for approximately 2 min or until finely minced. The triturated tissue was transferred to a 15 ml conical tube, washed once, resuspended in PBS/BSA containing collagenase B (2-4 mg/ml) (Roche, Mannheim, Germany) and DNase (2000 U/ml) (Calbiochem, Darmstadt, Germany), and incubated at 37°C for 20 min to yield a mostly single cell suspension. To hasten the digestion, tissue was gently disrupted every 5-10 min using a 16G syringe needle. Cells were then washed twice in cold PBS/BSA supplemented with 2000 U/ml DNase I, which was used to prevent re-aggregation of dissociated cells, and filtered through a 40 μm nylon mesh (BD Biosciences, San Jose, CA, USA) before antibody staining. Between 3-5 million cells per pancreas can be recovered from adult B6 mice.

Colony Assays
Sorted cells were resuspended at a density of 2.5×10 3 cells/well/0.5 ml for the Matrigel assay or 2.5×10 4 cells/well/0.5 ml for the laminin-hydrogel assay as described previously [18]. Culture media contained DMEM/F12 media, 1% methylcellulose (Sinetsu Chemical, Tokyo, Japan), 50% conditioned media from mouse embryonic stem cell derived pancreaticlike cells, 5% fetal calf serum (FCS), 10 mmol/l nicotinamide (Sigma, St. Louis, MO, USA), 10 ng/ml human recombinant activin B, 0.1 nmol/l exendin-4 (Sigma), and 1 ng/ml vascular endothelial growth factor-A (VEGF) (Sigma). When indicated, either 5% Matrigel or 100 μg/ml of laminin hydrogel [18] was added to media to generate Ring or Endocrine/Acinar colonies, respectively. Cells were plated in 24-well ultra-low protein-binding plates (Corning, Corning, NY, USA) and incubated in a humidified 5% CO 2 atmosphere. Ring or Endocrine/Acinar colony numbers were counted 3 weeks or 10 days after plating, respectively. For serial re-plating experiments, individual Ring colonies were handpicked and dissociated into single cells by incubating with 0.25% trypsin-EDTA at 37°C for 5 min, followed by gentle pipetting. Single cells were then re-plated in the Matrigel assay as described above.

Conventional or Microfluidic Quantitative Reverse Transcription-Polymerase Chain Reaction (qRT-PCR)
Total RNA extraction, reverse transcription, and conventional qRT-PCR analyses were performed as previously described [18]. Duplicate samples were used in all analyses. Microfluidic qRT-PCR was performed using the BioMark ™ 48.48 Dynamic Array system (Fluidigm, South San Francisco, CA, USA). Single colonies were lifted one by one from the methylcellulose-containing medium under direct microscopic visualization using a 10 μl Eppendorf pipette, collected in reaction buffer (10 μl), and pre-amplified (14 cycles) according to the manufacturer's instructions (Fluidigm). Amplified cDNA was loaded onto a 48.48 DynamicArray using the NanoFlex integrated fluidic circuit (IFC) controller (Fluidigm). Threshold cycle (C t ), as a measure of fluorescence intensity, was determined by the BioMark PCR analysis software (Fluidigm) and expressed as a heat map or delta C t compared to β-actin. All experiments were performed with negative (water) and positive (adult C57BL/6J pancreatic cells) controls. Taqman probes (Life Technologies, Grand Island, NY, USA) and their catalog numbers are listed in Supplementary Table 2.

Electron Microscopy
Single colonies were collected, pooled, and fixed in 2% glutaraldehyde + 0.1 M Cacodylate buffer [Na(CH 3 ) 2 AsO 2 ·3H 2 O; pH7.2] at 4°C overnight. Colonies were washed three times with 0.1 M Cacodylate buffer (pH 7.2), post-fixed with 1% OsO 4 in 0.1 M Cacodylate buffer for 30 min, and washed three times with 0.1 M Cacodylate buffer. The samples were then dehydrated, embedded in Eponate, and polymerized at 64°C for 48 h. Ultrathin sections (70-nm thickness) were cut using a Leica ultramicrotome (Leica Biosystems, Wetzlar, Germany) with a diamond knife, transferred to 200-mesh EM grids, and stained with 2% uranyl acetate in 70% ethanol for 1 min followed by treatment with Reynold's lead citrate for 1 min. Electron microscopy was performed on a Tecnai 12 transmission electron microscope (FEI, Hillsboro, OR, USA) equipped with an Ultrascan 2K CCD camera (Gatan, Pleasanton, CA, USA).

Immunostaining
For frozen sections, prefixed-pancreata were cryoprotected in 30% sucrose in PBS, embedded in O.C.T. (Sakura Finetek, Torrance, CA, USA), and sliced at 10 μm thickness. For paraffin-embedded formalin-fixed tissues, samples were cut to 5 μm thickness, dewaxed in xylene bath for 15 minutes, rehydrated in ethanol, and antigens retrieved by heating in a microwave oven for 10 min in 200 ml of sodium citrate buffer. Samples were incubated with blocking buffer supplemented with 5% donkey serum and 0.1% TritonX-100 at 4°C overnight before addition of primary antibodies. For whole-mount immunohistochemistry, colonies were manually picked, pooled, fixed in 4% paraformaldehyde at 4°C overnight, and incubated with blocking buffer supplemented with 5% donkey serum and 0.1% TritonX-100 at 4°C overnight. Primary antibodies were as listed in Supplementary Table 3 and detected with donkey-raised secondary antibodies conjugated to Cy3, Cy5, DyLight488, or AlexaFlour647 (Jackson Immunoresearch, West Grove, PA, USA). Images were captured on a Zeiss LSM510 META NLO Axiovert 200M inverted microscope, and figures prepared with LSM Image Browser software (Carl Zeiss, Germany).

In Vitro Glucose Challenge Assay and Intracellular C-peptide measurement
Primary Endocrine/Acinar colonies were pooled and incubated overnight with Krebs-Ringer solution (KRBH; 129 mM NaCl, 4.8 mM KCl, 2.5 mM CaCl 2 , 1.2 mM KH 2 PO 4 , 1.2 mM MgSO 4 , 5 mM NaHCO 3 , 10 mM HEPES, 0.1% bovine serum albumin) containing 10% FCS and 2.5 mM D-glucose. The next day, colonies were washed three times with KRBH containing 2% FCS and 2.5 mM D-glucose and then sequentially incubated with 2.5 mM Dglucose, 16.7 mM D-glucose, and 16.7 mM D-glucose plus 10 mM theophylline (37°C, 0.5 ml/well, 2 h). The C-peptide concentration in the buffer was measured using the murine Cpeptide ELISA kit, U-Type (Shibayaji Co., Gunma, Japan), which has a detection limit of 30 pg/ml. The stimulation was expressed as the fold change of C-peptide concentrations in buffer from the second and third treatments compared to the first treatment of the same well.
To measure intracellular C-peptide, cells were lysed in RIPA buffer (Sigma) and the resulting lysate was analyzed for levels of total protein by BCA protein assay (ThermoFisher Scientific, Waltham, MA, USA) and C-peptide by ELISA as described above.

In Vivo Transplantation
Primary Endocrine/Acinar colonies were handpicked, pooled, and incubated in recovery solution (BD Bioscience) for 1 h on ice. Intact colonies released from the semisolid media were washed and collected for transplantation. NOD-SCID mice (8 week-old males) were injected with streptozotocin (STZ; 50 mg/kg body weight; freshly made in 0.05 M citrate buffer, pH 4.5) for 3 consecutive days to induce diabetes [41]. Hyperglycemia was defined as >200 mg/dl and measured by OneTouch Ultra2 glucometer (LifeScan, Milpitas, CA, USA) starting 1 week after the last STZ injection. Approximately 12,000 Endocrine/Acinar colonies were placed under the renal capsule per mouse (n=2). Grafts were examined 5 weeks post-transplantation by immunohistochemical staining of formalin-fixed paraffinembedded tissue sections.

RNA-seq analysis
Reads were aligned to the mouse genome (build mm10). Reads per kilobase per million (RPKM) [42] were calculated for RefSeq transcripts [43] using an expectation-maximization algorithm in Partek Genomics Suite (Version 6.6) (Partek, St. Louis, MI, USA) [44]. RPKM values were then further normalized via log 2 transformation following the addition of a scaling factor of 0.1 [45]. Fold-change values were calculated on a linear scale based on the least-squares mean. P values were calculated by one-way ANOVA using the normalized RPKM values. False discovery rate (FDR) values were calculated using the method of Benjamini and Hochberg [46]. Transcripts were defined as differentially expressed if they showed a fold-change value higher than 1.5 and FDR lower than 0.05. Potential functional significance was assessed by Gene Ontology enrichment terms (Genomics Suite ™ ) (Partek). Genes differentially expressed by the CD133 high CD71 low compared to the CD133 + CD71 − cells are listed in Supplementary Table 1. Data were deposited to the repository system Gene Expression Omnibus.

PCFU
pancreatic colony-forming unit CD cluster of differentiation

Highlights
• CD133, a cell-surface marker expressed by pancreatic ducts, enriches colonyforming progenitor cells from adult mice but at a low efficiency (~5%).
• CD133 high CD71 low cells are most enriched for colony-forming progenitor cells with a colony-forming efficiency up to 30%.
• CD133 high CD71 low cells are capable of self-renewal and tri-lineage differentiation into duct, acinar, and endocrine cells in vitro-these are stem cell-like behaviors.
• CD133 high CD71 low cells give rise to insulin-expressing cells in vitro in the presence of a defined extracellular matrix protein. These insulin-expressing cells express glucagon simultaneously and secrete insulin in response to glucose.
• Transplantation into diabetic mice of colonies grown in the presence of a defined extracellular matrix protein leads to the formation of duct, acinar, and endocrine (e.g., insulin + glucagon − and insulin − glucagon + ) cells.     Upper panel: A schematic representation of a re-plating experiment. Sorted CD133 high CD71 low cells were plated into Matrigel-containing culture. Three weeks later, a primary Ring colony was hand-picked, dissociated into single cell suspension, and re-plated into Matrigel-containing culture for a total of 6 passages. Lower panel: Serial re-plating of an individually hand-picked primary Ring colony (n=8) generated secondary and subsequent Ring colonies in Matrigel-containing culture.