Conserved markers of fetal pancreatic epithelium permit prospective isolation of islet progenitor cells by FACS

Abstract

Prospective isolation and characterization of progenitor cells is a paradigmatic strategy for studies of organ development. However, extraction of viable cells, fractionation of lineages, and in vitro analysis of progenitors from the fetal pancreas in experimental organisms like mice has proved challenging and has not yet been reported for human fetal pancreas. Here, we report isolation of pancreatic islet progenitor cells from fetal mice by FACS. Monoclonal antibodies that recognize cell-surface proteins on candidate stem cells in brain, skin, and other organs enabled separation of major pancreatic cell lineages and isolation of native pancreatic cells expressing neurogenin 3, an established marker of islet progenitors. New in vitro cell culture methods permitted isolated mouse islet progenitors to develop into hormone-expressing endocrine cells. Insulin-producing cells derived in vitro required or expressed factors that regulate fetal β cell differentiation; thus, the genetic programs normally controlling in vivo mouse islet development are similarly required in our system. Moreover, antibodies that recognize conserved orthologous cell-surface epitopes in human fetal pancreas allowed FACS-based enrichment of candidate islet progenitor cells expressing neurogenin 3. Our studies reveal previously undescribed strategies for prospective purification and analysis of pancreatic endocrine progenitor cells that should accelerate studies of islet development and replacement.

Evidence that replacement islets for transplant-based diabetes treatment might be derived from multipotent cells by recapitulating steps that direct embryonic islet development has increased interest in elucidating the cellular mechanisms of islet development (1). Three principal cell types, endocrine islets, exocrine acini, and ducts, form in the developing pancreas, and our understanding of the basis for pancreas ontogeny in the embryo has recently grown (reviewed in refs. 2 and 3). By contrast, little is known regarding the molecular basis of islet formation in the postnatal pancreas, which only recently has been studied (4–6). In fetal mice, pancreatic islet cells, including insulin-producing β cells, derive from pancreatic endocrine progenitors expressing the transcription factor neurogenin 3 (NGN3; also known as Atoh5 and Relax; refs. 7–10). NGN3 expression in developing mouse pancreatic endocrine progenitor cells peaks at mid-gestation on embryonic day (E)15, then extinguishes before terminal differentiation into hormone⁺ cells (7, 8, 11–13). Thus, NGN3-expressing cells (abbreviated NGN3⁺) represent lineage-committed pancreatic islet progenitors.

To purify NGN3⁺ progenitor cells from fetal mice, prior studies adopted a transgenic strategy that resulted in expression of a fluorescent protein marker from ngn3 gene cis-regulatory elements (14, 15). Fluorescence-activated cell sorting (FACS) of dispersed pancreatic cells from these transgenic mice permitted enrichment of fluorescent protein-labeled NGN3⁺ cells, but perdurance of the fluorescent protein marker produced contaminating fluorescent protein-labeled hormone⁺ cells lacking NGN3. Thus, transgenic-marking methods did not separate NGN3⁺ progenitors from differentiated insulin-expressing and glucagon-expressing cells (refs. 14 and 15; data not shown). Moreover, a transgenic strategy to isolate specific pancreatic progenitor cell subsets may not be adaptable to all species, particularly humans. A successful precedent of stem cell isolation from bone marrow and other organs encouraged us to systematically survey monoclonal antibodies that recognize surface epitopes on a variety of stem cell populations (16–19). We identified two markers, CD133 and CD49f, that permit FACS-based isolation of pancreatic NGN3⁺ cells from fetal mice and human. Here, we report the use of FACS and in vitro cell culture methods to isolate and analyze these candidate islet progenitor cell populations.

Results

Identification of Cell-Surface Markers Expressed by Mouse Pancreatic Islet Progenitor Cells.

Stem and progenitor cells in distinct organs express a shared set of cell-surface markers (18–21). Thus, we systematically tested a panel of >30 stem cell markers to identify those expressed by NGN3⁺ islet progenitor cells in the fetal mouse pancreas [supporting information (SI) Table 1]. CD133 (also called prominin-1) is a transmembrane protein of unknown function previously reported to be a surface marker for hematopoietic progenitor and neural stem cells (18), and using immunohistology, we found CD133 was expressed by NGN3⁺ cells. CD133 appeared to be localized to the apical membrane of pancreatic ductal epithelial cells (Fig. 1A and B) and nascent acinar cells (data not shown). Colocalization of CD133 and NGN3 in a single cell was confirmed by iterative optical sectioning through multiple focal planes (“Z stack”; SI Movie 1). By contrast, insulin⁺ cells did not detectably express CD133 (Fig. 1C), suggesting that CD133 might be a cell-surface marker useful for separating NGN3⁺ cells from hormone⁺ cells by flow cytometry. To test this possibility, we used monoclonal antibodies that recognize the CD133 antigen to separate CD133⁺ and CD133⁻ cells from the pancreas of WT embryonic mice at E15.5 (Fig. 1D). Immunostaining of sorted cells revealed that the CD133⁺ fraction contained NGN3⁺ cells (Fig. 1F) and the exocrine cells that express carboxypeptidaseA (CarbA⁺). We calculated a modest total enrichment of NGN3⁺ cells in this CD133⁺ fraction of 1.7-fold (SI Table 2). However, the CD133⁺ fraction excluded insulin⁺ and glucagon⁺ cells that did not express NGN3, consistent with our histological data (Fig. 1 A–F). Thus, CD133-based sorting permitted complete separation of hormone⁻ NGN3⁺ mouse pancreatic cells from differentiated insulin⁺ or glucagon⁺ islet cells lacking NGN3. FACS-based analysis of CD133 expression in fetal pancreata from earlier stages (E12.5, E13.5, and E14.5) revealed a dynamic temporal expression pattern of this marker during pancreatic development (SI Fig. 6).

Fig. 1.

FACS-based isolation of NGN3⁺ endocrine progenitor cells from E15.5 mouse embryonic pancreas. (A–C) Immunohistology of E15.5 mouse pancreas. (A and B) NGN3⁺ cells are CD133⁺ (arrows). Images were taken by optical sectioning through multiple focal planes; representative images are shown. (C) Insulin⁺ cells are CD133^–. (D) FACS plot of sorted cells after exposure to anti-CD133 antibody. Percentages of CD133^– and CD133⁺ cells are indicated. (E) Subsets of isolated CD133^– cells express insulin and glucagon. Nuclei were visualized by DAPI staining (blue). (F) Subsets of CD133⁺ cells express NGN3. (G and H) Immunohistology showing expression of CD49f in E15.5 mouse pancreatic cells, including NGN3⁺ and insulin⁺ cells. (G) Magnified image of H. (I) FACS plot of cells sorted after exposure to anti-CD49f antibody. Percentages of CD49f^high, CD49f^low, and CD49f^– cells are indicated. (J) Subsets of isolated CD49f^−/low cells express insulin and glucagon. Cells expressing CarbA were not detected. (K) Subsets of isolated CD49f^high cells express CarbA (red). (L) FACS plot of cells sorted after exposure to antibodies recognizing CD133 and CD49f. Four fractions (labeled I–IV) are indicated. (M) CD49f^low CD133⁻ cells (fraction III) exclusively contain insulin⁺ or glucagon⁺ cells. (N) CD49f^low CD133⁺ cells (fraction II) exclusively contain NGN3⁺ cells. (E, J, and M) Insulin (white arrowheads); glucagon (green, arrows); (F and N) NGN3 (red, arrowheads).

To achieve further enrichment of CD133⁺ NGN3-expressing cells and to eliminate accompanying CD133⁺ CarbA⁺ cells, we examined additional markers of NGN3⁺ cells. We found that CD49f (also called α 6 integrin), a receptor subunit for laminin that is expressed by organ-specific and embryonic stem cells (19–22), was coexpressed by NGN3⁺ cells and other pancreatic epithelial cells. Immunohistology revealed three qualitatively distinct levels of CD49f expression in mouse E15.5 pancreas. CD49f expression appeared most intense in exocrine acinar cells expressing products like CarbA (data not shown). In NGN3⁺ cells and hormone⁺ cells including insulin-expressing cells, CD49f expression at E15.5 was clearly detectable but less intense than in CarbA⁺ cells (Fig. 1G and H). In remaining cells, including mesenchyme, CD49f expression was not detected (data not shown). In contrast to CD133, CD49f expression in epithelial cells surrounding ducts appeared to localize to the basolateral cell membrane.

Using antibodies that recognized CD49f, we separated E15.5 pancreatic cells by FACS into three different fractions: CD49f “high,” “low,” and “negative” (Fig. 1I). Consistent with our studies of fixed sectioned tissues, we found that the sorted CD49f^high fraction contained CarbA⁺ exocrine cells; >90% of CD49f^high cells obtained by cytospin were CarbA⁺ by immunostaining (Fig. 1K). The CD49f^low fraction contained endocrine cells including insulin⁺, glucagon⁺, and NGN3⁺ cells (Fig. 1J). By contrast, we found no endocrine or acinar cells in the CD49f⁻ fraction. In the CD49f^low fraction, we calculated that NGN3⁺ cells were enriched by 4-fold (SI Table 2). By combining antibodies that recognize CD133 and CD49f, we fractionated E15.5 pancreatic cells into four distinct viable cell populations (Fig. 1L and SI Table 2). Analysis by immunostaining and RT-PCR revealed that the CD49f^high CD133⁺ cell population (“fraction I,” 50% of recovered cells) was composed mainly of differentiated exocrine cells that express CarbA (Fig. 1K and data not shown). The CD49f^low CD133⁻ fraction (“fraction III,” 10% of recovered cells) included hormone⁺ cells expressing endocrine products like insulin and glucagon (Fig. 1M and 2G). By contrast, the CD49f^low CD133⁺ fraction (called “fraction II,” 13% of recovered cells) contained NGN3⁺ cells but not hormone⁺ cells (Fig. 1N and 2G). Approximately 8% of fraction II cells produced immunostainable NGN3 (SI Table 2). In the CD49f⁻ CD133⁻ fraction (“fraction IV”), which comprised 25% of recovered cells, we did not detect cells expressing NGN3, CarbA, or islet hormones (Fig. 2G and data not shown). FACS-based analysis of CD49f expression from earlier stages (E12.5, E13.5, and E14.5) revealed a dynamic expression pattern of this marker during pancreatic development (SI Fig. 6).

Fig. 2.

In vitro differentiation of NGN3⁺ cells. (A) Schematic of the coculture strategy for assessing developmental potential of sorted fraction II single cells. (B–D) Expression of indicated β cell markers after coculture on PA6 feeders of fraction II cells from WT mice. (E and F) Coexpression of insulin and EGFP after coculture on PA6 feeders of fraction II cells from MIP-EGFP mice. (G) RT-PCR analysis of sorted cells before in vitro culture. (H) RT-PCR analysis of fraction II cells after coculture with MEFs for 2 days.

This work shows the complete isolation of islet progenitor cells away from the differentiated endocrine and exocrine lineages in the embryonic pancreas. Prior studies have demonstrated similarities between developing islets and neurons (11, 23–25). Thus, we postulated that additional antibodies, particularly those recognizing epitopes in neural stem cells, might permit purification of NGN3⁺ cells to greater homogeneity. For example, we found that NGN3⁺ cells produced CD24, a glycophosphatidylinositol-anchored cell surface protein also found on brain-derived neural cells (18, 26–28). NGN3⁺ cells were enriched an additional 1.4-fold after FACS sorting of fraction II with antibodies specific for CD24 (SI Fig. 7). Because the yield of NGN3⁺ cells after use of anti-CD24 was relatively low, in the remainder of experiments described here, we focused on the use of anti-CD133 and anti-CD49f antibodies.

Isolation and Enrichment of NGN3⁺ Cells from Transgenic ngn3-EGFP Mice.

Next, we tested whether FACS enrichment for CD133 could separate hormone-expressing cells from NGN3⁺ cells derived from ngn3-EGFP transgenic mice (14). The ngn3-EGFP transgene in these mice contains cis-regulatory elements from the mouse ngn3 promoter adjacent to a cDNA encoding EGFP. Consistent with our results with WT mice (Fig. 1), the fraction of EGFP⁺ CD133⁺ cells isolated from ngn3-EGFP mice lacked detectable glucagon- or insulin-expressing cells (SI Fig. 8; data not shown). Eighty percent of EGFP⁺ CD133⁺ cells expressed immunostainable NGN3 in the absence of hormone⁺ islet cells, which represents a 110-fold enrichment and a high purification of NGN3⁺ cells without contaminating hormone⁺ cells (SI Fig. 8 and SI Table 2). We obtained similar results with ngn3-YFP mice (SI Fig. 8; ref. 15). Thus, a combination of transgenic and antibody-based strategies described here enabled purification of NGN3⁺ cells to near homogeneity.

Cultured NGN3⁺ Cells Differentiate into Insulin-Producing Cells with Features of β Cells.

To assess the developmental potential of NGN3⁺ cells isolated by our methods, we identified conditions that permitted in vitro differentiation of NGN3⁺ fraction II progenitor cells isolated from WT fetal pancreas. We surveyed multiple culture conditions and found that glucagon- and insulin-expressing cells developed when fraction II NGN3⁺ progenitors at low density were cocultured either with mitomycin C-treated mouse embryonic fibroblasts (MEFs) or with PA6 mouse stromal cells, a feeder cell layer previously shown to promote neural differentiation by embryonic stem cells (Fig. 2A; refs. 29 and 30). Input pancreatic cells were readily identified by the size of their nuclei (Fig. 2 B–E) and subsequent immunostaining. Within one day after culture of NGN3⁺ progenitors from fraction II on MEF or PA6 feeder cells, we observed the appearance of insulin⁺ cells (Fig. 2B and data not shown). The number of cultured insulin⁺ cells peaked after 4 days and remained stable through the fifth day (Fig. 2F and data not shown). The number of insulin⁺ cells was 10 times greater than the number of glucagon⁺ cells in these cultures, approximating the ratio found later in the adult mouse pancreas. We did not observe significant changes in cell survival or differentiation in these cocultures when we varied the density of input pancreatic cells, or by expressing Bcl-2, an anti-apoptotic factor that has improved in vitro survival of mouse hematopoietic stem cells (ref. 31; data not shown). More than 90% of the insulin⁺ or glucagon⁺ cells in fraction II cultures were single cells, and the remainder were found as clusters of two or three cells. We did not observe cell clusters containing mixtures of insulin⁺ and glucagon⁺ cells, consistent with the possibility that individual NGN3⁺ cells are unipotent, and not highly proliferative (8, 32). Confirmation of this hypothesis will require additional studies. During in vitro differentiation of these hormone-expressing cells, expression of NGN3 was extinguished (Fig. 2G-H), similar to the pattern of transient ngn3 expression by the fetal mouse pancreas in vivo (7, 8, 11–13).

To address the concern that insulin immunostaining of input pancreatic cells in our cultures might reflect uptake of culture media-derived bovine insulin (33), we performed experiments to verify de novo insulin synthesis. Insulin⁺ cells were immunostained by antibodies specific for insulin C-peptide, an internal portion of the preproinsulin primary translation product. Ninety-six percent of insulin⁺ cells coexpressed C-peptide, indicating de novo insulin synthesis by the majority of insulin⁺ cells (Fig. 2B). Consistent with these findings, insulin-1 mRNA was not produced by input CD49^low CD133⁺ pancreatic cells in fraction II (Fig. 2G) but was clearly detected after 2 days culture of these cells (Fig. 2H). In addition, transcripts for the other principal islet hormones, including glucagon, somatostatin, and pancreatic polypeptide (PP), were detected in these cocultures (Fig. 2H), but not before culture (Fig. 2G). To verify further our finding of de novo insulin expression and to confirm that sorted pancreatic cells (and not feeder cells) were the source of insulin, we isolated NGN3⁺ cells from embryonic transgenic mice containing a MIP-EGFP transgene (34). This transgene contains cis-regulatory elements from the mouse insulin promoter (MIP) adjacent to a cDNA encoding EGFP. In vivo EGFP expression was observed in pancreatic insulin⁺ cells (34), providing a useful marker to confirm insulin gene expression. We found that all insulin C-peptide⁺ cells derived from MIP-EGFP animals in our cultures coexpressed EGFP (Fig. 2E; data not shown). Collectively, these results demonstrate that individual NGN3⁺ cells can develop into cells that transcribe and translate insulin de novo.

Isolated adult pancreatic progenitor cells exhibit unexpected developmental plasticity during in vitro culture, producing differentiated hepatic, pancreatic, and neural cell types (5, 6). One way to assess the physiologic relevance of in vitro differentiation by pancreatic progenitor cell populations is to test whether the genetic programs controlling in vivo islet development are also required during in vitro differentiation. All islet cells, including insulin⁺ β cells and glucagon⁺ α cells, fail to form in mice lacking NGN3 (7); thus, we isolated CD133⁺ CD49f^low cells from ngn3 mutant mice to determine whether in vitro differentiation of insulin⁺ and glucagon⁺ cells required ngn3. The number of CD133⁺ CD49f^low cells (fraction II) from ngn3^–/– mice at E15.5 was reduced compared with WT controls (Fig. 3A) and did not contain NGN3-expressing cells detectable by immunostaining (data not shown). The number of CD133^– CD49f^low cells (fraction III) also was reduced. This population of cells contains insulin⁺ and glucagon⁺ cells when isolated from WT mice (Figs. 1M and 2G), but insulin⁺ and glucagon⁺ cells were undectectable in cytospin samples of fraction III from ngn3^–/– mice (data not shown), consistent with prior studies of ngn3^–/– mice (7, 9). Cocultures of CD133⁺ CD49f^low cells (fraction II) from ngn3^–/– mice with PA6 cells yielded a population of cells expressing the epithelial marker E-cadherin (Fig. 3B). Expression of insulin C-peptide or glucagon, however, was never detected by immunostaining (Fig. 3B) or by RT-PCR in these cultures (data not shown). These results confirmed the specificity of surface markers and physiological relevance of our culture system. Moreover, these results show that yields from our method are sufficient to permit analysis of a single fetal pancreas.

Fig. 3.

FACS and in vitro culture of E15.5 ngn3^–/– pancreatic cells. (A) FACS analysis of pancreatic cells from control (WT) and ngn3^–/– embryos revealed reduced cell numbers in fractions II and III (circled) from ngn3^–/– pancreas. (B) After 4 days of coculture, ngn3^–/– fraction II cells expressed detectable E-cadherin (green), an epithelial marker, but failed to produce detectable C-peptide (red).

During native β cell development at late gestational stages, NGN3 is required for subsequent expression of gene products that regulate or define mature β cells, like insulin C-peptide, MafA, and Nkx6.1 (2). To assess further whether insulin⁺ cells formed during in vitro culture were similar to pancreatic β cells, we performed immunohistology to detect expression of these established regulators of β cell development and function. Insulin⁺ cells derived from cocultures of CD133⁺ CD49f^low cells with PA6 cells also expressed Nkx6.1 and MafA (Fig. 2 C and D). Thus, established NGN3-dependent targets are expressed during in vitro differentiation of isolated endocrine progenitors. This is consistent with the genetic requirement for NGN3 during in vitro differentiation. To our knowledge, similar in vitro methods that permit differentiation of isolated islet progenitors from embryonic pancreas have not been reported.

Isolation of a CD133⁺ CD49f^high NGN3⁻ Pancreatic Compartment That Generates NGN3⁺ Cells.

Prior studies suggest that NGN3⁺ islet progenitor cells derive from a population of pancreatic cells that lack NGN3 (7–13). Thus, we next investigated whether our fractionation strategy also could be used to isolate a cellular precursor for ngn3-expressing islet progenitors. Because CD133⁺ cells expressing CD49f are enriched for populations of multipotent brainstem cells (18, 19) and stem cells in mammary gland are included in a CD49f^high population (35), we postulated that the population of CD133⁺ CD49f^high cells (fraction I) obtained from the fetal mouse pancreas might include precursors for NGN3⁺ cells. We prepared fraction I cells from ngn3-EGFP mice (14). Isolation of EGFP^– cells was confirmed by double sorting with fluorescence gating, visual inspection, and RT-PCR analysis (data not shown). To assess their potential to generate NGN3-expressing cells, coculture of these cells with MEFs or PA6 feeders gave inconsistent results. However, systematic testing led to the discovery that AC6 cells, a mouse bone marrow-derived stromal line used for in vitro culture of lineage-committed blood precursors (36), supported differentiation of ngn3-expressing cells from fraction I (Fig. 4). Like ngn3 expression in the fetal pancreas (7, 8, 11–13), ngn3-EGFP expression in vitro was transient (Fig. 4 B and C). After 2 days, the number of ngn3-EGFP⁺ cells peaked and the cells transcribed endogenous ngn3 mRNA (Fig. 4D); thereafter, the number of EGFP⁺ cells was decreased. Thus, the EGFP^– CD133⁺ CD49f^high NGN3^– cell fraction isolated here may contain a population of pancreatic progenitor cells capable of differentiating into ngn3-expressing cells.

Fig. 4.

Identification of a precursor cell population for NGN3⁺ cells. (A) Schematic of the coculture strategy for assessing developmental potential of sorted ngn3-EGFP^– fraction I single cells. (B) ngn3-EGFP⁺ cells (green) after 2 days of culture on an AC6 feeder layer. (C) Quantification of ngn3-EGFP⁺ cells during culture (n = 4). (D) RT-PCR analysis of ngn3 expression by fraction I cells before and after culture. Values represent mean ± SD.

FACS Studies Reveal That Human Fetal NGN3⁺ Pancreatic Cells Express CD133 and CD49f.

We postulated that flow cytometry-based methods stemming from studies of mouse pancreas might be used to isolate orthologous ngn3-expressing cells from human fetal pancreas. To test this possibility, we first assessed expression of ngn3, CD133 and CD49f. Quantitative RT-PCR studies showed that the relative abundance of ngn3 mRNA peaked at ≈14–15 weeks gestation (data not shown); thus, we focused subsequent experiments on pancreatic samples from this stage. Immunohistology of human pancreas from 14 weeks revealed that NGN3⁺ cells coexpressed both CD133 and CD49f (Fig. 5A and B and SI Movie 2). Like in fetal mouse pancreas, CD133 expression also appeared to localize to the apical portion of human pancreatic epithelial cells, whereas CD49f was found more prominently at the basolateral membrane (Fig. 5 A and B and SI Movie 2). With respect to these markers, we noted two qualitatively distinct populations of NGN3⁺ cells in human fetal pancreas that were not detected in our mouse studies. Expression of CD133 and CD49f appeared more prominent in NGN3⁺ cells in ductal epithelia that surrounded a lumen compared with lower intensity expression of these markers by NGN3⁺ cells not clearly associated with ducts (Fig. 5 A and B and SI Movie 2). This molecular heterogeneity also was detected by using FACS and PCR methods (see below). Thus, in addition to several conserved features, we also noted qualitative heterogeneity in human NGN3⁺ cells not detected in our studies of mouse pancreatic NGN3⁺ cells.

Fig. 5.

Immunohistology and FACS of human fetal pancreas. (A and B) Immunohistology of human fetal pancreas at 14 weeks. NGN3⁺ cells express both CD49f (arrowheads) (A) and CD133 (arrows) (B). (C–E) FACS analysis of human fetal pancreas at 15 weeks. Dissociated cells were stained with antibodies recognizing CD133 (C and E) and CD49f (D and E). Four populations are outlined. (F) Quantitative RT-PCR analysis of ngn3 mRNA levels in cells after sorting of 15-week human fetal pancreas. ngn3 expression level was normalized to β-actin and shown in arbitrary unit. The average of two samples is shown. The pie chart shows the percentage of recovered cells in each indicated fraction.

Using monoclonal antibodies that recognize human CD133 and CD49f, we next tested whether FACS could be used to isolate human ngn3-expressing cells prospectively. Similar to our findings with fetal mouse pancreas, analysis of dispersed pancreatic cells from fetal human pancreata at 14–15 weeks allowed separation of two pancreatic cell populations based on CD133 levels, CD133^high and CD133^low/– (Fig. 5C). Moreover, FACS analysis with anti-CD49f antibodies revealed three populations of CD49f expression, CD49f^high, CD49^low, and CD49f^– (Fig. 5D). Combining antibodies that recognize CD133 and CD49f, FACS permitted separation of dissociated human fetal pancreas cells into four distinct populations (Fig. 5E) called fraction A (CD133^high CD49f^high; 8% of recovered cells), fraction B (CD133^low/– CD49f^low; 6% of recovered cells), fraction C (CD133^low/– CD49f^high; 9% of recovered cells), and fraction D (CD133^low/– CD49f^–; 77% of recovered cells). In good agreement with our immunohistology, analysis of gene expression by real-time RT-PCR revealed that ngn3 mRNA levels were enriched in fractions A and B and nearly undetectable in fractions C and D (Fig. 5F). mRNA levels for insulin, however, were detectable in fractions A, B, or C after sorting; thus, we have not yet separated human ngn3-expressing cells from insulin-expressing endocrine cells (data not shown). Collectively, these data show that sorting based on CD133 and CD49f permitted enrichment of human fetal pancreas cells expressing ngn3.

Discussion

Isolation and characterization of cellular progenitors is a powerful general strategy that has accelerated growth in our understanding and uses of organs like bone marrow (16–19). Strategies to generate islet replacement cells for diabetes therapy focus on recapitulating embryonic islet development, but islet biology has been hindered by an inability to isolate islet progenitors from fetal pancreas. Prior attempts to purify fetal islet progenitors were based on transgenic marking strategies and yielded heterogeneous cell populations with incomplete separation of islet progenitors from differentiated hormone-producing cells (14, 15, 26). Moreover, in human tissue samples, a lack of surface markers precluded prospective isolation of progenitor cells. Here, we successfully separated fetal islet progenitor cells from differentiated islet cells without genetic modification.

In addition to revealing unique strategies for isolating islet progenitors, our experiments reveal conserved molecular markers useful for developmental studies of progenitor cells from the pancreas. Here, we show that CD133 is a marker expressed on the apical cell surface of fetal ductal epithelial cells (including NGN3⁺ cells) and nascent acini, whereas CD49f appears to be localized to the basolateral membrane in these cells. These observations provide molecular evidence of polarity in mouse pancreatic NGN3⁺ cells. Although the functions of this molecular polarization require further study, we note that prior reports have revealed similar apical localization of CD133 in periventricular brain cells thought to be neural progenitors (37). Thus, our work provides additional evidence of similarities between pancreatic and neural developmental programs (11, 23, 24).

Recent studies reported isolation of pancreatic progenitors from adult mice, but these putative progenitor cells could differentiate into both pancreatic cells and neural cells, raising concerns regarding the methods used to assess the developmental potential of these cells (6). We used cells from mice lacking NGN3 to show that known in vivo genetic programs directing islet development also are required for development of islet cells in our in vitro system. Thus, although FACS isolation of islet progenitor cells will accelerate molecular and cellular biological analysis of those cells, conclusions regarding the properties of isolated pancreatic cells from in vitro studies, in turn, benefit from validation with appropriate in vivo observations.

In studies of mice, targeted gene inactivation and lineage-tracing studies have established an essential role for NGN3 in mouse islet development (7, 8). By contrast, it is not yet known whether NGN3⁺ pancreatic cells in humans are islet progenitors (38). Based on the challenges of investigations with limited amounts of human fetal pancreatic tissue, we have not yet purified human NGN3⁺ cells sufficiently to test whether isolated cells differentiate into hormone-producing cells with islet phenotypes. If this limitation can be overcome, we speculate that our methods should provide opportunities to elucidate the roles of NGN3 in human islet cell development.

Our results also should accelerate identification of factors that regulate the growth, developmental potential, maturation, and survival of pancreatic islet progenitor cells. Because our cell fractionation methods are built on antibody-based FACS isolation, they could serve as the basis for general strategies to purify and study islet progenitors from sources other than fetal pancreas, like adult pancreas, or differentiating embryonic stem cells. Improved yields will permit future studies to test whether transplantation of isolated NGN3⁺ cells or their progeny can impact physiologic outcomes in diabetic animal models. Thus, knowledge from these studies may be useful for accelerating discovery of new cell-replacement strategies for diabetes mellitus.

Materials and Methods

Mouse Pancreatic Cell Dispersion and Flow Cytometry.

The mouse dorsal pancreas was manually dissected from E15.5 embryos. Tissue was washed in Dulbecco's PBS, then exposed to 0.05% trypsin/0.53 mM EDTA (Invitrogen, Carlsbad, CA) at 37°C for 5 min and triturated. Trypsin was neutralized with “FACS buffer” containing PBS, 10 mM EGTA, and 2% FBS. Cell viability at this stage exceeded 90%, as assessed by Trypan Blue exclusion. Dissociation into single cells was confirmed by light microscopy. Cells then were treated for 15 min in a blocking solution composed of FACS buffer containing 300 ng/ml rat IgG (Jackson ImmunoResearch, West Grove, PA). We used the following primary antibodies: biotin anti-CD133 (13A4, 1:100; eBioscience, San Diego, CA), PE anti-CD24 (M1/69, 1:100; eBioscience), FITC anti-CD49f (GoH3, 1:50; BD Biosciences, San Diego, CA), and PE anti-CD49f (GoH3, 1:50; BD Biosciences). Streptavidin-APC (1:100; eBioscience) was used to visualize biotinylated antibodies. PE-Cy5 anti-TER-119 (1:100; eBioscience), PE-Cy5 anti-CD45 (1:100; eBioscience), and propidium iodide (PI) were used to remove erythrocytes, leukocytes, and dead cells, respectively. Cell sorting was performed in the Stanford FACS facility by using a modified triple laser FACS instrument and Summit software (DAKO, Fort Collins, CO) or DIVA software (BD Biosciences, San Jose, CA). Drops containing more than one cell were removed by using pulse width. Cells were sorted twice serially to reduce contamination from other fractions. Sorted cells were fixed on Polysine glass slides (Erie Scientific, Portsmouth, NH) and stained or cultured as described below. FACS data were analyzed by using FlowJo software (Tree Star, San Carlos, CA).

Supporting Information.

Additional details can be found in SI Materials and Methods.

Abbreviations

En

embryonic day n

MEF

mouse embryonic fibroblast

NGN3

neurogenin 3

NGN3⁺

NGN3-expressing cell.