Exocrine-to-endocrine differentiation is detectable only prior to birth in the uninjured mouse pancreas

Background Histological evidence suggests that insulin-producing beta (β)-cells arise in utero from duct-like structures of the fetal exocrine pancreas, and genetic lineage tracing studies indicate that they are maintained in the adult by self-renewal. These studies have not addressed the origin of the new β-cells that arise in large numbers shortly after birth, and contradictory lineage tracing results have been published regarding the differentiation potential of duct cells in this period. We established an independent approach to address this question directly. Results We generated mice in which duct and acinar cells, comprising the exocrine pancreas, can be genetically marked by virtue of their expressing the mucin gene Muc1. Using these mice, we performed time-specific lineage tracing to determine if these cells undergo endocrine transdifferentiation in vivo. We find that Muc1+ cells do give rise to β-cells and other islet cells in utero, providing formal proof that mature islets arise from embryonic duct structures. From birth onwards, Muc1 lineage-labeled cells are confined to the exocrine compartment, with no detectable contribution to islet cells. Conclusions These results argue against a significant contribution by exocrine transdifferentiation to the normal postnatal expansion and maintenance of β-cell mass. Exocrine transdifferentiation has been proposed to occur during injury and regeneration, and our experimental model is suited to test this hypothesis in vivo.


Background
The origin of pancreatic islet cells has been the subject of study and controversy since before the discovery of insulin [1][2][3]. Histological and ultrastructural studies suggested that islets arise from exocrine ducts during embryogenesis, but whether such exocrine-endocrine conversion continued after birth remained a matter of controversy [4]. Nucleotide analogues have recently been used to identify and trace the fate of proliferating cells, but these studies have been interpreted both for and against the hypothesis of new islet cell differentiation, or neogenesis, in adulthood [5,6]. Similarly, cell culture studies have provided evidence for and against neogenesis, but extrapolating these findings in vivo remains a challenge [2]. Nonetheless, determining whether β-cell neogenesis occurs in vivo will inform efforts to replenish β-cells lost in diabetes: if neogenesis can occur in mice, it might be possible in the human organ as well. Evidence against neogenesis would encourage more aggressive efforts elsewhere, such as the derivation of β-cells from human embryonic stem cells.
Genetic lineage tracing techniques have transformed our understanding of pancreas developmental biology, providing insights either unavailable from or contradictory to prior studies of fixed tissue [2]. For example, using the Cre-loxP system to monitor the fate of cells expressing the acinar protein Carboxypeptidase A1 (Cpa1) has revealed that these cells comprise a selfrenewing, multipotent progenitor pool during midembryogenesis [7]. At later stages, Cpa1 + cells become restricted to the acinar compartment, and lineage tracing of adult acinar cells indicates that they do not contribute to β-cells [7,8]. Other lineage tracing studies similarly cast doubt on the neogenesis model; for example, it is now appreciated that all islet cell types arise from Neurog3 + precursor cells [9], yet Neurog3 + cells are usually not detected after birth [10][11][12]. Similarly, a key study of adult β-cell expansion and self-renewal suggests that these processes reflect division of existing β-cells rather than neogenesis [13], a conclusion reinforced by subsequent independent findings [5,14].
Nonetheless, without a Cre line capable of marking duct cells, these studies could not exclude a minor ductal contribution to β-cells. Furthermore, they have not addressed the rapid expansion of β-cell mass that occurs shortly after birth [15], which may derive in part from neogenesis [6]. The ideal tool to address these questions would be a mouse line in which ducts can be inducibly marked at a specific time point, and their ability to contribute to β-cells determined at later stages: a cellular "pulse-chase" experiment [13]. Such mice could also clarify the origin of new β-cells formed in regeneration models, such as duct ligation and partial pancreatectomy, which may involve neogenesis [16,17] Four transgenic mouse lines have recently been described in studies aimed at addressing this issue. Inada et al. [18] generated mice in which either Cre or its tamoxifen (TM)-inducible derivative CreER™ is driven by the promoter of Carbonic anhydrase II (CAII), a marker of adult duct cells. This study concluded that ducts continue to give rise to islets after birth, and that neogenesis is dramatically increased after duct ligation [18]. Means et al. [19] used a knock-in approach to target a similar tamoxifen-inducible CreERT molecule to the duct-specific cytokeratin-19 (K19) gene. K19 CreERT labels many duct cells in neonatal and adult mice, as well as a small fraction of islet cells due to "leaky" recombinase expression in islets themselves. When K19 CreERT was activated by TM treatment at birth, islet labeling at one week of age was no greater than expected from this background activity, suggesting that new islet cells generated in the interim did not derive from the more robustly-labeled duct population [19]. Finally, Solar et al. [20] generated a BAC transgenic in which TM-dependent CreERT2 is targeted to the first exon of Hnf1b, a transcription factor expressed selectively in embryonic and adult duct cells. This study indicated that Hnf1b + duct cells give rise to β-cells prior to birth but not thereafter, even in the context of injury models such as duct ligation [20]. The apparent contradiction between these studies, in particular those using CAII-and Hnf1b-driven Cre transgenes, suggests that additional markers should be sought out and exploited for their capacity to label duct cells, thereby providing independent evidence for or against postnatal neogenesis [21].
The gene mucin1, transmembrane (Muc1) is expressed throughout the ductal network in both the embryonic and adult pancreas, and excluded from islets [22,23].
We have generated mice in which the Muc1 gene is tagged with an IRES-CreERT2 cassette, permitting the inducible labeling of Muc1 + cells. We find that this line also marks embryonic and postnatal acinar cells, reflecting endogenous Muc1 expression, but is completely excluded from islets in short-term "pulse" experiments. In the embryo, Muc1 + cells give rise to endocrine cells as well as their Neurog3 + precursors, formally confirming that islets originate from fetal ducts. Following birth, however, we find that the Muc1 + lineage completely fails to contribute to new islet αor β-cells, indicating that both the expansion and homeostasis of these cell types occurs independent of contribution from exocrine ducts or acini. These results argue against a major role for neogenesis in the normal postnatal pancreas, and set the groundwork for studies of potential neogenesis during regeneration.

Targeting CreERT2 to the Muc1 locus
To determine if the Muc1 locus could be exploited to mark ducts, we evaluated its expression in the embryonic and adult pancreas. We find that Muc1 is widely expressed at E11.5, when most pancreatic cells are still undifferentiated progenitors ( Figure 1A-C). As embryogenesis proceeds, Muc1 expression persists in the branching epithelial network ( Figure 1D-F), and it is expressed by all ductal cells of the mature organ ( Figure  1G-I). Another duct marker, cytokeratin-19 (CK19), is undetectable prior to E17.5 ( Figure 2). Interestingly, Muc1 expression appears to decrease as duct caliber increases, opposite to CK19 ( Figure 2G-I). Nonetheless, we have not observed, at any stage, a Muc1-negative cell incorporated within a duct structure.
Embryonic islet precursors express the transcription factor Neurogenin3 (Neurog3), and are considered to arise from primitive duct-like cells [4,[9][10][11]. We find that Neurog3 expression localizes within or in close proximity to Muc1 + cells at E15.5, consistent with a ductal origin for islet cells ( Figure 3A). While Neurog3 expression is not detectable in adults, Muc1 remains expressed throughout the ductal network of the pancreas, including large ducts as well as fine terminal branches within acini ( Figure 3B). Importantly, Muc1 expression is excluded from islet cells during embryogenesis as well as in the mature organ (Figs. 1, 3C).
Altogether, Muc1 appears to satisfy our requirements for a Cre driver line to study islet neogenesis: it is expressed in all duct cells, embryonic and adult, and excluded from differentiated islets.
As detailed in the Methods, we generated a Muc1 IRES-CreERT2 allele (henceforth, Muc1 IC2 ) by gene targeting, introducing an IRES-CreERT2 cassette after the endogenous stop codon of the Muc1 locus ( Figure 3D-E).
The CreERT2 protein is a tamoxifen (TM)-dependent recombinase that provides temporal control of Cre activity [24], which should allow us to investigate the differentiation potential of embryonic and postnatal duct cells. Muc1 IC2 heterozygous and homozygous mice are viable and fertile (data not shown).

Muc1 IC2 marks exocrine cells in the adult pancreas
To determine whether Muc1 IC2 marks ducts, we performed short-term labeling experiments with the Credependent EYFP reporter strain, Rosa26 EYFP [25]. We induced recombination by treating adult (6-8 week-old, n = 3) Muc1 IC2/+ ;Rosa26EYFP /+ mice with a single dose of 10 mg tamoxifen. After a two-day "chase" period, we found the EYFP lineage label not only in cytokeratin-19 + duct cells (~6%), as expected, but also within amylase + acinar cells ( Figure 4A-E,~12%), indicating that this Cre line is active in both duct and acinar cells.
(Note that in this and other experiments, we find that Rosa26 EYFP appears to drive stronger EYFP expression in acinar cells than ducts.) An identical labeling distribution was observed using a different reporter strain, Rosa26 LacZ [26] (data not shown). In these and other experiments, all Muc1 IC2 -labeled cells were found to be E-cadherin + (data not shown), indicating that they represent pancreatic parenchymal cells rather than connective tissue or vasculature.
Having observed the lineage label in acinar cells, we wanted to confirm that these cells actually express Muc1. After enzymatic dissociation and immunostaining of isolated acini, we detect Muc1 not only within small acini, comprised entirely of amylase + cells, but also in single amylase + cells ( Figure 5A-F). Confocal imaging of wholemount stained pancreata confirmed that Muc1 protein localizes to the apical poles of acinar cells, and is readily detected within individual acini that lack    , independent field). As Rosa26 EYFP expression is lower in duct cells than acini, we have increased the green signal in these panels to make ductal staining more obvious. (F) 8-month old TMuntreated control mouse, stained for EYFP (green), amylase (red) and cytokeratin-19 (white). Less than 0.1% of duct and acinar cells are labeled in the absence of TM. Scale bars: 100 μm. Abbreviations: ac, acinus; du, duct; is, islet. centroacinar cells [27] ( Figure 5G-I). Since this antibody recognizes the intracellular domain of the Muc1 protein [28], the observed staining is unlikely to reflect shed or contaminating Muc1 from duct or centroacinar cells. In situ hybridization studies have similarly revealed MUC1 expression in both acinar and duct cells of the human pancreas [29], and we conclude that the acinar labeling observed with Muc1 IC2 reflects endogenous Muc1 expression.
We also determined the relative distribution of Muc1 IC2 -labeled cells between small (intercalated), medium (intralobular) and large (interlobular) ducts ( Figure  6A-C). We found that Muc1 IC2 -labeled cells were present in all three duct types, with slightly lower labeling within interlobular ducts than those of smaller caliber ( Figure 6D). This labeling distribution is consistent with the distribution of Muc1 protein itself, described above ( Figure 2), and suggests that Muc1 IC2 -driven recombination can occur in any Muc1-expressing cell.
Examining TM-untreated 4-8 month Muc1 IC2/+ ; Rosa26EYFP /+ and Muc1 IC2/+ ;Rosa26LacZ /+ mice (n = 3) revealed a background recombination rate of less than 0.1% ( Figure 4F and data not shown), confirming that this line exhibits stringent tamoxifen dependence. Furthermore, we never observed lineage-labeled endocrine cells in our short-term experiments. Muc1 IC2 labeling is thus TM-dependent, as expected, and extends throughout the exocrine pancreas (acini and ducts). Although TM-induced Muc1 IC2 labeling is relatively sparse (5-15% of exocrine cells, in this experiment), its distribution closely matches that of Muc1 itself, and appears to represent a random sampling of the exocrine pancreas. As Muc1 IC2 does not directly label islet cells, it can be used to detect potential exocrine-derived neogenesis.

Islet cells arise from embryonic Muc1 + cells
To determine whether Muc1 + cells contribute to islet neogenesis in utero, we administered tamoxifen to pregnant females carrying Muc1 IC2/+ ; Rosa26EYFP /+ embryos. These experiments yielded a very low overall labeling rate (≤1% of any cell type, data not shown), possibly due to reduced amounts of TM entering fetal circulation. To increase the cellular Cre concentration, therefore, we used homozygous Muc1 IC2/IC2 ; Rosa26EYFP /+ embryos for all in utero labeling experiments. Muc1 IC2/IC2 ;Rosa26EYFP /+ embryos were labeled with a single maternal dose of TM (5-10 mg) at different embryonic stages: E11.5, E13.5 or E15.5. All embryos were analyzed at E17.5 (n = 3 for each tamoxifen treatment group). As expected, Muc1 IC2 labels fetal ducts at all these timepoints ( Figure 7A and data not shown), albeit at low frequencies (2-6%, Figure 7B). Consistent with its labeling pattern in the adult pancreas, Muc1 IC2 marks a similar proportion of fetal acinar cells as well ( Figure 3B and data not shown). Furthermore, we observe EYFP lineage-labeling of insulin + β-cells and glucagon + α-cells in all of these pancreata ( Figure 7B-D), indicating an origin within the Muc1 + exocrine compartment. Our experimental design therefore identifies potential neogenesis even from a sparsely-labeled population. As β-cell numbers increase exponentially between E11.5 and E17.5 [30],~100-fold total, most of the β-cells analyzed in this experiment would have been born since tamoxifen was administered. That we find comparable Muc1 IC2 lineage labeling in β-cells and exocrine cells ( Figure 7B) suggests that most of the new β-cells arose via exocrine-derived neogenesis.
To further confirm that Muc1 IC2 is not active in fetal islets themselves, we performed short-term (12-14 hr) pulse-chase labeling of embryos at E13.75 or E14.75. We could easily detect the lineage label in the embryonic ductal system as well as in Neuog3 + cells ( Figure 8A-B), but failed to detect any labeled hormone-producing cells ( Figure 8C-D). The short-term labeling of Neurog3 + cells agrees with the co-expression of Neurog3 and Muc1 in embryonic ducts ( Figure 3A), and the lack of short-term islet labeling confirms that the later appearance of labeled endocrine cells reflects differentiation from exocrine tissue.
Finally, to determine whether fetal Muc1 + cells contribute to adult islets, we administered 5 mg TM to pregnant females at E14.5 and analyzed labeling at weaning age (P21). We found labeling of both β-cells (2.2 +/-0.4%) and α-cells (2.1% +/-0.2%), present in mature islet structures ( Figure 9). As expected, lineage label also persists in adult duct cells, and appears to be distributed equally among various classes of duct structures ( Figure 6D).
Altogether, these results strongly suggest that (a) embryonic Muc1 + duct cells give rise to all segments of the adult ductal network, (b) Muc1 is not expressed by endocrine cells and (c) mature islet cells arise from Neu-rog3 + precursors within the embryonic Muc1 + ductal network. These results represent direct evidence that endocrine cells arise from embryonic ducts, but they leave open the possibility that embryonic Muc1 IC2 labeling actually marks a ductal stem cell-like population, which continues to give rise to new islet cells after birth. We therefore turned our attention to the differentiation potential of postnatal Muc1 + cells.

Muc1 + exocrine cells do not undergo endocrine differentiation after birth
We performed three experiments to determine whether the Muc1 + exocrine compartment contributes to islet cells after birth. In experiment 1, we administered 10 mg tamoxifen to young adult (~P60) Muc1 IC2/+ ; Rosa26EYFP /+ mice, and assayed the potential contribution of labeled cells to either exocrine or endocrine cells after 7, 60 or 120 days ( Figure 10A-C and data not shown). We observed an overall labeling efficiency (i.e. proportion of EYFP + cells per field, without reference to cell type-specific markers) of~24% in these experiments, which included~30% labeling of acinar and 10% labeling of duct cells ( Table 1). As our short-term labeling experiments never revealed detectable lineage marking of islet cells (D.K., unpublished observations), we focused our quantitative analyses on the 60 d and 120 d chase periods. We counted randomly-chosen fields of β-cells and α-cells, scoring separately for number of insulin + or glucagon + cells, number of EYFP + cells, and number of double-positive (hormone + /EYFP + ) cells. In fact, after scoring several thousand cells positive for each marker (Table 2), we never observed a single βcell or α-cell positive for EYFP, suggesting that new islet cells do not arise in significant numbers from adult Muc1 + exocrine cells.
To determine if Muc1 + cells contribute to the rapid expansion of islet cell numbers after birth [15], we performed experiment 2, in which we induced recombination in Muc1 IC2/+ ; Rosa26 LacZ/+ neonates by administration of tamoxifen to nursing mothers (consecutive 10 mg doses on P0 and P1). We sacrificed mice 21 or 120 days after treatment (i.e. as weanlings or mature adults), and performed immunofluorescence to detect LacZ lineage marker within the exocrine and endocrine compartments ( Figure 10D-F and data not shown). We found an overall labeling efficiency of~10% in these experiments, including~15% of acinar cells and~4% of duct cells (Table 1). As in experiment 1, however, despite scoring several thousand cells for each marker, we did not observe a single LacZ + β-cell or α-cell (Table 2).
Previous studies indicate that acinar cells do not contribute to islets after birth [7,8,31], and experiment 2 suggests that neonatal duct cells are also excluded from the islet lineage. This interpretation hinges on relatively infrequent ductal labeling, which could have hidden a low level of duct-derived neogenesis. To increase the duct labeling frequency, we performed experiment 3, in which we directly injected newborn Muc1 IC2/+ ; Rosa26EYFP /+ pups with tamoxifen (2 mg per pup, delivered subcutaneously). Upon sacrifice, 21 days after TM administration, we found increased overall labeling compared to mice that received maternal TM (~30% EYFP + , Table 1). Importantly, the duct labeling Figure 7 Muc1 + cells give rise to endocrine cells during embryogenesis. (A) A single dose of 5 mg tamoxifen was given to pregnant mice at E11.5, and pancreata of Muc1 IC2/IC2 ; Rosa26 EYFP/+ offspring were analyzed at E17.5 by immunofluorescence for EYFP (green), DBA lectin (red) and DAPI (blue). Arrowheads indicate DBA + cells co-expressing EYFP. (B) Quantification of observed co-labeling at E17.5, following tamoxifen administration at indicated time points. Scatter plot depicts the percentages of DBA + (orange), amylase + (green), insulin + (red) and glucagon + (blue) cells co-expressing EYFP. Each point represents an independent embryo, and means are indicated by dotted lines. (C-D) Pancreata of Muc1 IC2/IC2 ; Rosa26 EYFP/+ embryos, administered TM at E11.5 as in (A), stained at E17.5 for EYFP (green), insulin or glucagon (red) and DAPI (blue). Inserts: confocal z-projection (1.36 μm) of the boxed areas, to confirm localization of EYFP to hormone-producing cells. Scale bars: 25 μm.
Kopinke and Murtaugh BMC Developmental Biology 2010, 10:38 http://www.biomedcentral.com/1471-213X/10/38 frequency was increased to 10%, with equal distribution among interlobular, intralobular and intercalated ducts ( Figure 6D). Nonetheless, we did not observe any labeled β-cells or α-cells, despite scoring several thousand cells positive for each marker ( Table 2). These analyses suggest an upper limit to the contribution of neogenesis to postnatal islet growth. β-cell mass has been reported to expand between 4-and 10-fold in the first 2-4 weeks after birth [32][33][34]. If we assume a five-fold expansion between P0 and P21, we can infer that~80% of the β-cells scored in experiment 3 were "new" since P0 (3600 of the~4500 β-cells counted, Table 2). If all of these had been derived from Muc1 IC2labeled duct cells, given a duct labeling index of~10% (Table 1), we would have expected to observe roughly 360 labeled β-cells. As we observed zero, we conclude that ≤1% of all β-cells generated after birth could have arisen from labeled ducts (1% neogenesis would have resulted in~4 labeled β-cells, which is probably near the limit of reliable detection). Altogether, experiments 1-3 fail to reveal duct-to-islet transdifferentiation after birth.

Discussion
At birth, the mammalian β-cell changes from a metabolic passenger to the driver of glucose homeostasis. Based on our results and those of Solar et al. [20], we propose that the mechanisms controlling β-cell mass also change at birth, from a fetal period of new differentiation, or neogenesis, to a mature state of self-renewal ( Figure 11). To detect this transition, we performed a direct comparison of duct and acinar cell lineages before and after birth. We provide formal proofconfirming prior studies of histology and gene expressionthat islets arise from embryonic Muc1 + ducts. From birth onwards, however, we find no evidence for a ductal   We had intended, in creating the Muc1 IC2 allele, to specifically address the differentiation potential of duct cells. Instead, we find that Muc1 IC2 labels both acinar and duct cells, at all stages examined, and that Muc1 protein is readily detected within acinar cells. Nonetheless, we can treat the labeling of postnatal acinar cells as "background," as acinar-to-islet transdifferentiation does not occur after birth [7,8,31]. Cells expressing the acinar enzyme Cpa1 do behave as multipotent "tip cell" progenitors prior to E13.5, but are thereafter restricted to the acinar lineage [7]. As Muc1 + cells still contribute to islets at E13.5 and E15. 5 (Figs. 7,9), we propose that islet differentiation competence normally shifts from Muc1 + / Cpa1 + tips to Muc1 + /Cpa1-negative "trunks" after E13.5, before being lost entirely at birth (Figure 11).
Another recently developed mouse line, K19 CreERT , in which CreERT is targeted to the cytokeratin-19 locus, drives TM-dependent recombination in inter-and intralobular ducts [19]. Unlike Muc1 IC2 , K19 CreERT does not label distal intercalated ducts, and is active in a small fraction of islet cells. Nonetheless, preliminary experiments reported using K19 CreERT provide independent evidence supporting our model: TM treatment at birth Three "pulse-chase" lineage-labeling experiments were performed, as described in the text, in which tamoxifen was administered (by oral gavage to adults, or subcutaneous injection to neonates) and mice sacrificed after chase periods of 7-120 days. For each sample, we counted total cells per field (DAPI), acinar cells (amylase + ) and duct cells (cytokeratin-19 + ), as well as the number of cells double-positive for these markers and for the lineage tracer. From these counts, we derived the labeling efficiency (expressed as a percentage) of total cells (DAPI), acinar (AMY) and duct (CK19) cells. Shown are summary data for every mouse analyzed, as well as mean values +/-standard error. ND, not determined. As described in the text, mice subjected to "pulse-chase" labeling with Muc1 IC2 were analyzed for potential contribution of lineage-marked cells to insulin + β-cells or glucagon + α-cells (exocrine labeling results in Table 1). Indicated are the total number of cells scored as positive for lineage label (EYFP or LacZ), as well as the total number of insulin + and glucagon + cells scored in the same fields. In no case did we observe a lineage label-positive endocrine cell. ND, not determined. Figure 11 Dynamic differentiation potential within theexocrine pancreas. Multipotent pancreatic progenitors (E11.5-E13.5) express the digestive enzyme Cpa1, which is later restricted to acinar cells (E14.5-adult), together with Muc1 and Hnf1β [20]. While Cpa1 + cells cease contribution to islets from approximately E13.5 [7], Muc1 + /Hnf1b + cells continue to give rise to Neurog3 + endocrine precursors (pink nuclei) and islets (blue and green) through at least E15.5. From around birth, Muc1 + and Hnf1b + cells no longer contribute to the endocrine lineage, and mature islets are maintained by self-replication of those pre-existing. However, it remains a formal possibility that a subpopulation exists within the ductal network (dark blue cytoplasm), expressing neither Muc1 nor Hnf1b, from which β-cells continue to arise after birth.
results in ≥10% labeling of ducts after one week, but <1% labeling of islets, equivalent to the direct activity of this line in islet cells themselves [19]. While this manuscript was in preparation, Solar and colleagues [20] published a study using another exocrine CreERT2 line, driven by the Hnf1b locus. Unlike Muc1 IC2 , this driver is not active in acini, and labels a higher fraction of duct cells postnatally (approximately 20% at birth and 40% in adults, compared to 10% labeling at either timepoint with Muc1 IC2 ). As with Muc1 IC2 , lineage-tracing of Hnf1b + cells revealed duct-to-islet differentiation prior to birth, but none thereafter. Further experiments by these investigators indicate that such differentiation does not occur in the context of injury and regeneration [20], as previously believed [16]. Our data provide further evidence against postnatal duct-toislet differentiation in the healthy pancreas, although it remains to be determined if injury can induce neogenesis from Muc1 IC2 -expressing population.
The Hnf1b-CreERT2 and Muc1 IC2 lineage tracing results contradict those obtained with a Cre transgene driven by the Carbonic anhydrase II promoter (CAII-Cre) [18]. Using Rosa26 LacZ reporter mice to detect recombination [26], these authors report that CAII-Cre drives duct-restricted recombination beginning at E18.5, but labels roughly 15% of β-cells at four weeks of age. We cannot offer an obvious explanation for this discrepancy; given the number of β-cells that we counted, we should have detected such a robust contribution from Muc1 IC2 -labeled duct cells. One possibility is that CAII-Cre-catalyzed recombination actually begins prior to birth, when Neurog3 + cells are still present [10,11], but that LacZ expression cannot be detected until one or more days after birth. In fact, half of the newborn pancreata examined in this study already exhibited at least some labeled β-cells (in one pup, as many as 70% of islets contained labeled β-cells), indicating prenatal recombination [18]. These authors have also generated a CAII-CreERT transgene, which could be used to follow postnatal labeling specifically, although their experiments with these mice revealed surprisingly high levels of tamoxifen and Cre-independent LacZ expression in adult islets [18].
Alternatively, a subpopulation of duct cells with the capacity for islet differentiation might escape labeling by Muc1 IC2 , K19 CreERT and Hnf1b-CreERT2, but not by CAII-Cre ( Figure 11). Indeed, as neither our Cre driver nor those described by others labels a majority of postnatal duct cells [19,20], the possibility of a substantial unmarked subpopulation is impossible to exclude formally. As three independent and distinct Cre transgenes have yielded identical conclusions, however, the burden of evidence would appear to weigh against postnatal islet neogenesis.
With respect to Muc1 IC2 in particular, we note that although its recombination efficiency in utero is even lower than after birth, our experimental approach still identifies islet cells arising from the sparsely labeled embryonic exocrine compartment. Furthermore, we have never observed a Muc1-negative duct cell (Figs. 1, 2, 3, and data not shown), nor is there evidence for anatomical exclusion of Muc1 IC2 -labeled cells within the ductal network ( Figure 6). We also do not observe any obvious change in Muc1 expression or distribution between embryonic stages, when Muc1 IC2 does label endocrine cells, and postnatal stages, when it does not. An obvious transition that does occur perinatally is the extinction of Neurog3 expression, which itself weighs against the persistence of duct-to-islet differentiation after birth [10,11].
From a physiological perspective, it might make sense that expansion of β-cells after birth involves a mechanism independent of neogenesis, as postnatal βcells must contend with metabolic demands from which embryonic progenitor cells are buffered. In fact, numerous knockout mouse studies indicate the existence of postnatal-specific mechanisms to control β-cell mass [32,[35][36][37][38][39]. Furthermore, recent studies suggest that expansion of β-cell mass in adults, in response to β-cell damage or increased insulin demand, occurs via proliferation rather than neogenesis [13,37,40]. Potential exceptions to this rule have been described, including partial pancreatectomy and duct ligation, in which development of new β-cells is accompanied by the re-appearance of Neurog3-expressing cells within the ductal epithelium [16,17]. Studies using Hnf1b-CreERT2 to mark pre-existing duct cells in such models did not detect contribution to new β-cells, however [20]. The Muc1 IC2 line is well suited for similar experiments and, as a tool to mark cells throughout the exocrine pancreas, it should complement and extend results obtained by others.

Conclusions
Our results constitute formal evidence that insulin-producing β-cells, and other endocrine cells of the mature pancreatic islet, derive from ductal cells of the embryonic organ. Furthermore, the ability to trace the lineage of cells expressing Muc1 at different timepoints allows us to compare their differentiation potential before and after birth. We find that Muc1-expressing cells lose the capacity for islet differentiation postnatally, prior to the major increase in β-cell numbers that occurs in juvenile mice. These data add to an emerging model for control of β-cell mass, driven by developmentally-programmed neogenesis in the womb and physiologically-regulated proliferation after birth.

Targeting CreERT2 to the Muc1 locus
We followed the procedure of Wu et al. [41] to generate a Muc1 IRES-CreERT2-neo (Muc1 IC2neo ) targeting vector. In brief, we recombineered a 9.2 kb fragment of the mouse Muc1 gene from a 129Sv BAC library (clone bMQ-356N19, from the Sanger Institute) into a conventional plasmid, introduced an IRES-CreERT2-FRT-neo R -FRT cassette after the endogenous Muc1 stop codon, and flanked the homology arms with thymidine kinase (tk) cassettes ( Figure 3D). The targeting vector was then electroporated into R1 ES cells [42], generously provided by Mario Capecchi, which were selected with G418 and FIAU [43]. 52/96 of the surviving clones exhibited homologous recombination upon Southern blotting with a probe located outside the 5' homology arm. Proper recombination was confirmed for 8/8 of these upon further Southern analyses, and one of these clones, DK5.25 ( Figure 3E), was used to generate chimeras (University of Utah, Transgenic Core Facility).
Tamoxifen (Sigma T-5648) was dissolved in corn oil, and administered by oral gavage at doses of 5-10 mg to adult mice, or 2 mg by subcutaneous injection of neonates. For timed-pregnancy studies, noon on the day after vaginal plugging was considered embryonic day 0.5 (E0.5). All animal procedures were approved by the Institutional Animal Care and Use Committee.

Immunostaining and analysis
Tissue fixation, processing and immunostaining were performed essentially as described [45]. Tissues were fixed with 4% paraformaldehyde (PFA) in PBS for 1-2 hrs at 4°C, embedded in OCT and cryosectioned at 7-8 μm thickness. Primary antibodies used in this study are listed in Table 3. Secondary antibodies were purchased from Jackson Immunoresearch. To calculate labeling efficiencies, we photographed 5-12 randomly selected 20× fields per stained specimen, across 4-8 sections separated by 100-150 μm. The total number of each cell type (DAPI for total cells per field, LacZ or GFP for Muc1 IC2 -labeled cells, insulin and glucagon for endocrine cells, amylase, cytokeratin-19 and DBA lectin for exocrine cells) was determined using the Analyze Particles function of ImageJ (NIH). Double-positive cells were detected by additive image overlay, in ImageJ, of the DAPI channel with lineage + and marker + staining. Accuracy of counts was confirmed by eye in Adobe Photoshop for random samples. Calculations and graphs were generated with Microsoft Excel and R http://www. r-project.org.

Acinar isolation and staining
Acini were isolated by sequential trypsin and collagenase P digestion of minced dorsal pancreas, as described [46], PFA-fixed for 15 min and adhered to microscope slides by cytospin (Thermo-Fisher). Cytospin slides were stained as per cryosections. For wholemount immunofluorescence of intact tissue, small pieces of the dorsal pancreas were excised and PFA-fixed as above, washed with PBS, permeabilized with 1% Triton X-100 in PBS and stained with primary and secondary antibodies (overnight incubations, followed by extensive PBS/0.1% Tween-20 washes).