Abstract
Notch signaling governs the proliferation of multipotent pancreatic progenitor cells (MPCs) and their segregation into proximal duct/endocrine bipotent progenitors (BPs), and distal unipotent pro-acinar cells (PACs). However, it is unclear which ligands are involved and when they act. Here we show antagonistic effects of Jag1 and Dll1 on MPC proliferation. Furthermore, blocking Notch signaling before E13 shunts MPCs into the distal PAC fate, while later inactivation shunts BPs to endocrine differentiation. All BPs are eliminated in Jag1, Dll1 double mutants, with Jag1 expression in PACs proving critical for specification all but the most proximal 5% of BPs. Hes1 expression is elevated in E12.5 Jag1 mutant pancreas and release from the multipotent state is delayed. However, by E14.5 Hes1 expression becomes attenuated, coincident with the biased adoption of a PAC fate. Remarkably, ductal morphogenesis and organ architecture are minimally perturbed in the absence of Jag1 until later stages, when ductal remodeling fails and signs of acinar-to-ductal metaplasia appear. Our study uncovers that an interplay between Jag1 and Dll1 control the multipotent state and that they together specify the entire pancreatic duct cell lineage.
Deciphering the mechanisms that control the differentiation of progenitor cells into one of several possible cell fates is crucial for understanding disease etiology and for using pluripotent stem cells in cell replacement therapy applications. The three principal cell lineages of the mammalian pancreas, acinar, duct and endocrine, arise from multipotent pancreatic progenitor cells (MPCs) through a series of binary cell fate choices1. MPCs are specified from the posterior foregut endoderm as dorsal and ventral anlagen at around embryonic day (E)8.52, and are distinguished by their expression of several transcription factors including Pdx1, Ptf1a, Hnf1β, Sox9 and Nkx6-1. The multipotentiality of these cells has been demonstrated at the population level by many lineage-tracing studies3-7 and, more recently, at the clonal level8.
The dorsal MPCs give rise to an early wave of endocrine cells from ∼E9.5 to ∼E11 while endocrinogenesis is delayed ∼36 hours in the ventral anlage9-11. The appearance of hormone-producing cells is preceded by, and dependent on, expression of Neurog3 (encoding Ngn3) in endocrine precursors, which initiates at E8.5 in the dorsal bud and E10.0 in the ventral bud3,12,13. Intriguingly, there is a transient decline in Neurog3 expression from E11 to E1212 commensurate with the segregation of Ptf1a+ Nkx6-1+ MPCs into proximal, Ptf1a− Nkx6-1+ bipotent progenitors (BPs) and distal, Ptf1a+ Nkx6-1− pro-acinar cells (PACs)14,15. This proximodistal (PD) patterning is regulated by Notch signaling. For example, forced expression of a Notch intracellular domain (NICD) prevents MPCs from adopting a PAC fate15-17 while dominant-negative Maml1 prevents a BP fate18,19. Total inactivation of Notch signaling in the endoderm, as seen in Foxa2T2AiCre/+; Mib1f/f (Mib1ΔFoxa2) embryos, results in a complete shift from BP-to PAC fate19. Downstream of Notch, PD patterning depends on mutually antagonistic interactions between Ptf1a and Nkx6-1/Nkx6-215.
Notch signaling, and its downstream target gene Hes1, have previously been shown to prevent precocious and excessive endocrine differentiation20,21, and independently of this also stimulate MPC proliferation22. After PD patterning is complete at around E14, Notch signaling and Hes1 expression persist in BPs where they maintain Sox9 expression and ductal fate, and inhibit endocrine differentiation by repressing Neurog323-27. Less is known about which ligands regulate the adoption of distinct cell fates. Dll1 is involved in the control of early endocrine differentiation20,22. Yet, PD patterning is unaffected in Dll1ΔFoxa2 embryos19 and the endocrinogenic phenotype of E10.5 Dll1ΔFoxa2 embryos is much weaker than that of E10.5 Mib1ΔFoxa2 embryos, in which essentially the entire dorsal bud is converted into glucagon cells19,28. Together, these findings show that additional Mib1 substrates, most likely other Notch ligands, are involved in these cell fate decisions. In zebrafish, intrapancreatic duct cells are depleted in jag1b/jag2b morphants29,30 and completely absent from jag1b−/−; jag2b−/− embryos31 with the latter also showing a reduction in endocrine cells while acinar cells were unaffected. However, in mice Jag1 is the only Jagged-type ligand expressed in the pancreatic endoderm32, and while conditional deletion of Jag1 using Pdx1-Cre causes ductal malformation and ductal paucity postnatally, this was attributed to reduced duct cell proliferation33. Recently, conditional double-deletion of Dll1 and Jag1 with Ptf1aCre was shown to yield a relatively modest phenotype with loss of only the terminal duct or centroacinar cells (CACs)34. Thus, which ligands regulate the various cell fate decisions during mammalian pancreatic development remain unclear and the time when Notch signaling transitions from regulating PD patterning to controlling duct-endocrine fate choice remains undefined.
In this study, we demonstrate that Jag1 is crucial for segregation of MPCs into the BP and PAC lineages of the secondary transition pancreas. We show that down-regulation of Notch activity and resolution of the MPC markers Nkx6-1, Ptf1a, and Sox9 into their distinct BP and PAC compartments is delayed by two days in Jag1ΔFoxa2 embryos. During the eventual segregation around E14.5 – E15.5, most of the prospective BPs are mis-specified into PACs such that Ptf1a+ PACs entirely replace the Sox9+ Nkx6-1+ BPs normally found in the branches of the epithelial tree. The few remaining BPs, located in the central core of the organ, depend on Dll1 function as they are lost in Dll1; Jag1ΔFoxa2 embryos, resulting in a pancreas essentially comprising acinar cells only. Remarkably, epithelial plexus formation and the gross architecture of the organ are remarkably unperturbed in spite of the profound changes in cell fate. However, at late stages larger ducts are missing or malformed, terminal ducts are morphologically abnormal and signs of acinar-to-ductal metaplasia (ADM) appear. Thus, cell fate allocation and organ architecture are achieved via independent pathways during mammalian pancreas morphogenesis.
Results
Jag1 expression marks nascent PACs
Since BPs are converted to PACs in Mib1ΔFoxa2 embryos, but not in Dll1ΔFoxa2 embryos19, we reasoned that an additional Mib1 substrate, most likely Jag120,22,32,34-36, must act during PD patterning. Because the reported expression patterns are not entirely consistent, we decided to re-analyze the expression of these two ligands by targeting fluorescent protein reporters to the Dll1 and Jag1 loci. The resulting Jag1J1 VmC and Dll1D1 VmC alleles comprise a Venus-T2A-mCherry cassette inserted in frame with the coding region of the last exons to generate Dll1-and Jag1-Venus fusion proteins that serve as dynamic reporters of ligand protein expression while mCherry acts as a more sensitive reporter owing to its longer half-life and as a consequence of being cleaved from the fusion protein (Supplementary Fig. 1). Co-immunofluorescence (IF) analysis with anti-RFP antibodies did not detect mCherry prior to E10.5 in Jag1J1VmC/+ pancreas, at which stage it was seen in most Pdx1+ Sox9+ MPCs and in the surrounding mesenchyme. At E11.5 mCherry expression appeared intensified in the distal epithelium, and by E12.5 expression was largely confined to distal cells and extinguished from many proximal Sox9+ emerging BPs. By E15.5 mCherry was prominently expressed in vascular cells and very weakly in acinar cells but, in spite of previous reports34,35, it was not detectable in Sox9+ BPs of the ducts (Supplementary Fig. 1). Anti-GFP antibodies reproduced the uniform expression pattern for the Jag1-Venus fusion protein in E10.5 Jag1J1 VmC/+ dorsal pancreas epithelium and surrounding mesenchyme (Fig. 1b) and the restriction to the distal epithelium at E12.5, with most of the centrally located Sox9+ emerging BPs being negative for Jag1-Venus (Fig. 1c). At E15.5, weak Jag1-Venus expression was detected apically in acinar cells and in non-parenchymal cells, including the vasculature, but expression was excluded from Sox9+ BPs (Fig. 1d). Since the Jag1 expression pattern we observe conflicts with previous work34,35, we validated two anti-Jag1 antibodies (see Methods). Both reproduced the Jag1-reporter expression pattern in wild type pancreas, and IF analysis for additional markers revealed expression in vascular cells and a very low level in acinar cells at E15.5, its absence from Sox9+ BPs at the later stages and from the endocrine lineage at all stages (Supplementary Fig. 2).
In Dll1D1VmC embryos anti-RFP antibodies detected scattered mCherry+ cells in E9.5 dorsal pancreas which were often also Sox9Lo/− and located adjacent to Sox9+ mCherryLo/− cells. A similar pattern of mCherry expression was seen at E10.5, E11.5 and E12.5, with the mCherry+ cells frequently seen in small clusters of 2-5 cells. At E15.5, many mCherry+ cells were found in forming acini, often interspersed with mCherry− cells (Supplementary Fig. 1). Anti-GFP antibodies reproduced this pattern with detection of Dll1-Venus fusion protein in dispersed cells throughout the E10.5 and E12.5 dorsal pancreatic epithelium, many of which were Sox9Lo/− (Fig. 1h,i). By E15.5, Dll1-Venus was found intracellularly in the apical pole of acinar cells, and in scattered Sox9Lo/− cells in the proximal epithelium that most likely represent endocrine precursors (Fig. 1j). A validated anti-Dll1 antibody reproduced this expression pattern in wild type tissue; co-staining for lineage markers showed that Dll1+ cells could be subdivided into several populations comprising Dll1+ Ptf1a+ MPCs or PACs present in E10.5, E12.5, and E15.5 pancreas epithelium as well as Dll1+ Ngn3+ endocrine precursors, evident at all stages, and Dll1+ Ngn3− cells in E12.5 proximal epithelium (Supplementary Fig. 2). Co-expression of Dll1 and Jag1 was evident in E10.5 Sox9+ MPCs (Fig. 1e,k) and in a subset of distal cells at E12.5 (Fig. 1f,l). IF for Jag1, Dll1, Ptf1a and Nkx6-1 revealed that these were Ptf1a+ and Nkx6-1− or Nkx6-1Lo, likely representing nascent PACs and MPCs, respectively (Fig. 1o-y). The few Nkx6-1Hi cells found in the distal epithelium were typically Jag1Lo/− Dll1− Ptf1a− (Fig. 1n-s). Conversely, in the Ptf1a− proximal region Dll1+ cells were Jag1− and either Nkx6-1+ or Nkx6-1− (Fig. 1t-y), which may represent either BPs or endocrine precursors derived from BPs.
Notch activity is suppressed in nascent PACs
To define the cells in which Notch receptors have been activated before and during PD patterning we first analyzed embryonic pancreata from the Notch1 activity-trap mouse line N1IP::CreHI; Rosa26LSL-Ai3 in which EYFP permanently labels the progeny of cells experiencing Notch1 activation (Fig. 2a)37. We detected YFP labeling in a few E10.5 Ptf1a+ Sox9+ MPCs and at E12.5 we found YFP expression in Ptf1a+ nascent PACs, Sox9+ nascent BPs and in Ptf1a+ Sox9+ MPCs (Fig. 2b,c). Quantification revealed that YFP+ cell distribution reflected no bias towards either nascent BPs or PACs at E12.5 (Fig 2d), consistent with the cells experiencing Notch1 activation in MPCs, from which both lineages derive, prior to the linage segregation occurring by E12.5. Since this reporter is specific to Notch1, we next analyzed Hes1 expression as a general, pan-Notch acute readout. We bred mice harboring the Hes1-EGFP reporter24 to the Jag1J1VmC and Dll1D1VmC lines and assayed for EGFP expression using an antibody that only detects the high levels of (cytoplasmic) EGFP derived from the Hes1-EGFP reporter, but not the low levels of (membrane-associated) Jag1-Venus or Dll1-Venus fusion proteins (Fig. 2e-n). We compared expression of EGFP to that of mCherry and Sox9 by IF. EGFP+ Sox9+ MPCs were seen as early as E9.5 in both reporters, and the Dll1D1VmC; Hes1-EGFP reporter revealed that these were adjacent to Dll1-expressing cells (Fig. 2e,j). The majority of cells in the E10.5 Jag1J1VmC; Hes1-EGFP dorsal bud were Sox9+ EGFP+ mCherry+ MPCs, although EGFP+ mCherry− cells were also seen (Fig. 2f) and again the EGFP+ cells were often Sox9Hi and adjacent to Sox9Lo mCherry+ cells in E10.5 Dll1D1VmC; Hes1-EGFP buds (Fig. 2k). In E11.5 Jag1J1VmC; Hes1-EGFP embryos, EGFP+ mCherry+ cells were seen mostly in the periphery while EGFP+ mCherry− cells were abundant in the center (Fig. 2g). By E12.5, most of the center is occupied by emerging EGFP+ mCherry− BPs while the periphery is mainly composed of nascent mCherry+ PACs interspersed with EGFP+ mCherry+ MPCs and rare EGFP+ mCherry− cells (Fig. 2h). We frequently observed Sox9Hi EGFP+ mCherry− cells intercalated with Sox9Lo/− EGFP− mCherry+ cells in E11.5 and E12.5 Dll1D1VmC; Hes1-EGFP pancreata (Fig. 2l,m), suggesting that high levels of Dll1 and Jag1 in PACs could mediate Notch trans-activation in neighboring Dll1Lo/− Jag1Lo/− BPs. In E15.5 Jag1J1VmC; Hes1-EGFP pancreata we found EGFPHi cells to be Sox9+ mCherry− BPs located in the proximal “trunk” epithelium or occasionally in forming acini (Fig. 2i). In E15.5 Dll1D1VmC; Hes1-EGFP pancreata, Sox9+ EGFPHi BPs were often in contact with Sox9Lo/− mCherry+ cells and in the forming acini they were often found sandwiched between EGFP− mCherry+ cells (Fig. 2n).
An important caveat to these analyses is the long half-life of the reporters. To overcome this, we used an anti-Hes1 antibody to map Notch activation in relation to ligand protein expression in E12.5 MPCs and emerging BPs and PACs, distinguished by their Ptf1a and Nkx6-1 expression status. We found most Hes1Hi cells to be Nkx6-1Hi Ptf1aLo/− Jag1Lo/− and most Hes1Lo/− cells to be Nkx6-1Lo/− Ptf1aHi Jag1Hi except for a few Hes1Lo/− cells that were Nkx6-1Hi Ptf1aHi Jag1Hi (Fig. 2o-q). In relation to Dll1, we found that Hes1Hi cells typically were Nkx6-1Hi Ptf1aLo/− Dll1Lo/− BPs, mostly emerging centrally but in rare cases as single cells in the periphery located next to Hes1Lo Nkx6-1Lo/− Ptf1aHi Dll1Hi cells (Fig. 2r-t).
As the ratio between ligands and receptors determines whether cells are prone to sending or receiving Notch signals we examined the distribution of Notch1 and Notch2, the two receptors known to be expressed in the pancreatic epithelium32, at E12.5 using previously validated antisera27. While Notch1 was expressed in PACs and BPs (Fig. 2u-w), Notch2 was specifically enriched in BPs (Fig. 2x-z), as previously noted in E15.5 pancreas27. Thus, BPs (Jag1− Dll1− Notch1+ Notch2+) may have receptors in stoichiometric surplus, favoring signal reception, while in PACs (Jag1+ Dll1+/− Notch1+ Notch2−) the ligands may be in surplus, favoring signal sending. Taken together, these data show that Notch receptor activation becomes suppressed in emerging Jag1+ Dll1+/− PACs, and support the notion that these, together with Ngn3+ Dll1Hi endocrine precursors in the central trunk epithelium, are activating Notch receptors in nascent, Jag1Lo/− Dll1Lo/− BPs.
Suppressing Notch before ∼E13 shunts progenitors to a PAC fate
To define the temporal window through which Notch signaling controls segregation of MPCs into BP and PAC fates we administered tamoxifen (Tam) to pregnant dams carrying either Hnf1b-CreERT2; Rosa26LSL-dnMaml1-eGFP/+ (hereafter referred to as R26dnMaml1) embryos or Hnf1b-CreERT2; Rosa26LSL-YFP/+ (hereafter referred to as R26YFP) control embryos at different timepoints and harvested embryos for analysis at E15.5 (Fig. 3a). As expected from a previous Hnf1b-CreERT2 lineage-tracing analysis6, IF examination of E11.5 Tam-treated R26YFP pancreata for Sox9 and Ptf1a revealed that ∼45% of the YFP+ cells expressed the BP marker Sox9, ∼25% expressed Ptf1a, and ∼30% expressed neither marker (Fig. 3b, n). In contrast, similarly treated R26dnMaml1 pancreata showed an ∼11-fold reduction of GFP+ cells adopting a BP fate, while the fraction of GFP+ cells allocated to a PAC fate had increased almost 3-fold. The fraction of YFP+ cells expressing neither marker was unchanged (Fig. 3e, n).
In R26YFP pancreata exposed to Tam at E12.5, about one third of the YFP+ cells were Sox9+ while ∼20% expressed Ptf1a, and ∼50% expressed neither marker (Fig. 3c,n). Tam injection at E13.5 revealed an even more pronounced distribution of control YFP+ cells towards BP fate: ∼50% of the YFP+ cells were Sox9+ and only ∼8% of the YFP+ cells were Ptf1a+, while ∼42% expressed neither marker (Fig. 3d,n). Thus, while some Hnf1b-expressing cells retain their multipotency at E13 and E14, they become progressively more biased towards the BP lineage, as also noted previously6. As observed for E11.5 Tam injections, significantly fewer EGFP+ cells co-expressed Sox9 in E15.5 R26dnMaml1 pancreata injected with Tam at E12.5 or E13.5 compared to controls. However, in sharp contrast to E11.5 Tam-induced pancreata, the proportion of EGFP+ cells expressing Ptf1a was unchanged following Tam treatment at E12.5 and E13.5. Instead, more EGFP+ cells were Sox9− Ptf1a−, compared to YFP+ cells in the controls (Fig. 3f,g,n). Together, these data show that suppression of Notch signal reception in MPCs and BPs before ∼E13 shunts the cells to a PAC fate. However, if Notch signaling is blocked in BPs after ∼E13, they adopt an alternative fate.
Suppressing Notch after ∼E13 shunts progenitors to an endocrine fate
The location of the Sox9− Ptf1a− EGFP+ cells seen in E15.5 R26dnMaml1 pancreata exposed to Tam at E12.5 or E13.5 suggests that these may represent cells of the endocrine lineage. However, it is less clear whether one would expect accumulation of endocrine precursors, endocrine cells or both. We therefore performed IF for EGFP/YFP, Sox9, Ngn3 and Chga. As expected, the fraction of labelled cells expressing Sox9 in E15.5 R26dnMaml1 pancreata was significantly lower than in controls after Tam injection at E11.5 and consistent with the increase of labelled Ptf1a+ cells seen above in the same embryos, significantly more labelled cells triple-negative for Sox9, Ngn3 and Chga in dnMaml1 embryos, compared to controls (Fig. 3h,k,o). Examining endocrine lineage markers we found that the fraction of labelled cells expressing either Ngn3 or Chga, was not significantly different between dnMaml1 embryos and controls (Fig. 3h,k,o). However, fewer labelled cells co-expressed Ngn3 in dnMaml1 embryos than in controls (Fig. 3p).
Tam injections at E12.5 and E13.5 gave a different result. While the fraction of labelled cells expressing Sox9 was still reduced in dnMaml1 embryos compared to controls, the fraction of labelled cells expressing endocrine markers was now markedly increased compared to controls, while the fraction of labelled cells triple-negative for Sox9, Ngn3 and Chga was unchanged between groups (Fig. 3i,j,l,m). Again, fewer labelled cells co-expressed Ngn3 in dnMaml1 embryos than in controls (Fig. 3p).
These findings were reproduced by lineage-tracing analyses in E15.5 Sox9-CreERT2; R26dnMaml1 embryos compared to stage-matched Sox9-CreERT2; R26YFP embryos. A qualitative analysis of these revealed the same shift in the fate of the labelled cells when comparing Tam injections at E11.5 to E12.5 and E13.5 as described above for Hnf1b-CreERT2 -driven recombination (Supplementary Fig. 3). Taken together, these results show that the time window through which prevention of Notch activation can shunt cells to a PAC fate closes by ∼E13 and confirms that prevention of Notch activation in BPs after ∼E13 induces endocrine differentiation26,27.
Jag1 is required to specify most BPs
To begin to understand the role of individual ligands in PD patterning we next generated conditional Jag1 and Dll1 mutants. Analysis of E12.5 Dll1; Jag1ΔSox17-TamE6.5 single and double mutant embryos revealed an increase of Ptf1a+ Nkx6-1+ MPCs and a reduction of Ptf1a− Nkx6-1+ BPs (Supplementary Fig. 4), suggesting that Jag1 and Dll1 are both involved in PD patterning. Due to the variable effect size and intrinsic mosaicism in such embryos we decided to employ Foxa2T2AiCre (hereafter referred to as Foxa2iCre) in further experiments to ensure efficient recombination prior to pancreas specification19. However, since Foxa2 is linked to Jag1 on mouse chromosome 2 at a distance of ∼5.7 cM (Supplementary Fig. 5) we first introduced a Jag1 null allele38 on the same chromosome as the Foxa2iCre allele via meiotic crossover. Animals carrying Jag1−; Foxa2iCre chromosomes were then backcrossed to homozygous Foxa2iCre/iCre animals to secure this chromosome from further crossover events. Timed matings of these mice with Jag1fl/fl; R26YFP/YFP animals generated Jag1ΔFoxa2/− embryos (referred to as Jag1ΔFoxa2) and Jag1ΔFoxa2/+ heterozygote littermate controls. Examination at E10.5 by whole-mount IF (WMIF) analysis revealed both the dorsal and ventral pancreata to be increased in size but with no evidence of excessive endocrine differentiation (Supplementary Fig. 5). At E12.5, the Jag1ΔFoxa2 mutant pancreas appeared grossly normal in size and morphology, yet IF analysis revealed that the number of Ptf1a+ Nkx6-1+ MPCs was significantly increased at the expense of Ptf1a− Nkx6-1+ BPs compared to controls, and the staining intensity of Ptf1a also appeared consistently brighter (Fig. 4a-c). While Ptf1a is essentially confined to the outermost epithelial cell layer in E12.5 controls (Fig. 4a, insert), Ptf1a expression was also seen in many proximal cells in Jag1ΔFoxa2 pancreata (Fig. 4b, insert). This shows that, similar to Jag1ΔSox17 -TamE6.5 embryos, the segregation of MPCs into PAC and BP domains is compromised in E12.5 Jag1ΔFoxa2 pancreata. IF analyses at later stages revealed that most Ptf1a+ cells co-expressed Nkx6-1 and Sox9 at E13.5, thus maintaining an MPC marker profile. Not until E14.5 did we observe a resolution into distinct Ptf1a+ Nkx6-1− PACs and Ptf1a− Nkx6-1+ BPs in the Jag1ΔFoxa2 pancreas. However, while BPs are normally extending all the way into the forming acini at this stage, the Jag1ΔFoxa2 pancreas was nearly devoid of such cells in the periphery, while the number of Ngn3+ endocrine precursors and insulin+ β-cells, appeared unaffected at these stages (Supplementary Fig. 6). Remarkably, one day later the Jag1ΔFoxa2 pancreas manifests a striking ∼85-90% decrease in Nkx6-1+ and Sox9+ BPs relative to littermate controls with the few remaining BPs being confined to the center of the organ (Fig. 4d-i). The total number of cells is unchanged due to an equivalent increase of Ptf1a+ cells, suggesting that MPCs normally fated to become BPs switch to a PAC fate.
The combined activities of Jag1 and Dll1 specify the entire BP population
The residual ∼10% proximal-most BPs observed in Jag1ΔFoxa2 pancreata may still require a Notch signal which Dll1 is likely to provide given its spatiotemporal expression pattern. To test this notion, we crossed Jag1+/−; Dll1fl/+; Foxa2T2 AiCre/T2AiCre with Jag1fl/fl; Dll1fl/fl; R26YFP/YFP mice to generate compound Jag1; Dll1ΔFoxa2 mutants. For comparison, we used stage-matched single Dll1ΔFoxa2 and Jag1ΔFoxa2 mutants as well as control mice without ligand deletions (Foxa2T2AiCre; R26YFP). Analysis of all four genotypes at E15.5 revealed no change in the numbers of Ptf1a+ Nkx6-1− PACs or Ptf1a− Nkx6-1+ BPs between single Dll1ΔFoxa2 mutant and control pancreata. As expected, Ptf1a− Nkx6-1+ BPs were severely depleted in Jag1ΔFoxa2 mutants and we noted a concomitant expansion in Ptf1a+ Nkx6-1− PAC, while compound Jag1; Dll1ΔFoxa2 mutants showed an even more profound loss of Ptf1a− Nkx6-1+ BPs and increase in Ptf1a+ Nkx6.1− PACs over either single mutant (Fig. 5a-e). The confinement of the remaining Nkx6-1+ BPs to the central core in Jag1ΔFoxa2 mutants, was confirmed and extended to Jag1; Dll1ΔFoxa2 mutants by quantifying the fraction of Nkx6-1+ and Ptf1a+ cells in the z-dimension (Fig. 5f). Since Nkx6-1 is expressed in both BPs and β-cells we next analyzed Sox9+ and insulin+ cell numbers to distinguish between the two. While Sox9+ BPs were unchanged in E15.5 Dll1ΔFoxa2 mutants compared to wild types, they were severely reduced in Jag1ΔFoxa2 mutants and even more so in compound Jag1; Dll1ΔFoxa2 mutants (Fig. 5g-k). As expected from the loss of BPs, assessment of endocrine differentiation by IF analysis of Ngn3 and insulin revealed a loss of BP progeny by E15.5. An ∼50% decrease in the number of insulin+ β-cells was seen in both Dll1ΔFoxa2 and Jag1ΔFoxa2 compared to littermate control pancreata, while compound Jag1; Dll1ΔFoxa2 mutants exhibited a more profound loss than either single mutant alone (Fig. 5g-l). Together, these data show that the combined activities of Jag1 and Dll1 are crucial for proper PD patterning of the mouse pancreas epithelium.
Impeding ligand trans-activation in late development only affects CACs
The strong phenotype we observe in Jag1; Dll1ΔFoxa2 mutant embryos contrasts with the rather modest phenotype seen in Jag1; Dll1ΔPtf1a mutant embryos, which only lack CACs34. This suggests that the timing of Cre-mediated recombination, which is mosaic and occurs considerably later with Ptf1aCre compared with Foxa2iCre 44,48, is critical for deciding the outcome. Considering that Mib1 is required for all Notch ligand trans-activation39-41 and that its elimination in Mib1ΔFoxa2 pancreata phenocopies the cell fate changes in Jag1; Dll1ΔFoxa2 embryos seen here19, we therefore asked whether Mib1ΔPtf1a pancreata would phenocopy Jag1; Dll1ΔPtf1a pancreata. Mib1-depleted cells and their progeny were identified by R26YFP recombination. Examination of Mib1ΔPtf1a pancreata prior to E13.5 failed to reveal any obvious defects. In contrast, analysis at E15.5 revealed the ductal tree to be truncated distally. The distal-most, Sox9+ prospective CACs which normally protrude into the nascent Ptf1a+ acini are specifically depleted in Mib1ΔPtf1a embryos (Supplementary Fig. 7). Notably, this phenotype closely resembles the reported loss of CACs following Ptf1aCre -driven compound deletion of Dll1 and Jag134. Taken together, these results suggest that the trans-activation of Notch receptors by Jag1 and Dll1 expressed on PACs and emerging acinar cells is required to specify and/or maintain adjacent terminal BPs/CAC precursors.
Early Jag1ΔFoxa2 mutants have normal plexus formation and organ architecture
We next analyzed how the BP-to-PAC fate switch affects ductal morphogenesis and overall organ development. In spite of the delayed PD patterning, we found that formation of the epithelial plexus occurred normally in E12.5 Jag1ΔFoxa2 mutants and overall organ size and morphology was comparable between Jag1ΔFoxa2 mutants and control littermates (Fig. 6a-d). Remarkably, even at E15.5 the overall organ architecture of Jag1ΔFoxa2 mutants is essentially unaffected. The Jag1ΔFoxa2 dorsal pancreas had a well-formed, anvil-shaped head and the body gradually tapered into the narrow connection with a normal-sized ventral pancreas located in the duodenal loop (Fig. 6e,f). In contrast, the Dll1ΔFoxa2 dorsal pancreas was hypoplastic with the head being malformed and the body of the pancreas being greatly shortened (Fig. 6g,h). However, closer inspection of Jag1ΔFoxa2 embryos revealed that although the ductal plexus appeared to have remodeled into a hierarchical tree-like structure, the smooth walls of the intercalated ducts seen in controls and Dll1ΔFoxa2 mutants, showed a more serrated appearance in Jag1ΔFoxa2 mutants (Fig. 6i-p). Nevertheless, acinar structure appears normal with apical localization of Muc1, ZO-1 and PKCζ, indicating that acinar cytoarchitecture is maintained in the E15.5 Jag1ΔFoxa2 pancreas (Supplementary Fig. 8). In the domain usually occupied by Sox9+ prospective ducts, we were able to identify elongated tubular structures expressing Ptf1a instead of Sox9 (Fig. 6q,r). Similar to normal PACs these mis-specified PACs also expressed Bhlha15/Mist1 (Fig. 6s,t). Taken together, these findings suggest that the overall organ architecture and remodeling of the ductal plexus is regulated independently from the differentiation programs that allocate MPCs to endocrine, duct and acinar lineages. In contrast, finer morphological features of the ductal tree are clearly perturbed by the BP-to-PAC fate switch.
Late Jag1ΔFoxa2 embryos show signs of acute pancreatitis and ADM
E18.5 Jag1ΔFoxa2 pancreata remained equivalent to controls in size and overall organ morphology. The dorsal and ventral pancreas appeared fused normally, the gastric lobe was present, and the normal “anvil” shape of the dorsal pancreas42 was evident (Fig. 7a,b). However, closer inspection revealed that the larger ducts were not formed properly. The main duct was disrupted, and interlobular as well as larger intralobular ducts were largely absent (Fig. 7c-f and Supplementary Fig. 9). Again, many of the terminal ducts appeared serrated, occasionally connected by ducts of relatively normal morphology despite being composed of Ptf1a+ Mist1+ cells (Fig. 7c-f). As expected, we saw a prominent loss of endocrine cells in E18.5 Jag1ΔFoxa2 pancreata, with scattered α-cells and a central cluster of β-cells. The late-arising, somatostatin+ δ-cells were nearly absent, even in the central cluster and examination of Muc1 and Krt19 expression and DBA lectin binding revealed an obvious paucity of the ductal tree and throughout the epithelium we observed numerous “ring-like” structures with perturbed apicobasal polarity and co-expression of amylase, Cpa1 and Krt19 (Fig. 7g-k, Supplementary Fig. 8, and Supplementary Fig. 9), reminiscent of the acinar-to-ductal metaplasia (ADM) associated with Cerulein-induced acute pancreatitis43.
Increased Hes1 expression in Jag1ΔFoxa2 mutant MPCs
To begin to understand the mechanism causing progenitors to almost exclusively adopt a PAC fate we assessed how loss of Jag1 impacts Notch activity in Jag1ΔFoxa2 pancreata by IF analysis of Hes1 expression. We found that Hes1 expression, and by inference active Notch signaling, was upregulated in the E12.5 Jag1ΔFoxa2 pancreas compared to control littermates (Fig. 8a,b). This was especially notable in the distal-most cells, in which average Hes1 levels were increased ∼2-fold (Fig. 8c). This finding suggests that Jag1 is required cell-autonomously to inhibit Notch activation in emerging PACs, which is somewhat surprising considering that Mib1ΔFoxa2, which lose Hes1 expression, and Hes1ΔFoxa2 pancreata show the same BP-to-PAC fate switch19. However, given that Notch signaling maintains the MPC fate22, it does explain the persistence of the MPC marker profile in most epithelial cells in E12.5 Jag1ΔFoxa2 pancreata.
We therefore analyzed Hes1 expression in Jag1ΔFoxa2 pancreata at E14.5, and saw fewer Hes1Hi cells in the epithelium of Jag1ΔFoxa2 pancreas compared to control (Fig. 8d, e). In contrast, Dll1 expression appeared similar in controls and mutants at these stages. Together, these findings suggest that Notch signaling is maintained in most of the epithelial cells in Jag1ΔFoxa2 pancreata, most likely by Dll1, but becomes suppressed around E14.5, which favors the PAC fate.
Discussion
In this study we show that the Notch ligand Jag1 is crucial for PD patterning of the developing mouse pancreas. The requirement for Notch signaling in MPC proliferation and choice of BP versus PAC fate has been comprehensively documented by previous work15-18,20-22,27,36,44,45. Here we extend these studies by defining the temporal window of Notch dependent MPC segregation and by uncovering Jag1 as a crucial ligand for regulating MPC proliferation as well as coordinating the timely exit from the multipotent stage and proper PD patterning. We first found Jag1 uniformly expressed in ∼E10.5 MPCs with a subset of MPCs showing heterogeneous co-expression of Dll1. Recently, initial experiments in our labs have shown that Dll1 (and Hes1) expression exhibits ultradian oscillations in the developing pancreas (data not shown) making this heterogeneity a likely result of catching oscillating cells in their peak phase46. We have previously shown that Dll1-deficient E10.5 pancreata are hypoplastic due to reduced proliferation19,22,28 and initial experiments have shown that E10.5 Dll1 Type 1 and Type 2 mutants, in which both Dll1 and Hes1 oscillations are dampened47, also present with pancreatic hypoplasia albeit less severe than in Dll1−/− and Dll1ΔFoxa2 mutants (data not shown). Here we found the opposite phenotype in E10.5 Jag1-deficient pancreata, which are hyperplastic, suggesting that Jag1 antagonizes Dll1 function at this stage, possibly by a cis-inhibitory interaction that sequesters a fraction of the available Notch receptors. We propose that Jag1 provides a dampening tone on Notch activation mediated by oscillating Dll1 expression. Ultradian Dll1 oscillations then enables a temporal symmetry where MPCs alternate between sending and receiving input via Notch that ultimately couples to the mitotic machinery (Fig. 9a).
In pancreata undergoing PD patterning, we found that emerging PACs (Ptf1a+) expressed high levels of ligands and little to no Hes1, indicating a state of low Notch activity. Conversely, adjacent MPCs (Ptf1a+ Nkx6-1+) or BPs (Nkx6-1+) were generally expressing no or low levels of ligand and high levels of Hes1, indicating Notch activation. This suggests that mutually inactivating cis-interactions between ligands and receptors48 are crucial for exiting the multipotent stage and for the cells to adopt either a PAC or BP fate. Such a notion is supported by the downregulation of Hes1 and absence of Notch2 expression in PACs and thus overall lower levels of Notch receptor expression than seen in emerging Notch1+ Notch2+ BPs. Conversely, Notch1/2 co-expression, and the absence of Jag1 expression in nascent BPs, would render these more sensitive to signal reception and less prone to signal emission due to cis interactions (Fig. 9b).
To test these ideas and to investigate the role of individual ligands in PD patterning we performed single and double Dll1/Jag1 loss-of-function experiments. Our marker analyses showed that most epithelial cells in the E12.5 Jag1ΔFoxa2 embryos maintained a Ptf1a+ Nkx6-1+ Sox9+ marker profile suggestion that they failed to exit the MPC stage. This correlated with increased Hes1 expression, suggesting that Notch activity is increased and that Jag1 normally acts cell-autonomously to inhibit Notch activation in emerging PACs. We suggest that upregulation of Jag1 cis-inhibits Notch receptors and thus acts as a symmetry breaker that terminates oscillatory Hes1 expression in nascent PACs. Loss of Notch activity may additionally downregulate Nkx6-118 and/or liberate Rbpj from N1ICD, which would then be free to complex with Ptf1a49. Both of these mechanisms would favor a PAC fate15,49. Concurrently, free ligand molecules would be able to convey trans-activation of receptors in neighboring cells if these are in a responsive, ligandLo /receptorHi state, and instruct these to adopt a BP fate (Fig. 9b).
In spite of increased Notch activity at early stages, Jag1ΔFoxa2 MPCs eventually adopt a PAC fate, which depends on the cells attaining a state of low Notch activation15,18,19. Indeed, we find that coincident with adoption of a PAC fate around E14.5, the number of Hes1Hi cells is reduced in Jag1ΔFoxa2 mutant epithelium, indicating that most of the cells have acquired a state of low Notch activation at this stage. The decline in Notch1 mRNA expression seen at this stage32 is also indicative of reduced Notch activation50. However, it remains to be determined what triggers a reduction in Notch activation, but it is noteworthy that the timing coincides with the onset of Lfng (Lunatic fringe) expression at E14.5 in PACs51. Lunatic fringe has been shown to inhibit Notch activity in presomitic mesoderm52 and could potentially strengthen cis-inhibitory activity of Dll153 in E14.5 Jag1ΔFoxa2 progenitors. This would attenuate expression of the BP-promoting Notch target genes Nkx6-118 and Sox927 in Dll1Hi cells allowing these to exit from the MPC state and ultimately adopt a PAC fate (Fig. 9c). Testing this hypothesis and identifying the precise role of different Notch receptors awaits future experiments. The strong effect on PD patterning we observe is surprising since previous analyses of pancreas-specific Jag1 or Jag1/Dll1 deletions did not uncover a prominent expansion of the PAC domain at the expense of BPs33-35. We suspect that this can be attributed to the different timing of efficient, non-mosaic recombination between different Cre lines. Conditional Jag1 deletion with Pdx1-Cre or Foxa3Cre driver lines occurs much later than with our targeted Foxa2Cre driver, which recombined with >99% efficacy prior to pancreas specification19. More recently, compound Dll1; Jag1ΔPtf1a mutants were found to have a loss of CACs, the terminal-most cell type in the ductal tree, while single mutant littermates did not show any phenotype. As our Mib1ΔPtf1a mutants phenocopy the loss of CACs reported in Dll1; Jag1ΔPtf1a mutants, this suggests that the Ptf1aCre -driver only becomes non-mosaic in PACs and their progeny and that CACs are specified by PACs late in pancreatic development.
In spite of the prolonged MPC state, the ductal plexus forms normally. However, at later stages the consequence of the BP-to-PAC switch becomes evident as normal remodeling of the ductal plexus into a well-structured ductal tree is disrupted. The tubular network making up the intercalated ducts seems to form but the normal smooth morphology of the ductal lumen is perturbed by E15.5. We also noted a complete absence of larger intralobular ducts and interlobular ducts at E18.5 and ductal structures in the area of the main duct appear interrupted as also previously noted in Jag1ΔPdx1 animals33. These perturbations are not surprising given that acinar cells are not designed to form cuboidal or columnar epithelia, but rather to adopt a pyramidal shape fitting for cells forming an acinus. However, in spite of these disturbances the overall organ architecture is surprisingly well preserved. We conclude that the regulatory principles governing the shape of the pancreas are highly resilient to cell fate changes, at least as long as these occur in the internal part of the organ.
Methods
Animals
Published mouse strains were genotyped according to the original work: R26LSL-dnMaml1 -EGFP 19, Gt(ROSA)26Sortm1(EYFP)Cos (R26LSL-YFP reporter54), Gt(ROSA)26Sortm3(CAG-EYFP)Hze (R26LSL-Ai3 reporter55), Mib1tm2Kong (floxed Mib141), Dll1tm1Gos (Dll1LacZ null allele56), Dll1tm1.1Hri (floxed Dll119), Jag1tm1Grid (Jag1 null allele38), Jag1tm2 Grid (floxed Jag157), Tg(Hes1-EGFP)1Hri (BAC transgenic Hes1-EGFP reporter24), Hes1tm1Fgu (Hes1 null allele58), Foxa2T2AiCre (Cre add-on allele19), Hnf1b-CreERT2 6, Sox9- CreERT2 4, Sox17T2 AiCre 59, Ptf1aCre 60 and Notch1tm4 (cre)Rko (N1IP::CreHI37). Additional genotyping primers are given in Supplementary Table 1. Homozygous Dll1D1VmC and Jag1J1VmC mice are viable and fertile, but were maintained and analyzed as heterozygotes due to Dll1D1VmC being a weak hypomorph evident by short, kinky tails in homozygote animals. The Sox17CreERT2 line, which was generated by cassette exchange in the Sox17LCA allele61, was a kind gift by Anne Grapin-Botton. Generation of Jag1 C-terminal Venus-T2A-mCherry fusion reporter knock-in construct was conducted using a BAC clone (RP23-173O12) from the BACPAC Resources Center at Children’s Hospital Oakland Research Institute. An frt-PGK-EM7-Neo-frt cassette was inserted downstream of a Venus-T2A-mCherry reporter in pBluescript II SK+, flanked by 300-500-bp homology arms from the Jag1 gene with the Jag1 stop codon removed. BAC targeting cassettes were excised and electroporated into competent SW105 cells containing the BAC clone of interest. Correctly targeted BAC clones were identified by a panel of PCR primers and restriction digestions. The knock-in cassette fragment was retrieved and cloned into pMCS-DTA (a kind gift from Dr. Kosuke Yusa, Osaka University, Japan). The 5′-and 3′-homology arms in the retrieval vector were designed such that between 2.5-and 7.5-kb DNA segments, flanking the Venus-T2A-mCherry reporter-frt-PGK-EM7-Neo-frt cassette in the BAC clone, were subcloned into pMCS-DTA. The shorter homology arm was used to design PCR-based screening for targeted ES cells (TT2). Chimeric mice were produced from successfully targeted ES cell clones by aggregation with ICR embryos. Germ line transmission of the targeted allele was assessed by PCR of tail DNA. pCAG-FLPe mice62 were used to remove the frt-PGK-EM7-Neo-frt cassette. The Dll1-Venus-T2A-mCherry knock-in mice were generated by a similar strategy using a Dll1 containing BAC clone (RP23-306J23).
Noon on the day of vaginal plug appearance was considered E0.5. Tamoxifen (Sigma) was dissolved at 10 mg/ml in corn oil (Sigma) and a single dose of 75 μg/g (for Hnf1b- and Sox9-CreERT2 -mediated R26YFP/dnMam1-eGFP induction) or 40 μg/g (for Sox17CreERT2 -mediated ligand deletion) body-weight administered by intraperitoneal injection at noon ± 1 hour. Jag1ΔFoxa2 and Dll1; Jag1ΔFoxa2: We first introduced a Jag1 null allele38 on the chromosome carrying the Foxa2iCre allele via meiotic crossover. Animals carrying Jag1−; Foxa2iCre chromosomes were then backcrossed to homozygous Foxa2iCre/iCre animals to secure this chromosome from further crossover events. Jag1+/−; Foxa2iCre/iCre animals were next crossed with Jag1fl/fl R26YFP/YFP animals to generate Jag1ΔFoxa2 embryos and to Dll1fl/+; Foxa2T2 AiCre/T2AiCre animals to generate Jag1+/−; Dll1fl/+; Foxa2T2 AiCre/T2AiCre mice. The latter was then crossed with Jag1fl/fl; Dll1fl/fl; R26YFP/YFP animals to generate Jag1; Dll1ΔFoxa2 embryos. Dll1ΔFoxa2 embryos and R26YFP controls were made as previously described19. All animal experiments described herein were conducted in accordance with local legislation and authorized by the local regulatory authorities.
Immunostaining
All primary antibodies are listed with dilution in Supplementary Table 2. Dissected whole embryos (E10.5-E12.5) and foregut preparations (E13.5-E18.5) were fixed in 4% paraformaldehyde in PBS, embedded in Tissue-Tek O.C.T. (Sakura Finetek) and cryosectioned at 10 μm. For immunofluorescence analysis, antigen retrieval was conducted in pH6.0 citrate buffer, followed by permeabilization in 0.15% Triton X-100 in PBS. After blocking in 1% normal donkey serum in PBS with 0.1% Tween-20, sections were incubated overnight at 4° C with primary antibodies diluted in the same buffer. Primary antibodies were detected with anti-rabbit, guinea pig, mouse, rat, goat, sheep or chicken donkey-raised secondary antibodies conjugated to either Cy5 (1:500), Cy3 (1:1,000), Alexa Fluor 488 (1:1,000) or DyLight 405 (1:200) (all Jackson ImmunoResearch Laboratories). Slides were mounted in Vectashield (Vector Laboratories) with or without DAPI for counterstaining nuclei. Whole-mount IF of E10.5 whole embryos and E12.5, E15.5 and E18.5 foregut preparations was performed as previously described63. Specimens were cleared with BABB (benzyl alcohol:benzyl benzoate 1:2) then scanned confocally for z-stack image acquisition. Images were captured on a Leica SP8 or Zeiss LSM780 confocal microscope and figures prepared using Adobe Photoshop/Illustrator CS6 (Adobe Systems, San Jose, CA, USA).
Antibody validation
Antisera against Dll1, Jag1, and Hes1 were validated by IF analysis of E10.5 neural tube from embryos that were either wildtype or null for the relevant gene (Supplementary Fig. 10). Two characteristic stripes of Jag1 was detected in the neural tube64 and uniform but weak Jag1 was detected in the E10.5 pancreas by both anti-Jag1 antisera in wild type tissue, but not in equivalent Jag1-null38 tissue. The anti-Dll1 antibody detected Dll1 in the expected reciprocal pattern (compared to Jag138) in the neural tube and scattered cells in the pancreas of wild type embryos, but not Dll1−/− embryos56. The rabbit anti-Hes1 monoclonal detected Hes1 in the expected patterns in both wild type tissues (i.e. prominently in floor plate and dorsal neural tube as well as pancreatic epithelium and weakly in the surrounding mesenchyme), but not in Hes1−/− tissues.
Cell quantification
∼400 cells YFP+ lineage-traced cells were counted on 9-11 evenly spaced optical sections from each of three E12.5 N1IP::CreHI; Rosa26LSL-Ai3 embryos and scored for co-expression of Sox9 and Ptf1a. YFP+ and GFP+ lineage-traced cells from Hnf1b-CreERT2; R26YFP and R26dnMaml1-GFP embryos, respectively, were counted on every fifth section throughout the pancreas, for a total of >200 cells/embryo, for each marker combination. Ptf1a+ Nkx6.1−, Ptf1a− Nkx6.1+ and Ptf1a+ Nkx6.1+ cells were manually scored on every fifth section throughout E12.5 dorsal pancreas from controls (R26Yfp/+; Sox17CreERT/+), Dll1ΔSox17Tam, Jag1ΔSox17Tam, Dll1/Jag1ΔSox17Tam using Imaris(tm) (Bitplane). E12.5 Jag1ΔFoxa2 embryos were quantified the same way but with their own controls (see below). For quantification of E12.5 Jag1ΔFoxa2 Hes1 IF signal intensity in distal-most Ptf1a+ cells, corrected total cell fluorescence (CTCF) was determined using FIJI65,66. Numbers of Ptf1a+ Nkx6.1−, Ptf1a− Nkx6.1+ and Ptf1a+ Nkx6.1+ cells as well as Sox9+ and insulin+ cells were manually scored from every tenth section of dorsal pancreata from E15.5 controls (R26Yfp/+; Foxa2iCre/+; Dll1+/+; Jag1+/+), Jag1ΔFoxa2, Dll1ΔFoxa2 and Jag1/Dll1ΔFoxa2 embryos and expressed relative to the area (in mm2) of the YFP+ dorsal pancreatic epithelium using FIJI.
Statistical analyses
The data were analyzed using Student’s t test using 2-tailed analysis (GraphPad), except where otherwise noted. All tests were unpaired except for Hes1 IF CTCF quantification, where a paired t test was used to compare Jag1ΔFoxa2 with control values collected from three separate pairs of mutants/littermate controls in three independent analyses. Data are presented as mean ± S.D. and the sample number (indicated in figure legends) was a minimum of three embryos per genotype, except for Jag1; Dll1ΔFoxa2 for which we so far have only been able to collect two embryos.
Author Contributions
P.A.S. and C.A.C designed, carried out experiments and wrote the manuscript. M.C.J. performed all staining, imaging and quantitative analysis of whole-mount specimens. K.H.L. contributed to phenotype analysis. I.I. and R.Ka. generated the Jag1J1VmC and Dll1D1VmC mouse lines. T.N. and R.Ko. generated the N1IP::CreHI; Rosa26LSL-Ai3 embryos. P.S. conceived the study, designed and interpreted experiments and wrote the manuscript. All authors revised and approved the manuscript.
Competing Interests
The authors declare no competing or financial interests.
Acknowledgements
We thank Ole D. Madsen and Jane E. Johnson for antibodies, Young-Yun Kong, Jorge Ferrer, Heiko Lickert, Chris V.E. Wright, Anne Grapin-Botton and Mark A. Magnuson for mouse lines, and Thi Nguyen for technical assistance. P.S. received grants from the Novo Nordisk Foundation (NNF16076 and NNF10717). The Novo Nordisk Foundation Center for Stem Cell Biology is supported by a Novo Nordisk Foundation grant number NNF17CC0027852.
Footnotes
↵† P.A.S. and C.A.C. are shared first authors.