Genetic evidence that Nkx2.2 acts primarily downstream of Neurog3 in pancreatic endocrine lineage development

Many pancreatic transcription factors that are essential for islet cell differentiation have been well characterized; however, because they are often expressed in several different cell populations, their functional hierarchy remains unclear. To parse out the spatiotemporal regulation of islet cell differentiation, we used a Neurog3-Cre allele to ablate Nkx2.2, one of the earliest and most broadly expressed islet transcription factors, specifically in the Neurog3+ endocrine progenitor lineage (Nkx2.2△endo). Remarkably, many essential components of the β cell transcriptional network that were down-regulated in the Nkx2.2KO mice, were maintained in the Nkx2.2△endo mice - yet the Nkx2.2△endo mice displayed defective β cell differentiation and recapitulated the Nkx2.2KO phenotype. This suggests that Nkx2.2 is not only required in the early pancreatic progenitors, but has additional essential activities within the endocrine progenitor population. Consistently, we demonstrate Nkx2.2 functions as an integral component of a modular regulatory program to correctly specify pancreatic islet cell fates. DOI: http://dx.doi.org/10.7554/eLife.20010.001


Introduction
The pancreas is an essential dual-functioning organ that contains an exocrine and endocrine compartment. The exocrine pancreas represents the majority of the pancreatic tissue and contains acinar cells that secrete digestive enzymes and the ductal cells through which the enzymes are transported into the duodenum. The endocrine pancreas only represents 1-2% of the organ, but is comprised of hormone-producing islet cells that are critical for the maintenance of glucose homeostasis. Interestingly, although these two compartments have highly disparate functions, they arise from a common multipotent progenitor population that gradually forms spatially and molecularly distinct progenitor subpopulations during development (Burlison et al., 2008;Offield et al., 1996;Pan and Wright, 2011;Zhou et al., 2007). A bi-potent duct/endocrine progenitor population is then derived from the multipotent progenitor population and subsequently differentiates into a highly restricted endocrine progenitor population that produces all islet cell types, including the functionally important insulin-producing b cell lineage.
The progressive restriction of progenitor-cell lineage potential to ultimately give rise to pancreatic islet cells is believed to be regulated by the sequential expression of several essential transcription factors. The multipotent progenitor population is defined by the expression of Pancreatic and duodenal homeobox 1 (Pdx1) and Pancreas transcription factor 1a (Ptf1a), transcriptional regulators that are essential for pancreatic formation; deletion of Pdx1 or Ptf1a leads to pancreatic agenesis (Burlison et al., 2008;Gu et al., 2003;Jonsson et al., 1994;Kawaguchi et al., 2002;Offield et al., 1996). A subset of cells from the Pdx1 + Ptf1a + multipotent progenitors gradually become restricted to the endocrine lineage with the expression of Neurogenin 3 (Neurog3), which delineates the endocrine progenitor population. Neurog3 is essential for the formation of all six embryonic pancreatic endocrine cell types: the insulin-producing b cells, glucagon-producing a cells, somatostatin-producing d cells, pancreatic polypeptide-producing PP cells, gastrin-producing G cells, and ghrelin-producing e cells (Gradwohl et al., 2000;Gu et al., 2002;Heller et al., 2005;Schwitzgebel et al., 2000;Suissa et al., 2013). Mice lacking Neurog3 almost completely fail to form these endocrine lineages (Gradwohl et al., 2000;Heller et al., 2005;Suissa et al., 2013).
Although it is well-established that b cells are derived from the Neurog3 + progenitor lineage, there is still some uncertainty about when and where the b cell lineage is specified during pancreas development. b cell fate may be influenced early within the Pdx1 + pancreatic progenitor to specify unipotent populations of Neurog3 + endocrine progenitor cells (Desgraz and Herrera, 2009) or it may be occurring within the Neurog3 + endocrine progenitor lineage after induction of Neurog3. Interestingly, Johansson et al. (2007) demonstrated that pancreatic progenitor cells transition through distinct competence windows in response to Neurog3 induction for each endocrine cell fate, with b cell competence being acquired during a window of development between embryonic day (E)10.5 and E16 (Johansson et al., 2007).
Many of the transcription factors that are essential for b cell specification have been identified, including Nkx2.2, Nkx6.1, Rfx6, Glis3, Insm1 and Neurod1; however, we still have an incomplete understanding of how these factors regulate the timing and mechanism of b cell fate induction since they are often expressed both early and throughout pancreas development and are not always restricted to the b cell lineage. For example, Nkx2.2, Nkx6.1 and Rfx6 are expressed in multipotent pancreatic progenitors and endocrine progenitors, in addition to their maintained expression in several of the mature endocrine lineages (Arnes et al., 2012b;Jørgensen et al., 2007;Pedersen et al., 2005;Soyer et al., 2010). Glis3 expression is first detected in the bipotent progenitor cells, but is then maintained in the preductal and endocrine progenitors, and ultimately restricted to the b, PP and ductal lineages (Kang et al., 2016). Insm1 is similarly expressed in the pancreatic pre-endocrine cells, endocrine progenitor cells and is maintained in all endocrine cell types (Gierl et al., 2006;Mellitzer et al., 2006;Osipovich et al., 2014). Lastly, Neurod1 can be detected as early as E9.5 in the early glucagon and ghrelin producing cells, but predominantly becomes restricted first to endocrine progenitors and then is maintained in all endocrine lineages (Anderson et al., 2009a;Naya et al., 1997). Null mutations of all six of these transcriptional regulators cause neonatal lethality likely due to their severe defects in b cell development (Gierl et al., 2006;Naya et al., 1997;Osipovich et al., 2014;Prado et al., 2004;Smith et al., 2010;Sussel et al., 1998); however, it is unknown exactly when each of these factors exerts their functional role in b cell fate specification.
To gain a better understanding of both the timing and mechanism of b cell fate induction we chose to focus on the essential islet transcription factor, Nkx2.2. Nkx2.2 is a critical regulator of appropriate islet cell lineage specification during pancreagenesis; mice carrying an Nkx2.2 null mutation form reduced numbers of a and b cells and instead form increased numbers of ghrelin-producing e cells due to defects in cell lineage specification (Prado et al., 2004;Sussel et al., 1998). Importantly, b cell fate specification is completely dependent on Nkx2.2 as no b cells are formed in mice carrying a null mutation in Nkx2.2 (Nkx2.2 KO ) (Sussel et al., 1998). Nkx2.2 is co-expressed with Pdx1 throughout the pancreatic progenitor domain, but its expression becomes restricted to the Neurog3 + endocrine progenitor population (Arnes et al., 2012b;Jørgensen et al., 2007;Sussel et al., 1998). Nkx2.2 also appears to function at or near the top of the regulatory cascade of transcription factors to specify b cell fate: The expression of Neurog3, Nkx6.1, Rfx6 and Neurod1 was significantly reduced in Nkx2.2 KO progenitor populations (Anderson et al., 2009a(Anderson et al., , 2009bChao et al., 2007) (Table 1), suggesting that Nkx2.2 functions within the pancreatic epithelium to influence the derivation of endocrine-committed cells that can enter the appropriate endocrine lineage pathways.
The early expression of Nkx2.2 within the multipotent progenitor population, upstream of Neu-rog3 prompted us to determine whether Nkx2.2 also functioned downstream of Neurog3 to influence islet cell specification. We used Neurog3-Cre; Nkx2.2 flox/flox (Nkx2.2 4endo ) mice to ablate Nkx2.2 specifically within the Neurog3 + endocrine progenitor lineage. Although Nkx2.2 is expressed upstream of Neurog3 and regulates Neurog3 expression (Anderson et al., 2009b;Sussel et al., 1998), deletion of Nkx2.2 using precise Neurog3-based floxed gene inactivation resulted in severe endocrine defects similar to the Nkx2.2 KO phenotype. Notably, unlike the Nkx2.2 KO mice, Neurog3 expression in the Nkx2.2 4endo mice was not affected. This suggests that Nkx2.2-mediated endocrine specification is established within the endocrine lineage, downstream of Neurog3 activation. Furthermore, unlike the Nkx2.2 KO mice, expression of several essential islet cell transcription factors, including Rfx6 and Neurod1, was unaffected. Co-occupancy analysis revealed significant overlap between Nkx2.2 direct targets and Rfx6 and Neurod1 bound genes. This indicates b cell formation may not be regulated by a simple linear transcriptional cascade that functions downstream of Nkx2.2, and implicates Nkx2.2 as an essential component of a modular b cell transcription network.

Results
Nkx2.2 is efficiently ablated in Nkx2.2 4endo mice Nkx2.2 is expressed at the onset of pancreas development throughout the multipotent progenitor domain, but subsequently becomes restricted to the Neurog3 + endocrine progenitor population (Jørgensen et al., 2007). Furthermore, null mutations in Nkx2.2 (Nkx2.2 KO ) lead to significantly reduced Neurog3 RNA expression, although there is no corresponding change in Neurog3 + cell numbers (Anderson et al., 2009b). To determine whether Nkx2.2 activity downstream of Neurog3 was also necessary for specific aspects of endocrine cell specification, Neurog3-Cre; Nkx2.2 flox/flox or Nkx2.2 flox/LacZ ; R26R-Tomato (hereafter referred to as Nkx2.2 4endo ) mice were generated to ablate Nkx2.2 specifically within the endocrine progenitor cells and their descendants (Madisen et al., 2010;Mastracci et al., 2013;Schonhoff et al., 2004). Efficient deletion of Nkx2.2 in the Nkx2.2 4endo mice was demonstrated by a reduction in Nkx2.2 mRNA expression at E15.5 and E18.5 ( Figure 1A). Residual Nkx2.2 gene expression at E15.5 likely reflects the Nkx2.2 expression in pancreatic progenitor cells that are still present at this stage of development. Consistent with the decrease in Nkx2.2 gene expression, Nkx2.2 protein was almost undetectable in the pancreas at E15.5 and postnatal day (P)0 ( Figure 1B,C, Figure 1-figure supplement 1). Activation of the R26R-Tomato allele was also used to monitor accurate Cre activity within the Nkx2.2 4endo mice and, accordingly, Tomato + (Tom + ) cells were found specifically within the endocrine compartment ( Figure     Deletion of Nkx2.2 did not appear to affect the earlier cell populations that give rise to endocrine progenitors since the expression of Sox9, an early progenitor marker that is expressed immediately upstream of Neurog3 in the bipotent progenitor pool from which endocrine cells are derived, was unaffected ( Figure 1F). Consistent with specific deletion of Nkx2.2 from the endocrine progenitor population, there were also no obvious changes in the exocrine lineage, as indicated by amylase expression (Figure 1-figure supplement 1). Expression of pan-endocrine genes, such as Chromog-raninA (Chga), Isl1, and Pax6 was also unchanged in the absence of Nkx2.2 at E15.5, and the area and number of Tomato-labeled endocrine lineage cells was also unaffected in the Nkx2.2 4endo mice at E15.5 and P0, suggesting endocrine compartment size was not decreased when Nkx2.2 was deleted from the endocrine progenitor cell ( Figure 1B Nkx2.2 4endo mice display similar endocrine defects as Nkx2.2 KO mice Since Nkx2.2 is necessary for the establishment of proper endocrine cell ratios, we assessed the status of endocrine cell differentiation in the Nkx2.2 4endo mice during the course of pancreas development. Similar to the Nkx2.2 KO mice (Prado et al., 2004;Sussel et al., 1998), Nkx2.2 4endo mice displayed a reduction of insulin + (Ins + ) cells and glucagon + (Gcg + ) cells, and increased ghrelin + (Ghrl + ) cells at E15.5 and P0 ( , which is possibly due to the inability of the Neurog3-Cre allele to completely inactivate Nkx2.2. Gene expression at E15.5 and E18.5 corresponded to changes in endocrine cell numbers. Insulin (Ins1 and Ins2) and glucagon expression were significantly decreased; there was increased ghrelin expression, and no change in somatostatin gene expression at E15.5 and E18.5 ( Figure 2M). Gast gene expression was also decreased at E15.5, but Gast expression was extinguished from both the wild type and mutant pancreata by E18.5 ( Figure 2M), as previously observed (Suissa et al., 2013). Similar to the Nkx2.2 KO mice, the Nkx2.2 4endo mice die shortly after birth between P2 and P4 ( Figure 2N).
Interestingly, we observed increased pancreatic polypeptide (Ppy) expression at E15.5 and E18.5 in Nkx2.2 4endo embryos ( Figure 2M). Although this was in contrast to previous studies in which we had observed decreased Ppy gene expression and PP + cell numbers (Chao et al., 2007;Prado et al., 2004;Sussel et al., 1998), we confirmed that there was similar upregulation of Ppy in the Nkx2.2 LacZ/LacZ (Nkx2.2 KO ) pancreas (Arnes et al., 2012b) compared to the Nkx2.2 4endo pancreas, although the magnitude of upregulation was greater in the Nkx2.2 4endo embryos ( Figure 2M, O). In Nkx2.2 4endo mice, upregulation of Ppy RNA correlated with changes in protein; immunostaining revealed increased PP + cells at both E15.5 and P0 (Figure 2-figure supplement 2). While some Ins + PP + cells could be observed (arrow in Figure 2-figure supplement 2), there was no significant difference between Nkx2.2 4endo mice and controls (Figure 2-figure supplement 2), suggesting that similar to the ghrelin-expressing cells, whose increased numbers were due to altered e cell specification at the expense of b cell specification, the excess PP cells did not arise from misexpression of PP in b cells or as a result of b cell reprogramming. We speculate that the novel Ppy phenotype in the Nkx2.2 4endo and Nkx2.2 KO mice may result from a change in mouse strain background (see Discussion).
Nkx2.2 4endo and Nkx2.2 KO mice share important gene expression changes It was somewhat unexpected that the restricted deletion of Nkx2.2 in the Neurog3 + progenitor cells resulted in a phenotype that was remarkably similar to the Nkx2.2 KO mice, which lacked Nkx2.2 throughout pancreas development. To more precisely determine the similarities and differences between the Nkx2.2 4endo and Nkx2.2 KO mice during the process of endocrine cell differentiation, we performed RNA-Seq analysis on whole pancreata derived from Nkx2.2 4endo embryos, Nkx2.2 KO embryos, and their respective littermate controls at E15.5, a time of relatively high Neurog3 + cell abundance. Consistent with their similar phenotypes, there was significant overlap (120 genes, p-value <2.2e-16, Figure 3A) between the differentially expressed genes (adjusted p-value <0.05), including genes encoding pancreatic hormones ( Figure 3B) and important a and b cell regulatory  Table 1). Furthermore, several transcription factor genes expressed in endocrine progenitors and involved in b cell specification, such as Nkx6.1, Pax4, and Pdx1, that were unchanged Nkx2.2 KO embryos at E15.5 were also not affected in Nkx2.2 4endo pancreas at this early time point (Figure 3-figure supplement 1; Table 1).

Nkx2.2 direct target genes are simultaneously bound by Rfx6 and/or Neurod1
The finding that b cell specification was disrupted, even though Neurog3, Neurod1 and Rfx6 expression remained intact, could indicate that Nkx2.2 functions as an obligate co-factor for these proteins in the regulation of b cell specification. This is supported by our previous observation that Nkx2.2 cooperated with Neurog3 to regulate the expression of Neurod1 (Anderson et al., 2009a). To determine whether there was a general requirement for a cooperative role for Nkx2.2 within the known b cell specification regulatory pathways, we assessed whether Nkx2.2 co-regulated the b cell program with Rfx6 and Neurod1. Nkx2.2 bound and regulated genes were identified by crossreferencing the differentially expressed genes in the Nkx2.2 4endo RNA-Seq (adjusted p-value <0.05, Figure 3) with an Nkx2.2 ChIP-Seq performed on the MIN6 b cell line (GSE79725) (Gutiérrez et al., 2017). Notably, the majority of the genes regulated by Nkx2.2 appeared to be primary targets: 169/ 175 of the significantly regulated genes were bound by Nkx2.2 within 100 kb of of their respective transcriptional start sites (TSS) ( Figure 3A; Figure 4A,C). We then compared the Nkx2.2 direct targets to genes bound by either Rfx6 (GSE62844; 9406 genes) or Neurod1 (GSE30298; 5003 genes) (Piccand et al., 2014) (Tennant et al., 2013). This analysis revealed significant overlap of Rfx6 binding near Nkx2.2 target genes (120 genes, p-value <2.2e-16, Figure 4A) and gene ontology (GO) analysis indicated significant representation of genes related to hormone secretion ( Figure 4B). Analysis of Neurod1 bound genes revealed similar overlap with Nkx2.2 targets (92 genes, p-value <2.2e-16, Figure 4C), and GO analysis identified significant representation of insulin secretion genes ( Figure 4D). Furthermore, greater than 75% of the 92 genes simultaneously bound by Nkx2.2 and Neurod1 in the same or nearby regulatory regions were also bound by Rfx6 (72 genes, Figure 4E), demonstrating novel regulatory co-operation between these three transcription factors.
Although it is possible that at some genomic loci the Nkx2.2/Neurod1/Rfx6 co-regulation of important b cell genes could be due to direct protein-protein interactions, the respective binding sites were often not closely clustered. Inspection of the Nkx2.2, Rfx6, and Neurod1 ChIP-Seq data revealed a variety of co-occupancy patterns in the regulatory elements associated with several candidate Nkx2.2-regulated genes ( Figure 4F). For example, all three transcription factors showed overlapping binding peaks at the Mlxipl locus ( Figure 4G). However, at a region upstream of Mafa, only Nkx2.2 and Neurod1 were present ( Figure 4H); whereas Nkx2.2 and Rfx6 co-bound at the Gast locus ( Figure 4I). In addition, Nkx2.2, Rfx6, and Neurod1 all bound near G6pc2; however, these peaks were present at distinct locations flanking the gene ( Figure 4J). These results suggest that while Nkx2.2, Rfx6, and Neurod1 are individually critical for the regulation of the islet cell program, it is their cooperative function that may be required for the appropriate activation of a number of essential b cell genes.

Discussion
The spatiotemporal progression of islet cell specification is not well understood. While many transcription factors that are essential for this process are well-characterized, their sustained expression throughout development and in several different cell types has made it challenging to identify the timing of their function. Nkx2.2 is expressed within the Pdx1 + pancreatic multipotent progenitor population and its expression is maintained throughout pancreas development, gradually becoming restricted to the endocrine cell lineages (Arnes et al., 2012b;Jørgensen et al., 2007;Sussel et al., 1998). Global deletion of Nkx2.2 causes the reduction of many essential regulatory factors, including Neurog3, Neurod1 and Rfx6 (Anderson et al., 2009a(Anderson et al., , 2009bChao et al., 2007) (Table 1) and misspecification of all endocrine lineages, except the somatostatin-expressing d cells. The fact that these mice completely fail to form b cells, combined with the reduced expression of many b cell regulatory factors provided a strong argument for positioning Nkx2.2 at or near the top of the hierarchy for regulation of b cell specification. In this study, we ablated Nkx2.2 specifically within the endocrine progenitor lineage to determine whether Nkx2.2 also functions downstream of Neurog3 in regulating islet cell specification. Surprisingly, we found similar endocrine lineage changes in Nkx2.2 4endo mice compared to Nkx2.2 KO mice, including the failure to form b cells (Sussel et al., 1998) (Schematized in Figure 5). However, unlike the Nkx2.2 KO mice, expression of the essential b cell transcription factors Neurog3, Rfx6, and Neurod1 was unaffected, suggesting that although Nkx2.2 regulates these factors in the pancreatic progenitor population, reduction of Neurog3, Neu-rod1 and Rfx6 expression is not an obligate step in the mis-specification of islet fates. This may indicate that although Neurog3, Neurod1 and Rfx6 are essential for specifying the b cell program, Nkx2.2 activity must be preserved to facilitate the functional integrity of the b cell regulatory pathway ( Figure 5). Furthermore, co-occupancy analysis revealed significant overlap of Nkx2.2 direct targets with Rfx6 and Neurod1 binding, suggesting cooperative transcriptional regulation of these factors.
Given its early expression in the multipotent progenitor population, it was unexpected that deletion of Nkx2.2 from the Neurog3 + cells was sufficient to recapitulate the Nkx2.2 KO phenotype.   However, it is consistent with a recent study showing that Neurog3-Cre; Rfx6 flox/flox mice display b cell loss similar to the Rfx6 null mice (Piccand et al., 2014). Similarly, Nkx6.1 functions downstream of Neurog3, with Neurog3-Cre; Nkx6.1 flox/flox mice displaying defective b cell differentiation (Schaffer et al., 2013). These studies provide evidence that the endocrine progenitor cell, downstream of Neurog3 induction, is the primary site of islet cell specification. However, it is still possible that preprogramming of this population does occur, but that the respective programs can be overwritten once the cell has formed. The importance of Nkx2.2 function downstream of Neurog3 expression is also consistent with the timing of NKX2.2 in human islets; unlike mice, NKX2.2 is only first detected after the appearance of NEUROG3 progenitors (Jennings et al., 2013). However, while Nkx2.2 function within the mouse Pdx1 + pancreatic progenitor was not sufficient to allow correct islet cell specification and b cell formation, further studies would be required to determine whether Nkx2.2 has additional functions at this earlier stage of pancreas development.
Although deletion of Nkx2.2 in the endocrine progenitor population appears to grossly phenocopy the Nkx2.2 KO phenotype, the phenotypes were not as severe as observed in the Nkx2.2 KO mice. Although this may be due to the efficiency of Cre-mediated deletion of the Nkx2.2 allele, it could also suggest that there are partially redundant functions of Nkx2.2 in the pancreas progenitor and endocrine progenitor populations. In addition, there were over 300 genes significantly altered in Nkx2.2 KO pancreas that were unchanged in the Nkx2.2 4endo mice ( Figure 3A; Table 4). Again, this could be due to the more penetrant phenotype associated with the Nkx2.2 KO allele or could suggest that Nkx2.2 may have additional non-redundant functions upstream of Neurog3. Future studies in which we are able to conditionally re-express Nkx2.2 in the Neurog3 population in an Nkx2.2 KO background, similar to the Neurog3 replacement studies by Johansson et al. (2007), would possibly reveal these functions. There were also a small number of genes whose expression only appeared to be changed in the Nkx2.2 4endo mice ( Figure 3A, Table 3). This is more difficult to explain unless The following source data is available for figure 4: Source data 1. Nkx2.2 bound and regulated genes that are also Rfx6 bound (120 genes) or NeuroD1 bound (92 genes  there are some novel, compensatory gene changes that can occur in the remaining cell populations. Assessment of the identity of the differentially affected genes in either the Nkx2.2 KO or Nkx2.2 4endo mice did not reveal any obvious mechanistic insights. Several previous studies of the Nkx2.2 KO mice demonstrated that Nkx2.2 is essential for the correct specification of all endocrine cell populations, except d cells (Arnes et al., 2012b;Mastracci et al., 2013Mastracci et al., , 2011Prado et al., 2004;Sussel et al., 1998). These earlier studies also revealed that in the absence of Nkx2.2 there was mis-specification of the endocrine lineages, rather than reprogramming of the endocrine populations following their formation. Alternatively, deletion of the conserved tinman (TN) domain of Nkx2.2 resulted in both aberrant endocrine cell specification and reprogramming of b cells to the a cell fate (Papizan et al., 2011). Furthermore, recent studies in which Nkx2.2 was deleted from b cells after their specification resulted in b cell reprogramming into alternative endocrine lineages and the appearance of islet cells expressing more than one hormone (Gutiérrez et al., 2017). Interestingly, the phenotype of the Nkx2.2 4endo mice most closely resembles the Nkx2.2 KO mice, and we do not see evidence of reprogramming or multi-hormonal expressing (transient or stable) endocrine cells in the mutant islets. However, since we are unable to genetically lineage trace the a or b cell populations in the Nkx2.2 4endo mice, we cannot completely rule out the possibility that reprogramming of the a or b cells contributes to the elevated number of e or PP cells.
Comparison of Nkx2.2 direct targets with Rfx6 and Neurod1 binding revealed co-occupancy of these factors in the same or nearby enhancer elements located within proximity of an endocrinerelated gene. While ChIP-Seq analyses can reveal novel insights into the transcriptional regulation of cell fate, these analyses were performed on chromatin isolated from the immortalized MIN6 b cell line in order to isolate sufficient material for the assay. It is important to note that although we only focused on potential gene targets that were dysregulated in the Nkx2.2 4endo pancreas, transcription factor binding in MIN6 cells may not accurately reflect binding in embryonic endocrine progenitor cells. Furthermore, our population-based transcriptome analysis is limited by the cell type heterogeneity present in the E15.5 pancreas, particularly when gene changes that are present in both progenitor populations and downstream sub-lineages may be represented. Future studies will benefit from single-cell RNA-Seq and ChIP-Seq analysis that will allow us to parse out the differential gene changes and transcription factor binding in the Nkx2.2-deficient endocrine progenitor cells.
We were surprised to observe an increase in the PP population of Nkx2.2 4endo mice. Since reevaluation of Ppy expression in the Nkx2.2 KO mice revealed a similar phenotype, the increase in PP is likely not due to a unique function of Nkx2.2 in the pancreatic progenitor to specify PP cells. The best explanation for the discrepant PP phenotype is that we moved the Nkx2.2 null allele to a C57Bl6/J background for this study. The upregulation of PP is also interesting since it parallels the increase we observed with e cells. Consistent with the observation that a subset of the PP population is derived from Ghrl-expressing e cells (Arnes et al., 2012a), PP + Ghrl + cells can be infrequently found in Nkx2.2 4endo mice at P0 (data not shown). Furthermore, we observe an upregulation of Arx expression, which has also previously been shown to induce enhanced PP cell formation (Collombat et al., 2007). We are unable to determine whether these altered islet cell ratios are similarly due to b cell transdifferentiation events since the Neurog3-Cre labels all islet cell lineages.
Similar to many other developmental pathways, b cell specification involves the function of numerous transcription factors, some of which are expressed early and throughout tissue formation. Our results identify a novel critical window for Nkx2.2-mediated regulation of islet cell specification within the Neurog3 lineage following induction of Neurog3. Furthermore, although Nkx2.2 appears to function near the top of the islet transcriptional hierarchy, we demonstrate that maintained Nkx2.2 activity is also a necessary component of the regulatory networks functioning downstream of Neurog3 to induce b cell specification. Our discovery that different combinations of the set of transcription factors analyzed in this study directly regulate important genes in the b cell pathways raises important questions about the complex nature of cell-specific gene regulation that may have important implications for b cell induction, maturation and function. Furthermore, while previous research has focused on the initial expression of b cell factors, future studies may benefit by ensuring the prolonged and combinatorial expression of these factors to recapitulate in vivo development and allow cooperative induction of target genes.

Immunohistochemistry Fixation
Samples were fixed in 4% paraformaldehyde (PFA). E15.5 embryos were fixed for 3 hr (h) at 4˚C, and P0 neonates were fixed overnight at 4˚C. Samples were washed in phosphate-buffered saline (PBS), and then transferred to 30% sucrose/PBS overnight at 4˚C. Samples were embedded and frozen in optimum cutting temperature (O.C.T.) and stored at À80˚C. 8 mm sections were taken and stored at À80˚C.

Staining
Sections were washed in PBS and 0.1% Triton/PBS (PBST) and blocked in 2% donkey serum (DS)/ PBST for 30 min (min) at room temperature (RT). The primary antibody was diluted in 2% DS/PBST and incubated on sections overnight at 4˚C. Sections were washed in PBS and PBST and incubated 2-3 hr at RT with secondary antibody (Jackson Immuno Research, West Grove, PA) diluted in 2% DS/PBST at 1:500. Sections were washed in PBS and PBST and incubated with DAPI (Invitrogen TD21490) diluted in PBS at 1:1000 for 15 min at RT. Sections were washed in PBS and mounted with fluorescent mounting medium (DAKO Cat# S3023; Agilent Technologies, Santa Barbara, CA).

Microscopy
Images were taken using a Zeiss Confocal LSM 710 microscope and processed with Zen, ImageJ, and Adobe Photoshop software.

Cell quantification
For each neonate, every twentieth section was analyzed for insulin + , pancreatic polypeptide + , glucagon + , somatostatin + , ghrelin + , and Tomato + cell quantification throughout the entire pancreas. Tiled images were taken at 20x objective with a Leica fluorescent microscope (DM5500/MZ16F) and processed with LAF software. Cells were counted and normalized to the total number of Tomato + cells or total pancreatic area measured in arbitrary units using ImageJ software. An n ! 3 biological replicates (individual pancreata from different animals) were used. All values are expressed as mean ± SEM. Statistical analysis was performed using two-tailed student's unpaired t-test since independent animals were used in the analysis. Significance was achieved with p-value <0.05.

RNA analysis RNA extraction
Whole pancreata were dissected in PBS and transferred to RNAlater (Ambion AM7021). Pancreata were incubated at 4˚C overnight and stored at À20˚C until RNA extraction. Pancreata were homogenized and total RNA was isolated following the Qiagen RNeasy Mini kit.
Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis A normalized starting quantity of RNA (200 ng-1 mg) isolated from whole pancreas and random hexamers were used to synthesize cDNA following the SuperScript III reverse transcriptase protocol (Invitrogen 18080-044). qRT-PCR was performed using 200 ng cDNA and master mix (Eurogentec) and analyzed using the BioRad CFX96 Real-Time System. Genes were normalized to CyclophilinB (probe 5': tggtacggaaggtggag, forward primer 5': gcaaagttctagagggcatgga, reverse primer 5': cccggctgtctgtcttggt) and to littermate controls. An n ! 3 biological replicates (individual pancreata from different animals) were used at all ages; technical replicates were also completed in triplicate during qRT-PCR plate preparation. All values are expressed as mean ± SEM. Statistical analysis was performed using two-tailed student's unpaired t-test since independent animals were used in the analysis. Significance was achieved with p-value <0.05.

RNA-sequencing (RNA-Seq)
RNA was isolated from whole pancreata at E15.5 from Nkx2.2 4endo (n = 3 biological replicates), Nkx2.2 KO (n = 3 biological replicates), and respective littermate control embryos (n = 3 biological replicates for each mutant). Total RNA concentration and quality were measured using the Agilent Bioanalyzer RNA Nano chip. Library preparation was performed using Illumina TruSeq RNA prep kit. Libraries were then sequenced using Illumina HiSeq2000 at the Columbia Genome Center. RNA-Seq was performed with 30 million single end 100 bp reads. RTA (Illumina) was used for base calling and CASAVA (version 1.8.2) was used for converting BCL to fastq format, coupled with adaptor trimming. Reads were mapped to a reference genome (Mouse: UCSC/mm9) using Tophat (Trapnell et al., 2009) (version 2.0.4) with four mismatches (-read-mismatches = 4) and 10 maximum multiple hits (-max-multihits = 10). Relative abundance (expression level) of genes and splice isoforms was determined using cufflinks (Trapnell et al., 2010) (version 2.0.2) with default settings. Differentially expressed genes were calculated using Deseq (Anders and Huber, 2010). Dataset overlap was determined with a Fisher's Exact test; significance was achieved with p-value <0.05. Data is available at GSE80444: reviewer link -(http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? token=izyriugqnbmdpmp&acc=GSE80444). implementation and interpretation of the data, Co-wrote the manuscript with AJC, Conception and design, Analysis and interpretation of data, Drafting or revising the article Author ORCIDs Lori Sussel, http://orcid.org/0000-0003-2832-1319 Ethics Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (AC-AAAG3206) of Columbia University. Every effort was made to minimize suffering.

Major datasets
The following datasets were generated: