Defining Pancreatic Endocrine Precursors and Their Descendants Peter White,1 Catherine Lee May,1,2 Rodrigo N

Home , CIZ1, CUTL1, DCTN2, DDEF2, DPF2, E2F5

ORIGINAL ARTICLE Defining Pancreatic Endocrine Precursors and Their Descendants Peter White,1 Catherine Lee May,1,2 Rodrigo N. Lamounier,1 John E. Brestelli,1 and Klaus H. Kaestner1

OBJECTIVE—The global incidence of diabetes continues to The Edmonton cadaveric islets transplantation protocol increase. Cell replacement therapy and islet transplantation offer (4) opened the door to improved treatment of the disease hope, especially for severely affected patients. Efforts to differ- but faces two significant challenges: the need for improved entiate insulin-producing ␤-cells from progenitor or stem cells immuno-regulatory measures and the necessity to increase require knowledge of the transcriptional programs that regulate the supply of islets or ␤-cells by a factor of at least 1,000 to the development of the endocrine pancreas. be able to treat even the most severely affected patients. RESEARCH DESIGN AND METHODS—Differentiation to- While we have gained significant insights into the tran- ward the endocrine lineage is dependent on the transcription scriptional programs and signaling mechanisms that con- factor Neurogenin 3 (Neurog3, Ngn3). We utilize a Neurog3– trol the differentiation of endocrine precursors to mature enhanced green fluorescent protein knock-in mouse model to hormone-expressing endocrine cells of the islet (rev. in isolate endocrine progenitor cells from embryonic pancreata 5–9), efforts to direct differentiation of various cells into (embryonic day [E]13.5 through E17.5). Using advanced genomic ␤ approaches, we generate a comprehensive gene expression pro- -cells have met with only limited success (2). file of these progenitors and their immediate descendants. Differentiation of the pancreatic endocrine lineage is dependent on Neurogenin 3 (Neurog3, Ngn3), a member of RESULTS—A total of 1,029 genes were identified as being the family of basic helix-loop-helix (bHLH) transcription temporally regulated in the endocrine lineage during fetal devel- factors. Neurog3-null pancreata lack all five mature endo- opment, 237 of which are transcriptional regulators. Through crine cell types (10,11), and Neurog3 is also required for pathway analysis, we have modeled regulatory networks involv- ing these proteins that highlight the complex transcriptional enteroendocrine cell development in the stomach and hierarchy governing endocrine differentiation. intestine (12,13). Notably, regulatory genes marking endocrine precursors, including Pax4, Pax6, Isl1, and Neurod1 CONCLUSIONS—We have been able to accurately capture the (14,15), are not expressed in the absence of Neurog3. gene expression profile of the pancreatic endocrine progenitors Ectopic expression of Neurog3 can initiate differentiation and their descendants. The list of temporally regulated genes of endocrine cells in mouse (16), humans (17), and pig identified in fetal endocrine precursors and their immediate descendants provides a novel and important resource for devel- (18). Furthermore, lineage tracing during mouse embryo- opmental biologists and diabetes researchers alike. Diabetes genesis has revealed that Neurog3-positive cells differen- 57:654–668, 2008 tiate exclusively into islet cells, suggesting that expression of this gene could be used as a marker for isolating these progenitors for further study (19). During pancreatic development in the mouse, expres- ignificant efforts to treat and potentially cure sion of Neurog3 is initiated during pancreatic budding, as diabetes have been focused on generating renew- early as embryonic day (E)9.5 in the dorsal primordium able sources of insulin-producing ␤-cells from (20,21). Relatively low levels of Neurog3 expression are their progenitors to be used in transplantation maintained until E13.5, when a dramatic upregulation is S observed, marking the beginning of the second transition (1–3). This effort is motivated by the increased incidence of both type 1 and type 2 diabetes and the limited (16). Expression is believed to peak at E15.5 and then effectiveness of pharmaceutical treatments for the disease. rapidly decrease to undetectable levels in juvenile and adult islets (10). Different transcriptional regulators par- ticipate in the activation of Neurog3, including Onecut1 From the 1Department of Genetics and Institute for Diabetes, Obesity and (Hnf6) as a direct activator of its transcription and mem- Metabolism, University of Pennsylvania School of Medicine, Philadelphia, bers of the FoxA family (7,22,23). Expression of Neurog3 Pennsylvania; and the 2Department of Pathology, The Children’s Hospital of Philadelphia Abramson Research Center, University of Pennsylvania School is restricted to a relatively small population of cells of Medicine, Philadelphia, Pennsylvania. destined for the endocrine lineages (19). After differentia- Address correspondence and reprint requests to Klaus H. Kaestner, PhD, tion into hormone-expressing cells has occurred, expres- University of Pennsylvania, Medical School, Department of Genetics, 415 Curie Blvd., Philadelphia, PA 19104. E-mail: [email protected]. sion of Neurog3 ceases. This is most likely due to negative Received for publication 26 September 2007 and accepted in revised form feedback, as Neurog3 has been shown to repress its own 30 November 2007. expression (24). Homeodomain factors such as Pax4 (25) Published ahead of print at http://diabetes.diabetesjournals.org on 10 De- cember 2007. DOI: 10.2337/db07-1362. and Nkx2–2 (26) as well as the bHLH factor NeuroD1 (27) Additional information for this article can be found in an online appendix at are targets for Neurog3 activation. However, the majority http://dx.doi.org/10.2337/db07-1362. of downstream genes dependent on Neurog3 and the bHLH, basic helix-loop-helix; EDGE, Extraction of Differential Gene Ex- mechanisms through which this factor regulates differen- pression; EGFP, enhanced green fluorescent protein; GO, gene ontology; HLH, helix-loop-helix; qRT-PCR, quantitative real-time RT-PCR. tiation of endocrine precursors into islet cells are un- © 2008 by the American Diabetes Association. known. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance In the present study we set out to identify potential with 18 U.S.C. Section 1734 solely to indicate this fact. novel targets of Neurog3 and to determine the transcrip-

654 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES tome of the endocrine lineage during fetal development. incubated in a Corning Hybridization chamber (Corning Incorporated Life Taking advantage of the fact that enhanced green fluores- Sciences) overnight at 42°C. The Mouse PancChip 6 contains ϳ13,000 mouse cent protein (EGFP) persists in cells for several days after cDNAs chosen for their expression in various stages of pancreatic development, many of which are not found on commercially available arrays. Detailed the promoter driving its expression has been turned off, information on this array and full protocols are available at http://www.cbi- we were able to sort Neurog3-EGFP endocrine precursors l.upenn.edu/EPConDB. After hybridization, the coverslips were removed in and their immediate descendants from fetal pancreas. By 2ϫ SSC, 0.1% SDS, and the arrays were washed for 5 min in 0.2ϫ SSC, 0.1% determining the gene expression profile of these develop- SDS at 42°C and 5 min in 0.2ϫ SSC at room temperature. The arrays were ing cell populations, we derived gene signatures that can immediately scanned with the Agilent DNA Microarray Scanner, Model be used to guide future efforts to enhance differentiation of G2565B (Agilent Technologies) at a resolution of 5 ␮mol/l. Data processing. The median Cy5 (red) and Cy3 (green) intensities of each endocrine precursors. element on the array were determined by processing the array images with GenePix Pro version 5.1 (Molecular Devices). All subsequent steps were performed using scripts we developed for this analysis in the R open-source RESEARCH DESIGN AND METHODS language environment for statistical computing (http://www.r-project.org/). Heterozygous male mice containing an EGFP-marked null allele of Neurog3 Positive control elements were removed from the dataset, and the expression (Neurog3ϩ/EGFP) (12) were mated with CD1 females and pregnant females ratio for each element on the array was calculated in terms of M [log2(red/ killed at either 12.5, 13.5, 14.5, 15.5, 16.5, or 17.5 days of gestation. Embryos green)] and A {[log2(red) ϩ log2(Green)]/2}, without local background signal were rapidly dissected and the pancreata removed and placed into PBS. subtraction. The data were normalized by the print tip loess method using the Neurog3ϩ/EGFP embryos (at the expected 50% Mendelian ratio) were easily BioConductor package “marray” (31). Quality control diagnostic plots were identified by their green fluorescence and transferred into a separate buffer prepared for each array, and those failing to exhibit high-quality hybridiza- ready for dissociation into single cells. A total of 70 pregnant females, tions were excluded from further analysis. This resulted in the final dataset producing 785 embryos, were required to produce three biological replicates containing three biological replicates for each embryonic time point and four for each time point. Adult islets were prepared from eight CD1 female mice as biological replicates for the adult islet time point, giving a total of 19 samples described previously (28). Islet preparations from two animals were pooled in the time course. for RNA isolation using the RNeasy kit (Qiagen), giving a total of four Quantitative real-time RT-PCR. Gene expression profiles were confirmed biological replicate samples for expression analysis. using quantitative real-time RT-PCR (qRT-PCR). cDNA was prepared from Preparation of single-cell suspensions and flow cytometry. EGFPϩ each replicate at the five developmental time points, from the adult islet pancreata were separated from EGFPϪ pancreata and pooled in 500–1,000 ␮l samples, and from E14.5 and E16.5 EGFPϪ cells that were collected along prewarmed Trypsin at 37°C in a standard scintillation vial. The number of with the EGFPϩ cells. cDNA was synthesized from ϳ20 ng total RNA using the pancreata needed to form a biological replicate depended upon the develop- WT-Ovation RNA Amplification System (NuGEN Technologies). PCRmixes mental stage and the number of Neurog3ϩ/EGFP embryos harvested. On were assembled using the Brilliant SYBR Green qRT-PCR Master Mix (Strat- average 25 embryos from six females were required for each E13.5 and E14.5 agene) according to the manufacturer’s instructions. Reactions were per- replicate, 19 embryos from five females were required for each E15.5 and formed using the SYBR Green (with dissociation curve) program on the E16.5 replicate, and 12 embryos from three females were required for each Mx3000 Multiplex Quantitative PCR System (Stratagene). Cycling parameters E17.5 replicate. The pool of pancreata was minced in the trypsin solution by were 95°C for 10 min and then 40 cycles of 95°C (30 s), 60°C (1 min), and 72°C repeatedly chopping the samples using very fine surgical scissors (Fine (30 s) followed by a melting curve analysis. All reactions were performed with Science Tools) for 1–2 min. A small stir-bar was added to the vial, and the three biological replicates and three technical replicates with reference dye sample was incubated at 37°C on a stir plate at low speed for 7 to 15 min, until normalization. Actb, Hprt, Gapdh, Tbp, and Ubc were tested for their no clumps of cells were visible. The disassociation was terminated with the suitability as a housekeeping gene using the geNorm analysis package (32). addition of an equal volume of RPMI-1640 containing 10% FBS, transferred to Expression of Actb was shown to be unstable across the samples being a 5-ml polystyrene round-bottom Falcon tube through a 35-␮m nylon mesh cell analyzed and as such was not included in the qRT-PCR normalization set. The strainer cap (BD Biosciences), and placed immediately on ice ready for median cycle threshold (CT) value was used for analysis, and all CT values sorting. were normalized to expression of four housekeeping genes: Tbp, Hprt, Ubc, Neurog3ϩ/EGFP-positive and -negative cells were sorted by the University of and Gapdh. Primer sequences are available from the authors upon request. Pennsylvania Flow Cytometry and Cell Sorting Resource Laboratory using the Statistical analysis. All statistical analysis of the array data were performed FACSVantage SE (BD Biosciences). Cells were gated for viability and nonag- using the normalized M values for all noncontrol elements on the array. This gregates to achieve a high-purity sort. Both positive and negative cells were M value represents the ratio of the test sample over the reference sample. separated and collected directly into sterile 1.5-ml microcentrifuge tubes Identification of temporally regulated genes was performed using the R-based containing 500 ␮l of the denaturing buffer RLT from the RNeasy mini kit application Extraction of Differential Gene Expression (EDGE version (Qiagen). No more than 70,000 Neurog3ϩ/EGFP-positive cells were collected to 1.1.291) (33), which uses the newly developed Optimal Discovery Procedure prevent over-dilution of the denaturing buffer used for the RNA extraction. (ODP) statistical theory (34). For direct comparisons between any two Once the sort was complete, samples were snap frozen in liquid nitrogen and conditions, EDGE was implemented with static settings to identify differential stored at Ϫ80°C for further processing. expression. Clustering and principal component analysis was performed using Probe preparation and microarray hybridization. Total RNA was pre- GeneSpring GX (Agilent Technologies). pared from the sorted cells or adult islets using the RNeasy Mini Kit (Qiagen) Time-course analyses of microarray data are considerably more difficult and eluted in water. The quality of each RNA sample was assessed using either than identification of differentially expressed genes in a two-state comparison, the RNA 6000 Pico or Nano LabChip Kit with a 2100 Bioanalyzer (Agilent and there are limited tools available to achieve this. Of the available statistical Technologies). Only samples of high RNA quality with a 28S-to-18S RNA ratio tools, four were compared for their ability to identify temporally regulated Ͼ2 and with RNA integrity number Ͼ7 were used. genes: ANOVA (GeneSpring GX), SAM (35), PaGE (36), and EDGE (33). It was Approximately 20 ng total RNA from each sample was used for labeling and very clear to us that EDGE was the superior method for identification of hybridization. Samples were labeled using the Ovation Aminoallyl RNA temporally regulated genes in this time course (data not shown). EDGE was Amplification and Labeling System (NuGEN Technologies), which uses the executed with 1,000 iterations, using a natural cubic spline, with a dimen- rapid and sensitive Ribo-SPIA RNA amplification process. After amplification, sional basis for the spline of 3. Analysis of the P value versus q value plot and the reference sample was created by pooling amplified cDNA in equal aliquots the q value versus the number of significant tests plot clearly indicated that a of each of the samples used in the hybridization. To analyze changes in gene P value cutoff of 0.05, corresponding to a q value of 0.23, provided an ideal expression occurring over time, a “reference” experimental design was used point to return the maximum number of significant genes with fewest for the microarray analysis. Cyanine dyes (GE Healthcare Bio-Sciences) were false-positives (Supplementary Data Fig. 1 [available in an online appendix at chemically attached (coupled) to 2 ␮g amplified cDNA in 50 mmol/l sodium http://dx.doi.org/10.2337/db07-1362]). A total of 1,170 distinct genes were bicarbonate, pH 8.5, in the dark at room temperature for 1 h. All test samples returned with a P value Յ 0.05. This list was filtered to remove any elements were coupled to Cy5 (red), whereas the reference sample was coupled to Cy5 whose fold change was Ͻ1.2 between any two time points, giving a final list of (green). Each test sample was combined with an equal amount of the 1,029 genes. reference sample and purified using the MinElute Reaction Cleanup Kit The complete list of targets organized into these broad functional groups (Qiagen). The eluted sample of fluorescently labeled cDNA probe was mixed can be found in Supplementary Data Table 1. To aid the reader in exploring with hybridization buffer (2.5 ␮g Cot1 DNA, 2.5 ␮g oligo-dT, 25% formamide, this results table, Web links to the appropriate Entrez Gene description page 5X SSC, and 0.1% SDS), denatured at 95°C for 5 min and applied to the Mouse are provided by clicking on the identifier. In addition, expression profiles were PancChip 6 spotted cDNA array (29,30), covered with glass coverslips, and generated for the significant genes and are available in Supplementary Data

DIABETES, VOL. 57, MARCH 2008 655 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

E13.5

E14.5

E15.5

E16.5

E17.5

(FIG. 1. Development of the endocrine pancreas. Heterozygous male mice containing an EGFP-marked null allele of Neurog3 (Neurog3؉/EGFP were mated with CD1 female mice. Pregnant females were killed at either 13.5, 14.5, 15.5, 16.5, or 17.5 days of gestation. Photographs represent bright field images of wild-type (left) and Neurog3؉/EGFP (right) pancreata. Panels on the right show pancreata from Neurog3؉/EGFP embryos .identified by their green fluorescence. The Neurog3؉/EGFP cells are located near the center of the organ

656 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES

Expression Proﬁles. The natural cubic spline used by EDGE to ﬁt the data are 6 shown in blue. A Gene annotation enrichment analysis. The resulting gene list was anno- 5 tated with functional categories using the National Institute of Allergy and Infectious Diseases Database for Annotation, Visualization, and Integrated Discovery (DAVID) bioinformatics resource, using all of those elements on the 4 array annotated with an Entrez Gene ID (37). Enrichment of these functional categories was determined using the Fisher’s exact test through a comparison 3 of the categories found in our list when compared with the background list of all genes found on the PancChip. The complete results of this analysis can be 2 found in Supplementary Data Table 4.

Ingenuity pathway analysis. Biologically relevant networks were drawn cells) total (% of EGFP 1 from the list of 1,029 temporally regulated genes identified by EDGE analysis. These data were generated through the use of Ingenuity Pathways Analysis, a 0 web-delivered application (www.Ingenuity.com) that enables the visualization and analysis of biologically relevant networks to discover, visualize, and 6000 explore therapeutically relevant networks. The application and a detailed B explanation of significance scoring were described previously (38). Of this list, 950 genes could be mapped to elements in the Ingenuity Pathways Analysis 5000 database. To focus the analysis on those genes with most significant changes in expressions, pathway analysis was performed using a subset of 334 genes 4000 (focus genes) in this list that showed a fold change Ͼ1.5 between any two time points. 3000

RESULTS 2000

Isolation of endocrine precursors. To profile the gene 1000 expression changes in fetal endocrine pancreas precursors cells per EGFP pancreas and their immediate descendants, we used Neurog3-EGFP 0 E12.5 E13.5 E14.5 E15.5 E16.5 E17.5 E18.5 Islets knock-in mice (12). Mice homozygous for this allele de- (n=1) (n=7) (n=5) (n=15)(n=5) (n=8) (n=3) (n=4) velop diabetes and die shortly after birth, whereas het- 25 75 121 11773 37 6 8 erozygous mice (Neurog3ϩ/EGFP) show no apparent differences from wild-type littermates as assessed by over- FIG. 2. The proportion of endocrine precursors in the pancreas is all development, growth characteristics, and glucose ho- highest at E14.5, and descendants of Neurog3-expressing cells can be marked by the persistence of EGFP. Neurog3؉/EGFP pancreata were meostasis. Use of a “knock-in” allele is advantageous disassociated into a single-cell suspension after trypsinization (RE- compared with a Neurog3 promoter-driven transgene, SEARCH DESIGN AND METHODS). The number of biological replicates (indi- which might be missing essential cis-elements or which vidual sorts) used is shown on the x-axis under the developmental stage, and the number below this represents the total number of might be influenced by integration site effects. Heterozy- embryos that were used at that stage. Neurog3؉/EGFP-positive and gous male mice were mated with CD1 female mice, and -negative cells were FACS sorted and used for RNA extraction. A: The embryos were harvested daily from E13.5 to E17.5, with a percentage of EGFP-positive cells was determined in reference to the ϩ/EGFP total number of cells sorted. EGFP expression began at E13.5 and total of 377 Neurog3 embryos required to complete peaked the following day. EGFP expression declined rapidly thereaf- the study. Figure 1 highlights the development and differ- ter, but positive cells could still easily be sorted until E17.5. B: The entiation of the pancreas throughout the secondary tran- total number of EGFP-positive cells per pancreas was determined. This figure clearly demonstrates the ability of EGFP to mark descendants of -sition, with EGFP levels peaking at E14.5. This figure also Neurog3؉ cells, with the long half-life of EGFP resulting in an accumu illustrates the fact that EGFP is persistent in the fetal lation of positive cells. At E18.5, 4 days following the peak in Neurog3 pancreas due to the long half-life of the protein. Thus, we expression, a decline in the number of marked cells was apparent. were able to isolate the direct descendants of Neurog3ϩ cells at later stages of pancreas development, i.e., cells in of the time-course, and protein levels are not detectable at which Neurog3 transcription and new protein synthesis these late stages (10). had been extinguished. Identification of temporally regulated genes. Total For optimal preparation of high-quality RNA required RNA samples isolated from the sorted Neurog3-EGFP for expression analysis, 30–70,000 cells were sorted from cells and passing stringent QC measures were used for pools of 12–25 pancreata to produce each biological expression profiling. Furthermore, unlike previous expres- replicate. At least three biological replicates were pre- sion array–based studies in which only one or two arrays pared for each time point. The procedure was attempted at were used per time point, we used three or four biological E12.5 using 25 Neurog3ϩ/EGFP embryos, but only 1,000 replicates for each time point, giving this study increased Neurog3ϩ cells were obtained, representing Ͻ0.03% of the statistical power. Time-course analysis of microarray data total cell number. This low level of expression is consis- are considerably more difficult than identification of dif- tent with expression studies using in situ hybridization ferentially expressed genes in a two-state comparison, but (10,21). the EDGE program proved to be powerful in identifying The greatest percentage of cells expressing Neurog3- temporally controlled gene expression (RESEARCH DESIGN EGFP occurred on E14.5, with 5.2% of the sorted popula- AND METHODS). Using EDGE, 1,029 temporally regulated tion positive for EGFP (Fig. 2A). Subsequently, there is a genes were identified (complete list in Supplementary steady decrease in the percentage of positive cells, which Data Expression Profiles). For the purposes of highlight- likely reflects the differentiation and expansion of the ing this data, genes were grouped based upon peak population of non-Neurog3–expressing cells at a higher expression at each of the six developmental stages inves- rate than those expressing the gene (Fig. 2B). The persis- tigated and ranked according to their EDGE P value. The tence of 3,000 to 4,000 EGFPϩ cells up to E18.5 is the top five most significantly temporally regulated genes with result of the long half-life of EGFP; we observed Neurog3 expression levels peaking at each developmental time mRNA levels to be considerably reduced in the latter half point are shown in Fig. 3.

DIABETES, VOL. 57, MARCH 2008 657 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

E13.5 E14.5 E15.5 E16.5 E17.5 ISLETS Gpc3: 1 (P=2.7e−06) Haghl: 17 (P=7.2e−05) D15Ertd621e: 47 (P=0.00026) Slc7a2: 5 (P=9e−06) Hsp90b1: 45 (P=0.00025) Slc40a1: 10 (P=2.7e−05) . . . . . 1 0 1 0 1 0 1 0 0 5 . 0 5 ...... 0 0 0 0 − 0 5 − 0 5 ...... −1 5 −1 5 −1 0 −1 0 −1 0 −1 0 Rcn1: 2 (P=4.2e−06) Chgb: 19 (P=8.7e−05) Gap43: 62 (P=0.00049) Zhx2: 33 (P=0.00018) Vegfa: 68 (P=6e−04) Iapp: 11 (P=3e−05) . . . . 1 0 1 0 1 0 1 0 . 1 1 0 . . . . 0 0 . 0 0 0 0 . . . . . −2 −1 −1 0 −1 0 −1 0 −1 0 −1 0 Igfbp1: 3 (P=5.6e−06) Nedd4: 27 (P=0.00014) Hsd3b7: 102 (P=0.0014) Zbtb20: 117 (P=0.0017) Hspa5: 91 (P=0.001) Id1: 14 (P=4e−05) ...... 1 0 1 0 1 0 1 0 1 0 1 0 ...... 0 0 0 0 0 0 ...... −1 0 −1 0 −1 0 −1 0 −1 0 −1 0 Scn2b: 4 (P=8.3e−06) Spon1: 36 (P=2e−04) Pgr: 106 (P=0.0015) St6gal2: 122 (P=0.0018) Tgfb2: 93 (P=0.0011) Hey1: 16 (P=5.8e−05) ...... 1 0 1 0 1 0 1 0 1 0 1 0 ...... 0 0 0 0 0 0 ...... −1 0 −1 0 −1 0 −1 0 −1 0 −1 0 Ddef2: 6 (P=1e−05) Dpf2: 46 (P=0.00025) Pgea1: 121 (P=0.0018) Hbp1: 128 (P=0.0019) Nnat: 123 (P=0.0018) Hadh: 22 (P=0.00011) ...... 1 0 1 0 1 0 1 0 1 0 1 0 ...... 0 0 0 0 0 0 ...... −1 0 −1 0 −1 0 −1 0 −1 0 −1 0

FIG. 3. Expression profiles of the five most significantly temporally regulated genes at each of the six developmental time points studied. A total of 1,029 genes was found to be temporally regulated during development of endocrine precursors. Genes were observed to peak in expression at each of the six time points investigated. The top five genes, ranked by their EGDE P value, at each time point are depicted. Plots are headed with official gene symbol: EDGE rank (P value). The individual data points for each sample are represented by the circles and, left to right, represent E13.5, E14.5, E15.5, E16.5, E17.5 (black), and adult (gray) islets. The natural cubic spline used by EDGE to fit the data is shown by a dashed line. Expression in adult islets is shown for reference. The intent of this figure is to highlight some of the most significant expression profiles; the entire set of temporally expressed genes can be found in Supplementary Data Expression Profiles.

Expression profile data were confirmed via qRT-PCR cation of the ␤-cell, and Mycl1, a gene with links to lung (Fig. 4). For the 18 genes tested, the results of the RT-PCR and colorectal cancer, but there is no literature with analysis closely mirrored those of the array analysis. Little regard to its role in the pancreas. or no transcript was detected in EGFPϪ cells for the Principal component analysis was performed to identify majority of the genes tested. The presence of Hmga2, Id1, predominant gene expression patterns. Three principal and Id2 mRNA in these nonendocrine precursor cells may components were identified that accounted for 95% of the indicate an additional role for these genes in development significant variability in the data (Fig. 5D). Over half of the of the exocrine pancreas. As expected, Neurog3 was not temporally regulated genes were observed to cluster with expressed in either the EGFPϪ cells or adult islets, indi- Neurog3, with a pattern of decreasing expression over cating the high level of purity of the sorted cells. Ghrl and time. Of the genes whose expression profile correlated Pou3f4 also had no detectable mRNA in the mature islet, most significantly with Neurog3, 50% were found to be with expression profiles closely matching that which had expressed in the nucleus (P Ͻ 0.007), and gene ontology been described previously (16,39). (GO) functions of development (P Ͻ 0.011) and transcrip- Clustering analysis. Hierarchical clustering was per- tion (P Ͻ 0.048) were significantly overrepresented. This is formed using the 1,029 genes identified to be temporally consistent with a model in which subdivision and differ- regulated (Fig. 5A). Biological replicates for each time entiation of the endocrine lineage is dependent on a point were observed to cluster together, and a clear cascade of multiple transcription factors. progression of similarity between time points is apparent Differential gene expression in endocrine precursors (E13.5 most similar to E14.5, which together were similar and their descendants. In addition to the temporal to E15.5 and so forth). Clustering of the genes revealed analysis performed with EDGE, a direct comparison anal- that many genes were expressed in similar patterns during ysis was performed to highlight differences between the the time course. The cluster containing Neurog3 identified endocrine precursors and their immediate descendants. 21 genes whose expression profile was strikingly similar to Based on the clustering analysis described above (Fig. 5), this master regulator (Fig. 5B). Finally, hierarchical clus- the endocrine precursors sorted from E13.5 and E14.5 tering was performed on the subset of 64 genes with pancreata had very similar expression profiles. Likewise, transcription factor activity (Fig. 5C). Three patterns were the descendants of these precursor cells, which were evident among these genes: those with increasing expres- marked by EGFP and sorted from later time points (E16.5 sion over time (Sox15, Pax4, Myc), those with peak and E17.5), were also observed to cluster closely together expression in the middle of the time course (Onecut1, and clearly had an expression profile that was distinct Mafb, Isl1), and then those with decreasing expression from the precursors. Therefore, to gain statistical power, over time (Gata4, Neurog3, Foxa2). Neurog3 was ob- we performed a direct comparison between the six E13.5 served to cluster most closely with Nkx2–2, which is and E14.5 biological replicates, representing endocrine known to function downstream of Neurog3 in the specifi- precursors, and the six E16.5 and E17.5 biological repli-

658 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES

0.8 A. Neurog3 0.8 B. Neurod10.8 C. Pou3f4 0.7 0.7 0.7 0.6 0.6 0.6 0.5 0.5 0.5 0.4 0.4 0.4 0.3 0.3 0.3 0.2 0.2 0.2 0.1 0.1 0.1 0.0 0.0 0.0

0.9 D. Ins1 0.7 E. Gcg 1. 0 F. Sst 0.8 0.6 0.9 0.7 0.8 0.5 0.6 0.7 0.5 0.4 0.6 0.5 0.4 0.3 0.4 0.3 0.2 0.3 0.2 0.2 0.1 0.1 0.1 0.0 0.0 0.0

1. 2 G. Ghrl 0.8 H. Ppy 0.9 I. Iapp 0.8 1. 0 0.7 0.7 0.6 0.6 0.8 0.5 0.5 0.6 0.4 0.4 0.4 0.3 0.3 0.2 0.2 0.2 0.1 0.1 0.0 0.0 0.0

0.9 J. Pcsk2 0.6 K. Id1 0.7 L. Id2 0.8 0.5 0.6 0.7 0.5 0.6 0.4 0.5 0.4 0.3 0.4 0.3 0.3 0.2 0.2 0.2 0.1 0.1 0.1 0.0 0.0 0.0

0.9 M. Sim10.9 N. Nkx2-2 1.2 O. Pax6 0.8 0.8 1.0 0.7 0.7 0.6 0.6 0.8 0.5 0.5 0.6 0.4 0.4 0.3 0.3 0.4 0.2 0.2 0.2 0.1 0.1 0.0 0.0 0.0

0.6 P. Nr2fc 0.9 Q. Nrld2 1.0 R. Nnat 0.8 0.9 0.5 0.7 0.8 0.4 0.6 0.7 0.5 0.6 0.3 0.5 0.4 0.4 0.2 0.3 0.3 0.2 0.1 0.2 0.1 0.1 0.0 0.0 0.0 E14.5 E16.5 E14.5 E16.5 E14.5 E16.5 E13.5 E14.5 E15.5 E16.5 E17.5 Islets E13.5 E14.5 E15.5 E16.5 E17.5 Islets E13.5 E14.5 E15.5 E16.5 E17.5 Islets -ve -ve -ve -ve -ve -ve

FIG. 4. Patterns of temporal regulation are consistently conﬁrmed using qRT-PCR. cDNA was prepared from each replicate used for the analysis ,shaded bars) and from E14.5 and E16.5 EGFP-negative cells (open bars) that were collected along with the EGFP؉ cells. For the 18 genes tested) the results of this qRT-PCR analysis closely mirrored the results of the array analysis. For the majority of genes tested, little or no transcript was .present in EGFP؊ cells cates, representing their descendants. It should be noted population of descendants and cells continuing to express- that although we observed Neurog3 mRNA levels to be ing Neurog3. A total of 648 genes were found to be substantially reduced in this population, levels were not differentially expressed with an absolute fold change Ն1.3 completely extinguished (Fig. 4A). As such, the cells and a P value Ͻ0.05 (Supplementary Data Table 2). Of sorted at these latter time points may represent a mixed these genes, 182 (28%) have higher expression in endo-

DIABETES, VOL. 57, MARCH 2008 659 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

Marcksl1 Nkx2-2 Pspc1 IMAGE=4481345 Igf2bp2 Grb10 Neurog3 Rap1b 2010107G23Rik Dctn2 Hmgn2 Npm1 Pdk3 Myo1d Hdac1 Rbm8a Mrpl1 Hnrpr Snx15 Mycl1 0610007L01Rik

FIG. 5. Hierarchical clustering analysis identifies three principal components within the time course and highlights several novel genes that were observed to cluster closely with Neurog3. Hierarchical clustering was performed using the subset of 1,029 genes identified to be temporally regulated using EDGE. Clustering was performed using the average linkage clustering algorithm and the Pearson correlation as a measure of similarity (A). Biological replicates for each time point were observed to cluster together. Clustering of the genes revealed that many genes were expressed in similar patterns during the time course. The cluster containing Neurog3 identified 21 genes whose expression profile was strikingly similar to that of Neurog3 (B). Hierarchical clustering using the subset of 64 genes with transcription factor activity identified several transcription factors whose expression profile closely resembled that of Neurog3 (C). Three principal components were identified that accounted for 95% of the significant variability in the data. As such, K means clustering was performed using the Pearson correlation to cluster the genes into three groups with a high degree of similarity within each cluster and a low degree of similarity between clusters (D). Over half of the genes identified were observed to cluster with Neurog3, with a pattern of decreasing E13.5 E14.5 E15.5 E16.5 E17.5 Islets expression over time. crine precursors (E13.5/E14.5) and 466 (72%) higher ex- Using this approach, an additional 338 differentially pression in the descendants (E16.5/E17.5). Over 100 expressed genes were discovered that had not been iden- transcriptional regulators were identified in this analysis, tified in the time course analysis described earlier. These 33 of which were expressed at highest levels in endocrine include Insm1, Nars2, Foxp1, Hhex, and Cd14. Among the precursors. genes that were most highly expressed in the E16.5/E17.5

660 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES

10 C D

Sox15 Nr1d2 Mitf Pax4 Onecut2 Nr4a3 Myc Nr4a1 1 Ankib1 Nfya Trps1 Fos Rora Nfat5 Nr3c1 Pitx1 Hod Fosl2 E13.5 E14.5 E15.5 E16.5 E17.5 Adult Foxd3 Set 1: 318 genes Klf5 Sox7 10 Onecut1 Bcor Hnf4g Vps72 Preb Gata3 Stat1 Satb1 Zhx2 Egr1 Zfp263 Cebpb 1 Tcf7l2 Purb Mycn Klf5 E2f5 Cutl1 Hey1 Asb4 E2f5 Sox12 E13.5 E14.5 E15.5 E16.5 E17.5 Adult Nr2f6 Set 2: 522 genes Etv5 Mrg1 10 Mafb Isl1 Tbx19 Pgr Runx2 Irx3 Sox11 Zfp445 Gata4 Foxa1 Zfp207 Yy1 1 Nkx2-2 Neurog3 Mycl1 Ahr Tcfec Gabpb1 Mta2 Btf3 Foxa2 E13.5 E14.5 E15.5 E16.5 E17.5 Adult

E13.5 E14.5 E15.5 E16.5 E17.5 Adult Set 3: 189 genes

Fig. 5. Continued

DIABETES, VOL. 57, MARCH 2008 661 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

TABLE 1 Temporally regulated transcriptional genes in the development of the endocrine pancreas Transcriptional regulators (transcription factors, transcriptional regulators, and potential regulators of transcription) Developmental stage E13.5 Zfp445, Asb4, Nkx2–2, Etv5, Hnf4g, Sox11, Gata4, Yy1, Cutl1, C230097I24Rik, Zfp207, E2f5, Irx3, Nr2f6, Id2, Tcf12, Rbm14, Prdm4, Cnot7, Zfp143, Inppl1, Sfpq, Gmcl1, Basp1, Nono, Pspc1, Hmga2, Csde1, Rbl1, Mcm4, Abcb4, Stau1, Hnrpm, Paip1, Hmgn1, Dek, Kin, Nudt21, Mrpl2, Rbm12, Rps3, Pcbp2, Hmgn2, Hnrpa1, Rpl39, H2afy2, H2afz, Igf2bp2, Zcchc3, Prdm15, Nap1l4, Zcchc11, Ncoa6ip, Xab1, Mtpn, Bud31 E14.5 Klf5, Neurog3, Tcf7l2, Mycn, Gabpb1, Mta2, Runx2, Tcfec, Ahr, Hmgb2, Psmc5, Jarid2, Mxi1, Eaf1, Smarce1, Litaf, Ezh2, Sfrs6, Baz1a, Morf4l1, Dpf2, Hmgb3, Neo1, Ing4, Rab1, Polr2f, Khsrp, Tsnax, Npm1, Arid1a, Rpl37, Rpa2, Rps20, Rps24, Lsm2, Snrpd1, Tarbp2, Rpl9, Zfp521, Cbx5, Slbp, 2610209M04Rik, Rps18, 2610101N10Rik, Cstf2, Hnrpa2b1, Rbmxrt, Rps11, Trove2, Rps7, Aplp2, Rps6, Rps14, Sfrs2, Rps9, Rps23, Cbx1, Snrpe, Rps4x, Rpl28, Prpf40a, Parp1, Uba52, Eid1 E15.5 Isl1, Foxa2, Foxa1, Btf3, Mrg1, Pgr, Fubp1, Aebp2, Pqbp1, Neurod1, Hdac1, Eya2, Zfp710, 1810035L17Rik, Rpo1–1, Ciz1, Mrpl1, Msh3, Hnrpr, Rbm7, Rbm8a, H3f3a, Sf3a3, Srp14, Rps29, U2af2, Rap1b, Top2b E16.5 Mycl1, Tbx19, Zhx2, Sox7, Zfp263, Preb, Tshz1, Satb1, Bcor, Stat1, Mafb, Vps72, Onecut1, Sec14l2, Mlxipl, Grlf1, Rfx1, Ncor1, Med12, Brca1, Rbpsuh, Hbp1, Mybl2, Hes2, Zfp652, Zfp9, A430033K04Rik, Ddx5, Zcchc9, Hist2h2aa1, Nuﬁp2, Zc3h7a, Tiparp, Sfrs7, Zbtb20, Isg20, Ptma E17.5 Fosl2, Cebpb, Gata3, Egr1, Smarca4, Zfp579, 1300003B13Rik, Strbp, Thoc1, H2afv Islets Rere, Foxd3, Pitx1, Sox12, Ankib1, Pax4, Nfya, Trps1, Sox15, Hey1, Nfat5, Mitf, Mafa, Creb3l2, Fos, Onecut2, Myc, Hod, Nr3c1, Rora, Nr4a3, Nr4a1, Nr1d2, Klf9, Id1, Pcaf, Baz1b, Mll3, Zfp715, Sf1, Med28, Ctnnd1, Pde8b, Zfp36l2, Ankrd57, Arid5a, Bicc1, Hist1h3f, Prpf19, Rod1, Prkra, Mbnl1, Atxn2, Rbpms, Nucb2, Snrpb2, Cpeb1, Ddx50, G3bp2, Hexim1, Ncoa4

Of the 1,029 temporally regulated genes, 246 transcription factors, transcriptional regulators, and potential transcriptional regulators were found to be differentially expressed during development of the endocrine precursors. Transcription factors and transcriptional regulators are shown in bold face type and potential regulators of transcription in non–bold face type.

Neurog3 descendants were markers of the ␤-cell (Ins1, development were highly enriched biological processes in Ins2, Glut2) and ε-cell (Ghrl), indicative of differentiation endocrine precursors and their descendants. Gene anno- toward the mature endocrine cell types. Overrepresented tation enrichment analysis using the list of temporally functions and canonical pathways were determined via regulated genes revealed that the functions of nucleic acid pathway analysis (Supplementary Data Fig. 2). Pathways binding and transcriptional regulation were significantly involved in cellular movement, integrin signaling in rela- enriched. Detailed analysis and annotation with GO function to cytoskeletal rearrangements, protein ubiquitina- tions determined that this list contained 69 transcription tion, and insulin receptor signaling were all observed to be factors, 71 other transcriptional regulators (such as cofac- significantly enriched in the progenitor cells when com- tors and elements of the basal transcriptional machinery), pared with the descendants. Strikingly, carbohydrate me- and 97 other genes potentially involved in transcriptional tabolism was absent from the progenitors but was regulation. Together, these genes represented 27% of the observed to be highly enriched in the descendants, along with lipid metabolism and the Ppar␣/Rxr␣ activation path- entire list of temporally regulated genes. Table 1 lists way. These observations reflect the differentiation of en- these 246 transcription factors, transcriptional regula- docrine precursors into more mature endocrine cells. tors, and potential transcriptional regulators found to be Finally, we compared the expression profile of endo- temporally regulated during development of the endocrine precursors (E13.5/E14.5) directly to that of adult crine precursors. islets. Over 2,000 genes were found to be significantly Several of these transcription factors have been shown differentially expressed. Of these genes, 103 were found to by genetic means to function in the development of the be highly expressed in the precursor cells, with a fold endocrine pancreas, including Neurod1, Mafa, Mafb, change more than two when compared with the adult Nkx2–2, Foxa1, and Foxa2 (Fig. 6). Moreover, markers of islets (Supplementary Data Table 3). A total of 25 of these mature islets cells, such as Ins1, Ppy, Ghrl, and Iapp were genes were classified as transcriptional regulators, many observed to have increasing expression levels over time. of which have an unknown role in the developing endo- Interestingly, Gcg, a marker of the mature ␣-cell, was not crine pancreas, e.g., Nars2, Mycl1, Sox11, Foxb1, and significantly differentially expressed in these cells, but as Zfp306. expected high levels were observed in adult islets (Fig. Gene annotation enrichment analysis. Temporally reg- 4E). It was recently demonstrated that the role of Neurog3 ulated genes were mapped to protein families using the in the development of the ␣-cell lineage occurs much Ingenuity Pathways Knowledge Base and to enriched GO earlier than the time frame studied in the present work, biological process and molecular function categories (Sup- with the induction of Neurog3 in Pdx1ϩ progenitors at plementary Data Fig. 3). Protein metabolism and biosyn- E8.7 resulting in an almost exclusive induction of gluca- thesis, cellular organization and differentiation, and gon-positive cells by E14.5 (40).

662 DIABETES, VOL. 57, MARCH 2008 tde 5–1,fu ftegnswr bevdt edfeetal xrse nortm course: time our in expressed differentially be to observed were genes the Gcg. of not four but Ppy, (59–61), and studies Ghrl, Ins1, of regulation pancreas. endocrine the of development IBTS O.5,MRH2008 MARCH 57, VOL. DIABETES, headed are Plots islets. reference. (gray) for adult shown and each is (black), for islets E17.5 points E16.5, data adult E15.5, individual in E14.5, The Expression ( E13.5, pancreas. rank line. represent endocrine EDGE right, dashed the to symbol: of left gene development ofﬁcial and, during circles with factors by transcription represented of are proﬁles sample expression Temporal 6. FIG. BC

−1.5 0.0 1.0 −3 −1 1 2 A

−1.0 0.0 1.0 −1.0 0.0 1.0 −1.5 −0.5 0.5 −1.0 0.0 1.0 −1.0 0.0 1.0 Ins1: 261(P=0.0056) Ppy: 1231(P=0.04) Neurod1: 556(P=0.01) Onecut1: 838(P=0.02) Hey1: 16(P=5.8e−05) Foxa1: 1101(P=0.03) Isl1: 487(P=0.01) P

B −1.0 0.0 1.0 au) h aua ui pieue yEG ofi h aaaddtriesgicnei hw ya by shown is significance determine and data the fit to EDGE by used spline cubic natural The value). −1.0 0.0 1.0 xrsinpolso maueadmtr omnso h norn se,dmntaigtemporal demonstrating islet, endocrine the of hormones mature and immature of profiles Expression : Ghrl: 125(P=0.0018) C Gcg: 8684(P=0.49) ftesxdaee-soitdlc eetyietfidi eiso hl-eoeassociation whole-genome of series a in identified recently loci diabetes-associated six the Of :

−1.5 −0.5 0.5 −2.0 −1.0 0.0 −1.0 0.0 1.0 −1.0 0.0 1.0 −1.0 0.0 1.0 Onecut2: 1336(P=0.047) Neurog3: 476(P=0.01) A Foxa2: 780(P=0.02) Mafa: 492(P=0.01) Id1: 14(P=4e−05) rncito atr nw n pcltdt lya motn oeduring role important an play to speculated and known factors Transcription :

−1.0 0.0 1.0 −1.5 −0.5 0.5 Slc30a8: 42(P=0.00023) Cdk5rap1: 507(P=0.01)

−1.0 0.0 1.0 2.0 −1.0 0.0 1.0 −2 −1 0 1 −1.0 0.0 1.0 −1.0 0.0 1.0 Slc30a8 Nkx2−2: 135(P=0.002) Gdf11: 1018(P=0.03) Id2: 26(P=0.00014) Mafb: 410(P=0.01) Pax4: 664(P=0.01)

−2.0 −1.0 0.0 −1.0 0.0 1.0 , Tcf7l2 Tcf7l2: 271(P=0.0058) Igf2bp2: 870(P=0.02) , .WIEADASSOCIATES AND WHITE P. Cdk5rap1 and , Igf2bp2 663 . EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

E PP RB PP NEDD4 YY1 BMP4 EIF5 SST GRB10 PP IRS1 MXI1 CREBBP PTF1A

A, PP PD, PP PD PD T T PD PP E E I TR , M, PD A RB P MYC EIF3S10 P A, E HES1 , , P PD D, T ID2 A PP D , PP, RB IGF2R M, P, A E, PDX1 P PP TR D INSM1 PP P E T E E, , L P E, A, E, L, P PD P A, O E, L, RB, TR I, TCF3 D , PP E A, E P A, PP, RB PP E Notch P HEY1 E IGF1 E E , E I P A PD NR3C1 D, P, RB I, IGF2 , P Rb E E PP T E, LO A, PP P , RB P E P NEUROG3 PD Delta/Jagged , P PP PP M T I , SMARCA4 PD 6.780E-3 PP , I, , PD H19 E A PD E RB PP E P PD

T P I, , E, E PPY E ID1 IGFBP1 E ONECUT1 E E PP E A P P PAX4 P P E GPC3 E PD PP , A E , E E FOXA1 E D , E, P P L E RB,TPP, D O PP , PD, D P E, P E, P D, PP MAFA ABCC8

P E, R NEUROD1 E, PD E PGR T B A, PP E A, E, , PD PP, LO E, RB D PD A P E E , E, P , FOXA2 D A PD E RB PD, PP, T E D GCG INS1 HMGB2 E, P PP NKX6-1 PD, PP POU3F4 E, D ISL1 P E E PD A, PD E NR4A1 T PD , A, T T PD PD , T PD MAFB SIM1 FOXD3

E E PAX6 PD EL NKX2-2 MYCL1 NKX6-2 ONECUT2 I GHRL HADH PD IAPP PD

A Activation / deactivation Growth Transcription Peptidase RB Regulation of Binding factor factor PR Protein-mRNA binding PP Protein-Protein binding Cytokines Enzyme Phosphatase PD Protein-DNA binding EExpression IInhibition binding only Kinase L ProteoLysis acts on M Biochemical Modification PPhosphorylation inhibits Other TTranslocation LO LOcalization

FIG. 7. Pathway analysis identifies a complex regulatory network controlling endocrine system development and function in the pancreas. The network is displayed graphically as nodes (genes/gene products) and edges (the biological relationships between the nodes). Nodes were colored to highlight those genes whose differential expression was significant in either the temporal regulation analysis or static analysis. Color intensity indicates significance, from the most significant (dark red) to least significant (light pink) genes. Nodes are displayed using various shapes that represent the functional class of the gene product. Edges are displayed with various labels that describe the nature of the relationship between the nodes. The above network was produced by expanding the most significant network (34 focus genes and a score of 60) with additional information in PubMed. To simplify the image, many of those interactions not demonstrated in the pancreatic tissues were removed.

Identification of regulatory networks. Gene annota- Nkx2–2, Mafa, and Pax4, are part of this network derived tion enrichment analysis provided information regarding from our expression data. However, several genes that had categorical changes with regard to the biological function limited prior evidence regarding their role in this develop- and metabolic activity of the temporally regulated genes in mental program were observed, such as Sim1, Insm1, Id2, this time course. However, our specific interest in this and Nr3c1 (glucocorticoid receptor). Furthermore, genes study was to understand how individual genes were inte- with no previously reported role in this development grated into specific regulatory and signaling networks. process were identified, such as Id1, H19, Yy1, and Mycl1. This type of analysis has not been reported in microarray Pathway analysis was also used to examine the relation- studies of the developing endocrine pancreas and revealed ship between our gene set and existing canonical signaling several new findings. pathways. Four pathways in particular showed dramati- Several major networks were identified, but by far the cally significant enrichment (Supplementary Data Figs. most significant was a complex regulatory network con- 4–7). A total of 22 genes were associated with Igf1 trolling endocrine system development and function cen- signaling (P value ϭ 1.24E-09), 7 with the endoplasmic tered around Neurog3 (Fig. 7). Many of the transcription reticulum stress pathway (P value ϭ 2.65 E-05), 21 with factors previously reported to play a role in endocrine cell Wnt/␤-catenin signaling (P value ϭ 1.54 E-04), and 17 development, such as Pdx1, Foxa2, Neurod1, Isl1, genes with insulin receptor signaling (P value ϭ 4.51 E-04).

664 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES

DISCUSSION ing adenovirus (44). Less than 20% of the genes claimed as Expression profiling of endocrine precursors and Neurog3 targets from adenoviral overexpression were also their descendants. We identified over 1,000 genes that identified in the present study and were restricted to those are temporally regulated during development of the endo- genes whose interaction with Neurog3 has been reported crine pancreas. Strikingly, a large number of the genes elsewhere (Supplemental Data Fig. 8). This striking differ- identified in our analysis have no prior reports of a role in ence between the two studies most likely reflects a signif- the developing pancreas and many represent novel genes icant disadvantage in studies where Neurog3 gene with no prior art. The most significantly temporally regu- expression is artificially induced, in a cell line far removed lated gene was Glypican 3 (Gpc3), with a marked decrease from the fetal pancreas. Furthermore, there is limited in expression over time. Glypicans are cell-surface hepa- evidence to show that pancreatic duct cells can function in ran sulfate proteoglycans that are bound to the exoplasmic vivo as endocrine progenitors or be induced to differenti- surface and are thought to play a role in the control of cell ate into mature islets cells (46). division and growth regulation, possibly via modulation of More recently, microarray analysis was performed using Wnt signaling (41). Gpc3 is known to be highly expressed a Neurog3-deficient mouse model in an attempt to identify during embryogenesis and was recently demonstrated to Neurog3-dependent genes expressed in whole pancreas (47). The use of whole pancreas tissue, as opposed to be a marker of hepatic progenitor/oval cells with an ϩ sorted Neurog3 expression profile remarkably similar to that observed in cells, severely limited the power of this approach, as at most 5% of the cells in the total pancreas our analysis of pancreas progenitors (42). express Neurog3 (Fig. 2). Therefore, the approach of using A total of 246 transcription factors, transcriptional reg- total pancreas to investigate changes resulting from the ulators, and potential transcriptional regulators were deletion of Neurog3 will result in at least a 20-fold reduc- found to be temporally controlled during development of tion in the power to detect alterations in gene expression. the endocrine precursors, several of which are well estab- Consequently, this study only identified 52 differentially lished and important regulators of the endocrine pancreas, expressed genes, many of which were genes expressed at for instance Neurod1, Mafa, Mafb, Nkx2–2, Foxa1, and high levels in mature endocrine cells, like glucagon and Foxa2. As expected, markers of mature islets cells, such as insulin, which were already known to be lacking in Neu- Ins1, Ppy, Ghrl, and Iapp, were observed to have increas- rog3Ϫ/Ϫ pancreas (10). ing expression levels over time, highlighting the compre- Regulatory networks in the developing endocrine hensiveness of our analysis. pancreas. Our pathway analysis identified several net- Previous attempts to describe transcriptional regulation works of genes regulating endocrine precursor specifica- of the developing endocrine pancreas have failed to pro- tion, development, and expansion. The centrality of duce the in-depth transcriptional profile of endocrine Neurog3 in this process is highlighted in Fig. 7. The precursors and their descendants presented in the current signaling factors involved in specifying the developmental study. Gu et al. (43) produced transcriptional profiles from decision between endocrine and exocrine tissue during four stages of endocrine pancreas development: E7.5 organogenesis of the pancreas are of considerable inter- unspecified endoderm, E10.5 Pdx1-positive cells, E13.5 est. Whereas insulin- and glucagon-producing cells are not Neurog3-positive cells, and adult islets. Expression profil- related to a hormone coexpressing precursor cell, a com- ing of Neurog3-precursor cells at E13.5 identified 71 genes mon origin of endocrine cells does exist at the non– that were temporally enriched in this population, of which hormone-expressing precursor level, and both ␣- and only 7 were transcription factors: Pou3f2 (Brn2), Isl1, ␤-cells develop from Pdx1- and Neurog3-positive cells Mycl1, Mafb, Myt1, Neurod1, and Pax4. With the excep- (19,48). Exocrine pancreatic cells arise from Ptf1a-ex- tion of Myt1, all of these transcription factors were among pressing precursors, although it was recently demon- the 246 transcription factors, transcriptional regulators, strated that mature pancreatic cells develop through a and potential transcriptional regulators identified by our very early common progenitor expressing Pdx1, Ptf1a, analysis. We observed expression of Myt1 to remain cMyc, and Cpa1 (49). In those cells not fated to become constant throughout the fetal time course, with a twofold part of the islets, the transcriptional repressor Hes1, a reduction of expression in the adult islet (Supplementary main effector of Notch signaling, strongly inhibits Neurog3 Data Table 3). Several factors may explain this marked gene promoter activity through a mechanism known as difference between this present study and the previous lateral inhibition, which occurs via Notch receptor signal- report. Gu et al. (43) marked E13.5 endocrine precursors ing (20,50). Indeed, in mice containing a homozygous using a Neurog3-EGFP transgene that may be missing mutation of the Hes1 gene, misexpression of Ptf1a essential elements of the Neurog3 promoter and is less throughout the gut epithelium results in ectopic pancreas likely to represent the true expression pattern of Neurog3 formation (62). that we obtained through the use of a knock-in approach. Temporal regulation of Id1 and Id2 was highly signifi- Furthermore, only one or two biological replicates were cant and followed a pattern of downregulation similar to analyzed at each time point, resulting in significant statis- that of Neurog3 as development progressed. The Id1 and tical limitations. Finally, their comparisons were per- Id2 proteins contain a helix-loop-helix (HLH) domain but formed using different GeneChips for each time point, the not a basic domain and lack DNA binding activity and most comprehensive of which lacked ϳ4,000 of the genes therefore can inhibit the DNA binding and transcriptional found on the PancChip array (30). activation ability of bHLH transcription factors, such as An alternative approach to the use of transgenic mice to Neurog3 and Neurod1. The spatial and temporal control of mark Neurog3ϩ cells has been to transduce immortalized the proneural bHLH factors (Neurog1, 2, and 3) and cell lines with recombinant viral vectors expressing Neu- inhibitory HLH factors (Id1 and Hes1) was recently shown rog3 (44,45). Microarray analysis demonstrated increased to coordinate the timing of differentiation of neurons and expression of 51 genes in an in vitro study where pancre- glia (51). However, the role of inhibitory HLH factors atic duct cell lines were infected with a Neurog3-express- during development of the pancreas has received little

DIABETES, VOL. 57, MARCH 2008 665 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS attention. Unlike expression of Id2, expression of Id1 was E15.5 (58). Again, this represents a novel observation for observed to be highest in the adult islet, and there is the endocrine pancreas, suggesting a mechanism to tem- evidence to suggest that this gene may play a role in porally limit the rate of proliferation of endocrine precur- promoting ␤-cell function (52). Recent in vitro studies sors. demonstrated that Bmp4 promotes the heterodimerization Conclusions. Through the use of the Neurog3-EGFP of Id2 to Neurod1, with a resultant decrease in Pax6 knock-in model, combined with careful cell sorting, daily expression (53). We observed significant expression of sampling throughout the secondary transition and beyond, Bmp4 during development of endocrine precursors with a and the use of powerful statistical and analytical tools, we peak in expression at E16.5. Our data support a model in have been able to accurately capture the gene expression which high expression of Bmp4 at E16.5 triggers the profile of the pancreatic endocrine progenitors and their switch that inhibits continued transcriptional activation by descendants. Furthermore, through the use of the mouse bHLH factors (Neurog3, Neurod1) via regulation of inhib- PancChip array, we have been able to capture information itory HLH factors (Id1, Id2). The result of this switch will on numerous genes known to be expressed in the pan- be to block further differentiation of endocrine precursors creas but not found on commercially available arrays, and instead promote their expansion into mature islets many of which are novel and will lead to further investi- cells. gation and discovery. The list of temporally regulated In addition to the several factors with well-documented genes identified in fetal endocrine precursors and their roles in the development of endocrine precursors, our immediate descendants provides a novel and important analysis confirmed several of the more recently discovered resource for developmental biologists and diabetes re- and less well-documented potential targets of Neurog3. searchers alike. Furthermore, from our attempt to model Insulinoma-associated 1 (Insm1 or Ia1) was found to be the regulatory networks that control development of the significantly downregulated in the descendants of endo- endocrine pancreas, it is apparent that the complex inter- crine precursors (E16.5/E17.5). This transcriptional reactions between these genes, and their requirements for pressor was recently reported to have an essential role in carefully controlled spatial and temporal expression, goes ␤- and ␣-cell differentiation, acting downstream of Neu- far beyond what has traditionally been presented in sim- rog3 and parallel with Neurod1 (45). Moreover, as is plified models of transcriptional hierarchy. shown in Fig. 7, Neurog3 is believed to form a heterodimer The identification of so many transcription factors, with Tcf3 (E47) on the E-box3 of the Insm1 promoter and which are typically expressed at relatively low levels and to recruit the Creb-binding protein (Crebbp) (54). Expres- as such often not detected in array analysis, lends signifi- sion of Sim1, which is involved in the differentiation of cant credence to the quality of this dataset and the neuroendocrine cells of the hypothalamic-pituitary axis statistical tools used to analyze it. A series of whole- (55), was recently shown to be under the control of genome association studies, in which over 380,000 single Neurog3 in a murine embryonic stem cell line in which nucleotide polymorphisms were analyzed in over 1,400 Neurog3 was overexpressed (56). Our data provide evi- patients with type 2 diabetes, recently identified six novel dence to suggest that this in vitro observation may be loci strongly associated with the disease (59–61). The biologically relevant during development of the endocrine authors hypothesize that these six genes have a primary pancreas in vivo. Although profiling of Sim1 expression by role in the ␤-cell. Our data provides strong evidence in qRT-PCR (Fig. 4M) supports the observations that this support of this notion for four of the genes associated with gene is expressed in the developing endocrine pancreas, these loci, as we found them to be differentially expressed and at much greater levels than in Neurog3Ϫ cells, we in our time course: Slc30a8, Tcf7l2, Cdk5rap1, and observed static expression profile across the time course. Igf2bp2 (Fig. 6). Moreover, Igf2bp2 was one of the genes This may suggest, as does Fig. 7, that Sim1 is not a direct whose expression profile clustered most closely to that of target of Neurog3. Neurog3 (Fig. 5B). Very little is known about the function Use of pathway analysis to examine the relationship of these genes, but our observations of their expression between our gene set and existing canonical signaling profile in murine endocrine precursors, combined with the pathways strongly suggested that Igf1 and insulin receptor genetic findings in humans, provides strong evidence to signaling pathways play a significant role in this process. their playing a key role not only in development of the Strikingly, Grb10 plays a critical role in both of these endocrine pancreas but also in long-term islet function and pathways, and along with Igfbp2 was observed to be one the development of diabetes. of the genes clustering most closely with Neurog3 (Fig. 5B). This gene encodes a growth factor receptor–binding protein that interacts with insulin receptors and insulin- ACKNOWLEDGMENTS like growth-factor receptors, and is imprinted in a highly This study was supported by the U.S. National Institute of isoform- and tissue-specific manner. Grb10 was shown to Diabetes and Digestive and Kidney Diseases of the Na- form a complex with Nedd4, which was also highly tional Institutes of Health Functional Genomics of the significantly downregulated over the time course (57,58). Beta Cell Grant UO1-DK56947 and the University of Penn- The role of this protein-protein complex in the Igf1 signal- sylvania Institute of Diabetes, Obesity and Metabolism ing pathway is in regulating ubiquitination and stability of Diabetes and Endocrinology Research Center Grant the Igf1 receptor (Igf1r). In response to Igf1, Grb10 asso- P30DK19525. ciates with Igf1r and brings Nedd4 into the vicinity of Igf1r, Microarray data for this study have been deposited at leading to its ubiquitination, which in turn would result in Array Express with the accession number E-CBIL-40. The the internalization and degradation of the receptor. There- data and MIAME (Minimum Information About a Microar- fore, the high levels of Grb10 and Nedd4 expression we ray Experiment) compliant annotation can also be queried observed at E14.5 could represent a cellular mechanisms through the user-friendly interface RAD (RNA Abundance used to prevent continuous and enhanced activation in Database) (www.cbil.upenn.edu/RAD). The final anno- response to Igf1, expression of which was seen to peak at tated gene lists and numerous additional analysis tools are

666 DIABETES, VOL. 57, MARCH 2008 P. WHITE AND ASSOCIATES available through the Betacell Biology Consortium Web ter G, Madsen OD, Serup P: Independent development of pancreatic ␣- and site (www.betacell.org) via the Endocrine Pancreas ␤-cells from neurogenin3-expressing precursors: a role for the notch Consortium Database (EPConDB: www.cbil.upenn.edu/ pathway in repression of premature differentiation. Diabetes 49:163–176, 2000 EPConDB). 22. Jacquemin P, Durviaux SM, Jensen J, Godfraind C, Gradwohl G, Guillemot Thanks to Dr. Joshua R. Friedman for critical reading of F, Madsen OD, Carmeliet P, Dewerchin M, Collen D, et al: Transcription the manuscript; Dr. Marco Z. Vatamaniuk for preparation factor hepatocyte nuclear factor 6 regulates pancreatic endocrine cell of adult islets; Ryan D. Wychowanec and Dr. Jonni S. differentiation and controls expression of the proendocrine gene ngn3. Mol Moore, University of Pennsylvania Flow Cytometry & Cell Cell Biol 20:4445–4454, 2000 Sorting Facility, and Drs. Elisabetta Manduchi, Joan Maza- 23. Lee JC, Smith SB, Watada H, Lin J, Scheel D, Wang J, Mirmira RG, German MS: Regulation of the pancreatic pro-endocrine gene neurogenin3. Diabe- relli, Jonathan Schugg, and Chris Stoeckert for their advice tes 50:928–936, 2001 on computational analysis. 24. Smith SB, Watada H, German MS: Neurogenin3 activates the islet differentiation program while repressing its own expression. Mol Endocrinol 18:142–149, 2004 REFERENCES 25. Smith SB, Gasa R, Watada H, Wang J, Griffen SC, German MS: Neurogenin3 1. D’Amour KA, Bang AG, Eliazer S, Kelly OG, Agulnick AD, Smart NG, and hepatic nuclear factor 1 cooperate in activating pancreatic expression Moorman MA, Kroon E, Carpenter MK, Baetge EE: Production of pancre- of Pax4. J Biol Chem 278:38254–38259, 2003 atic hormone-expressing endocrine cells from human embryonic stem 26. Watada H, Scheel DW, Leung J, German MS: Distinct gene expression cells. Nat Biotechnol 24:1392–1401, 2006 programs function in progenitor and mature islet cells. J Biol Chem 2. Bonner-Weir S, Weir GC: New sources of pancreatic beta-cells. Nat 278:17130–17140, 2003 Biotechnol 23:857–861, 2005 27. Huang HP, Liu M, El-Hodiri HM, Chu K, Jamrich M, Tsai MJ: Regulation of 3. Madsen OD: Stem cells and diabetes treatment. APMIS: Acta Pathol the pancreatic islet-specific gene BETA2 (neuroD) by neurogenin 3. Mol Microbiol Immunol Scand 113:858–875, 2005 Cell Biol 20:3292–3307, 2000 4. Shapiro AM, Lakey JR, Ryan EA, Korbutt GS, Toth E, Warnock GL, 28. Vatamaniuk MZ, Gupta RK, Lantz KA, Doliba NM, Matschinsky FM, Kneteman NM, Rajotte RV: Islet transplantation in seven patients with type Kaestner KH: Foxa1-deficient mice exhibit impaired insulin secretion due 1 diabetes mellitus using a glucocorticoid-free immunosuppressive regi- to uncoupled oxidative phosphorylation. Diabetes 55:2730–2736, 2006 men. N Engl J Med 343:230–238, 2000 29. Kaestner KH, Lee CS, Scearce LM, Brestelli JE, Arsenlis A, Le PP, Lantz KA, 5. Edlund H: Pancreatic organogenesis–developmental mechanisms and im- Crabtree J, Pizarro A, Mazzarelli J, Pinney D, Fischer S, Manduchi E, plications for therapy. Nat Rev Genet 3:524–532, 2002 Stoeckert CJ Jr, Gradwohl G, Clifton SW, Brown JR, Inoue H, Cras-Meńeur 6. Habener JF, Kemp DM, Thomas MK: Minireview: transcriptional regulation C, Permutt MA: Transcriptional program of the endocrine pancreas in mice in pancreatic development. Endocrinology 146:1025–1034, 2005 and humans. Diabetes 52:1604–1610, 2003 7. Lantz KA, Kaestner KH: Winged-helix transcription factors and pancreatic 30. Scearce LM, Brestelli JE, McWeeney SK, Lee CS, Mazzarelli J, Pinney DF, development. Clin Sci 108:195–204, 2005 Pizarro A, Stoeckert CJ Jr, Clifton SW, Permutt MA, Brown J, Melton DA, 8. Wilson ME, Scheel D, German MS: Gene expression cascades in pancreatic Kaestner KH: Functional genomics of the endocrine pancreas: the pan- development. Mech Dev 120:65–80, 2003 creas clone set and PancChip, new resources for diabetes research. 9. Jensen J: Gene regulatory factors in pancreatic development. Dev Dyn Diabetes 51:1997–2004, 2002 229:176–200, 2004 31. Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis 10. Gradwohl G, Dierich A, LeMeur M, Guillemot F: neurogenin3 is required B, Gautier L, Ge Y, Gentry J, et al.: Bioconductor: open software develop- for the development of the four endocrine cell lineages of the pancreas. ment for computational biology and bioinformatics. Genome Biol 5:R80, Proc Natl Acad SciUSA97:1607–1611, 2000 2004 11. Heller RS, Jenny M, Collombat P, Mansouri A, Tomasetto C, Madsen OD, 32. Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, Mellitzer G, Gradwohl G, Serup P: Genetic determinants of pancreatic Speleman F: Accurate normalization of real-time quantitative RT-PCR data epsilon-cell development. Dev Biol 286:217–224, 2005 by geometric averaging of multiple internal control genes. Genome Biol 12. Lee CS, Perreault N, Brestelli JE, Kaestner KH: Neurogenin 3 is essential 3:0034.1–0034.12, 2002 for the proper specification of gastric enteroendocrine cells and the 33. Leek JT, Monsen E, Dabney AR, Storey JD: EDGE: extraction and analysis maintenance of gastric epithelial cell identity. Genes Dev 16:1488–1497, of differential gene expression. Bioinformatics 22:507–508, 2006 2002 34. Storey JD, Dai JY, Leek JT: The optimal discovery procedure for large- 13. Jenny M, Uhl C, Roche C, Duluc I, Guillermin V, Guillemot F, Jensen J, scale significance testing, with applications to comparative microarray Kedinger M, Gradwohl G: Neurogenin3 is differentially required for endo- experiments. Biostatistics 8:414–432, 2007 crine cell fate specification in the intestinal and gastric epithelium. EMBO 35. Tusher VG, Tibshirani R, Chu G: Significance analysis of microarrays J 21:6338–6347, 2002 applied to the ionizing radiation response. Proc Natl Acad SciUSA 14. Naya FJ, Huang HP, Qiu Y, Mutoh H, DeMayo FJ, Leiter AB, Tsai MJ: 98:5116–5121, 2001 Diabetes, defective pancreatic morphogenesis, and abnormal enteroendo- 36. Grant GR, Liu J, Stoeckert CJ Jr: A practical false discovery rate approach crine differentiation in BETA2/neuroD-deficient mice. Genes Dev 11:2323– to identifying patterns of differential expression in microarray data. 2334, 1997 Bioinformatics 21:2684–2690, 2005 15. Sosa-Pineda B, Chowdhury K, Torres M, Oliver G, Gruss P: The Pax4 gene 37. Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki is essential for differentiation of insulin-producing beta cells in the RA: DAVID: Database for Annotation, Visualization, and Integrated Discov- mammalian pancreas. Nature 386:399–402, 1997 ery. Genome Biol 4:P3, 2003 16. Schwitzgebel VM, Scheel DW, Conners JR, Kalamaras J, Lee JE, Anderson 38. White P, Brestelli JE, Kaestner KH, Greenbaum LE: Identification of DJ, Sussel L, Johnson JD, German MS: Expression of neurogenin3 reveals transcriptional networks during liver regeneration. J Biol Chem 280:3715– an islet cell precursor population in the pancreas. Development 127:3533– 3722, 2005 3542, 2000 39. Prado CL, Pugh-Bernard AE, Elghazi L, Sosa-Pineda B, Sussel L: Ghrelin 17. Heremans Y, Van De Casteele M, in’t Veld P, Gradwohl G, Serup P, Madsen cells replace insulin-producing beta cells in two mouse models of pancreas O, Pipeleers D, Heimberg H: Recapitulation of embryonic neuroendocrine development. Proc Natl Acad SciUSA101:2924–2929, 2004 differentiation in adult human pancreatic duct cells expressing neurogenin 40. Johansson KA, Dursun U, Jordan N, Gu G, Beermann F, Gradwohl G, 3. J Cell Biol 159:303–312, 2002 Grapin-Botton A: Temporal control of neurogenin3 activity in pancreas 18. Harb G, Heremans Y, Heimberg H, Korbutt GS: Ectopic expression of progenitors reveals competence windows for the generation of different neurogenin 3 in neonatal pig pancreatic precursor cells induces (trans)dif- endocrine cell types. Dev Cell 12:457–465, 2007 ferentiation to functional alpha cells. Diabetologia 49:1855–1863, 2006 41. Song HH, Shi W, Xiang YY, Filmus J: The loss of glypican-3 induces 19. Gu G, Brown JR, Melton DA: Direct lineage tracing reveals the ontogeny of alterations in Wnt signaling. J Biol Chem 280:2116–2125, 2005 pancreatic cell fates during mouse embryogenesis. Mech Dev 120:35–43, 42. Grozdanov PN, Yovchev MI, Dabeva MD: The oncofetal protein glypican-3 2003 is a novel marker of hepatic progenitor/oval cells. Lab Invest 86:1272–1284, 20. Apelqvist A, Li H, Sommer L, Beatus P, Anderson DJ, Honjo T, Hrabe de 2006 Angelis M, Lendahl U, Edlund H: Notch signalling controls pancreatic cell 43. Gu G, Wells JM, Dombkowski D, Preffer F, Aronow B, Melton DA: Global differentiation. Nature 400:877–881, 1999 expression analysis of gene regulatory pathways during endocrine pancre- 21. Jensen J, Heller RS, Funder-Nielsen T, Pedersen EE, Lindsell C, Weinmas- atic development. Development 131:165–179, 2004

DIABETES, VOL. 57, MARCH 2008 667 EXPRESSION PROFILING OF THE ENDOCRINE PANCREAS

44. Gasa R, Mrejen C, Leachman N, Otten M, Barnes M, Wang J, Chakrabarti 54. Breslin MB, Wang HW, Pierce A, Aucoin R, Lan MS: Neurogenin 3 recruits S, Mirmira R, German M: Proendocrine genes coordinate the pancreatic CBP co-activator to facilitate histone H3/H4 acetylation in the target gene islet differentiation program in vitro. Proc Natl Acad SciUSA101:13245– INSM1. FEBS Lett 581:949–954, 2007 13250, 2004 55. Michaud JL, Rosenquist T, May NR, Fan CM: Development of neuroendo- 45. Mellitzer G, Bonne S, Luco RF, Van De Casteele M, Lenne-Samuel N, crine lineages requires the bHLH-PAS transcription factor SIM1. Genes Dev Collombat P, Mansouri A, Lee J, Lan M, Pipeleers D, et al: IA1 is 12:3264–3275, 1998 NGN3-dependent and essential for differentiation of the endocrine pan- 56. Treff NR, Vincent RK, Budde ML, Browning VL, Magliocca JF, Kapur V, creas. EMBO J 25:1344–1352, 2006 Odorico JS: Differentiation of embryonic stem cells conditionally express- 46. Dor Y, Brown J, Martinez OI, Melton DA: Adult pancreatic beta-cells are ing neurogenin 3. Stem Cells 24:2529–2537, 2006 formed by self-duplication rather than stem-cell differentiation. Nature 57. Murdaca J, Treins C, Monthouel-Kartmann MN, Pontier-Bres R, Kumar S, 429:41–46, 2004 Van Obberghen E, Giorgetti-Peraldi S: Grb10 prevents Nedd4-mediated 47. Petri A, Ahnfelt-Ronne J, Frederiksen KS, Edwards DG, Madsen D, Serup vascular endothelial growth factor receptor-2 degradation. J Biol Chem P, Fleckner J, Heller RS: The effect of neurogenin3 deﬁciency on pancre- 279:26754–26761, 2004 atic gene expression in embryonic mice. J Mol Endocrinol 37:301–316, 58. Vecchione A, Marchese A, Henry P, Rotin D, Morrione A: The Grb10/Nedd4 2006 complex regulates ligand-induced ubiquitination and stability of the insu- 48. Herrera PL: Adult insulin- and glucagon-producing cells differentiate from lin-like growth factor I receptor. Mol Cell Biol 23:3363–3372, 2003 two independent cell lineages. Development 127:2317–2322, 2000 59. Saxena R, Voight BF, Lyssenko V, Burtt NP, de Bakker PI, Chen H, Roix JJ, 49. Zhou Q, Law AC, Rajagopal J, Anderson WJ, Gray PA, Melton DA: A Kathiresan S, Hirschhorn JN, Daly MJ, et al: Genome-wide association multipotent progenitor domain guides pancreatic organogenesis. Dev Cell analysis identiﬁes loci for type 2 diabetes and triglyceride levels. Science 13:103–114, 2007 316:1331–1336, 2007 50. Artavanis-Tsakonas S, Rand MD, Lake RJ: Notch signaling: cell fate control 60. Scott LJ, Mohlke KL, Bonnycastle LL, Willer CJ, Li Y, Duren WL, Erdos MR, and signal integration in development. Science 284:770–776, 1999 Stringham HM, Chines PS, Jackson AU, et al: A genome-wide association 51. Sugimori M, Nagao M, Bertrand N, Parras CM, Guillemot F, Nakafuku M: study of type 2 diabetes in Finns detects multiple susceptibility variants. Combinatorial actions of patterning and HLH transcription factors in the Science 316:1341–1345, 2007 spatiotemporal control of neurogenesis and gliogenesis in the developing 61. Zeggini E, Weedon MN, Lindgren CM, Frayling TM, Elliott KS, Lango H, spinal cord. Development 134:1617–1629, 2007 Timpson NJ, Perry JR, Rayner NW, Freathy RM, et al: Replication of 52. Wice BM, Bernal-Mizrachi E, Permutt MA: Glucose and other insulin genome-wide association signals in UK samples reveals risk loci for type 2 secretagogues induce, rather than inhibit, expression of Id-1 and Id-3 in diabetes. Science 316:1336–1341, 2007 pancreatic islet beta cells. Diabetologia 44:453–463, 2001 62. Fukada A, Kawaguchi Y, Furuyama K, Kodama S, Horiguchi M, Kuhara T, 53. Hua H, Zhang YQ, Dabernat S, Kritzik M, Dietz D, Sterling L, Sarvetnick N: Koizumi M, Boyer DF, Fujimoto K, Doi R, et al: Ectopic pancreas formation BMP4 regulates pancreatic progenitor cell expansion through Id2. J Biol in Hes1-knockout mice reveals plasticity of endodermal progenitors of the Chem 281:13574–13580, 2006 gut, bile duct, and pancreas. J Clin Invest 116:1484–1493, 2006

668 DIABETES, VOL. 57, MARCH 2008