Research Resource: The Pdx1 Cistrome of Pancreatic Islets

Abstract

The homeodomain transcription factor pancreas duodenal homeobox 1 (Pdx1, also known as insulin promoter factor 1) is a master regulator of pancreas development, as mice or humans lacking Pdx1 function are a pancreatic. Importantly, heterozygous mutations in Pdx1 cause early and late onset forms of diabetes in humans. Despite these central roles in development and adult β-cell function, we have only rudimentary knowledge of the transcriptome targets of Pdx1 that mediate these phenotypes. Therefore, we performed global location analysis of Pdx1 occupancy in pancreatic islets. We used evolutionary conservation of target genes to identify the most relevant Pdx1 targets by performing chromatin immunoprecipitation sequencing on both human and mouse islets. Remarkably, the conserved target set is highly enriched for genes annotated to function in endocrine system and metabolic disorders, various signaling pathways, and cell survival, providing a molecular explanation for many of the phenotypes resulting from Pdx1 deficiency.

Diabetes is one of the great public health challenges of our time. Patients with type 2 diabetes are twice as likely to die as nondiabetics of the same age (1). In addition to the obvious environmental component, complex genetic factors contribute to the insulin resistance and β-cell failure that characterize type 2 diabetes. Although the vast majority of type 2 diabetes cases are polygenic in nature, monogenic forms of diabetes have provided insights into disease pathogenesis. Maturity onset diabetes of the young (MODY) is a form of monogenic diabetes caused by autosomal dominant mutations that lead to early onset diabetes. The majority of MODY mutations occur in genes encoding transcription factors, including MODY1, MODY3, MODY4, MODY5, and MODY6, which are caused by mutations in HNF4A, HNF1A, PDX1, HNF1B, and NEUROD1, respectively. Homozygous mutation in PDX1 in humans and mice causes pancreatic agenesis, and heterozygous mutations cause glucose intolerance and early and late onset forms of diabetes (2–8).

Pancreas duodenal homeobox 1 (Pdx1) is a homeobox transcription factor expressed throughout the pancreatic endoderm during development and restricted to high-level expression in β-cells postnatally (9). Studies of Pdx1 regulation of insulin gene transcription suggested that Pdx1 is involved in recruiting the histone acetyl transferase p300, which mediates histone H4 acetylation, and the histone methyltransferase Set9, which leads to H3K4 methylation (10, 11). These chromatin modifications allow the activation of RNA polymerase II to induce transcription. Pdx1 regulates target genes through interaction with a number of cofactors. Pdx1 is part of the Parahox gene cluster related to the homeobox (Hox) factors, which are necessary for patterning across the anterior-posterior axis in vertebrates (12). Hox factors have been shown to cooperatively bind DNA with Pbx to regulate target genes; similarly, Pdx1 has been found to interact with Pbx (13, 14). Pdx1 has been shown to synergize with Pbx1, Prep1, and paired box (Pax) 6 in the regulation of somatostatin; to form a complex with Pbx1b and Meis2 on the elastase 1 promoter in acinar cells; to function with Pbx1b and Meis1a to repress keratin 19 in ductal cells; and to interact with β2/NeuroD1 to regulate insulin in β-cells (15–19). During development, Pdx1 also interacts with Hnf6 to regulate Neurogenin 3, a transcription factor required for endocrine cell specification (20).

Pdx1^+/− mice have impaired insulin secretion and increased β-cell apoptosis, factors contributing to β-cell failure during type 2 diabetes (21, 22). Transcriptional targets of Pdx1 that mediate its roles in insulin secretion include insulin, the glucose transporter Glut2, and the transcription factor MafA (23–25). To identify additional targets dysregulated during Pdx1 deficiency that contribute to the diabetic phenotype, several studies have used gene expression arrays. Microarray analysis of embryonic d10.5 pancreatic buds from Pdx1^−/− embryos found down-regulation of numerous targets, including the transcription factors Pax6, Nkx6.1, and Ptf1a; the fibroblast growth factor receptor FgfR2IIIb; and the extracellular matrix protein Spondin1 (26). Microarray analysis of rat islets infected with adenoviruses encoding a dominant negative version of Pdx1 showed down-regulation of several genes involved in metabolism, including nd1, a mitochondrially encoded subunit of the reduced nicotinamide adenine dinucleotide dehydrogenase complex 1 of the mitochondrial respiratory chain (27). Pdx1 was shown to regulate nd1 indirectly through regulation of the mitochondrial transcription factor TFAM (27). Dramatic reduction of Pdx1 expression in the doxycycline-regulated Pdx1-rtTA; TetO-Pdx1 genetic model led to reduced expression of TFAM, a regulator of mitochondrial DNA copy number and insulin secretion (28, 29). In mouse insulinoma, MIN6 cells infected with lentivirus encoding short hairpin RNA targeting Pdx1, several proapoptotic genes were up-regulated, including Nix, although the mechanism of regulation has not been established (30). Pdx1 deficiency increased Nix, a mediator of mitochondrial permeability transition-dependent necrosis, and genetic disruption of Nix rescued β-cell mass and survival in Pdx1^+/− mice (30). We previously found increased β-cell apoptosis in high-fat diet-fed Pdx1^+/− mice compared with Pdx1^+/+ littermates. In MIN6 cells infected with adenovirus encoding short hairpin RNA targeting Pdx1, genes expressed in the endoplasmic reticulum (ER) were found to be down-regulated, indicating that Pdx1 may regulate ER homeostasis (31).

High-throughput chromatin occupancy arrays have also been used to identify direct Pdx1 targets. Although limited to the cis-regulatory elements represented on the array, chromatin immunoprecipitation (ChIP)-on-chip analysis in the mouse insulinoma cell line NIT1 identified Pdx1 occupancy of numerous metabolic genes, and Pdx1 and NeuroD were found to cooperate in the regulation of microRNA, miR-375, which regulates insulin secretion (32, 33). ChIP-on-chip also identified Atf4 and Wfs1 as direct Pdx1 targets (31).

To expand our understanding of genes directly regulated by Pdx1, we performed ChIP sequencing (ChIPSeq) in human and mouse islets. Pdx1 occupancy in human islets was compared with occupancy in mouse islets to focus on genes that may contribute to the phenotype shared by both species. We analyzed gene ontology, performed de novo motif analysis, assessed enrichment for motifs that bind neighboring factors, and compared occupancy with genes differentially regulated in Pdx1^+/− islets. Remarkably, the conserved target set is highly enriched for genes that function in endocrine system and metabolic disorders, various signaling pathways and cell survival, providing a molecular explanation for many of the phenotypes resulting from Pdx1 deficiency.

Results

Identification of genome-wide Pdx1 binding sites in human and mouse islets

Pdx1 is a homeodomain transcription factor with critical roles in β-cell function and survival in mature islets. Several direct and/or indirect Pdx1 targets explaining some of the phenotypes associated with Pdx1 deficiency have been identified by high-throughput cDNA microarrays (26, 27, 30, 31). Whether these changes in gene expression associated with Pdx1 deficiency involve direct occupancy or indirect regulation remains to be elucidated. Therefore, to expand the understanding of direct Pdx1 targets that mediate vital functions in β-cells, we analyzed genome-wide Pdx1 chromatin occupancy by ChIPSeq. Because human Pdx1 mutations cause type 2 diabetes and MODY4, we analyzed Pdx1 occupancy in human islets harvested from nondiabetic donor pancreata. Chromatin harvested from islets was immunoprecipitated with Pdx1 antiserum, and the resulting DNA was used to construct libraries, which were sequenced with the Illumina Genome Analyzer (Illumina, San Diego, CA). Pdx1 ChIPSeq was performed on three biological replicates, and the sequence tags were aligned to the human genome (hg18). The pooled data resulted in 15 million reads, which were analyzed by GLobal Identifier of Target Regions (GLITR), a peak calling program that evaluates fold change over input and peak height to identify significant regions and calculate false discovery rate (FDR) (34). This was followed by profile smoothing and splitting to identify individual peaks. At an FDR of 0.5%, 29,410 peaks were identified and further filtered to the top 15,000 high-quality peaks based on a rigorous threshold defined by motif analysis. Analysis of the locations of Pdx1 binding sites in human islets showed that 5% of Pdx1 binding regions were in the proximal promoter within 1 kb of the transcriptional start site (TSS), 48% at intronic sites, 35% at intergenic locations greater than 5 kb from the nearest gene, and approximately 3% at distal sites 1–5 kb from the TSS, in the 5′UTR (untranslated region), in the 3′UTR, or in coding sequences (Fig. 1A).

Fig. 1.

Identification of genome-wide Pdx1 binding sites in human and mouse islets. A, Distribution of Pdx1 binding sites in human islets relative to genomic landmarks. Pooled data from three ChIPSeq (chromatin immunoprecipitation sequencing). B, Distribution of Pdx1 binding sites in mouse islet ChIPSeq relative to genomic landmarks. C, Venn diagram of Pdx1 binding regions in human islets compared with mouse islets. D, Venn diagram of Pdx1 occupied genes in human islets compared with mouse islets. CDS, Coding sequence.

Pdx1-deficient mouse models have contributed significantly to our understanding of Pdx1 function in vivo. Islets from Pdx1^+/− mice have impaired insulin secretion, reduced insulin content, and increased β-cell apoptosis (21, 22, 31). Pdx1 ChIPSeq was performed in mouse islets to identify potential direct targets mediating these effects. The sequence tags from the mouse islet ChIPSeq were aligned to the mouse genome (mm8). Pdx1 ChIPSeq in mouse islets resulted in 7 million reads and 48,299 peaks at an FDR of 0.5% by GLITR analysis, which were filtered to the top 15,000 high-quality peaks using motif analysis. In contrast to the location of Pdx1 binding sites in human islets, Pdx1 binding in mouse islets occurred more often (18%) at the proximal promoter and less often (22%) at intergenic sites, with the remaining distributed 41% at intronic sites, 6% at distal sites, 4% in the 5′UTR, 1% coding sequences, and 1% in the 3′UTR (Fig. 1B).

To identify Pdx1 direct target genes most likely to contribute to in vivo phenotypes, we compared Pdx1 occupancy in human and mouse islets by utilizing the UCSC liftOver tool to map 50% (7583 of 15,000) high-quality human peaks to mouse genomic sites. We observed overlapping Pdx1 occupancy of 1206 sites, which corresponds to approximately 8% of the high-quality mouse peaks, leaving 13,794 high-quality sites that are specifically bound in each species (Fig. 1C). We also compared the overlap between human and mouse islet occupancy at the gene level and found Pdx1 occupancy of 2824 genes in both species, 1646 genes specifically in human islets, and 5052 genes specifically in mouse islets (Fig. 1D). Comparison of the Pdx1 binding regions without the motif filter showed 21% of the binding regions in human islets overlap with mouse islets, and 6% of binding regions in mouse islets overlap with human islets, whereas comparison at the gene level shows 84% of occupied genes in human islets overlap with mouse islets and 42% of occupied genes in mouse islets overlap with human islets (Supplemental Fig. 1, published on The Endocrine Society's Journals Online web site at http://mend.endojournals.org). The greater overlap at the gene level than the peak level suggests that Pdx1 binding regions are not well conserved from mouse to human, which is consistent with the observation of greater Pdx1 binding in proximal promoter regions in mouse islets and greater intronic binding in human islets. Previous ChIPSeq analyses of PPARG and FOXA2 in human and mouse tissues also observed limited binding site conservation (∼9%) and much greater conservation at the gene level (∼50%) (35).

These ChIPSeq analyses indicated that Pdx1 occupies a surprisingly large number of genes. We performed Ingenuity analysis of the genes occupied in human and mouse islets to broadly examine the functional categories represented by the Pdx1 target genes. Consistent with the known functions of Pdx1, this analysis showed very high P values for association with diabetes, which is categorized under endocrine system, gastrointestinal, metabolic, and genetic disorders, as well as for molecular and cellular functions in gene expression and cell death (Fig. 2). In contrast, functional characterization of the genes specifically occupied in human islets showed lower P values and involvement in infectious disease. Similarly, the genes specifically occupied in mouse islets showed involvement in genetic disorders without clear enrichment for diabetes-associated genes, suggesting that conservation of gene occupancy between species may be generally useful to focus analysis on biologically relevant transcription factor targets. Pdx1 occupancy of genes involved in diabetes is well conserved between human and mouse islets, consistent with the similar phenotype between species.

Fig. 2.

Pdx1 occupancy conserved in genes associated with diabetes. Ingenuity analysis of human islet specific (left), mouse islet specific (right), and conserved genes (center) occupied by Pdx1. Top pathways with P < 10⁻⁵.

Motifs in Pdx1 binding regions

Pdx1 has been shown to bind to the consensus motif TAAT (9, 24). To determine the motifs enriched in the Pdx1 binding regions throughout the mouse and human genomes, we performed a BioProspector analysis to identify motifs overrepresented within 30 bp of the peak centers. Peaks were grouped based on fold change, and each band of 500 peaks was analyzed for motifs separately. Two main motifs emerged from this analysis. The most prominent motif in mouse islets was T(A/G)AT except in the top band containing the highest enrichments for Pdx1 occupancy, where a sequence similar to the somatostatin UE-A site, TGATTGAT, was identified (Fig. 3A and Supplemental Fig. 2A) (15). De novo motif analysis of the Pdx1 binding regions in human islets identified highest enrichment of the TGATNNAT, as well as significant enrichment of the TAAT motif (Fig. 3B and Supplemental Fig. 2B). Pdx1 interaction with a cofactor may contribute to binding to this motif in human islets, because previous studies have shown that this consensus sequence binds to Pbx-Hox heterodimers (13).

Fig. 3.

Pdx1 binding motif in mouse and human islets. A, BioProspector de novo motif analysis of Pdx1 binding sites in mouse islets. B, Motif analysis of Pdx1 binding sites in human islets. C, EMSA showing in vitro-translated (IVT) human Pdx1 and Pbx1a binding of probes containing the motifs present in human sequences from Pdx1 binding regions in glucokinase and Tcf7l2. Lanes contain radiolabeled probes incubated with 1) IVT control, 2) IVT Pdx1, and 3) IVT Pbx1a and 4–7) IVT Pdx1 and Pbx1 with antisera to show specificity: 4) control serum, 5) anti-Pdx1 serum, 6) control IgG, and 7) anti-Pbx1 IgG. Arrows show probe complex with Pdx1 with or without Pbx1, asterisks show supershift. D, Pdx1 and Pbx1 ChIP in human islets. Fold enrichment compared with control IgG, enrichment of the albumin promoter was measured as a negative control. Representative of three independent ChIP.

Next, we investigated the representation of binding sites [position weight matrices (PWM)] of known transcription factors near Pdx1 targets in mouse islets (Table 1). Interestingly, several critical pancreatic transcription factors with developmental and postnatal roles were represented among the most significant PWM, including Pbx, Pax4, Nkx6, and GATA (25, 36–41). Analysis of PWM in human islets identified similar factors, and Pbx had the highest area under the curve (AUC), consistent with the Pbx-Hox motif identified above (Table 2). Other potentially interesting PWM include factors that may regulate β-cell mass, for example, cAMP response element-binding (CREB) protein, and factors that regulate pancreas development, such as Hnf1, MafB, and Pax2 (42–48). The complete list of PWM in mouse islets and human islets with AUC values greater than 0.6 is presented in Supplemental Table 1.

Table 1.

Mouse islet neighboring factors

Transcription factor	AUC
IPF1/PDX1	0.81
OG2	0.80
PBX	0.78
CDPCR1	0.78
S8	0.78
PAX4	0.74
CHX10	0.73
OCT1	0.73
CDPCR3HD	0.73
LHX3	0.72
TCF11	0.71
DTYPEPA	0.71
HOXA4	0.71
NKX62	0.70
NKX25	0.70
SOX17	0.70
CRX	0.70
AFP1	0.70
GATA6	0.69
KID3	0.69

Consensus binding sites enriched near mouse Pdx1 targets. Top 20 PWM based on AUC identified in Pdx1 binding regions in mouse islets.

Table 2.

Human islet neighboring factors

Transcription factor	AUC
PBX	0.82
IPF1/PDX1	0.81
AP1	0.79
S8	0.77
CDPCR1	0.77
TCF11	0.76
AP1FJ	0.75
BACH2	0.75
LHX3	0.75
OG2	0.75
NRF2	0.74
PAX2	0.73
CREB	0.73
NFE2	0.72
HNF1	0.72
GATA6	0.72
MAFB	0.71
CDPCR3HD	0.71
CRX	0.71
GATA	0.71

Consensus binding sites enriched near human Pdx1 targets. Top 20 PWM based on AUC identified in Pdx1 binding regions in human islets. CREB, cAMP response element-binding protein.

To determine whether PDX1 can interact with PBX1 to bind regions that contain the de novo motif identified in human islets, we performed EMSA using probes containing variations of the motif that were present under Pdx1 ChIPSeq peaks. Radiolabeled probes were incubated with in vitro-translated human PDX1 alone or in combination with human PBX1a. We observed binding by both PDX1 and PBX1a to a sequence from the glucokinase gene containing the motif TGATGGAT (Fig. 3C). Specificity of binding was confirmed by using Pdx1 antiserum and Pbx1 antiserum to supershift the complexes. Similarly, we identified PDX1 and PBX1a binding to probes generated from the human TCF7L2 gene with binding motifs TGACTAAT and TGATGTAT. TCF7L2 is a Wnt signaling mediator with single nucleotide polymorphisms strongly associated with type 2 diabetes (49, 50). The TCF7L2 sequence containing TGACTAAT demonstrated strong binding with PDX1 alone, consistent with the presence of the TAAT motif, and formed a new complex with the addition of PBX1a, whereas the TGATGTAT sequence required addition of PBX1a for strong binding by PDX1. Both motifs were derived from the same region of the TCF7L2 gene, and PDX1 and PBX1 ChIP in human islets showed 3.7- and 2.8-fold enrichment of that region, respectively (Fig. 3D). These results confirm that PDX1 indeed binds to the motif TGATNNAT identified by our ChIPSeq analysis.

Integration of global occupancy data with gene expression profiling of Pdx1^+/− islets

To identify functional Pdx1 targets that are both occupied and differentially regulated, we compared the global occupancy data with gene expression profiling of Pdx1^+/− islets (31). In mouse islets, 121 genes were found to be altered more than 1.5-fold between Pdx1^+/+ and Pdx1^+/− islets on the microarray. When compared with the 2824 genes occupied in human and mouse islets, 33 genes were occupied and differentially regulated (Fig. 4A). It should be noted that we could not analyze Pdx1 null β-cells for expression profiling, because Pdx1^−/− mice are not viable. It seems likely that the overlap between occupancy and differential gene expression would be much larger if Pdx1 were absent. Two well-established Pdx1 targets, MafA and Glut2 (Slc2a2), as well as ER oxidoreductin 1-like β, a direct Pdx1 target that regulates insulin content and susceptibility to ER stress-induced cell death, were present on this list (Fig. 4B) (24, 25, 51). Among the novel potential targets, urocortin 3 (Ucn3) is a peptide in the corticotropin releasing factor family that amplifies insulin secretion, and tetraspanin 8 (Tspan8) was identified in genome-wide association studies as a type 2 diabetes risk factor (52, 53). To assess regulation of Ucn3 and Tspan8, we used MIN6 cells, which by ChIPSeq analysis is a highly similar model of Pdx1 occupancy compared with mouse islets (Supplemental Fig. 3). Pdx1 ChIP, using antisera distinct from those used for the ChIPSeq, resulted in 7.4-fold enrichment of Ucn3 and 3.4-fold enrichment of Tspan8 compared with IgG control in MIN6 cells (Fig. 4C). To confirm that the genes were differentially regulated after Pdx1 silencing, Ucn3 and Tspan8 transcript levels were measured in MIN6 cells nucleofected with pooled small interfering RNA (siRNA) duplexes targeting Pdx1. Ucn3 transcript levels were significantly reduced by 40%, whereas Tspan8 transcript levels were increased 3.3-fold (Fig. 4D). In human islets, Pdx1 ChIP showed 2-fold enrichment of Ucn3 and 6-fold enrichment of Tspan8 (Fig. 4E). In sum, analysis of global occupancy and gene expression uncovered Pdx1 regulation of novel targets, Ucn3 and Tspan8, which are potentially important for β-cell function and relevant to the pathogenesis of diabetes.

Fig. 4.

Integration of global occupancy results with Pdx1^+/− expression array. A, Genes occupied in human and mouse islets compared with genes significantly differentially regulated in Pdx1^+/− mouse islets compared with Pdx1^+/+ islets. B, Gene list of potential direct Pdx1 targets with absolute fold change greater than 1.5. C, Pdx1 ChIP in MIN6 cells. Fold enrichment compared with control IgG. Representative of three independent ChIP. D, Amaxa nucleofection of MIN6 cells with Pdx1 siRNA duplexes leads to decreased Ucn3 and increased Tspan8 expression compared with nucleofection of nontargeting siRNAs (siNT) (n = 5). **, P < 0.005. E, Pdx1 ChIP in human islets, fold enrichment compared with IgG, representative of three independent ChIP.

PDX1 occupancy of targets involved in cell death

Although informative, gene expression arrays are limited by the genes represented on the array and by the level of sensitivity of the assay. To broaden our analysis of Pdx1 occupied genes, the genes directly occupied in human and mouse islets were analyzed by Ingenuity software and by Database for Annotation, Visualization, Integration, and Discovery (DAVID). Because Pdx1 deficiency results in increased β-cell apoptosis, we focused on genes occupied by Pdx1 that are potentially involved in cell death (Supplemental Table 2). DAVID analysis drew our attention to the proapoptotic BH3-only Bcl2 family members Noxa (also known as PMAIP1) and Puma (also known as BBC3), which were not represented on the Pdx1^+/− islet microarray. Puma is regulated by cytokines and ER stress in β-cells and was previously shown to be up-regulated by lentiviral silencing of Pdx1 (30, 54). Puma and Noxa mediate apoptosis by binding antiapoptotic Bcl2 family members, causing cytochrome c release and caspase cleavage (55, 56). We confirmed Pdx1 occupancy of Puma and Noxa by ChIP and found 16- and 3.3-fold enrichment, respectively (Fig. 5A). Furthermore, Puma and Noxa were up-regulated by 1.3- and 4.6-fold, respectively, in MIN6 cells nucleofected with siPdx1 (Fig. 5B). Pdx1 ChIP in human islets showed 7-fold enrichment of Puma and 2.8-fold enrichment of Noxa, suggesting relevance for the in vivo phenotype (Fig. 5C). These findings demonstrate Pdx1 occupancy and regulation of the proapoptotic factors Puma and Noxa, as well as the broader utility of our dataset for identifying Pdx1 targets that mediate critical phenotypes of Pdx1 deficiency.

Fig. 5.

Pdx1 directly regulates proapoptotic factors Puma and Noxa. A, Pdx1 ChIP in MIN6 cells, representative of three independent ChIP. B, Silencing of Pdx1 in MIN6 cells leads to up-regulation of Puma and Noxa (n = 6). *, P < 0.05; **, P < 0.0005. C, Pdx1 ChIP in human islets, representative of three independent ChIP. siNT, Small interfering nontargeting oligonucleotide.

Discussion

Pdx1 is necessary for pancreas development and postnatal β-cell function and survival. Studies attempting to reprogram other cell types into β-cells have determined that Pdx1 expression, in concert with Neurog3 and MafA, is sufficient for the generation of insulin-expressing cells (57). Here, we expand our understanding of Pdx1 transcriptional targets by genome-wide analysis of occupancy. To our knowledge, this is the first characterization of global Pdx1 occupancy in human islets. Previous analysis comparing Pdx1 and Foxa2 occupancy in mouse islets has been reported (58). Although we find many of the same regions (9997 of 12,454 or 80%), our mouse dataset provided about 3.7 times more targets and slightly more high-quality peaks. In addition, our datasets contain approximately 1000 sites that are conserved between mouse and human. The comparison between Pdx1 occupancy in human and mouse islets has provided insights into the conserved and divergent aspects of Pdx1 regulation between species. We observe that Pdx1 occupancy in humans differs from occupancy in mouse in terms of the location relative to the gene, with more frequent Pdx1 occupancy at the promoter, which is the classical model for transcriptional regulation, in the mouse islets, and more intronic occupancy in human islets; possible explanations requiring further investigation include differences in the cis-regulatory regions or interaction with cofactors. In both mouse and human islets, the largest fraction of binding sites occurs in the intronic regions. The functional importance of Pdx1 binding at distant intergenic regions remains to be elucidated, because microRNA or other noncoding transcripts may be present in the vast uncharacterized stretches of the genome.

De novo motif analysis identified at least two distinct motifs enriched in Pdx1 binding regions, the canonical TAAT motif and the TGATNNAT motif, which we showed to bind Pdx1-Pbx heterodimers in EMSA studies. In vitro-translated human PDX1 and PBX1a formed a complex with sequences identified from Pdx1 binding regions in the human TCF7L2 gene, suggesting that these factors may interact to regulate this important type 2 diabetes susceptibility gene. Interestingly, in human islets, the TGATNNAT motif was more prominent across various peak strengths, whereas in mouse islets, the motif was enriched in the top band of 500 peaks but was surpassed by the TAAT motif in the remaining bands. The mouse islet de novo motif results lead us to speculate that Pdx1 may cooperate with Pbx to bind highly enriched regions, and the potentially broader contribution of Pbx to Pdx1 binding is highlighted by the de novo motif identified in human islets. Additional studies are required to analyze global Pbx1 binding and correlate ChIP enrichment to functional outcomes and pathways.

We compared Pdx1 occupied genes in human islets with mouse islets to identify directly occupied targets that may be relevant for the β-cell failure phenotype present in both species. We found greater overlap at the gene level compared with the peak level, suggesting that conservation of Pdx1 binding across species may not necessitate binding region conservation. An alternative mode of conservation of function that requires further investigation may involve regulation of different genes within the same pathways leading to similar phenotypes in humans and mice. We found substantial overlap between Pdx1 occupied genes in human and mouse islets and high prevalence of genes involved in diabetes only in the conserved genes and not the species-specific genes, suggesting that most functions of Pdx1 in β-cells are conserved across species, and comparison of occupied genes between species is a valuable means of focusing the broad number of genes identified by ChIPSeq.

Based on the overlap with a Pdx1^+/− islet microarray, we identify Ucn3 and Tspan8 as direct targets of Pdx1. Ucn3 has been shown to potentiate insulin secretion. Therefore, the reduction in Ucn3 in the setting of Pdx1 deficiency may contribute to impaired insulin secretion (52, 59). Tspan8 was identified in type 2 diabetes genome-wide association studies and was found to be up-regulated in β-cell-enriched samples from type 2 diabetic patients, correlating with the up-regulation that we observed in the setting of Pdx1 deficiency (53, 60). Tspan8 was initially characterized as being associated with tumor metastasis and is thought to promote cell motility through interactions with the integrins and angiogenesis through regulation of vascular endothelial growth factor and matrix metalloproteinases (61–63). Tspan8 null mice have a limited phenotype of decreased body weight, but Tspan8 levels appear to differ between mouse and human pancreata (64). The precise function of Tspan8 and the effects of up-regulation in β-cells remain to be elucidated. It is notable that similar numbers of Pdx1 occupied genes were down-regulated and up-regulated; the function of Pdx1 as a transcriptional activator is well accepted, whereas its role as a transcriptional repressor is less well understood.

There is a notable discrepancy between the vast number of Pdx1 occupied genes and the number of regulated genes identified by our previous microarrays and by others (27, 30, 31). There are several possible explanations, including the limited number of genes on a microarray compared with the entire genome, the limitation of statistical significance in an array study, and potential promiscuous binding with gene expression being modulated by interaction with other factors. These limitations may be overcome by rapidly advancing technology of RNA sequencing and characterization of other transcription factors by ChIPSeq to provide a more comprehensive understanding of Pdx1 transcriptional regulation in β-cells. It is likely that additional direct functional targets will emerge from a comparison of our occupancy data with greater degrees of Pdx1 loss of function, for example, by conditional homozygous Pdx1 deletion in the adult β-cell.

Pdx1 has potent prosurvival roles in pancreatic β-cells; thus, it is not surprising that Pdx1 occupied genes were associated with cell survival. Analysis of cell death genes led to the identification of Noxa and Puma as direct targets of Pdx1, and Puma and Noxa were found to be up-regulated in the setting of Pdx1 deficiency in MIN6 cells. Although Pdx1 acts as a transcriptional activator of many of its target genes, our findings highlight a role for Pdx1 as a transcriptional repressor, as previously indicated by Pdx1 repression of keratin 19 in ductal cells and glucagon in β-cells (18, 65). Because Puma and Noxa are potent mediators of apoptosis, up-regulation of Puma and Noxa in the setting of Pdx1 deficiency may contribute to increased β-cell apoptosis, although certainly other factors are involved as well. Global analysis of Pdx1 occupancy expands our understanding of Pdx1 binding motifs, potential cofactors, and direct targets. These findings facilitate an integrated understanding of potential mediators of the β-cell failure leading to diabetes in MODY4 and other settings of Pdx1 deficiency.

Materials and Methods

Islets

Human islets were harvested from three nondiabetic donor pancreata using the modified automated Ricordi method by the Islet Biology Core at the University of Pennsylvania (66). The purity and viability of the islet preparation was assessed, and 20,000 islets were harvested for chromatin. Mouse islets were harvested from 6- to 8-wk-old male CD1 mice as previously described (31). Briefly, the pancreata were inflated with collagenase in Hanks' balanced salt solution, then digested in collagenase for 16 min at 37 C and separated by Ficoll gradient (67). Islets were handpicked, and 3000 islets were harvested for chromatin preparation and immunoprecipitation.

Cell culture

MIN6 cells were grown in DMEM with 25 mm glucose supplemented with 10% fetal bovine serum, 1% penicillin/streptomycin, and 140 μm β-mercaptoethanol. Cells were maintained at 37 C in a 5% CO₂ incubator. To determine genes differentially regulated by Pdx1, MIN6 cells were Amaxa nucleofected with 0.3 nmol of ON-TARGETplus nontargeting siRNA pool or 0.3 nmol of Pdx1 ON-TARGETplus SMARTpool (Dharmacon, Lafayette, CO). MIN6 cells were harvested 72 h after nucleofection with TRIzol reagent (Invitrogen, Carlsbad, CA). RNA isolation was performed according to the TRIzol protocol, and Superscript II (Invitrogen) was used reverse transcription. Quantitative real-time PCR was performed using SYBR green with a Bio-Rad iCycler (Bio-Rad, Hercules, CA). Primers were designed using Primer3, tested for linearity, and have the following sequences: Ucn3 [forward (F), AGCACCCGGTACAGATACCAA and reverse (R), GGCCTTGTCGATGTTGAAGAG], Tspan8 (F, CTAGGAGCCGCTTTCAAACCT and R, CAGCACTTGAACTCCGACTGA), Puma (F, CAAGAAGAGCAGCATCGACAC and R, TAAGGGGAGGAGTCCCATGAA), and Noxa (F, AGTTCGCAGCTCAACTCAGGA and R, GCGCCAGAACCACAGTTATGT). Primers for Pdx1 and Hprt were previously described (31).

ChIP and ChIPSeq

Pdx1 ChIP was performed as previously described (18). For ChIPSeq, cells were cross-linked with 1% formaldehyde, quenched with glycine, washed with PBS, and lysed to isolate chromatin. A Bioruptor sonication device (Diagenode, Denville, NJ) was used to sonicate chromatin fragments to approximately 500 bp, and samples were precleared with normal rabbit IgG (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) overnight. An aliquot was removed for the input control, and the chromatin was immunoprecipitated with Pdx1 antiserum (2). DNA was prepared according to the Illumina protocol as previously described (68). T4 DNA polymerase, Klenow polymerase, and T4 polynucleotide kinase were used to blunt the DNA fragments, then Klenow exo⁻ was used to add an adenine to the 3′ end. DNA was ligated to adapters (Illumina), sized selected by 2% agarose separation, amplified by PCR, purified with QIAquick PCR purification kit (QIAGEN, Valencia, CA), and analyzed by Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA) to determine quantity and integrity. The Illumina protocol was followed for cluster generation and sequence alignment to the human (hg18) and mouse (mm8) genomes.

Data analysis

Data will be available for download at ArrayExpress MEXP-NNNN (http://web.me.com/kaestnerlab1/GLITR/). GLITR analysis was used for peak calling (34). The GLITR algorithm compares ChIP and input (pseudo-ChIP) tags to tags in a large input pool to determine fold change to eliminate false positives that arise from sequencing bias. The fold change and peak height are integrated into calculating the FDR. Pdx1 binding in mouse islets and MIN6 cells was compared with determine overlap; a fold change difference of less than 10 was considered conserved. GLITR segments the genome into contiguous regions covered by at least two reads. To identify the center of peaks within each region, we identified local maxima in a smoothed version of the profile. The extended read profile of each GLITR region was smoothed by removing the high frequency components. The power spectrum was calculated using a fast Fourier transform, then the power of frequencies with a wavelength under a threshold was set to zero. The smoothed profile was created by back-transforming the reduced spectrum. Maxima in the smoothed profile were identified by fitting a spline, then using the locations of maxima in the spline segments. We used the height of the spline peak as the score of the peak. Peaks were grouped based on score into bands of 500 and analyzed by BioProspector to identify the de novo motifs (34) within 30 bp of the peak. A background set for each band was constructed by randomly selecting regions of the genome such that the overall distribution of CG content in the peaks and background set matched. Selected PWM were used to scan for sequences that matched the motif within 30 bp of the peaks. We computed a receiver operating characteristic curve for these PWM, then plotted the AUC as a function of the band number. In both species, this curve decreased until band 30, at with point it leveled off for the remaining bands. The top bands contained distinct motifs, and the enrichment of the motif correlated with peak strength. We used thresholds of score 26.6 in mouse and 12.2 in humans, to select the top 30 bands (15,000) peaks from each species. BioProspector was also used to determine PWM within 100 bp of the peaks.

To determine alignment between human islets and mouse islets, the UCSC liftOver tool (http://genome.ucsc.edu/cgi-bin/hgLiftOver) was used to convert hg18 peak locations to mm8 locations (35). Peaks within 50 bp were considered to be conserved between human and mouse. Pdx1 binding regions were connected to genes, to determine human islets, mouse islets, and conserved Pdx1 occupied genes. A peak is connected to any gene that is contained in. It is also connected to the gene with the closest TSS (within 100 kb). Any additional genes with TSS within 1.5 times this nearest gene were also candidates for direct regulation. The gene lists were functionally categorized by IPA (Ingenuity Systems, Redwood City, CA), and cell death targets were further analyzed by DAVID analysis (http://david.abcc.ncifcrf.gov/).

Statistical significance

Right-tailed Fisher's exact test was used to determine P value in the IPA functional analyses. Two-tailed Student's t test was used to determine statistical significance, and differences of P < 0.05 were considered significant.