Transcription factor occupancy of the insulin gene in vivo. Evidence for direct regulation by Nkx2.2.

Consensus-binding sites for many transcription factors are relatively non-selective and found at high frequency within the genome. This raises the possibility that factors that are capable of binding to a cis-acting element in vitro and regulating transcription from a transiently transfected plasmid, which would not have higher order chromatin structure, may not occupy this site within the endogenous gene. Closed chromatin structure and competition from another DNA-binding protein with similar nucleotide specificity are two possible mechanisms by which a transcription factor may be excluded from a potential binding site in vivo. Multiple transcription factors, including Pdx-1, BETA-2, and Pax6, have been implicated in expression of the insulin gene in pancreatic beta cells. In this study, the chromatin immunoprecipitation assay has been used to show that these factors do, in fact, bind to insulin control region sequences in intact beta cells. In addition, another key islet-enriched transcription factor, Nkx2.2, was found to occupy this region using the chromatin immunoprecipitation assay. In vitro DNA-binding and transient transfection assays defined how Nkx2.2 affected insulin gene expression. Pdx-1 was also shown to bind within a region of the endogenous islet amyloid polypeptide, pax-4, and glucokinase genes that were associated with control in vitro. Because Pdx-1 does not regulate gene transcription in isolation, these sequences were examined for occupancy by the other insulin transcriptional regulators. BETA-2, Pax6, and Nkx2.2 were also found to bind to amyloid polypeptide, glucokinase, and pax-4 control sequences in vivo. These studies reveal the broad application of the Pdx-1, BETA-2, Pax6, and Nkx2.2 transcription factors in regulating expression of genes selectively expressed in islet beta cells.


Transcription of the insulin (INS)
gene is restricted to pancreatic islet ␤ cells. The 5Ј-flanking sequences within 350 base pairs of the transcription start site contain the binding sites for the factors that control cell type-specific expression (1)(2)(3)(4). An extensive series of in vitro gel shift and cell transfection experiments indicates that regulation is mediated by islet-enriched DNA-binding transcription factors, including Pdx-1 (5-7), BE-TA-2 (8), and Pax6 (9). Strikingly, each of these factors is also a key regulator of pancreas development. Thus, homozygous loss of Pdx-1 leads to pancreatic agenesis in mice (10,11) and humans (12), whereas there are severe defects in islet cell formation in BETA-2 (i.e. ␤, ␣, and ␦ cells; Ref. 13) and Pax6 (i.e. ␣ cells; Ref. 14) null mice.
Islet cell development and function is also affected by other islet-enriched transcription factors (15,16). For example, insulin is not made in ␤ cells of mice lacking Nkx2.2 (17). Interestingly, this appears to be a rather specific effect as other ␤ cell-enriched markers continue to be expressed (e.g. Pdx-1 and islet amyloid polypeptide (IAPP). Nkx2.2 and insulin are also co-expressed during embryogenesis, although Nkx2.2 expression is not restricted to ␤ cells in adult islets (17). The mechanism by which Nkx2.2 regulates insulin gene transcription has not been resolved. Indeed, no direct gene targets have yet been reported for Nkx2.2 in the pancreas.
Assays such as in vitro DNA binding, co-transfection of reporter constructs, and transgenic knock-out animals provide evidence for the involvement of a factor in transcriptional control. However, such experiments are indirect measures of regulation and do not account for conditions like competition with other binding proteins in vivo, chromatin structure of the endogenous gene, and cascades of regulatory interactions. In contrast, the chromatin immunoprecipitation (ChIP) assay (18 -20) is a powerful tool for directly analyzing transcription factor site occupancy in vivo. Importantly, one can determine the physical association of a specific DNA-binding factor with potential control sequences in intact cells with this assay. This technique has been applied to many experimental systems, including in ␤ cells to examine the in vivo occupancy of control sequences within the pdx-1 gene by Pax6 and HNF3␤ (21), within the glut2 glucose transporter gene by HNF1␣ (22), and within several ␤ cell genes by Pdx-1 (23,24).
In this study, the ChIP assay was used to assess the in vivo occupancy of insulin 5Ј-flanking sequences by Pdx-1, BETA-2, Pax6, and Nkx2.2. In addition to directly demonstrating binding of Pdx-1, BETA-2, and Pax6 to these sequences in intact ␤ cells, Nkx2.2 was also found to interact. Analysis of the insulin 5Ј-flanking region target sequences by gel shift and transfected insulin-driven reporter assays suggests that Nkx2.2 reduces transcription upon binding at Ϫ140/Ϫ119 bp in ␤ cell lines. Our findings that BETA-2, Pax6, and Nkx2.2 bind within the transcription control region of other islet-enriched genes also highlights how the ChIP assay can be used to identify targeted genes.

MATERIALS AND METHODS
Cell Culture and Transfection Analysis-Monolayer cultures of insulin-producing pancreatic ␤ cell lines (␤TC-3, HIT T-15 2.2, and MIN6) were maintained as described previously (25). The Ϫ238 WT CAT (chloramphenicol acetyltransferase) plasmid contains sequences from Ϫ238 to ϩ2 bp of the rat insulin II enhancer/promoter region linked to the CAT reporter gene (26). Construction of the insulin mutant (MT) expression vector, Ϫ238 CAT Ϫ108,107 MT, has been described (25). The Nkx2.2-binding element mutant was generated in Ϫ238WT CAT using the QuikChange mutagenesis kit (Stratagene, La Jolla, CA) with an oligonucleotide containing the Ϫ133,Ϫ132 mutation (5Ј-CCTAG-CACCAGGCccGTGTTTGGAAACTGCAGC-3Ј; mutated nucleotides are underlined). The LipofectAMINE reagent (Invitrogen) was used to transfect 1 g of the Ϫ238 CAT and Rous sarcoma virus (RSV):luciferase (LUC) plasmids. The RSV enhancer-driven LUC expression plasmid, pRSVLUC, was used as a recovery marker. Extracts were prepared 40 -48 h after transfection and assayed for LUC (27) and CAT (28) enzymatic activity. The CAT activity from the test construct was normalized to RSV-driven LUC activity. Each experiment was carried out at least three times.
ChIP Assay-ChIP assays were performed on ␤TC-3 cells as described (24) with the following modifications. Protein A-Sepharose was replaced with protein A/G-Agarose (Santa Cruz Biotechnology, Inc., Santa Cruz, CA), and the following antibodies were used: 1 l of Nterminal (amino acids   Electrophoretic Mobility Shift Assay-Nuclear extracts were prepared from ␤TC-3 and MIN6 cells as described previously by Zhao et al. (25). The TNT-coupled reticulocyte lysate system (Promega, Madison, WI) was used to in vitro transcribe and translate the Nkx2.2 encoding plasmid, pNkx2.2 (17). Ten g of nuclear extract protein or 2 l of in vitro-translated protein was added to the binding buffer containing 10 mM HEPES (pH 7.4), 1 mM EDTA, 10% (v/v) glycerol, 100 mM NaCl, 2 mM dithiothreitol, and 1 g poly(dI-dC). The double-stranded oligonucleotide probes were labeled with [␥-32 P]ATP with T4 polynucleotide kinase. Binding reactions, containing 40 fmol of radiolabeled probe, were incubated at 4°C for 30 min. The competitor DNA was added at 10 -50-fold molar excess prior to the addition of probe. Anti-Nkx2.2 antibody was pre-incubated with protein at 4°C for 20 min before adding probe to the binding reaction. The protein-DNA complexes were resolved on a 6% non-denaturing polyacrylamide gel, which was then dried and exposed to autoradiographic film.

Transcription Factor Occupancy of Insulin Control Region
Sequences in Vivo-The ChIP assay was first used to investigate binding of Pdx-1 to 5Ј-flanking control sequences of the insulin gene in an islet ␤ cell line, ␤TC-3. Two distinct antibodies to Pdx-1 were used to immunoprecipitate formaldehyde cross-linked chromatin. Insulin 5Ј-flanking sequences (INS; nucleotides Ϫ378/Ϫ46) were selectively amplified by PCR from chromatin precipitated by anti-Pdx-1 antisera made to either the N-or C-terminal region, but not from chromatin treated with normal rabbit IgG or in the absence of antiserum (Fig.  1A). In contrast, Pdx-1 antisera did not immunoprecipitate control sequences from the PCK gene (nucleotides Ϫ434/Ϫ96), which is not expressed in ␤ cells (Fig. 1A). Significantly, the amplified region of the mouse PCK promoter contains two TAAT motifs that can compete for Pdx-1 binding in vitro (data not shown). Thus, although Pdx-1 can bind to both the insulin and PCK control sequences in vitro, the ChIP assay demonstrates that only Pdx-1 interacts with the endogenous insulin gene in ␤ cells.
To further characterize the in vivo occupancy of the INS transcription unit, ChIP analysis was performed with antibodies to other potential transcriptional regulators, specifically BETA2 (8) and Pax6 (9). Both anti-BETA2 (Fig. 1B) and anti-Pax6 (Fig. 1C) antibodies effectively immunoprecipitated INS control sequences from formaldehyde cross-linked ␤TC-3 chromatin, but not PCK. Neither the INS nor PCK regulatory region was amplified from chromatin isolated in control immunoprecipitations.
Finally, we tested if the endogenous INS promoter is occupied by Nkx2.2, a factor that was proposed to be involved in expression as a result of its dramatic and selective effect on insulin levels in Nkx2.2 null mice (17). Anti-Nkx2.2 antibodies, but not control reactions, immunoprecipitated INS sequences from ␤TC-3 cells (Fig. 1D), whereas the PCK promoter was not immunoprecipitated by anti-Nkx2.2. The same patterns were observed with antisera to Nkx2.2, BETA2, Pdx-1, and Pax6 Identification of an Nkx2.2-binding Site between Nucleotides Ϫ140 and Ϫ119 in the Insulin Gene-The insulin 5Ј-flanking region (Ϫ378/Ϫ46 bp) analyzed in the ChIP studies was found to contain potential Nkx2.2-binding elements at nucleotides Ϫ140 to Ϫ119 (site 1) and Ϫ193 to Ϫ172 (site 2) upon compar-ison to the Nkx2.2 consensus-binding site (29) (Fig. 2A). To determine whether Nkx2.2 binds to these INS elements, gel mobility shift assays were performed with in vitro-translated Nkx2.2 and probes to site 1, site 2, and the consensus element (Fig. 2B). Two major protein-DNA complexes were commonly formed with each element in the Nkx2.2 and control in vitrotranslated reactions (see asterisk-labeled complexes in Fig. 2A). Significantly, a unique faster mobility complex was detected in the Nkx2.2 reaction with the INS site 1 and consensus element probes, although not with INS site 2 (Fig. 2B). To determine whether Nkx2.2 was involved in formation of this faster mobility complex, antiserum specifically raised against this protein was added to the binding reactions. This complex was selec- tively super-shifted in the reactions performed in the presence of Nkx2.2 with the consensus and site 1 probe (Fig. 2B). A complex of identical mobility was formed with the INS site 1 probe in ␤TC-3 and MIN6 ␤ nuclear extracts and was also super-shifted with the anti-Nkx2.2 antibody (Fig. 2C). In addition, competition assays demonstrated that Nkx2.2 bound effectively to the consensus-binding element and INS site 1 and not to site 2 or a site 1 element with a two base pair mutation in the core-binding sequence (INS site 1 Ϫ133,Ϫ132 MT) (Fig.  2D). Collectively, these data indicated that Nkx2.2 binds to site 1 at Ϫ140 to Ϫ119 bp of the insulin gene in ␤ cells.
Nkx2.2 Appears to Inhibit Insulin Gene Expression-To understand how Nkx2.2 influences insulin transcription, the Ϫ133/Ϫ132 mutant of site 1 that disrupted binding was introduced into an insulin enhancer/promoter-driven CAT reporter. The effect on activation was compared with a mutant in the RIPE3b1/C1 activator site at Ϫ108/Ϫ107 bp and the wildtype Ϫ238/ϩ2bp CAT construct in transfected HIT T-15, ␤TC-3, and MIN6 ␤ cell lines. Regulatory patterns for both the INS site 1 and RIPE3b1/C1 mutated constructs were similar in all three ␤ cell lines. Thus, activation was stimulated 1.8-(HIT T-15), 3.3-(␤TC-3), and 1.8-(MIN6) fold in the site 1 mutant and, as expected, reduced in the RIPE3b1 activator-binding site mutant (25) (Fig. 3). These results suggest that Nkx2.2 functions as a negative regulator of insulin expression in ␤ cell lines.
Pdx-1, Nkx2.2, BETA2, and Pax6 Also Bind in Vivo within the Control Region of Other ␤ Cell-enriched Gene Products-Because ␤ cell-specific expression of the insulin gene appears to be controlled by the combinatorial actions of Pdx-1, BETA2, and Pax6 (30 -32), the ChIP assay was used with ␤TC-3 cells to determine whether these factors also bound within control region sequences of other islet-enriched gene products, specifically IAPP, GK, and pax-4. The Pdx-1-specific antibodies were found to selectively immunoprecipitate the control sequences of the IAPP (33-35), GK (36), and pax-4 (37) genes that bound Pdx-1 in vitro and were mutationally sensitive in transfection analyses (Fig. 4A).
IAPP and pax-4 control sequences were also immunoprecipitated from ␤TC-3 cells by anti-BETA2 antibodies but not by control antibodies (Fig. 4B). These results confirm the in vitro studies that indicated that BETA2 was involved in pax-4 expression (37). BETA2 has also recently been shown to bind to the GK promoter in vivo and stimulate expression (38). In addition, antisera to Pax6 (Fig. 4C) and Nkx2.2 (Fig. 4D) selectively immunoprecipitated IAPP, GK, and pax-4 control sequences. Interestingly, the pax-4 promoter has been shown to contain a mutationally sensitive binding site that may be regulated by a factor in the Nkx2 family (39), although the precise binding protein had not been identified. In sum, these results suggest that Pdx-1, BETA2, Pax6, and Nkx2.2 define the core of a ␤ cell-enriched transcription complex. DISCUSSION Though assays such as in vitro binding, co-transfection with reporter constructs, and transgenic knock-out animals provide evidence for the involvement of a factor in transcriptional control, they do not prove this directly. This is especially problematic for factors, such as Pdx-1, that recognize relatively nonselective binding sites. The consensus Pdx-1-binding element, TAAT(T/G) (40), is expected to occur once in every 512 bp. As control region sequences are often several hundred bases in length, the presence of this motif is not uncommon and would not unequivocally demonstrate Pdx-1 regulation even if it were capable of binding to a cis-acting control element in vitro. Likewise, transfection experiments do not account for the native chromatin structure of endogenous genes, which may preclude physiological regulation by a factor that, nonetheless, can bind in vitro. The ChIP assay combines two highly specific techniques, antibody-specific protein-DNA precipitation and PCR amplification, to demonstrate a physical association between a transcription factor and specific control region se- quences in intact cells. The ChIP protocol can establish direct interactions of a transcription factor and a gene that are known to be functionally related and, importantly, can also be used to identify interactions with candidate target genes.
Using the ChIP assay, we observed occupancy of insulin 5Ј-flanking control region sequences in ␤TC-3 and MIN6 cells by Pdx-1, BETA2, Pax6, and Nkx2.2 (Fig. 1). Our demonstration of Pdx-1 occupancy of insulin control sequences corroborates a recent study that was published while our work was in progress (23). Though there was evidence to suggest a direct role for Pdx-1 (5-9, 13, 41), BETA2 (8,13), and Pax6 (9) in insulin gene expression, it has been unclear whether Nkx2.2 directly or indirectly contributed. Thus, the evidence supporting an involvement was based upon the loss of insulin expression in Nkx2.2 null mice, under circumstances where other ␤ cell markers continued to be expressed (17). However, a control site for Nkx2.2 within the insulin gene had not been characterized previously. Based on the reported consensus Nkx2.2binding element (29), we have localized an Nkx2.2-binding site between Ϫ140 and Ϫ119 bp of the insulin gene by in vitro gel shift analysis (Fig. 2). Importantly, this site is mutationally sensitive in transfection assays, indicating that Nkx2.2 binding can affect insulin gene activity (Fig. 3). Collectively, our results demonstrate that Nkx2.2 can directly regulate insulin transcription in ␤ cells.
One would have expected that Nkx2.2 would have had a positive effect on insulin transcription from the results of the Nkx2.2 gene ablation studies in mice (17). However, Nkx2.2 appears to act as a weak repressor in ␤ cell transfection assays performed with either an insulin enhancer (Fig. 3) or consensus site-driven reporter construct (29). Nkx2.2 is also known to function as a transcriptional repressor in the developing central nervous system, a function necessary in cell fate specification in the ventral neural tube (42). Interestingly, structurefunction analysis of the Nkx2.2 protein has identified a strong C-terminal transactivation domain whose function is masked by the highly conserved NK2-specific domain (NK2-SD) in ␤TC-3 cells (29). As a consequence, it is possible that a co-factor expressed during pancreatic development or a stage-specific protein modification might interfere with the NK2-SD-mediated repressor function of Nkx2.2 and allow it to function as a transcriptional activator during development. In contrast, Nkx2.2 repressor activity would result if ␤ cell lines or mature ␤ cells lack this effector activity. By extension of this reasoning, we presume that Nkx2.2 also inhibits pax-4, GK, and IAPP gene expression upon binding in ␤ cells. Although the Nkx.2.2 regulatory site(s) within these genes has not been defined precisely, it is likely to be found among the consensus-like binding sites within their control region sequences (i.e. pax-4, Ϫ1940/Ϫ1933 bp, Ϫ1876/Ϫ1869 bp; GK, Ϫ542/Ϫ536 bp, Ϫ486/ Ϫ479 bp; IAPP, Ϫ325/Ϫ319 bp, Ϫ286/Ϫ280 bp, Ϫ267/Ϫ261 bp, Ϫ146/Ϫ140 bp).
Having used the ChIP assay to demonstrate binding of Nkx2.2 to several genes expressed in an islet-enriched manner, we extended this study to examine if other key ␤ cell regulatory factors also bound within their regulatory domains. Previously, Pdx-1 had been shown to bind to its own promoter region in the ChIP assay (23,24). Two distinct antisera raised against Pdx-1 were found to immunoprecipitate the IAPP and pax-4 control sequences. Each of these regions has been reported to be a target of Pdx-1 function (IAPP, Refs. 34,35,and 41;Ref. 37) and was independently shown to be occupied in the ChIP assay (23). However, whereas we observed a clear interaction of Pdx-1 and the GK gene using antisera to both the N and C termini of Pdx-1, another group was unable to detect this association using a different Pdx-1 antiserum (23). This differ-ence highlights an important limitation of the ChIP assay, which despite the power of this technique to firmly establish a physical association between a transcription factor and a control region with immunoprecipitating antisera, a negative result does not provide definitive evidence for the lack of such interactions. Another example involves the inability to detect the occupancy of the GLUT2 promoter by Pdx-1 in the ChIP assay (data not shown and Ref. 23), despite a great deal of evidence supporting a role in activation (15,16). There are two likely explanations for these negative results. The microenvironment of the GLUT2 control region might preclude the exposure of Pdx-1 antigenic determinants required for immunoprecipitation, possibly due to interactions with adjacent binding proteins. Alternatively, it is possible that Pdx-1 does not directly regulate GLUT2 transcription or does so through elements that are located outside of the analyzed region. Both studies focused on the proximal region of the GLUT2 promoter found to bind Pdx-1 in vitro (data not shown and Ref. 23) and, as a result of methodological limitations, Pdx-1-GLUT2 interactions that lie at Ն1000 bp away would not have been detected.
Interestingly, the Pdx-1, Pax6, BETA2, and Nkx2.2 regulatory factors of the insulin gene were all found to bind in intact ␤ cells to the regulatory control regions of other genes selectively transcribed in islet cells. How each of these factors precisely mediates control within the context of the insulin, pax-4, IAPP, and GK transcription units is still unclear. As islet cell development was so profoundly affected in the null mutants of these factors, the knock-out studies in mice cannot be used to access how each factor controls islet ␤ expression post-natally under normal or stressed states. Our data strongly suggests that Pdx-1, Pax6, BETA2, and Nkx2.2 represent the core components of a transcription complex of an islet-enriched gene, and presumably also contribute in their expression during development.