Functional Genomics of the Endocrine Pancreas: The Pancreas Clone Set and PancChip, New Resources for Diabetes Research

The bacterial clones selected for the array were grown to confluence overnight in flat bottom 96-well square block plates with a volume of 1 ml LB-Amp medium per well. Plasmid DNA was prepared for each clone using a Qiagen 3000 robot, and the purified DNA used as template to amplify the inserts with PCR. The clones were assigned a PCR score of “pass” indicating a single strong band and “fail” if there were no bands or multiple bands. For the subgroup of genes named “select array,” primer pairs were synthesized for each gene and the cDNA amplified from a mixture of intestine, liver, and pancreas RNA using RT-PCR. All PCR products were purified using Millipore Mutiscreen PCR Cleanup 96 well plates, resuspended in 50 μl deionized sterile water, diluted with an equal volume of DMSO (Sigma), and printed on poly-l-lysine–coated slides with an Affymetrix 417 arrayer. There are 3,840 elements on the PancChip (Table 1), including pancreas specific clones, clones representing signal transduction pathways, controls, and blanks (9). Information about the PancChip (Version 2.0) is available at http://www.cbil.upenn.edu/EPConDB/pancChip.html.

Six adult CD1 female mice were killed, and the pancreas and heart were immediately homogenized in 10 ml denaturing solution (4 mol/l guanidium thiocyanate, 0.1 mol/l Tris-Cl pH 7.5, 1% β-mercaptoethanol) per organ. Total RNA was extracted using the acid-phenol extraction method (10). Mouse islets, as well as embryonic and newborn pancreas and liver tissues, were immediately homogenized in 1 ml TRIzol Reagent (Gibco). RNA was purified following the manufacturer’s protocols with the exception that 20 μg glycogen (Roche) was added to each sample. Subsequently, RNA pellets were washed with 75% ethanol and resuspended in 300 μl TES (10 mmol/l Tris pH 7.5, 1 mmol/l EDTA, 0.1% SDS). The RNA was re-extracted with 600 μl phenol:chloroform:isoamyl alcohol (25:24:1) and precipitated with 1/10 volume 3 mol/l sodium acetate and 3 volumes ethanol and stored at −80°C until use. For the developmental time course, a common control was prepared by pooling samples. The common control pool consisted of 2.5 μg of each of the E16.5 and E18.5 samples for a total of 15 μg from each of these time points, as well as 5.3 μg of each of the newborn, P7, and adult samples for a total of 32 μg from each of these later time points. For our initial expression survey with Incyte mouse GEM 1 (1.12 for newborn pancreas) and human GEM arrays, labeling, hybridization, and signal quantification were performed by the manufacturer using 200 μg total RNA supplied by us. For the experiments performed using the PancChip, these steps were performed as described below.

cDNAs were labeled with the 3DNA Submicro Array kit (Genisphere) according to the manufacturer’s protocol and recommendations. Total RNA (2.0 μg) and 2 pmoles Cy3 capture sequence primer or Cy5 capture sequence primer were brought to 10 μl with diethyl pyrocarbonate (DEPC)-treated water and incubated for 10 min at 80°C. The RNA mixture was then cooled to 42°C. An equal volume of reaction mix (2× first-strand Buffer [InVitrogen], 1 mmol/l dATP, 1 mmol/l dGTP, 1 mmol/l dCTP, 1 mmol/l dTTP, 20 mmol/l dithiothreitol [DTT], 40 units RNasin [Promega], and 200 units Superscript II reverse transcriptase [InVitrogen]) was added, and the reaction was incubated for 2 h at 42°C. The reaction was terminated by bringing it to 0.074N NaOH and 7.4 mmol/l EDTA and incubated at 65°C for 10 min. Finally, the reaction was neutralized by adjusting it to 0.175 mol/l Tris-Cl, pH 7.5. The Cy3 and Cy5 reactions were combined and precipitated with 20 μg linear polyacrylamide (Ambion), 1 volume 7.5 mol/l ammonium acetate, and 9 volumes ethanol at −20°C for 30 min. Following precipitation the pellet was air dried.

Prehybridization was performed for all arrays (9). A coplin jar containing 50 ml prehybridization buffer (5× SSC, 0.1% SDS, and 1% BSA) was brought to 45°C. The arrays were incubated for 45 min at 45°C, rinsed five times in deionized water at room temperature, rinsed once in isopropanol, and then placed into a 50-ml conical tube and centrifuged 1 min at 1,000 rpm. The prehybridization was done no more than 1 h before hybridization.

In preparation for hybridization, the cDNA pellet was resuspended in 2 μl sterile deionized water, 3.0 μl oligo dT₇₀ blocker (0.25 μg/μl), 2.5 μl of the Cy3 dendrimer, 2.5 μl of the Cy5 dendrimer, and 1 μl high-end differential enhancer (all from Genisphere) were added to the cDNA. Mouse Cot1 DNA, 2.5 μg (1 mg/ml, Gibco-BRL), and 1 μl anti-fade reagent (Genisphere) were added to 100 μl hybridization buffer (40% formamide, 4× SSC, 1% SDS; Genisphere), and the hybridization buffer was warmed to 45°C. The prepared hybridization buffer (19 μl) was added to the cDNA/dendrimer mix and incubated at 45°C for 15 min. This hybridization mix was added to a prehybridized glass microarray, covered with a glass coverslip (22 × 40 mm), and incubated in a Corning hybridization chamber overnight at 45°C. The labeled arrays were washed three times for 10 min each: once at 55°C in 2× SSC, 0.2% SDS, once in 2× SSC at room temperature, and once in 0.2× SSC at room temperature and then dried by centrifugation in a slide rack for 3 min at 1,000 rpm in a Sorvall SH-3000 rotor.

All slides were scanned immediately following hybridization and washing using an Affymetrix 418 scanner. The laser power was set to 100%, and the gain of the photomultiplier-tube was varied to avoid signal saturation in any spots. The image analysis was performed with ArrayVison 6.0 (Imaging Research). Signal and background pixel classification were determined by segmentation limited to 75–125% of the set spot diameter. Signal and background intensities were determined by the median pixel values. All of the array data are available through http://www.cbil.upenn.edu/EPConDB/query.html.

In the arrays used for the developmental time course, 2,914 genes had a passing PCR score (see “Glass Microarray Preparation”) and were considered for further analysis. Genes with high variation among the six replicates (coefficient of variation [CV] > 0.7) were examined for outliers. The data were normalized using the statistical software package R (11). For the comparisons across time-points, scaled print-tip group lowess normalization was used to remove spatial effects and other systematic variation (12). The local background intensities as calculated by ArrayVision were not subtracted from the data intensities. The subtraction of local background significantly increases data variance (E. Manduchi, L.M.S., J.E. Brestelli, G.R. Grant, K.H.K., C.J.S. Jr., Physical Genomics, In Press) and is not necessarily a good control for nonspecific binding of the labeled nucleic acids.

The normalized data were analyzed for trends over time using vector projection. Vector projection is a method that allows identification of these trends in gene expression data and incorporates all of information from all time points (Terry Speed, Department of Statistics, University of California at Berkley, and Genetics and Bioinformatics, Walter and Eliza Hall Institute Australia; and Ingrid Lönnstedt, Department of Mathematics, Uppsala University, personal communication). Specifically, vector projection facilitates quick identification of genes that match the predetermined patterns of interest shown in Fig. 3 (e.g., “late” gene expression defined as peak expression during the adult time point). Each gene has a vector of its normalized expression values across time. These values are projected onto the space spanned by the pattern of interest (vector of coefficients or weightings for each time point). Examination of the Normal QQ-plot of all projection values allows identification of the extreme projection scores. Genes with strong patterns will have the largest (or smallest) values of the inner product.

To produce a custom cDNA microarray, the PancChip, we first needed to identify a set of clones that represented genes expressed in the pancreas. Experimental and bioinformatics approaches were used to identify these clones. First, expression studies were performed with Incyte GEM arrays using RNA from newborn and adult mouse pancreas, newborn mouse liver, a mouse insulinoma cell line, and human islets. In each case, we identified the clones in the top 15% expression level as measured by fluorescence intensity, which given the random collection of cDNAs on the Incyte GEM array, represents genes expressed at low, moderate, and high levels. Second, we identified additional clones for the PancChip from dbEST cDNA libraries nos. 185 (13), 422 (14), 1,144, 1,880, and 2,712 (15). These libraries were chosen because they were all prepared from RNA isolated from pancreatic tissue. Library 1,870 was prepared with C57BL/6J adult mouse pancreas. All others were prepared from human pancreatic islets; no. 2,712 also included RNA from human total pancreas.

To identify and select a nonredundant set of clones, the expressed sequence tags (ESTs) and clones identified above were mapped to IMAGE clones using the Genomics Unified Schema (GUS) data system (16) accessible through AllGenes (http://www.allgenes.org). RNA entries contained within GUS are organized as Database of Transcribed Sequences (DoTS) assemblies; each entry or assembly represents a consensus of overlapping—confirmed and putative—transcribed sequences. The mouse clones identified by expression analysis described above could be directly mapped to DoTS assemblies. For human cDNA libraries, EST sequences over 100 bp were first retrieved and trailing polyA and leading polyT regions removed. Mouse orthologs for these ESTs and for the genes identified by the expression analysis of human islets were identified through stringent BLASTX similarity against the nonredundant protein database (NRDB) at the National Center for Biotechnology Information (NCBI) using a cut off P value of 1 × 10⁻⁵⁰. A set of IMAGE clones was chosen from combined groups of nonoverlapping mouse DoTS assemblies. Finally, one IMAGE clone was chosen to represent each of the DoTS assemblies identified, with preference given to clones containing the 3′ end of the assembly.

Using this combination of expression analysis and database mining, we obtained a nonredundant core clone set of 3,139 mouse IMAGE clones, each representing a unique assembly (“Pancreas clone set,” Table 1). Most of the genes represented in this set were identified by several paradigms; for example, hundreds were found in an expression array with mouse insulinoma RNA and were also present in an EST library derived from human pancreatic islets. Overall, ∼60% of the clones in the set, or 1,900 elements, were defined to be present in pancreatic islets or insulinoma cells. Of this core set, 2,369 clones showed >95% identity to known protein sequences, and 310 showed “no nonredundant (NR) protein similarities” and represent unique protein coding regions or untranslated regions. In addition, 1,898 of the clones were sequence verified at the Genome Sequencing Center at Washington University. The remaining 1,241 clones were not verified, as they did not match with a parent sequence. These clones are not necessarily incorrectly identified, due to the fact that many parent clones had previously only been sequenced from one end. Of the nonsequence verified clones, 830 or 69% were shown to be expressed during one of the stages of pancreatic development contained in this study, and, therefore, these clones were maintained as part of the 3.4K clone set. The corresponding PancChip information, including annotation, for the 3.4K clone set and the verification status of all clones is available at http://www.cbil.upenn.edu/EPConDB/pancChip.html.

The cDNA clones in the core pancreas clone set were categorized using a directed acyclic graphical (DAG) classification system defined by the Gene Ontology (GO) Consortium (17,18) (http://www.geneontology.org). The assignments of GO functions to the proteins represented by the cDNA clones in the core pancreas clone set (Fig. 1) were made computationally using an algorithm associating the translated protein domains with GO functions (25). Briefly, to assign GO function(s) to the translated sequences, it is assumed that the sequence of a previously characterized functional domain always functions as characterized. For example, a translated sequence having a domain with a BLAST similarity meeting a P value threshold with a previously characterized DNA binding domain is assigned the GO function “DNA binding.” Not every translated sequence will contain a domain meeting the criteria to be assigned a GO function. Thus, not every clone was assigned a GO function, and some sequences were assigned more than one top-level function. The distributions of the top-level GO function assignments for the core pancreas clone set are shown in Figs. 1 and Table 2.

To further increase the usefulness of the information obtained from the core pancreas clone set, we identified representative genes from various signal transduction pathways relevant to pancreatic development, which resulted in the identification of an additional 231 IMAGE clones (“Pathways” in Table 1). A group of 108 clones relevant to pancreatic development were added to the collection from laboratory stocks (“In-house” in Table 1). Using an alternative approach to obtain PCR products to spot onto microarrays, we designed primer pairs for a complementary group of 153 genes of importance to pancreatic development and amplified PCR products from a cDNA pool derived from intestinal, liver, and pancreas RNA (“Select” in Table 1). To control for the possibility of residual cDNA being contained in the purified PCR products, we performed PCR reactions without primers (cDNA controls). Finally, all PCR products were purified and spotted on the array. Additional controls included 30 housekeeping genes (19) and 8 yeast intergenic sequences in duplicate (Incyte Genomics).

The cDNA inserts of the clone sets were amplified by PCR and analyzed individually by agarose gel electrophoresis. As can be seen in Table 1, the overall rate of PCR success as defined by the presence of a single band on a gel was ∼80%. An additional 15% of the clones were amplified but contained more than one band. Therefore, of the 3,674 clones in the collection, 2,912 are represented as single-band products on the array. To demonstrate the basic characteristics of the PancChip, we labeled the same total pancreas RNA with both Cy3 and Cy5 using the dendrimer method (20) and hybridized both cDNAs on the same array. Idealized “same versus same” hybridization would show all the points falling perfectly on a line with the deviation of the slope from one reflecting the channel-to-channel differences. As can be seen in a scatter plot of the median intensities of this array (Fig. 2A), the intensities show excellent correlation with a linear regression with R² = 0.9809. Second, we compared two very different RNA samples on the array. In this experiment, islet RNA was labeled with Cy3 and heart RNA was labeled with Cy5. As we expected, the scatter plot of the median intensities clearly showed two populations (Fig. 2B). The larger population with 92% of the differentially expressed spots was labeled primarily with the islet RNA (Cy3 channel). The other population of only 8% of the differentially expressed spots showed hybridization primarily with the heart RNA (Cy5 channel), indicating that indeed our clone set is enriched for cDNAs expressed in pancreatic islets.

To demonstrate the value of the 3.4K clone set and the PancChip, we used them to investigate gene expression patterns during pancreatic development in mice. We dissected the pancreata from mice representing developmental stages from E14.5 through adulthood, isolated total RNA from the pancreata, and labeled the RNA for hybridization using a dendrimer labeling system (Genisphere; E. Manduchi, L.M.S., J.E.B., G.R. Grant, K.H.K., C.J.S. Jr., Physiological Genomics, In Press). For E16.5 though adulthood, each sample represents a single individual. Six replicates (each with RNA from a distinct individual) were performed for each time point from E16.5 through adulthood. Due to the low amount of RNA from each pancreatic primordium in E14.5 samples, we pooled several pancreata from individual embryos to prepare one RNA sample. Four replicates of this time point were done (each with pools from different groups of individuals). The experimental samples were all labeled with Cy3. We used a pool of all the samples in the study (except samples from E14.5) as the reference RNA and labeled it with Cy5. The labeled RNA was hybridized to the arrays and scanned. The data were analyzed as the ratios of the intensities of the individual time points for each gene over the intensities of the common control for the same gene as has been used for other two-channel microarray experiments with multiple comparisons (11,21).

In the developmental time course, only the 2,914 genes with a passing PCR score were considered for further analysis. Gene expression levels were normalized, and the normalized data were analyzed for trends over time as described in research design and methods. Three predetermined patterns were of interest: “late” expression (with peak expression at adult), “early” expression (peak is at E14.5), and genes that have peak expression at birth (Fig. 3). Genes with the most extreme vector projections for each trend were identified (see research design and methods). These genes are also listed in supplemental materials (http://www.cbil.upenn.edu/EPConDB/pancChip.html). We used the GO functions, discussed earlier, to categorize the genes with the top vector projection scores for each trend. There is a shift in the distribution of GO functions across the developmental stages examined (Fig. 4). During the early expression period, the predominant group of genes being expressed fall into the “Nucleic acid binding” category. At birth, the predominant group of genes being expressed has “No NR protein similarities.” Finally, in the late period of expression, the predominant group is “Enzyme,” as we would expect since our samples included the exocrine pancreas whose primary function in the adult is the secretion of digestive enzymes.

When we examined the range of intensities across time for a given trend, we found dramatic shifts in the intensities for a gene across the different time points (Fig. 5). The y-axis is the log₂ ratio of the average median intensities of the given developmental time point versus the average median intensities of the pooled common control. Due to the log nature of the scale, the changes seen in the early period (panel A) relate to an eightfold scale in the expression of the identified genes when compared with the common control. Included among the early expression genes for nucleic acid binding proteins are 1) a clone with high identity to heterogeneous nuclear ribonucleoprotein A3 (hnrnpA3) and 2) nucleophosmin, as well as transcription factors. Both hnrnpA3 and nucleophosmin are thought to play roles in cell growth and proliferation. In addition to the nucleic acid binding proteins, genes active in membrane transport functions, including thyroid receptor activator molecule (TRAM1) and lysosomal-associated protein transmembrane 4α (LAPTM4A), are expressed in early pancreatic development.

The scale of the graph for the expression-at-birth panel (Fig. 5B) shows an approximate eightfold difference in gene expression between birth and E14.5 and an approximate twofold difference between birth and adulthood. Fifty percent of the genes identified as highly expressed during the perinatal period are genes with no GO function (genes with no known protein similarities and clones with similarity to other uncharacterized ESTs). Other genes identified as highly expressed during the perinatal period include BAT2 (a voltage sensitive calcium channel) and proline oxidase 1.

The late expression proteins show up to a 20-fold change in expression between our earliest time point (E14.5) and adulthood (Fig. 5C). As can be seen on the graph, many of these genes are digestive enzymes one would expect to be expressed in the adult pancreas, including elastase-2 and amylase. Trefoil factor 2 (TFF2), which has previously been shown to be expressed in the pancreas (22–24), also increases expression in the adult versus the earlier stages.

We have assembled a set of clones that represent genes primarily expressed in the pancreas and have used this clone set to construct a glass cDNA microarray as a resource to the diabetes research community. Our core pancreas clone set has been extensively annotated and represents a varied spectrum of cell functions. All cDNAs of the core pancreas clone set have been resequenced, and the majority of the clones were sequence-verified. Most of the nonconfirmed clones are expressed at some point during the developmental stages studied. These clones will, however, be replaced with new cDNAs from Endocrine Pancreas Consortium libraries as they become available.

We have demonstrated the usefulness of the clone set by preparing a glass cDNA microarray, the PancChip, and by profiling pancreatic gene expression patterns from midgestation through adulthood with the PancChip. At the earliest stage in this study (E14.5), nucleic acid binding proteins were the largest group of differentially expressed genes as assigned by GO functions. The nucleic acid binding proteins of the GO function classification include transcription factors and other proteins involved in cell fate decisions. This result fits with expectations of gene expression during embryonic development. During the adult stage of pancreatic development, the largest group of genes being expressed encodes enzymes. Again, this is an expected result that confirms the validity of the PancChip, as we had used total pancreas for our time course, which in the adult consists largely of exocrine tissue whose main function is the secretion of digestive enzymes.

Previous gene profiling experiments made use of commercial arrays to study both type 1 and type 2 diabetes (1,2,5,6,8). None of these arrays were prepared with the intention to focus on diabetes or the pancreas. Because the pancreas clone set is enriched for genes that are expressed in the pancreas or were present in cDNA libraries prepared from pancreatic tissue, it offers the diabetes research community the opportunity to investigate genes of interest at a higher density. Ultimately, it will be desirable to utilize a complete genome-wide cDNA array representing all 20,000–40,000 mammalian genes for all expression profiling experiments, including those related to diabetes. However, until such a resource is available at a low cost, the PancChip will provide a valuable and affordable resource for the diabetes research community. The 3.4K pancreas clone set used to make the PancChip will be distributed through the NIDDK-funded biotechnology centers. Current efforts of the Endocrine Pancreas Consortium are centered on the generation and sequencing of pancreas-specific cDNA libraries from various stages of development from both mouse and human. These new libraries will allow us to dramatically increase the number of clones represented in the clone set and on the PancChip in the future.

We gratefully acknowledge support in the form of NIDDK Grant 56947 to K.H.K and NIDDK 56954 to A.M.P. D.M. acknowledges support from the Juvenile Diabetes Research Foundation.

We thank Phillip Phuc Le for computer support and for the design of the web interfaces and Jian Wang for initial clone selection. We also thank Elisabetta Manduchi and members of the Computational Biology and Informatics Laboratory (CBIL) in the Center for BioInformatics for helpful discussions.