DTT/EDTA treatment was employed to strip the epithelial layer from patient intestinal biopsies. Magenetic cell sorting was then used to deplete CD3⁺ cells, and enterocyte suspensions with purities of 98–99% were routinely obtained (Figure 1). Using Atlas Glass Human 3.8I oligonucleotide micoarrays (BD Biosciences Clontech, UK), we analysed the gene expression profile of a homogeneous population of duodenal epithelial cells taken from patients with active coeliac disease, in comparison to control patients. Of the 3,800 genes present on the array (all of which have been previously annotated), 3549 had sufficient data across the five experiments for comparison. Many of these genes showed fold-change ratios with little or no deviation from 1. Thus, to focus on only differentially expressed genes, the list was filtered on a fold-change of 1.25-fold. A fold-change of 1.25 has been described in previously published microarray experiments 2324 and has been shown to indicate reliable differences in gene expression 23. This filtering yielded a list of 1,256 genes with which to perform analysis. Of these 1,256 genes, 827 were up-regulated and 429 were down-regulated. A p-value of 0.05 was used as a cut-off to distinguish significantly expressed genes, which yielded a gene list of 102 (68 up-regulated and 34 down-regulated) genes which are presented in Table 1. These 102 genes were grouped according to functional categories, including protein transport, ion transport, proliferation, differentiation, anti-apoptosis/survival, structural, adhesion, metabolism, transcription, protein biosynthesis, signal transduction, cell cycle and DNA repair, and immune response and inflammation (categories of genes and up/down-regulation are summarised in Figure 2 and Table 2 respectively).

In order to corroborate the microarray gene expression results genes were selected for validation by real-time RT-PCR using the same patient samples. Miron et al have demonstrated that the popular strategy of selecting genes with the greatest fold-increase generally fails as a global validation strategy while a random selection of genes for validation (10–25 genes) is a more robust technique 25. We therefore used random-stratified sampling as a gene selection method. Ten genes representing a range of p-values were randomly selected for quantitative RT-PCR analysis.

The RT-PCR expression values were calculated from a standard curve and the mean of the values calculated for the coeliac samples was divided by the mean for the control samples to give a ratio. The RT-PCR ratios were then compared to the microarray ratios (Table 3). Eight of the ten genes showed up-regulation of expression in coeliac disease with both microarray and real-time RT-PCR analysis. One gene showed up-regulation of expression in coeliac disease with microarray analysis and non-differential expression with real-time RT-PCR. One gene showed down-regulation of expression in coeliac disease with microarray analysis and up-regulation of expression with real-time RT-PCR analysis. Thus, 80% of the genes tested in this study were found to be validated, which compares well to a reported average RT-PCR confirmation rate of 70% 26.

Two of the genes validated by real-time RT-PCR, were selected for immunohistochemical examination based upon their expression in the gastrointestinal tract and potential interest with respect to coeliac disease pathophysiology. Small proline-rich protein 3 (SPRR3) was selected for qualitative analysis because of its role as a structural protein and its potential to act as a substrate for tissue transglutaminase 27. Gap junction protein alpha 4 (GJA4) was selected because of recent literature suggesting the importance of gap junctions in the spreading of immune signals between epithelial cells, including small peptides 28.

Protein expression of SPRR3 was examined in biopsies from five further untreated coeliac patients and in corresponding biopsies taken from these same patients at a later date while consuming a gluten-free diet (Figure 3). Histology reports showed that while consuming gluten, all patients had Marsh 3 type lesions (one Marsh 3a, 2 Marsh 3b and 2 Marsh 3c). Upon adoption of a gluten-free diet, two patients had recovered to normal duodenal histology while three patients had recovered to type 3a lesions with only mild or variable villous blunting. These patients had been following a gluten-free diet for a mean of 7 years (range 4–11 years).

In the untreated coeliac tissue the common finding was intense cytoplasmic staining of SPRR3 throughout the epithelial cell, with strongest staining observed on the apical side of the cell. Perinuclear staining was also detected, although less visible due to the intensity of the cytoplasmic staining. The staining was observed all along the epithelium with the top part of the villous showing the greatest intensity. Staining for SPRR3 was increased in the untreated coeliac disease samples when compared to the gluten-free diet samples in 4 out of 5 cases. In the fifth matched sample, staining was equivalent pre and post-treatment with the gluten-free diet. The treated coeliac tissue generally showed faint staining in the villous-tip epithelial cells, with occasional punctate staining of moderate intensity also occurring only in the villous-tip region. Biopsies from the five untreated coeliac patients who showed up-regulation of SPRR3 mRNA expression by microarray and RT-PCR analysis were also examined for protein expression of SPRR3. All five samples showed a similar intense staining picture as seen in the untreated coeliac biopsies described above. Biopsies from ten normal control patients and ten disease control patients (consisting of five with peptic duodenitis and five with Crohn's disease with duodenal involvement) showed minimal staining of the epithelium with only the villous-tip involved, in a similar pattern to that found in the treated coeliac tissue (data not shown).

Expression of GJA4 protein was compared between the matched pairs of untreated and treated coeliac biopsies (Figure 4). In the untreated epithelial tissue, granular staining was predominantly seen in the cytoplasm of the epithelial cells, with an increasing intensity towards the brush border. Perinuclear staining was visible in some cells although nuclear staining was generally absent. In contrast, in the treated biopsies, nuclear staining of epithelial cells was common while weak cytoplasmic staining was only evident in the villous tips and brush border staining was absent. The intra-epithelial lymphocytes within the treated tissue showed the most intense nuclear staining while the staining of enterocytes appeared to be due to strong perinuclear staining. In biopsies from the five untreated coeliac patients, who showed up-regulation of GJA4 mRNA expression by microarray and RT-PCR analysis, staining was similar to that seen in the untreated coeliac biopsies described above. Biopsies from control patients, both normal and diseased, showed a similar staining pattern to that observed in treated coeliac tissue although fewer cells appeared to stain positively in these control tissues (data not shown). In particular, nuclear staining was less frequent than in the treated coeliac tissue while perinuclear staining remained a common finding.

Duodenal biopsy specimens were obtained from each patient via oesophago-gastro-duodenoscopy. Five coeliac patients had active, untreated disease and five age and sex-matched patients undergoing endoscopy for investigation of upper gastrointestinal symptoms were used as a control group. The demographic and clinical details of these patients are given in Table 4. Eight biopsy samples from each patient were used in microarray and RT-PCR experiments and a further two biopsies were processed for routine histological evaluation.

In the immuno-histochemical studies, duodenal biopsy samples from three further groups of patients were employed. Archived tissue blocks in St. James's Hospital, Dublin were the source of these samples. The study subjects included five additional coeliac patients (3 males, 2 females, mean age 60 years): samples taken before they had commenced a gluten free diet and after dietary treatment (mean 7 years) were investigated. Two disease control patient groups were also studied: these included five patients (2 female, 3 male, mean age 51 years) with Crohn's disease involving the duodenum; and five patients (2 female, 3 male, mean age 43 years) with peptic duodenitis. Finally, a further ten patients (6 female, 4 male, mean age 59 years) undergoing endoscopy for investigation of upper gastrointestinal symptoms were also studied: this entire latter group had normal duodenal histology.

Ethical approval for this study was granted by the Joint Ethics Committee of St James's Hospital and Tallaght Hospital, Dublin, Ireland.

Enterocyte isolation was perfomed based upon previously published methods 20. The eight biopsies taken from each patient were transferred into RPMI culture medium containing 10% fetal calf serum and processed immediately. Biopsies were agitated in calcium and magnesium free HBSS (Gibco BRL, Scotland) containing 1 mM dithiothreitol (DTT) (Sigma, USA) and 1 mM EDTA (Sigma, USA) and incubated at 37°C for 40 minutes. The supernatant containing the epithelial cells was removed, washed twice and centrifuged at 800 rpm for 10 minutes. Cells were magnetically sorted to deplete CD3⁺ T-cells, using MACS CD3 microbeads (Miltenyi Biotech, Germany) and an LD depletion column according to manufacturer's specifications (Miltenyi Biotech, Germany). Employing the anti-epithelial cell monoclonal antibody Ber-EP4 and the anti-T-cell monoclonal antibody CD3 (DAKO, Denmark), FACS analysis was performed on the eluted fraction in order to determine purity and values of 98–99% were repeatedly observed.

Total RNA was extracted from the epithelial cells using the NucleoSpin^® RNA II kit (BD Biosciences Clontech, UK) according to manufacturer's instructions with the following exceptions: vigorous vortexing of cells in the lysis buffer was performed, after which samples were frozen and then thawed before continuing with extraction; RNA was eluted from the column in 40 μl RNase-free water and the eluate reloaded onto the column twice more in order to collect the maximum yield of RNA. The RNA was concentrated further using Microcon YM-30 concentrators (Millipore, Ireland). Total RNA was quantified at 260 nm, and the 260/280 nm ratio was measured to calculate purity of RNA from contaminating protein. Agarose gel electrophoresis was also carried out to assess the quality of RNA. The same RNA samples were used for both microarray and RT-PCR analysis.

Fluorescent-labelled cDNA probes were synthesised using the BD Atlas™ SMART™ Fluorescent Probe Amplification Kit (BD Biosciences Clontech, UK). Briefly, cDNA was reverse-transcribed from total RNA and purified from unincorporated nucleotides before amplification using the CyScribe GFX Purification Kit (Amersham Biosciences, UK). From each patient specimen, two probe samples were synthesised. For each sample, the optimum number of PCR cycles was determined in order to ensure the amplification process was stopped while still in the exponential phase. This was essential as over-cycled cDNA which has reached the plateau phase of amplification could result in a less representative probe when examining differential gene expression. Once complete, an aliquot of each sample was analysed on a 1.2% agarose/EtBr gel under UV light, to ensure the reactions were successful. The PCR product was purified using the CyScribe GFX Purification Kit. Absorbance was read at 260 nm to calculate quantity and the 260:280 nm ratio was measured to calculate purity. Purified PCR product was fluorescently labelled, according to manufacturer's instructions. Single-use aliquots of monoreactive Cy3-NHS ester and Cy5-NHS ester dyes (Amersham Biosciences, UK) dissolved in DMSO were used. The labelled probe was purified from unreacted dye using the CyScribe GFX Purification kit and further purified from particulate matter by filtering through a 0.22 μm spin filter. Quantity and quality of the labelled probe was determined by UV/visible spectrophotometry using a Genesys 5 spectrophotometer (Thermo Electron Corporation, US). The optimal volume of labelled probe to use in the hybridisation reaction was determined on the basis of absolute optical units (OU_λ), using the following formula:

V opt ( μ l ) = 1 , 000 × OU λ A λ MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaeeOvay1aaSbaaSqaaiabb+gaVjabbchaWjabbsha0bqabaGccqGGOaakiiaacqWF8oqBcqqGSbaBcqGGPaqkcqGH9aqpjuaGdaWcaaqaaiabigdaXiabcYcaSiabicdaWiabicdaWiabicdaWiabgEna0kabb+eapjabbwfavnaaBaaabaGae83UdWgabeaaaeaacqqGbbqqdaWgaaqaaiab=T7aSbqabaaaaaaa@45E7@

where OU_λ for Cy3 and Cy5 is 0.01 (determined by BD Biosciences Clontech) and A_λ is the measured absorbance maxima for each dye; 550 nm for Cy3 and 650 nm for Cy5.

The microarrays used in this study were Atlas Glass Human 3.8 I microarrays (BD Biosciences Clontech, UK). Five biological replicate experiments were conducted, each one comparing one untreated coeliac sample to one control sample. In each experiment, two technical replicates were performed. Onto one microarray, Cy3-labelled coeliac cDNA and Cy5-labelled control cDNA were co-hybridised. Onto the other microarray, Cy5-labelled coeliac cDNA and Cy3-labelled control cDNA were co-hybridised. In this manner any discrepancies in rates of incorporation of the different dyes during the labelling step are controlled for. The appropriate coeliac and control probe were combined together and hybridised onto the microarray according to manufacturer's specifications. The slides were scanned using an Affymetrix^®428™ Array Scanner. Fluorescence was measured after excitation at 532 nm and 635 nm. Separate raw images for each dye were acquired and images were analysed using the software package ImaGene^® 5.0 (BioDiscovery, California). Quality control measures were performed on all spots to identify empty, poor and negative spots.

The raw data from image analysis was normalised using the free software ArrayNorm available at http://genome.tugraz.at (Graz University of Technology Bioinformatics group, Austria). All the data was subjected to a background subtraction followed by a Lowess normalisation performed separately on each block (subgrid) on each microarray. Once normalised, the technical replicates were averaged and the data saved as a text file. The normalised data was analysed using the software package GeneSpring^® 7 (Silicon Genetics, California). A custom genome was created using the genes present on the microarrays used. The data was analysed using log of ratio and the cross gene error model was turned off. The data was filtered on confidence using the t-test with a p-value of 0.05 considered significant. Significantly differentially expressed genes were grouped into functional categories using MAPPFinder http://www.GenMapp.org – a program that works with GenMAPP and the annotations from the Gene Ontology (GO) Consortium 66.

Genes to be validated were selected on the basis of significance and potential interest. The genes were chosen to represent a spread of p-values up to 0.1. Assays-on-Demand™ Gene Expression products (Applied Biosystems, UK) containing forward primer, reverse primer and probe labelled with FAM dye and MGB quencher in a single tube were used. These primer/probe mixtures were pre-designed by the manufacturer and pre-optimised to ensure high amplification efficiency (proprietary sequences).

Total RNA was reverse transcribed by standard methods. cDNA samples were quantified using the Fluorescent DNA Quantitation Kit (Bio-Rad, UK) according to manufacturer's specifications, and 250 pg of cDNA sample was used per reaction. Fluorescence was measured on a Tecan GENios Fluorimeter (Tecan, UK). Quantification of gene expression was carried out in the ABI TaqMan 7000 (Applied Biosystems), using purified PCR products as standards 67. In order to compare the expression values to those obtained from the microarray analysis, the mean expression value of the five coeliac samples was divided by the mean expression value of the five control samples to give a single ratio value.

Immunohistochemical staining was performed on 3 μm thick sections, cut from formalin-fixed, paraffin-embedded biopsies, using the avidin-biotin-peroxidase complex detection method. Rabbit polyclonal anti-human SPRR3 (small proline-rich protein 3) (Apotech UK), was applied at a concentration of 1 in 1000 and rabbit polyclonal anti-mouse connexin 37 (Alpha Diagnostic, US) was used at a concentration of 1 in 150. The mouse connexin 37 immunogenic peptide has an 87% homology to human connexin 37. The selection of these two antibodies was based on their known ability to react with human tissue 6869. The staining procedure was carried out using the VECTASTAIN^® Universal Elite ABC-Peroxidase Kit (Vector Labs, USA), according to manufacturer's specifications. Slides were visualised using a Nikon eclipse TE300 microscope attached to a computer. Images were captured using Leica DC100 and Adobe Photoshop software.