Transcriptome signatures in Helicobacter pylori-infected mucosa identifies acidic mammalian chitinase loss as a corpus atrophy marker

Patient recruitment and sample collection

Microarray experiment

For the microarray experiments, three patients were randomly selected for analysis from the Hp- and Hp + groups. From the Atr group, the three patients with most progressed lesions according to the histopathology analysis were selected. These three patients had moderate or marked atrophy (score 2 or 3) and mild, moderate or marked intestinal metaplasia (score 1 to 3) in their corpus tissue. The Illumina’s BeadArray technology was used for transcriptome analysis. cRNA amplification and biotin labeling from 500 ng of total RNA were performed using Illumina® TotalPrep RNA amplification kit (Ambion, Inc., Austin, TX). cRNA yields were quantified with NanoDrop® (Thermo Scientific, Wilmington, DE). Labeled cRNAs (1.5 μg) were hybridized to the Human-8 v2 Expression Beadchips (Illumina, Inc., San Diego, CA) then incubating for 19 h at 58°C before being washed and stained following the manufacturer’s protocol. Beadchips were then scanned using the IlluminaBeadArray-Reader scanner. Signal extraction and primary quality control was carried out with BeadStudio 3.1.7 software (Illumina Inc., San Diego, CA).

Microarray data acquisition and analysis

All raw signal intensity files from the BeadStudio were processed together by R software equipped 'lumi’ package [7] using quantile normalization. We used different methods to analysis the microarray data based on different hypotheses. First, the differential gene expressions between the corpus and antrum samples of each group of patients were evaluated by Student’s t test ('limma’ package) [8]. The results are illustrated in global view in Figure 1A and the numbers of differential expressed genes (cutoff p-value < 0.05) of each group are illustrated in Figure 1B. Thereafter, the gene ontology (GO) enrichment of the antrum and corpus samples in each group of patients was revealed by a reporter algorithm [9, 10]. The GOs that had enrichment p-value < 0.001 were illustrated in Figure 2 as a heat-map of significant values. Second, the gene expression was analyzed by one-way ANOVA to identify differences among the three patient groups. This was done separately for antrum and corpus samples. The gene expressions that had p-value < 0.001 were further selected for consensus-clustering analysis [11]. Hierarchical clustering of gene expression profiles were performed separately for corpus and antrum samples. The results are presented as a heat-map of gene expression similarity, using a cluster dendrogram height of 1.3. Functional enrichments of each cluster were then evaluated by module enrichment analysis [12] of its gene members, and illustrated using Cytoscape software [13] as illustrated in Figure 3. Third, linear regression analysis was performed over the selected genes based on ANOVA, as described above, to estimate the level of gene expression responses along the three groups of patients. An overview of the analysis layout, and detailed results are provided in Supplementary information (Additional file 1: Figure S1 and Additional file 2).

The microarray results were not subjected to FDR-analysis or multiple-test correction. This is due to the fact that we were limited in the number of well-characterized tissue samples we could obtain from each patient group. However, the key findings in this study were validated with real-time PCR as well as the protein abundance level. The microarray data are deposited and publicly available in Gene Expression Omnibus (GEO; Accession number GSE27411).

Tissue and cell type specific evaluation of gene clusters

To investigate whether the list of genes up-regulated in atrophic gastritis were enriched in genes with tissue-specific expression patterns, we used a publicly available reference dataset “Human Body Index – Transcriptional Profiling” (GEO-accession number GSE7307). All data were normalized together with the qspline algorithm [14] with signal processing method by Probe Logarithmic Intensity Error [15]. Genes were assayed for tissue specificity compared to small intestine, colon, antrum (“pyloric”), and cardiac/fundus tissue after verifying the validity of the data through hierarchical clustering of the tissues. Genes were analyzed using Wilcoxon rank-sum test and considered tissue specific if they had a p-value < 10^-3 and a fold change of > 2. The tissue specific genes were then tested for enrichment in the different clusters of Figure 3 using Fisher exact test. The genes in different clusters also tested against the list of genes differentially expressed in IM, SPEM and both IM and SPEM from the recent publication by Lee et al. 2010 [16].

RT-qPCR assay of selected genes

Selected genes from the microarray analysis were independently validated by RT-qPCR analysis. cDNA synthesis from 600 ng of RNA were performed by Omniscript RT kit (Qiagen, Germany). Each 20 μl RT-qPCR reaction mixture included 2 μl of the cDNA product, 17 μl SYBR Green master mixture (Applied Biosystems) and primers (Additional file 1: Table S1). All assays were performed in technical duplicates using a 7500 Real Time PCR System (Applied Biosystems). Hypoxanthine-phophoribosyl-transferase (HPRT1), which has previously been observed to be a good reference gene for normal stomach tissue [17], was used as a reference gene for all assays to normalize expression by the ΔCt method [18].

Western blot assay of acidic mammalian chitinase protein in stomach

One or two biopsies from antrum and corpus mucosa, respectively, were collected and stored for protein extraction as described above.

At the time of protein extraction, biopsy specimens were incubated in 400 μl PBS, containing 2% saponin, 100 mg/ml soybean trypsin inhibitor, 350 mg/ml phenylmethylsulfonylfluoride and 0.1% bovine serum albumin (all purchased from Sigma-Aldrich, St.Louis, MO, USA), over night at 4°C. The mixtures were then centrifuged at 13000 × g for 10 minutes, and the supernatants were collected and stored frozen at -70°C until used. To ensure equal loading of total protein in the gels, the extracts were run through Zeba Micro Desalt Spin Columns (Nordic Biolab labdesign, Täby, Sweden), whereafter their protein concentrations were determined using BCA Protein Assay reagent (Thermo Fisher Scientific, Rockford, USA). Three μg protein of each lysate was loaded per lane in 4-12% BisTris NuPAGE gels (Invitrogen, Carlsbad, CA) and gels were run in MOPS (3-[N-morpholino]propane sulfonic acid) buffer at 200 V. Proteins were transferred to 0.22 μm nitrocellulose membranes at 30 V using 25 mM Tris–HCl, 192 mM Glycine and 10% methanol for 1 h. Membranes were blocked for 1 h in 2% BSA-PBS/Tween. Rabbit anti-AMCase (ab72309, Abcam, Cambridge, UK) was used as primary antibody, with chrompure Rabbit IgG (Jackson Immuno, Suffolk, UK) as unspecific control. After washes in PBS + 0.1% Tween-20, Goat anti-Rabbit Immunoglobulins-alkaline phosphatase (Southern Biotech, Birmingham, AL) was added and after subsequent washes the blots were developed using 5-bromo-4-chloro-3-indole-phosphate/nitro blue tetrazolium (Southern Biotech, Birmingham, AL). The membranes were scanned with a Geldoc (Bio-Rad Laboratories, Hercules, CA) scanner, and band intensities were analyzed using Quantity One 4.6.1 software (Bio Rad).

Confirmation of differences between antrum and corpus samples

Depletion of differential gene expression between antrum and corpus mucosa in atrophic gastritis

In order to understand the characteristics of the atrophic corpus mucosa, we applied a global analysis of differential gene expression between antrum and corpus samples of the three patient groups. This analysis demonstrated that the most strongly significant differences between antrum and corpus gene expression are found in the Hp- group, followed by the Hp + gastritis and atrophy groups (Figure 1A). There was also a dramatic reduction in the number of differentially expressed genes in antrum vs. corpus samples from corpus atrophy patients: 90% and 75% reduction compared to Hp- and Hp + groups, respectively (Figure 1B). Among these genes, around equal numbers were over-expressed in antrum and in corpus. Over-expressed genes in corpus amounted to 54%, 60% and 41% of the differentially expressed genes in the Hp-, Hp + and Atr groups, respectively. Thus, our data thus demonstrates molecular evidence for a strong similarity of global gene expression between corpus and antrum tissue of patients suffering from corpus atrophy. This is also supported by the result of a transcriptome based hierarchical clustering of samples, as shown in Additional file 1: Figure S2.

Corpus-specific genes in the un-infected stomach include genes related to acid secretion, energy metabolism, and blood vessel activity

We then performed an integrated analysis to assess over-representation of different biological processes among the genes differentially expressed in antrum compared to corpus mucosa. This analysis demonstrated that the absolute majority of gene ontologies coupled to differential expression between antrum and corpus mucosa in Hp- individuals were associated with genes over-expressed in the corpus tissue (Figure 2). Thus, in all 45 ontologies there was significant enrichment of corpus over-expressed genes, and in 7 ontologies there was also enrichment of over-expression of genes in the antrum tissue.

Groups g1 through g4 (Figure 2) encompass ontologies over-expressed in corpus mucosa in the Hp- group. Therefore, these groups yield important information about the functions of genes specific for the corpus mucosa of Hp- non-atrophic stomach. G1 includes gene-groups related to gastric acid secretion and H⁺/K⁺-ATPase activity. A large part of the corpus-specific gene ontologies concern energy turnover and mitochondrial function (g2) as well as amino acid and lipid metabolism (g3). The g4 group indicates that an important function of the corpus mucosa is related to blood vessels and blood clotting. In this group, genes related to fibrinogen complex, platelet activation and blood vessel growth were found (see Additional file 1: Table S2 for details). This is illustrated most clearly by FGA and FGB, which encode the alpha and beta chains of fibrinogen, respectively. These genes were both among the 15 most highly over-expressed in the corpus compared to antrum of Hp- patients (20- and 31-fold over-expression, respectively). High FGA-expression in the corpus mucosa is confirmed by the Human Protein Atlas [19], where very strong patchy staining of FGA is observed in subepithelial cells of the foveolar region, while FGB is stained strongly by both epithelial and subepithelial cells (Additional file 1: Figure S3). The majority of gene ontologies associated to over-expression in corpus atrophy mucosa are related to immune response and inflammation (Figure 2; g6). There were, however, no differences in either active or chronic inflammation scores between the Hp + and Atr groups (Table 1), which indicate that the inflammatory association detected by the global analysis is related to more subtle inflammatory changes in the corpus atrophy mucosa.

The similarities of gene expression between antrum and corpus mucosa in atrophic gastritis is caused by transcriptional loss of corpus-specific gene groups

In agreement with data shown in Figure 1, the gene ontology (GO) analysis revealed a relative down-regulation of a large number of genes in the atrophic corpus mucosa with the loss of nearly all corpus-specific gene ontologies in the corpus atrophy group (Figure 2, g1 through g4). Strikingly, corpus-specific gene groups related to the acid secretion and the presence of parietal cells showed no difference between antrum and corpus samples of atrophy patients.

To enable a more detailed comparison between the different patient groups, an analysis of gene expression was performed using hierarchical clustering. While the previous analysis (Figures 1 and 2) compared differences between antrum and corpus samples within individual patients, the hierarchical clustering approach compared the gene expression of antrum and corpus samples separately, across the different patient groups. While only 356 genes were captured by the ANOVA cut-off (p-values < 0.001) in antrum samples, there were 1395 differentially expressed genes in the corpus samples. A clustering dendrogram clearly further divided the expression pattern into 4 and 3 clusters for corpus and antrum samples, respectively (Figure 3).

Detailed molecular evidence for simultaneous IM and SPEM in atrophic corpus tissue, as well as for SPEM development in non-atrophic H. pyloriinfected corpus tissue

In addition to a decrease of a substantial number of corpus-specific genes in atrophic mucosa, there was an almost equally large set of genes that were markedly up-regulated in the corpus tissue of atrophy patients. These genes grouped into the C2.1 cluster, which contains 442 genes (Figure 3). In addition, there were 217 genes that were up-regulated in both atrophic corpus mucosa and Hp + non-atrophic corpus mucosa (C2.2 cluster). A further analysis of C2.1 revealed that a proportion of C2.1 genes were antrum-specific. Using the non-conservative cut-off of p < 0.05 used in the initial differential analysis (Figure 1), 18% of the C2.1 genes were significantly over-expressed in antrum compared to corpus samples of Hp- subjects. Thus, 80 of the genes up-regulated in corpus atrophy mucosa were antrum-specific. An example of such a gene was gastrin (GAST), which showed a strongly increased expression in corpus mucosa of the Atr group compared to the Hp- group and Hp + group (253 fold difference; p = 2.68 × 10^-4 and 221 fold-difference; p = 2.68 × 10^-4 respectively).

The functional analysis of C2.1 genes revealed that the majority of ontologies in this cluster were related to functions of the intestinal epithelium, including lipid, cholesterol and sugar transport and metabolism (Figure 3). Indeed, enrichment analysis of a publicly available dataset, GSE7307, demonstrated that there was a robust enrichment of small intestinal-specific genes in the C2.1 cluster (p = 2 × 10^-26; Table 3). This strongly indicated that the C2.1 cluster was related to IM development in the atrophic mucosa, which agreed with the fact that the three atrophy patients analyzed by the microarray exhibited at least mild IM of the corpus according to histopathology assessment (Table 1). To further investigate the relation to IM, we performed an additional enrichment analysis compared to a recently published dataset of IM as well as SPEM marker genes, obtained by microarray analysis of microdissected human SPEM and IM tissue [16]. Using this data, we observed that the C2.1 cluster exhibited an extraordinary enrichment of IM up-regulated genes (p = 2 × 10^-111, Table 3). Importantly, there was also a significant enrichment of genes observed to increase in both IM and SPEM tissue (p = 5 × 10^-4), indicating that the antralization of the corpus was associated with SPEM. Enrichment analysis of the C2.2 cluster, which contained genes increased in the corpus mucosa of both Hp + and Atr groups, revealed an equally high association to genes up-regulated in both SPEM and IM (p = 6 × 10^-7), as with genes increased in IM alone (1 × 10^-7, Table 3), demonstrating a stronger association to SPEM genes compared to the C2.1 cluster. Furthermore, 43% of the genes in C2.2 were antrum-specific, as determined by a p-value cut-off of 0.05 for increased expression in antrum tissue of Hp- patients. The 5 genes of the C2.2 cluster with the highest antrum-specificity in uninfected mucosa were MUC17, CEACAM6, CLDN7, SLC6A14 and XDH (55-fold, 27-fold, 24-fold, 15-fold and 11-fold over-expression in normal antrum vs corpus, respectively). Some of the gene groups most highly associated with the C2.2 cluster are related to cell cycle regulation (Figure 3 “bubbles”), which shows that even in the pre-atrophic Hp + corpus, there are alterations in proliferative regulation, which may be involved in SPEM development. Strikingly, there was a strong association with DNA repair genes within the C1 cluster, i.e. genes up-regulated in Hp + corpus and then again decreased in the atrophic corpus mucosa. Thus, while proliferative regulation was altered in both non-atrophic and atrophic corpus mucosa, DNA repair genes were increased only in the non-atrophic tissue.

Analysis of corpus-specific genes lost in the atrophic mucosa reveals acidic mammalian chitinase as a signature for atrophy

We performed RT-PCR analysis of AMCase expression in order to confirm results obtained by the microarray analysis. RT-PCR confirmation was also performed for the ATP4B gene, as well as for PGA5, which are both known to be down-regulated in atrophic corpus tissue. RT-PCR analysis indeed confirmed that AMCase expression was down-regulated, as was expression of ATP4B and PGA5 (Figure 4). To further verify the loss of AMCase, we performed Western blot analysis of tissue lysates from antrum and corpus tissue from a number of patients of the three groups. This showed that there was an almost complete loss of AMCase in the atrophic corpus mucosa, confirming that loss of AMCase expression is a signature for atrophy. On the individual patient level, there was a very strong correlation (r² = 0.92; p < 0.0001) between mRNA and protein levels, supporting the specificity of the western blot assay used (Additional file 1: Figure S5).