Changes of Gene Expression in Gastric Preneoplasia following Helicobacter pylori Eradication Therapy

Gastric cancer is the second leading cause of cancer death and the fourth most common cancer in terms of new cases worldwide (1). In year 2005 alone, an estimated 21,860 incident gastric cancer cases and 11,550 deaths may occur in the United States (2). Despite the declining incidence in gastric cancer worldwide, due to the aging of the world population, gastric cancer is projected to be the eighth leading cause of all deaths by year 2010 (3). Helicobacter pylori infection is a major risk factor for gastric cancer. Based on epidemiologic studies (4-6) and animal models (7), the IARC classified H. pylori as a group 1 carcinogen, a definite cause of human cancer, in 1994 (8). The link between H. pylori infection and gastric cancer provides scientists with hope that eradicating H. pylori, a curable infection, can arrest carcinogenesis and thus prevent cancer.

Randomized clinical trials of H. pylori eradication therapy using cancer as an outcome are challenging due to the changing epidemiology of infection, requirement of large sample size, and necessity of long follow-up period. Consequently, investigators (ourselves included) have tried to show that eradication of H. pylori can reverse preneoplastic conditions (9, 10). These studies of “intermediate end points,” however, have their own problems. To date, studies of H. pylori eradication and gastric cancer regression have suggested that preneoplasia can regress, but this improvement is far from complete (10, 11). If regression of histologic preneoplasia is complete, then most people will agree that prevention of gastric cancer is likely. In the absence of complete regression, a strong probability remains that only the innocent lesions reverse and those leading to cancer do not. It is thus important to know the molecular underpinning of incomplete histologic regression.

To this end, we employed cDNA microarray technology to analyze gastric tissue biopsies from a randomized, placebo-controlled trial of H. pylori therapy conducted in subjects with chronic gastritis, atrophy, and/or intestinal metaplasia from Chiapas, Mexico. By identifying changes of gene expression due to H. pylori eradication, we hope to dissect the molecular pathways to cancer precursor lesions and assess their reversibility following H. pylori eradication.

Gastric tissue biopsies were selected from a randomized, double-blinded, placebo-controlled trial conducted in Chiapas, Mexico. The trial was approved by the Human Subject's Committees at Stanford University (Stanford, CA), Colegio de la Frontera Sur (Chiapas, Mexico), and Instituto Nacional de Cancerologia (Mexico City, Mexico). Details of the clinical trial design and subject recruitment were described elsewhere (9). In brief, we recruited healthy Chiapas residents over 40 years old and without any of the following conditions: current pregnancy, known alcoholism, allergic reactions to study medication, history of malignancy, gastrectomy or debilitating illness, recent antibiotic use, previous H. pylori eradication therapy, and specific medication use. We then screened the participants for their serum antibodies to H. pylori's CagA protein and fasting serum gastrin concentrations. We invited those who were CagA antibody positive and with fasting serum gastrin levels ≥25 μg/mL to participate in this study because, according to previous studies of the same population by our group, such subjects were more likely to have gastritis and/or gastric preneoplastic conditions than those with cagA-negative H. pylori and gastrin levels <25 μg/mL (12).

After screening, 327 subjects were invited to participate in the trial, and 316 agreed to enroll. The 316 subjects were then randomized to receive either treatment (20 mg of omeprazole twice a day, 1,000 mg of amoxicillin twice a day, 500 mg of clarithromycin twice a day) or placebo for a week. All 316 subjects had signed the first consent form at the time of initial screening and the second consent form before randomization. The third consent requesting the subjects' permission to use their tissues for further research was obtained from 213 (102 from the treatment group and 111 from the placebo group) of the 316 subjects. There was no statistically significant difference in age and sex distributions between those who signed the third consent and those who did not (data not shown). Upper endoscopy was done before the intervention, and at 6 weeks and 1 year after the intervention. For each subject, seven biopsies (three each from the antrum and body and one from the incisura angularis) were systematically collected from prespecified locations and embedded in paraffin. Histopathology of the seven biopsy samples was used to diagnose the presence of H. pylori in the study subjects at baseline and 1 year. An additional sample was taken from a site immediately adjacent to one of the antrum biopsies and snap frozen in liquid nitrogen. It is the frozen samples that were available for microarray analysis.

For this study, eligible subjects were those who signed all three consent forms. In addition, because the purpose of this study was to evaluate changes of gene expression associated with H. pylori eradication, those in the treatment group who remained infected after H. pylori therapy and those in the placebo group whose H. pylori disappeared without medication were not eligible. Of the 102 subjects from the treatment group who signed all three consent forms, 68 had H. pylori infection at baseline and no H. pylori at 1 year after treatment. Of the 111 subjects from the placebo group, 94 had H. pylori infection at both baseline and 1 year.

For microarray analysis, we included the biopsies of 13 eligible subjects from the treatment group and 14 from the placebo group. We chose subjects with the best preserved, frozen tissue samples and also balanced the age and sex distributions of the two groups. For immunohistochemistry, we selected the biopsies of all the subjects included in the microarray analysis plus 15 additional subjects from the treatment arm and 17 from the placebo arm. Tissues used for immunohistochemistry were the paraffin-embedded antrum biopsies taken from the sites immediately adjacent to where we took the snap-frozen biopsies. For both microarray and immunohistochemistry, we analyzed a pair of biopsy samples from each subject, one taken before the intervention and the other at 1 year after the intervention.

The histologic diagnosis was based on the same biopsies as those used for immunohistochemistry. The biopsies were read by a pathologist blinded to the treatment assignment and time of tissue sampling (baseline and 1 year). To evaluate change of histologic diagnosis between biopsies taken at baseline and those taken at 1 year, we assumed the following ascending order of preneoplastic conditions: no atrophy < moderate or severe atrophy < mild, moderate, or severe intestinal metaplasia < moderate dysplasia (9). The diagnosis of mild atrophy was ignored because it is prone to misclassification (9). Any change from a lower to higher order over time was defined as “progression,” whereas the reverse was defined as “regression”.

Total RNA was extracted by using Trizol (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. To obtain sufficient RNA for microarray analysis, we used Ambion MessageAMP aRNA kit to amplify extracted mRNA. First, the extracted RNA was resuspended in DEPC water and T7 bacteria phage promoter was incorporated in a reverse transcription reaction using an oligo dT(15)-T7 primer [5′-AAACGACGGCAAGTGAATACGACTCACTATAGGCGCT(15)-3′]. The master mix included 2 μL 10× first-strand buffer, 1 μL template-switch oligo primer (5′-AAGCAGTGGTAACAACGCAGAGTACGCGGG-3′), 1 μL RNase inhibitor, 4 μL deoxynucleotide triphosphate mix, and 1 μL reverse transcriptase. First-strand cDNA synthesis was then done at 42°C for 120 minutes. Full-length second-strand synthesis was subsequently accomplished by adding 63 μL nuclease-free water, 10 μL 10× second strand buffer, 4 μL deoxynucleotide triphosphate mix, 2 μL DNA polymerase, and 1 μL RNaseH. The mixture was then incubated for 120 minutes at 16°C. The resulting double-stranded cDNA was purified using the supplied cartridge, eluted in 100 μL DEPC water, and concentrated to the final volume of 8 μL using speed vacuum. Following cleaning/concentration of the double-strand cDNA, a mixture of 2 μL of T7 ATP, T7 CTP, T7 GTP, T7 UTP, T7 10× reaction buffer, and T7 enzyme mix was added to the double-strand cDNA solution and incubated in 37°C for 6 to 12 hours to synthesize aRNA. The resulting aRNA was then cleaned using the supplied cartridge and cleaning solution.

cDNA was synthesized and labeled by reverse transcription from aRNA in the presence of Cy5 using the Superscript II reverse-transcription kit (Life Technologies, Gaithersburg, MD). For all experiments, the comparator mRNA, labeled with Cy3-dUTP, was a reference pool of mRNA from 10 cancer cell lines manufactured by Stratagene (La Jolla, CA). Cy3-labeled and Cy5-labeled probes were concentrated to 16 μL by speed vacuum. Hybridization is conducted at 65°C with 3.4 × SSC, 0.3% SDS and 20 μg each of polyadenylated RNA, 1 μL yeast tRNA (Sigma, St. Louis, MO), and human Cot1 DNA (Life Technologies). After washing, the arrays were scanned on an Axon Instruments GenePix scanner and then gridded using GenePix 4.0 Software (Axon Instruments, Union City, CA). Spots with obvious blemishes were manually flagged. These data were entered into the Stanford microarray database (http://genome-www5.stanford.edu/). The microarray chips used in the study contained ∼44,500 cDNA clones that represented ∼30,300 unique genes and were produced at Stanford Microarray Core. The full list of clones used and their nominal identities are available at http://genome-www4.stanford.edu/cgi-bin/sfgf/home.pl. Gene expression data from all of the arrays used in this study are available at http://genome-www5.stanford.edu/.

We used immunohistochemistry to verify significant changes of gene expression observed by microarray. Three-micrometer formalin-fixed, paraffin-embedded tissue sections were used for the FABP1 protein immunohistochemical assay. Sections were placed in pretreated slides, deparaffinized with xylene, and rehydrated with ethanol and distilled water. Antigen retrieval was done using 10 mmol/L EDTA for 10 minutes in a microwave oven. Then, cool tissue sections were placed in a DAKO autostainer (DAKO Corp., Carpinteria, CA) where sections were washed with TRIS-buffered-saline Tween 20 for 5 minutes and incubated with 1:50 dilution anti–NH₂ terminus of FABP1-purified goat polyclonal primary antibody (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) for 30 minutes at room temperature. The sections were then washed, incubated with a biotinylated secondary antibody (K0690, DakoCytomation, Carpinteria, CA), streptavidin-peroxidase, and colorimetric detection was done with diaminobenzidine. Sections were counterstained with Harris hematoxylin. A sample of jejunum from a patient with a previous gastro-jejunum anastomosis was used as a positive control because of its high FABP1 expression in normal human small intestine. Evaluation of staining was done by a pathologist blinded to the treatment assignment and time of tissue sampling and was based on fraction of positive cells and the staining intensity, as described by Hashimoto et al. (13). Three compartments were also considered: gastric pits, gastric glands, and gastric intestinal metaplasia. Fraction of positive cells was scored as 1 for <5%, 2 for 5% to 50%, and 3 for >50%. The staining intensity was scored as 1 for no or low intensity, 2 for moderate intensity, and 3 for high intensity. The summary of the two scores was used to determine the extent of immunoreactivity. For each subject, we then compared the staining results of the pre-intervention biopsy to those of the post-intervention biopsy to evaluate change of FABP1 staining over time.

We selected nonflagged array spots with a regression correlation of >0.6 and fluorescent intensity in either channel greater than twice the local background. Only genes that met the abovementioned criteria and were available for >80% of arrays were included. We then used a scaling factor to set the mean sample/reference ratio for all the included spots to one. The normalized sample/reference ratios were next transformed to log2 scale and mean centered for each clone across all arrays to reduce potential printing and array-specific bias.

For each intervention group, we first used Fisher's exact test to test for difference in variable distributions between baseline and 1 year. Unsupervised hierarchical clustering algorithm (14) was used to group genes and biopsy samples. For each individual, we first calculated the net difference in gene expression levels for each gene by subtracting the gene expression level at baseline from that at 1 year. The value of this net difference represented the changes of gene expression over the 1-year period. Next, we filtered the genes by including only those with a net difference >3 SDs away from the mean of the net difference in at least three arrays. The filtering criteria left the data set with 207 cDNA clones for hierarchical clustering analysis.

To identify genes whose expression changed significantly over the 1-year period, we used significance analysis of microarrays (SAM algorithm; ref. 15). SAM computes a nonparametric t statistic for each normalized log ratio of gene expression level, and the t statistic was compared with a chosen threshold value to determine if the gene was “significant.” Each threshold value corresponded to a false discovery rate (the percentage of genes identified as “significant” by chance alone), which was generated by randomly permuted data. Three comparisons were made.

(a) For each person in the treatment group, we used SAM's paired analysis function to compare the gene expression levels of biopsies taken at baseline to their counterparts in biopsies taken at 1 year after H. pylori eradication. This comparison allowed us to evaluate the molecular changes over time due to H. pylori eradication.

(b) For each person in the placebo group, we used SAM's paired analysis function to compare the gene expression levels of biopsies taken at baseline to their counterparts in biopsies taken at 1 year after placebo. This comparison allowed us to evaluate the molecular changes over time in those with no treatment but possible progression of infection.

(c) We used the two-group analysis function of SAM to compare the baseline gene expression between biopsies with no intestinal metaplasia with those with intestinal metaplasia, regardless of their treatment/placebo assignment, to evaluate expression of genes associated with intestinal metaplasia.

We categorized changes of FABP1 expression into subgroups after completing microarray analysis and subsequent confirmatory immunohistochemistry assay. For the qualitative results of FABP1 immunohistochemistry, the changes of FABP1 expression were classified as increased staining, no change, or decreased staining. A change from a lower intensity category or smaller fraction of positive cells to a higher intensity category or larger fraction of positive cells is considered an increased staining, and the reverse is considered a decreased staining. For the continuous microarray results on FABP1 expression, the changes of FABP1 expression were classified as either up-regulation or down-regulation. Because of the continuous nature of gene expression data, even a small change of log2 (sample/reference) ratio can be categorized as either up-regulation or down-regulation. Next, we used Fisher's exact test to compare changes of expression of FABP1 between the treatment and placebo groups using data from both immunohistochemistry and microarray.

Table 1 shows the characteristics of the subjects selected for microarray analysis. The treatment and placebo groups had similar median ages at study entry (P = 0.89) and sex distribution (P = 0.58). Corresponding to our subject inclusion criteria, everyone in the treatment group had H. pylori infection at baseline but no infection at 1 year, whereas all the subjects in the placebo group remained infected with H. pylori at 1 year. The treatment group showed significant improvement of gastritis after H. pylori eradication (P = 0.0022), whereas the placebo group had no significant regression of gastritis over time. We observed no significant histopathologic change in atrophy and intestinal metaplasia over the 1-year period, in either the treatment or the placebo group.

Table 2 shows the baseline characteristics of subjects included in immunohistochemistry analysis. The variable distributions in subjects selected for immunohistochemistry were similar to those listed in Table 1, because all subjects analyzed by microarray were also included in the immunohistochemistry analysis.

We applied hierarchical clustering algorithm to group samples and genes based on the similarities of their expression patterns (Fig. 1). Unsupervised hierarchical clustering based on the 207 selected genes with the most changes of expression showed a major division of samples into treatment and placebo groups. A few biopsies from the placebo group were clustered with those from the treatment group; this overlap was not associated with histologic diagnosis.

Based on the results of SAM, the expression of 30 genes in biopsies taken at baseline differed significantly from their counterparts in biopsies taken at 1 year after treatment (0 up-regulated and 30 down-regulated; Table 3). The median false discovery rate of this analysis was 13.9%. The majority of the down-regulated genes were associated with immune response and inflammation (CXCL1, CXCL14, IGLC2, LOC400986, TNFSF10, and OAS1). Other down-regulated genes were those associated with cell-cell adhesion (LGALS4, TACSTD1, and VCAM1); cell surface receptor and lining (GPA33, ITGB8, and MUC13); signal transduction (GRK5 and TXNRD1); cell-cycle progression and differentiation (S100A10); transcription (MAD and XBP1); cytoskeleton (KRT19); protein synthesis and metabolism (GCNT3 and MMP12); protein modification (TRIM31 and TRA1); lipid synthesis, transport, and metabolism (FABP1, FABP4, CES2, and MTP); transport (LCN2); metabolism (PP); and other functions (CAB and ANXA4).

To investigate the underlying mechanism of H. pylori–induced molecular changes of premalignant lesions over time, we used SAM to compare the gene expression levels of biopsies taken at baseline to their counterparts in biopsies taken at 1 year after placebo. SAM identified 32 up-regulated and 23 down-regulated genes, with a median false discovery rate of 12.1% (Table 4). The commonly up-regulated genes in the placebo group included genes involved in immune response (ADA, C4BPB, and NT5E); cell-cell adhesion, surface receptor, and lining (CDH17, TACSTD1, MUC13, and CLDN7); cell cycle progression and differentiation (S100A10); transcription (TBX3, HNF4G, and HTATIP2); cytoskeleton (KRT20); protein biosynthesis and metabolism (TM4SF5, PGA5, PRSS3, ANPEP, and PEPD); lipid synthesis, transport, and metabolism (FABP1, DGAT1, HMGCS2, APOA1, CRYL1, AZGP1, HSD17B2, and MTP); transport (SLC5A1, VDAC2, HBB, and LAPTM4B); and other functions (CaMKIINα, LOC339321, and OLFM4).

Many of the genes down-regulated over time in the placebo group were associated with immune response (TNFSF10, CSF2RB, SERPING1, C1S, CD53, NCF1, IGJ, IGLC2, IGLL1, and TNFRSF17). Other down-regulated genes included those involved in cell-cell adhesion (VCAM1, PCDH18), cell proliferation and differentiation (NHP2L1 and LIMS1), signal transduction (PTPRC, RAB31, and PDK1), cytoskeleton and skeletal development (STOM and LUM), protein binding (PHC2), actin binding (LCP1), peptide catabolism (KYNU), and other functions (MANEA and FGL2).

Of particular interest are the common genes that showed significant changes of expression over time in biopsies from both the treatment and placebo groups. Five of the genes down-regulated over time in samples from the treatment group (FABP1, MTP, MUC13, S100A10, and TACSTD1) were up-regulated in samples from the placebo group. In a separate analysis that compared the changes of gene expression over time between the placebo and treatment groups, the abovementioned five genes consistently showed greater changes in expression in the placebo subjects than in the treatment subjects (data not shown). This suggests that H. pylori eradication might have stopped or reversed an ongoing molecular process that otherwise would have progressed.

We compared the baseline gene expression of subjects who had both gastritis and intestinal metaplasia (5 subjects) with those with only gastritis (22 subjects), regardless of their intervention group assignment. From the results of SAM, tissues with intestinal metaplasia had up-regulation of 11 genes and down-regulation of one gene when compared with those without intestinal metaplasia. The 11 up-regulated genes included those involved in defense/immune response (TFF3, DAF, DPP4, and NT5E), surface antigen/receptor activity (GPA33), fatty acid metabolism (FABP1), protein biosynthesis and metabolism (PEPD, PRSS2, and ASS), cytoskeleton (VIL1), and metabolism (AGXT1). The down-regulated gene (ABCC5) is associated with cellular transport. Some of the up-regulated genes found in intestinal metaplastic tissues were also found in our previous analyses. For examples, the gene GPA33 was significantly down-regulated over time in tissues from the treatment group (Table 3), whereas NT5E and PEPD were up-regulated in tissues from the placebo group (Table 4). In addition, the gene FABP1 became down-regulated over time in biopsies from the treatment group but up-regulated in those obtained from the placebo group (Tables 3 and 4).

To verify the changes of gene expression observed in the microarray data, we applied immunohistochemistry to stain the FABP1 protein using all the samples analyzed by microarray and the additional 31 samples selected from the Chiapas trial participants. The representative positive and negative staining of FABP1 are shown in Fig. 2. We chose to stain the FABP1 protein because FABP1 was the only gene significantly down-regulated over time in the treatment group but up-regulated in the placebo group and those with intestinal metaplasia. Table 5 shows the results of FABP1 immunostaining and corresponding microarray data on FABP1 expression. Based on the immunohistochemistry results, 5 people (17.2%) from the treatment group and 2 (6.5%) from the placebo group had decreased FABP1 staining over time. In contrast, none in the treatment group had increased FABP1 staining, whereas 5 (16.1%) in the placebo group had increased staining of FABP1. Taken together, using microarray or immunohistochemistry, treatment and placebo groups were significantly different in terms of their changes of FABP1 staining, even in the absence of significant changes of histopathologic diagnosis.

Studies of intermediate end points of gastric cancer have shown that some precursor lesions can regress after H. pylori eradication, but the improvement is not complete (9, 10, 16). The only randomized trial of H. pylori eradication using cancer as an outcome indicated that only a subset of patients with no discernable premalignant conditions could benefit from H. pylori eradication, and those with preneoplastic conditions retained a high risk of developing gastric cancer even after H. pylori eradication (17). To delineate the mechanisms and reversibility of gastric preneoplastic lesions, and to define the subset of patients who may benefit most from screening and treatment, it is important to know the molecular underpinning of H. pylori pathogenesis.

Changes of gene expression related to any environmental perturbation can theoretically be captured on these arrays even before any phenotypic change occurs. Thus, several studies have used microarray technology to evaluate H. pylori–associated gene expression in gastric cancer cell lines or tissues from gastric cancer patients (18-31). Cancer cell lines, however, do not fully represent normal human cells. Studies using tissues from human population face another challenge: it is difficult to control for unknown baseline differences between people from heterogeneous clinical populations. To our knowledge, this study is the first to investigate changes of gene expression profiles associated with H. pylori infection and eradication using samples from a double-blinded, placebo-controlled clinical trial. Studying the genes that showed significant changes of expression over time may help reveal molecular mechanisms of H. pylori–induced preneoplastic conditions.

In this study, H. pylori infection and eradication resulted in alterations of gene expression associated with cell damage, inflammation, proliferation, apoptosis, and intestinal differentiation. No gene was up-regulated over time in tissues from the treatment group. More changes of gene expression, possibly associated with persistent H. pylori infection and progression of preneoplasia, were observed in the placebo group. Furthermore, the fact that five genes (TACSTD1, MUC13, S100A10, FABP1, and MTP) changed expression in opposite directions in the two intervention groups suggests that H. pylori eradication might have stopped or reversed an ongoing molecular process that otherwise would have progressed. This observation is consistent with current knowledge that H. pylori infection induces cell hyperproliferation, inflammation, and genomic instability (32), thereby resulting in the up-regulation of genes related to transcription, cell-cycle progression, and metabolism. Eradicating the infection thus leads to down-regulation of these genes.

Two genes (MUC13 and TACSTD1) involved in cell adhesion and lining became down-regulated in the treatment group but up-regulated in the placebo group. Mucins are glycoproteins expressed by all mucosal epithelial tissues and can serve as protective layer of the mucus. Overexpression of mucins is common in gastrointestinal adenocarcinomas, and differential up-regulation of MUC13 can be found in tissues of colorectal carcinomas (33, 34). The protein TACSTD1, also called EpCAM, functions as a calcium-independent homophilic adhesion molecule that modulates cadherin-mediated intercellular adhesions (35). In addition, TACSTD1 plays a direct role in regulating cell cycle and proliferation through up-regulating the proto-oncogene c-myc (36). The expression of TACSTD1 is found in many normal epithelial cells, and up-regulation of TACSTD1 is seen in gastrointestinal carcinomas (31, 37-39) and other tumors of epithelial origin (40). In normal gastric epithelium, up-regulation of TACSTD1 expression has putatively been associated with early-stage intestinal metaplasia development (35). Increasing expression of TACSTD1 is positively correlated with dedifferentiation and proliferation of malignant epithelial cells (41). Clinically, TACSTD1 has been used as a diagnostic and prognostic marker for carcinomas (42, 43) and a target in colorectal cancer therapy trials (44).

H. pylori eradication and persistent infection were accompanied by changes of expression of genes involved in lipid transport and metabolism. Of these, FABP1 (fatty acid-binding protein 1, liver) showed the most significant change. FABP1 was down-regulated in those who received H. pylori eradication but up-regulated in those without H. pylori eradication and those with intestinal metaplasia. This observation was coherent with the results of FABP1 immunohistochemistry. FABPs belong to a family of small cytoplastic proteins that bind to long-chain fatty acids and other hydrophobic ligands and are commonly expressed in tissues with active fatty acid metabolism (45). FABP1 has a multi-tissue expression property and is found in liver, intestine, colon, and kidney (46). Pelsers et al. have shown that FABP1 can be used as a plasma marker for the detection of intestinal injury in patients with intestinal ischemia and other intestinal diseases (47, 48).

Although several significant genes may be involved in H. pylori pathogenesis, we do not know whether the expression of these genes are regulated directly by H. pylori or indirectly by other host factors. We also do not know if these genes are in the causal pathway of H. pylori carcinogenesis. In addition, because only a small fraction of the genes profiled showed significant changes over time, it is possible that the significant genes observed in this study do not fully represent the spectrum of genes involved in H. pylori pathogenesis. Also unknown is the interaction between these genes. Extrapolation of results based on a limited number of genes observed in this study can be misleading. Furthermore, the significant genes observed in this study can be population specific. With a different population, it is possible to have different significant genes.

To show that the observed alterations of gene expression are reproducible and can be used as markers for disease progression, several further studies are needed. First, it would be desirable to perform a similar study in a different population to see if specific patterns of gene expression are consistent. Second, physiologic studies on the mechanisms of these genes are important to evaluate the role they play in H. pylori–induced gastric preneoplasia and neoplasia. Third, it is of interest to confirm the association of specific gene expressions and gastric cancer using archived tissues from other population-based studies of gastric cancer (nested case-control studies or prospective cohort studies). In addition, using both normal and malignant tissues collected at different points of time from the same individual can help us delineate the association between changes of gene expression and cancer risk.

This study has some limitations. Because only a part of the frozen tissues yielded good-quality RNA, our sample size was limited. A larger sample size would give more statistical power to detect significant changes of gene expression with a lower false-discovery rate. In addition, as addressed by Mannick et al., H. pylori infection causes extensive inflammation, and the inflammatory cells may hamper the study of gene expression in gastric mucosa (24). It is thus difficult to determine the direct effect of H. pylori on gastric epithelial cells. Furthermore, the observed changes of expression in the treatment subjects might have been the side effect of antibiotics, rather than the consequence of H. pylori eradication. To help clarify the direct effect of H. pylori eradication, it would be useful to have gene expression data on placebo subjects whose H. pylori disappeared spontaneously during the 1-year period and treatment subjects whose H. pylori infection persisted 1 year after the triple therapy. Other limitations of this study include sampling bias because of the focal nature of preneoplastic and neoplastic lesions. Although biopsy samples for genetic and histopathologic analysis and before and 1 year after treatment were taken from the same general area in the stomach, they could not have been taken from the exact same location. Additional limitations when obtaining biopsy samples include the misidentification of biopsy sites by endoscopists, problems in identifying mucosal textures, and normal anatomic variations in each patient.

The potential limitations are counterbalanced by several study strengths. This is the first study to investigate the alterations of gene expression patterns associated with H. pylori eradication in human population. Its strengths include (a) use of a placebo-controlled design, with the placebo group serving as a reference group to show natural history of persistent H. pylori infection; (b) each person serves as his/her own comparison, thus minimizing potential selection bias and confounding; and (c) results from immunohistochemistry further confirmed the significant change of FABP1 expression detected by microarray.

In summary, this study shows an association of changes of gene expression and H. pylori eradication. The genes represent a wide spectrum of functional categories, such as immune response, inflammation, cell cycle regulation, and lipid metabolism. With further verification, alterations of expression of the significant genes may potentially be used as markers for identifying a subset of people at a higher risk of disease progression who may benefit most from screening.

We thank Raul Belmonte, Cecilia Limón, Juan Antonio Moguel, and Rosario Moreno for assistance with data collection; Amanda Allen, Marc Welsh, and William Lee for their help in testing these samples; and El Centro de Investigaciones en Salud de Comitan and El Colegio de la Frontera Sur in Chiapas, Mexico, for the use of their facilities. We also thank the Stanford Microarray Database Staff for their technical assistance.