Identification of Protein Expression Signatures Associated with Helicobacter pylori Infection and Gastric Adenocarcinoma Using Recombinant Antibody Microarrays*

Antibody microarray based technology is a powerful emerging tool in proteomics, target discovery, and differential analysis. Here, we report the first study where recombinant antibody fragments have been used to construct large scale antibody microarrays, composed of 127 different antibodies against mostly immunoregulatory antigens. The arrays were based on single framework recombinant antibody fragments (SinFabs) designed for high on-chip stability and functionality and were used for the analysis of malignant and normal stomach tissue samples from Helicobacter pylori-positive and -negative patients. Our results demonstrate that distinct tumor- as well as infection-associated protein expression signatures could be identified from these complex tissue proteomes, as well as biomarkers such as IL-9, IL-11, and MCP-4, previously not found in these diseases. In a longer perspective, this study may improve the understanding of H. pylori-induced stomach cancer and lead to development of improved diagnostics.

Gastric adenocarcinoma is often associated with Helicobacter pylori infection, and 1-2% of all infected individuals subsequently develop stomach cancer (6). The immune response to H. pylori results in regulation of different cytokines, chemokines, and other immunomodulatory molecules and is one of the most important cofactors for tumor progression (7). Consequently, to study the components of the immune response that are associated with the initiation and progression of gastric adenocarcinoma requires analysis of a large number of different molecules, often using only small tissue samples. Antibody microarray-based analysis of relevant immunomodulatory proteins in stomach tissue offers a potential approach to address these basic questions as well as to detect diseaseassociated protein signatures and biomarkers (1)(2)(3).
Antibody microarrays, pioneered by MacBeath and Schreiber (8) and Haab et al.(9), have emerged as a sensitive and high throughput method for detection of proteins and peptides (1,3). The prospect of using this technology for cancer research was rapidly realized (10 -12), and antibody microarrays more recently have been utilized for analysis of larger sets of clinical samples from tissue and serum (13)(14)(15)(16)(17)(18). Although the platforms used in these studies have been useful, the content, i.e. commercial off-the-shelf monoclonal and polyclonal antibodies, confer several limitations (19 -21). First, different monoclonal antibodies are derived from different germ line genes and therefore have different frameworks. Consequently their on-chip performance will display an unacceptable variation (22), resulting in tedious validation of each new individual antibody. Second, scaling up the arrays would require production of thousands of monoclonal antibodies, which is both time consuming and costly. To overcome these limitations, we have developed antibody fragments by molecular design. These microarray adapted single framework antibody fragments (SinFabs) 1 are single chain variable fragments (scFv) obtained from a phage display library (23), where all individual scFv are built around one single, stable framework (VH3-23/VL1-47) (23). These SinFabs have been used to develop an antibody microarray platform (19,24,25) suitable for sensitive analysis of complex proteomes, such as serum and tissue extracts. 2,3 In this study, we have constructed an array using 127 different SinFab probes against both low and high abundant proteins, such as cytokines, chemokines, and complement factors. Protein extracts from three different tissue types such as H. pylori-positive gastric adenocarcinoma, H. pylori-positive normal gastric epithelium, and H. pylori-negative normal gastric epithelium has been analyzed. The aim of the study was to validate the platform in oncoproteomics and to demonstrate that specific protein signatures associated with both tumor as well as infection could be identified.

EXPERIMENTAL PROCEDURES
Patient Specimens and Protein Extraction from Tissue Samples-In total, 35 samples from 20 patients (median age 75, range 55-87, 7 female, 13 men) undergoing gastrectomy at Sahlgrenska University Hospital, Gö teborg, Sweden, were included in the study. The patients were either suffering from non-cardia gastric adenocarcinoma (GA) (15 patients) or from pancreatic adenocarcinoma (5 patients). The study was approved by the ethical review board at Gö teborg University, and informed consent was obtained from each volunteer before participation. None of the patients had undergone radiotherapy or chemotherapy prior to gastrectomy.
The H. pylori-positive gastric adenocarcinoma samples (GA/Hpϩ) were obtained during surgery, and the H. pylori-positive normal gastric tissue samples (N/Hpϩ) were obtained as a strip of tumor-free tissue at least 5 cm from the tumor, whereas all H. pylori-negative normal gastric tissue samples (N/HpϪ) were obtained from pancreatic cancer patients. The normal gastric tissue samples from both H. pylori-positive and H. pylori-negative retained their normal architecture and showed no sign of tumor cells. Tumor-free mucosa from both antrum and corpus of the same patient was obtained in 3 of 5 H. pylori-negative normal samples and 6 of 15 H. pylori-positive normal samples. However, we found no statistically significant differences in expressed antigens between the antrum and corpus populations (data not shown). From the tissue samples, biopsies with a size of ϳ10 -20 mm 2 were cut and stored at Ϫ70°C for subsequent protein extraction. Proteins were extracted from the biopsies by incubating them in PBS containing 2% saponin, 100 g/ml soybean trypsin inhibitor (Sigma Chemical Co.), 350 g/ml phenylmethylsulfonyl-fluoride (Sigma Chemical Co.), and 0.1% bovine serum albumin at 4°C overnight. The samples were then centrifuged (13000 ϫ g for 5 min), and the supernatants were stored at Ϫ70°C until biotin labeled as described below.
Diagnosis of H. pylori Infection-The H. pylori infection status of the patients was analyzed by bacterial culture on Columbia-Iso plates and by serology. The serological analysis was performed using in-house ELISA methods detecting IgG or IgA antibodies to crude membrane preparations of H. pylori as described previously (28) as well as with a commercial ELISA kit (EIA-G III; Orion Diagnostics). Patients were considered to be H. pylori-positive if they were positive in culture and/or in at least 2 of 3 serology tests and considered to be H. pylori-negative if they were negative in culture and in all three serology tests.
Labeling of Protein Samples from Tissue Extracts-The extracted protein samples were desalted on a Zeba desalt spin column (Pierce) and eluted in PBS. The protein concentration was determined using a Micro BCA TM Protein Assay Reagent Kit (Pierce), and each sample was diluted to 0.5 mg protein/ml in PBS. For normalization purposes, cholera toxin subunit B (Sigma Chemical Co.) was added to all samples at a final concentration of 280 nM. Next, the samples was biotinylated with Nhydroxysulfosuccinimide-biotin (Pierce) by adding N-hydroxysulfosuccinimide-biotin to a final concentration of 0.6 mM for 2 h on ice, with careful vortexing every 20 min. Unreacted biotin was removed by extensive dialysis against PBS at 4°C, and the samples were then aliquoted and stored at Ϫ80°C prior to use.
Fabrication and Processing of Antibody Microarrays-127 different antibodies selected from the n-CoDeR library (23) were kindly provided by BioInvent International AB (Lund, Sweden). Briefly, the scFv antibodies were produced in Escherichia coli and purified by affinity chromatography on nickel-nitrilotriacetic acid matrix (Qiagen, Hilden, Germany). Bound molecules were eluted with 250 mM imidazole, extensively dialyzed against PBS, concentrated (average concentration 277 g/ml), and stored at 4°C until used. The integrity and purity of the scFv antibodies were confirmed by 10% SDS-PAGE (Invitrogen).
For production of the antibody microarrays, we used a set up previously optimized. 2, 3 Briefly, 1 to 30 femtomoles (7 fmol average) of each scFv were spotted in eight replicates using a non-contact Biochip Arrayer1 (PerkinElmer Life Sciences, Inc.). The scFvs were arrayed by spotting 2 drops (333 pL/drop), which were dried in between, in each position onto black polymer MaxiSorp microarray slides (NUNC, Roskilde, Denmark). To assist the alignment of the grid during the subsequent quantification, a row containing Cy5-conjugated streptavidine (2 g/ml) was spotted for every tenth row. To facilitate chip to chip normalization, a dilution series (23-370 g/ml) of an anti-cholera toxin antibody (CT17) was spotted on each array. The arrays were blocked with 5% (w/v) fat-free milk powder (Semper AB, Sundbyberg, Sweden) in PBS over night at room temperature. All incubations were conducted in a humidity chamber at room temperature. The arrays were washed four times with 0.05% Tween-20 in PBS (PBS-T) and incubated with 250 l of the biotinylated sample (0.1 mg/ml) in 1% (w/v) fat-free milk powder and 1% Tween in PBS (sample buffer) for 1 h. Next, the arrays were washed four times with PBS-T and incubated with 1 g/ml Alexa Fluor 647-conjugated Streptavidin in sample buffer for 1 h. Finally, the arrays were washed four times with PBS-T, dried under a stream of nitrogen, and scanned using the confocal ScanArray Express microarray scanner (PerkinElmer Life Sciences) using four different scanner settings. The ScanArray Express software V2.0 (PerkinElmer Life Sciences) was used to quantitate the intensity of each spot and to compensate for background. The two highest and the two lowest replicates were automatically excluded, and each data point represents the mean value of the remaining four replicates.
Data Analysis-Background corrected data from the different scanner settings from each sample were scaled to one particular setting, and only data from unsaturated spots were used for further analysis. Chip to chip normalization of the datasets was performed by multiplying the signal data from each sample (i) by a factor N I , which was calculated by n ϭ S (CT17) / (CT17) . S (CT17) is the signal intensity from CT17 (within the linear range of the dilution series) for each sample (i), and CT17 is the average signal intensity of CT17 from all samples.
The log2 values were then calculated, and the differences between the sample classes were determined using significance of microarray analysis (SAM) (29). Antibodies that were found to give significantly different signal intensities between samples, using Wilcoxon two class unpaired test, were listed (see Tables I-III) and used for cluster analyses of the sample classes. The antigens that were differentially expressed between the sample classes, (i) GA/ Hpϩ versus N/HpϪ were filtered at a false discovery rate of 0% and a delta value of 0.811,(ii) N/Hpϩ versus N/HpϪ were filtered at a false discovery rate of 0% and a delta value of 0.505, and (iii) GA/Hpϩ versus N/Hpϩ were filtered at the false discovery rate 3.5% and delta value of 0.546 using the SAM method. The SAMbased approach was compared with analysis of variance-based methods, which generated similar results. Furthermore, normal probability plots that were generated for representative datasets demonstrated that they were typically normal distributed (data not shown). Principal component analysis (PCA) was performed using Spotfire software (version 8.0, Spotfire AB). The data was Z-score normalized by antigens prior to the hierarchical clustering analysis, which was performed using the unweighted pair-group method with arithmetic mean using Spotfire 8 software.

RESULTS AND DISCUSSION
Here, we report the first study where recombinant, microarray-adapted antibody fragments have been used to construct large scale antibody microarrays for analysis of clinical tissue samples. This antibody microarray technology platform 2,3 has several unique features because it is based on molecularly designed recombinant antibody fragments resulting in high    antigen, thus avoiding dependence on single antigenic epitopes, minimizing the risk of label induced artifacts.
Validation of the Microarray Platform-The image of a representative microarray (Fig. 1) demonstrates the homogenous spot morphology as well as the low background obtained with this platform. Analysis of the scanned images demonstrated that the correlation of the replicate spots within the arrays was high, displaying an average correlation coefficient of Ͼ0.99 ( Fig. 2A). Furthermore, the reproducibility of duplicate experiments was analyzed by calculating the average correlation coefficient and average coefficient of variance). The average correlation coefficient, based on samples run 2-3 weeks apart, was 0.94 (Fig. 2B), and the average interarray coefficient of variance was 18%, which compares well with other approaches (15,16), demonstrating that the platform handled analysis of complex clinical proteomes well.
The normalization procedure is vital for correct data interpretation (15), and we have investigated two different strategies 3 and concluded that chip to chip normalization, using a spike-in protein (cholera toxin, subunit B) works very well. 3 In Fig. 1, the magnification shows the dilution series of the SinFab clone (CT17) used for detection of the spiked-in cholera toxin. Moreover, most SinFab clones against different epitopes of the same antigen, e.g. IL-13 in Fig. 3, displayed a very similar overall pattern.
Identification of Protein Expression Signatures Associated with H. pylori Infection and Gastric Adenocarcinoma-Next, we identified disease-associated protein signatures, using the SAM method (29). First, the combined infection and tumor signature was determined by comparing the expression profiles of the gastrointestinal adenocarcinoma tissue from H. pylori-positive (GA/Hpϩ) samples with tumor-free normal stomach tissue from H. pylori-negative (N/HpϪ) samples. This analysis resulted in 30 significantly differentially expressed antigens detected by 43 different antibody clones (Table I). Second, the signature associated only with infection was determined by comparing H. pylori-positive tumor-free normal tissue (N/Hpϩ) samples with N/HpϪ samples (Table II), resulting in 14 significantly differentially expressed antigens recognized by 17 different antibody clones. Finally, the signature associated with only the tumor was determined by comparing GA/Hpϩ samples with N/Hpϩ samples resulting in 29 significantly differentially expressed proteins (Table III)  The protein expression signatures were further evaluated by two-way hierarchical cluster analysis (Figs. 3-5). Analysis of the protein expression signature associated with both tumor and infection (GA/Hpϩ samples versus N/HpϪ samples) (Fig. 3) revealed two well separated clusters, demonstrating both high sensitivity (100%) and specificity (88%). Next, the signature associated only with infection was analyzed by comparing the N/Hpϩ samples with the N/HpϪ samples, resulting in two well defined hierarchical clusters (Fig. 4), showing 95% sensitivity and 88% specificity. Finally, the protein signature associated only with the tumor (GA/Hpϩ samples versus N/Hpϩ samples) was examined, resulting in high sensitivity (100%) but a lower specificity (52%) (Fig. 5), resulting in more false positives compared with the infection signature. This observation might be explained by the fact that the tissue samples in the normal H. pylori-positive (N/Hpϩ) group was obtained from tumor-free tissue at least 5 cm from the primary tumor. It is possible that gastric tumors have global effects on the immune response also affecting the surrounding gastric tissue, or that the normal samples are affected by the so called "field effect," which refers to changes in non-malignant tissue associated with a cancer (30). This latter explanation is supported by very recent data obtained from the same patient material, which shows that both T-cell activity and antibody levels are altered both in the tumor area as well as in the tumor-free mucosa of H. pylori-infected gastric cancer patients, as compared with H. pylori-infected asymptomatic individuals. 4,5 To further analyze the protein infection signatures and examine the differences between the sample classes, PCA was used to generate a two-dimensional view of the data (Fig. 6). Plots 6A and 6B show that both the combined infection and tumor signature as well as the infection signatures classes were well separated, with the exception of one outlying N/Hpϩ sample (Fig. 6B). Moreover, the PCA representation of the tumor-associated signature comparing GA/Hpϩ samples versus N/Hpϩ samples demonstrated that these sample classes could also be separated, although with lower specificity (Fig. 6C). These data further supported the conclusions from the hierarchical cluster analysis (Figs. 4 and 5).
Analysis of the Antigens in the Protein Signatures-Several cytokines were significantly up-regulated in the infection signature, including typical TH2 cytokines such as IL-5, IL-6, and IL-13, as well as TH1 cytokines including IFN-␥ and IL-2 (Table  II). In addition, IL-10 and TGF-␤, which are associated with regulatory T cells, were also up-regulated (Table II). This confirmed previous studies that demonstrated an increased frequency of regulatory T cells in H. pylori-infected stomach mucosa (33) and reflects the complex balance of the immune response in stomach tissue, which has been suggested to fall between a fully polarized TH1 or TH2 profile (7). In fact, TH1, TH2, and regulatory T cell cytokines have all previously been shown to be associated with H. pylori infection (34,35). In particular, IL-10, which suppresses TH1 cytokines, has been suggested to be responsible for the persistency of the infection (36). Additionally, granulocyte-monocyte-colony stimulating factor and IL-1␤, as well as the chemokine eotaxin (Table II), were also detected, confirming previous observations of H. pylori infection (34,(37)(38)(39)(40).
In the tumor-associated protein signature, the typical TH2 cytokine IL-4 was up-regulated (Table III), which might indicate that the tumor progression is associated with a local TH2 skewing of the immune response (41). Furthermore, the cytokine profile of tumor tissue was associated with increased levels of IL-10, as compared with both N/Hpϩ samples and N/HpϪ samples. Interestingly, increased IL-10 production was also recently observed in H. pylori-stimulated cultures of T cells from gastric cancer patients compared with asymptomatic H. pylori-infected individuals. 5 Because IL-10 is associated with immune suppression, the combination of TH2 skewing and IL-10 production might facilitate tumor progression. In addition, the TH1 cytokine IL-12 was also up-regulated (Table III), again demonstrating the complex balance of the immune response in the stomach (7). In comparison to the infection signature where C1s was the only up-regulated complement factor, several complement factors, such as C1q, C3, C5, factor B, (FB), and esterase inhibitor (EI) ( Table  III) were up-regulated in the tumor-associated signature. These complement factors might be produced by the tumor cells as a result of TNF␣ induction, as was recently shown for C3 and factor B on gastric cancer-derived cell lines (42). Of note, additional chemokines, such as MCP-1, MCP-3, and IL-8 (Table III) were up-regulated, which could reflect a more severe inflammatory status and an increased tissue vascular- ization in the tumor (43)(44)(45)(46). The latter suggestion was further supported by the up-regulation of vascular endothelial growth factor and angiomotin (Table III), which mediates angiogenesis and has previously been shown to be expressed by carcinomas (26,47,48). Finally sialyl-Lewis X, known to be highly expressed in gastric adenocarcinomas (27), and leptin, a possible neoplastic growth factor (31,32), were also found to be up-regulated in the tumor-associated protein expression profile (Table III).
Overall, the proteins of the infection signature and the tumor signature displayed distinct differences, although common features were also evident. The Venn diagram in Fig. 7 provides an overview of the infection and tumor signature as well as the combined signature. All but one of the proteins in the infection signature were also found within the combined signature. Several proteins in the tumor signature were also found within the combined signature, although 10 uniquely regulated antigens were detected. There was a relatively small overlap between the infection and the tumor-associated signature, which would be important for diagnostic applications. In fact, several of the proteins in the different signatures have the potential to be used as biomarkers and for development of novel diagnostic methods for H. pylori infection and gastric adenocarcinoma, although further studies will be needed to validate our findings. In addition, some of the proteins associated with the infection signature such as IL-9 and IL-11 as well as with the tumor signature, e.g. MCP-4, which plays a role in accumulation of leukocytes during inflammation, have not previously been associated with H. pylori infection or gastrointestinal carcinomas, and further investigations into these findings may advance the understanding of these diseases.

CONCLUSIONS
In this study, we have demonstrated the first example of an antibody microarray based on recombinant, microarray-adapted antibody fragments used for analysis of protein expression signatures in complex cancer proteomes. The results showed that we could identify protein expression signatures associated with either H. pylori infection or gastric adenocarcinoma, thus allowing us to distinguish a general disease signature as compared with a cancer associated signature.