First, we studied the differential expression of mRNA in a series of 6 intestinal type gastric cancer patients (3 with and 3 without lymph node metastases) using the GeneChip® Human Genome U133 Plus 2.0 Array from Affymetrix which detects 47,000 transcripts and variants as well as 38,500 well characterized human genes. mRNA was extracted and transcribed only from tissue samples obtained from the primary tumours. A total of 115 transcripts were found to be up- and 219 to be down-regulated in node positive gastric cancer compared with node negative gastric cancers (table S1). Next we searched for differentially expressed GPCRs. We identified 52 GPCRs and GPCR-related genes, which were up- or down-regulated with a fold change factor of >1.5 (table S2).

We then searched the ENTREZ data base of the “National Center for Biotechnology Information (NCBI)” for entries of the GPCRs and GPCR-ligands and their putative role in tumour biology. The significance of the angiotensin II type 1-receptor in gastric cancer was previously verified by our own group [10]. Several studies had shown that the expression of the duffy blood group chemokine receptor (DARC) correlated inversely with the prevalence and prognosis of prostate cancer [11]–[13]. In the animal model of breast cancer the expression of EDG2 correlated with the incidence of lung metastases [14]. FPRs mediate the effect of annexin 1 on cell motility and invasion, which are important for the metastatic potential of tumour cells [15]. LGR5 was recently shown to be a stem cell marker of cells of the small intestine and colon and stem-cell specific loss of Apc results in progressively growing neoplasias [16]. Collectively, these data provide evidence that our approach identified GPCRs and GPCR-ligands that may be involved in gastric cancer biology. Concerning chemokine receptors and chemokines, the expression of the chemokines CCL2 and CXCL12 were increased in nodal positive gastric cancer compared to nodal negative cases (supplementary table S2). Since the CXCL12-CXCR4 axis plays a prominent role in tumourigenesis, promoting angiogenesis and migration of tumour cells to metastatic sites [17]–[19], we selected CXCL12 and its receptor CXCR4 for further analyses.

The differential expression of CXCL12- and CXCR4-mRNA was validated by real-time RT-PCR in an independent set of 37 intestinal type gastric carcinoma samples. We compared non-neoplastic mucosa with the primary tumour as well as primary tumours of node negative with primary tumours of node positive cancers.

CXCL12 expression was significantly increased in gastric carcinoma compared with non-neoplastic mucosa (p = 0.033). Confirming the array data, CXCL12 expression was also up-regulated in nodal positive gastric carcinoma compared with nodal negative cases. However, this difference did not reach statistical significance (p = 0.132; figure 1a ). Concerning local tumour growth, there was no significant difference in CXCL12 expression in pT1/T2 stage tumour versus pT3/T4 tumours. But interestingly, CXCL12 expression showed a significant association in pT1/pT2a versus pT2b/T3/T4 (p = 0.049; figure 1b ).

There was neither a difference of CXCR4 expression in gastric carcinoma versus non-neoplastic tissue (p = 0.229) nor in nodal negative versus nodal positive gastric carcinoma (p = 0.22; figure 1c ). Comparing CXCR4 expression with the local tumour extent, CXCR4-mRNA levels increased with the local tumour growth (p = 0.079; figure 1d ).

The translation and histoanatomical distribution of CXCL12 was subsequently studied by immunohistochemistry in 347 gastric carcinoma samples. In 291 cases, CXCL12 immunoreactivity was assessable. Tumour cells expressed CXCL12 in 244 of 291 (84%) samples. A strong cytoplasmic and membranous immunoreaction was observed in 143 (49%) cases and a weak staining in 101 (35%). 47 tumours (16%) lacked CXCL12-immunoreactivity. All tumour samples showed a distinct CXCL12 positivity of the vascular endothelial cells, which served as an internal positive control ( figure 2a-c ). The statistical analyses showed a significant correlation between the expression of CXCL12 in tumour cells and distant metastases (p = 0.043) as well as tumour grade (p = 0.0064). All other clinicopathological parameters showed no association with CXCL12 expression in tumour cells ( table 1 ). When dividing the cohort into two groups, i.e. intestinal type and diffuse type gastric carcinoma, no correlation was found between CXCL12 expression and any clinicopathological parameter of either group (data not shown). The CXCL12 expression in tumour cells had no impact on patient survival (entire group: p = 0.830; intestinal type carcinoma: p = 0.766; diffuse type carcinoma: p = 0.817).

Translation of CXCR4 was also studied by immunohistochemistry. Immunoreactivity in tumour cells was assessable in 293 tumour samples, of which only 6 (2%) showed an unequivocal membranous staining (category 2+; figure 2d ). Most of the CXCR4 positive tumour specimens only revealed a faint cytoplasmic CXCR4 immunoreactivity (category 1+, 44 cases, 15%; figure 2e ). Nuclear CXCR4 immunoreactivity was not observed in any case and any cell type. An overall of 83% (243 of 293) of the gastric carcinomas were immunonegative for CXCR4, most of them showing concomitant CXCR4 positive leucocytes as an internal positive control ( figure 2f ). Interestingly, as previously observed in colorectal carcinomas, 86 of 293 gastric carcinomas (29%) showed CXCR4 positive microvessels in the tumour stroma with a strong CXCR4 immunoreactivity of endothelial cells ( figure 3a ). The vascular nature of these delicate structures was confirmed by a CD34 immunostaining ( figure 3b ).

When correlating CXCR4 expression in tumour cells with various clinicopathological parameters, CXCR4 expression was significantly associated with higher local tumour extent (T-status; p = 0.030). However, no further associations of tumoral CXCR4 expression and other clinicopathological variables were found ( table 2 ). When analyzing the subgroup of intestinal type gastric carcinomas, no associations were found with any clinicopathological parameter (data not shown).

Then we studied the correlation between CXCR4 expression in endothelial cells (vascular CXCR4 expression, vCXCR4) of tumour microvessels and various clinicopathological parameters. Interestingly, the expression of CXCR4 in microvessels correlated highly significantly with the local tumour growth (T-category; p = 0.0001) as well as with the UICC-tumour stage (p = 0.0059). Even in the subgroups of intestinal type and diffuse type gastric carcinoma, vCXCR4 expression was significantly associated with local tumour extent (intestinal type: p = 0.004; diffuse type: p = 0.030) and UICC-tumour stage (intestinal type: p = 0.020). Furthermore vCXCR4 expression was significantly associated with patient age (p = 0.0148) in the entire group ( table 3 ).

Survival analysis showed that CXCR4 expression in tumour cells of gastric carcinoma as well as in tumour microvessels had no impact on survival.

Since the CXCL12-CXCR4 axis has been shown to be involved in tumour progression [17]–[19], we investigated the correlation of the concomitant expression of CXCL12 and vCXCR4 with clinicopathological parameters. Indeed, CXCL12-vCXCR4 positive tumours were associated with higher local tumour extent (p = 0.0014) and higher UICC stages (p = 0.017). However, it had no impact on patient survival, even in the subgroups of intestinal and diffuse type gastric carcinoma.

The CXCL12-CXCR4 axis has been reported to be involved in metastatic processes in various tumour entities. Therefore we examined CXCL12 and CXCR4 immunoreactivity in a subset of 61 matching lymph node metastases. The CXCL12 expression pattern was available for 46 metastases. Overall, 4 lymph node metastases were CXCL12 negative like their corresponding primary tumour. 40 lymph node metastases showed a clear CXCL12 positivity according to the primary tumour. The staining intensity was very heterogeneous showing strongly positive tumour cells adjacent to faintly stained tumour cell clusters. However, two metastases revealed CXCL12 immunoreactivity although no CXCL12 expression has been detected in the primary tumour.

CXCR4 immunoreactivity was assessed in 54 lymph node metastases. Interestingly, none of them showed any CXCR4 expression. Even those tumours (n = 6), showing a faint CXCR4 positivity in the primary tumour, lacked CXCR4 expression in the corresponding lymph node metastases. However, all lymph node metastases revealed clearly CXCR4 positive lymphocytes, which served as internal positive control. Interestingly, in three cases intratumoural CXCR4 positive microvessels were detectable.

Tissue samples of gastric cancer were obtained surgically at the Charité University Hospital Berlin (1995–2008). Fresh frozen tissue of 6 cases of intestinal type gastric carcinoma was used for GeneChip analysis (nodal negative: 3 patients; nodal positive: 3 patients; female-male-ratio: 1∶2). An independent series of paired cancerous and tumor-adjacent normal tissues from 37 intestinal type gastric carcinomas were examined by real-time RT-PCR (nodal negative: 12 patients; nodal positive: 25 patients; female-male-ratio: ∼1∶1, for detailed patient characteristics see table S3). For immunohistochemical analyses, a patient cohort of 347 consecutive patients with gastric cancer was examined, comprising 194 intestinal type and 122 diffuse type carcinomas according to the Laurén classification. 31 samples showed other histological subtypes (i.e. mucinous, tubular, undifferentiated). Briefly, the cohort consisted of 220 men and 127 women. The mean age of the patients at the time of diagnosis was 64 years. Survival data was available from 196 of these patients. Follow-up data for the other patients was missing because these patients were not resident near the hospital and were lost to follow-up. Out of 196 patients, 124 died during follow-up. Median follow-up for those patients still alive at the endpoint of analysis was 33 months. Of 61 patients, tissue of matching lymph node metastases was available (27 intestinal type, 26 diffuse type, 8 other histological subtypes). Only patients with histologically confirmed gastric cancer and adequate tissue available were included. Patients with neoadjuvantly treated gastric carcinoma or other known malignancies were excluded from the study. This project was approved by the local ethics committee (ref. number EA1/06/2004).

Total RNA was isolated with phenol-chloroform using the mirVana™ miRNA Isolation Kit (Ambion, Austin, USA). Contaminating DNA was removed by DNase treatment (Turbo DNAfree kit; Ambion, Austin, USA) at 37°C for 30 min. We used the GeneChip® Human Genome U133 Plus 2.0 arrays (Affymetrix, Santa Clara, CA, USA) according to the manufacturers protocol to analyze mRNA expression levels. Affymetrix GeneChip® Operating Software (GCOS 1.4) automates the control of GeneChip® Fluidics Stations and GeneChip® Scanner 3000.

Raw data were analyzed with the Affymetrix GeneChip Operating Software (GCOS 1.4). The detection p-value of a transcript determines the detection call, which indicates whether the transcript is reliably detected (p<0.05; present) or not detected (absent). To enable the comparison between chips the data were scaled to a global intensity of 500. The Data Mining Tool 3.0 (Affymetrix) and GeneSpring software package 7.2 (Silicon Genetics, Redwood City, CA) were used to average results from different samples and perform statistical analysis to compare between gastric cancer with (N1) and without (N0) lymph node metastases. The data of six arrays were normalized to account for variability in hybridization for probe pairs and other hybridization artefacts. The normalization consists of the following three steps: first, data transformation (set measurements less than 300.0 to 300.0); second, per chip (normalize each chip to the 50th percentile of the measurements taken from that chip); and third, per gene (normalize each gene to the median of the measurements for that gene). The fold change was calculated for each gene as the arithmetic mean of the normalized expression values of N1 divided by the arithmetic mean of the normalized expression values of N0. Raw data from microarray experiments were uploaded to the Gene Expression Omnibus Database (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=pzmjlcoscakugfc&acc=GSE17187).

For cDNA synthesis, 2 µg of total RNA was reverse transcribed using the Omniscript RT Kit (Qiagen). The gene-specific primers were designed by the BioTeZ Berlin-Buch GmbH (Berlin, Germany). Primer sequences were as follows: CXCR4 5′ CAG CAG GTA GCA AAG TGA CG, 3′ CAG GGT TCC TTC ATG GAG TC; CXCL12 5′ CGA TTC TTC GAA AGC CAT GT, 3′CAC TTG TCT GTT GTT GTT CTT CAG; beta2-microglobulin 5′ ACC CCC ACT GAA AAA GAT GA, 3′ ATC TTC AAA CCT CCA TGA TG. Real-time reverse-transcriptase polymerase chain reaction (Real-time RT-PCR) was carried out using the QuantiTect™ SYBR® Green PCR Kit (Qiagen) and the LightCycler System (Roche Diagnostics, Mannheim, Germany). The comparative C_t values were normalized to that of the housekeeping gene beta2-microglobulin. No template controls (no cDNA in PCR) were run for each gene to detect unspecific or genomic amplification and primer dimerization.

For histological analyses, tissue samples were fixed in 10% neutralized formalin and embedded in paraffin. Deparaffinized sections were stained using hematoxylin and eosin. Gastric carcinoma was classified according to the WHO classification [1]. The pTNM stage was determined according to the UICC guidelines.

Formalin-fixed and paraffin-embedded tissue samples were used to generate tissue microarrays as described previously [40], [41]. Briefly, three to six morphologically representative regions of the paraffin ‘donor’ blocks were chosen. Tissue cylinders were punched from these areas and precisely arrayed into a new ‘recipient’ paraffin block using a customer built instrument (Beecher Instruments, Silver Spring, MD, USA). A minimum of three tissue cylinders of 0.6 mm diameter were punched from each sample. After completing the block construction, four micrometer sections of the resulting tumour tissue microarray block were cut for further analysis.

Immunostaining was carried out with an anti-CXCR4-antiserum (dilution 1∶100; rabbit polyclonal antiserum;[38]) and an anti-CXCL12-antibody (dilution 1∶100; mouse monoclonal IgG1; R&D Systems, Minneapolis, MN, USA). Following antigen retrieval (sodium-citrate, 4×5 min, 600 W, microwave oven), incubation with the primary antibodies was performed in a moist chamber at 4°C overnight. Slides were washed between steps with Tris-buffered saline. Immunoreactions were visualized with the Super Sensitive Link Label Detection System (BioGenex Laboratories, San Ramon, CA, USA) combined with the SIGMAFAST™ kit (Sigma-Aldrich, St. Louis, MO, USA). The specimens were counterstained with hematoxylin. Immunostaining with an anti-CD34-antibody (dilution 1∶100; mouse monoclonal IgG1, kappa; DAKO, Carpinteria, CA, USA) used the Ventana Benchmark XT automated staining system (Ventana Medical Systems, Tucson, AZ, USA). Omission of primary antibodies served as negative controls. Normal human adrenal gland tissue served as positive control for CXCR4, normal human tonsil tissue for CXCL12. The CXCR4 immunoreactivity in tumour cells was categorized as absent (0), faint cytoplasmic staining (1+), clear cytoplasmic and/or clear membranous staining (2+). CXCR4 expression in tumour microvessels was recorded as positive or negative. CXCL12 immunoreactivity in primary tumours was scored as absent (0), weak cytoplasmic (1) or strong cytoplasmic (2). CXCL12 expression in lymph node metastases was scored as positive or negative. All samples were evaluated by one pathologist (BI). When evaluating the sections, discrepancy in sample number was related to tissue loss during the transfer of the TMA sections onto the slides.

For statistical analyses 1+ and 2+ tumour samples were considered as CXCR4 positive (1) whereas tumours with lack of immunoreactivity were scored as negative (0). Vascular CXCR4 expression always showed a strong signal and was recorded as positive (1) or negative (0). 1+ and 2+ CXCL12 immunoreactivity was scored as positive, whereas tumour samples lacking CXCL12 immunoreactivity were scored as negative. Significance of correlations between protein expression (CXCR4 and CXCL12) and clinicopathological parameters was assessed by Fisher's exact test for 2×2 tables and by the chi squared test for larger tables. Survival curves were fitted with the Kaplan-Meier method. Differences in survival were assessed by the log rank test.

Real-time RT-PCR data was logarithmized to obtain approximately normally distributed data. Results were evaluated with an unpaired two-sided t-test. P-values<0.05 were considered as statistically significant. Statistical analyses were performed using the SPSS 17 statistical package (SPSS Inc., Chicago, IL, USA) or the GraphPad Prism statistical software (GraphPad Software, Inc. La Jolla, CA, USA).