Hypoxia increases membrane metallo-endopeptidase expression in a novel lung cancer ex vivo model – role of tumor stroma cells

Background Hypoxia-induced genes are potential targets in cancer therapy. Responses to hypoxia have been extensively studied in vitro, however, they may differ in vivo due to the specific tumor microenvironment. In this study gene expression profiles were obtained from fresh human lung cancer tissue fragments cultured ex vivo under different oxygen concentrations in order to study responses to hypoxia in a model that mimics human lung cancer in vivo. Methods Non-small cell lung cancer (NSCLC) fragments from altogether 70 patients were maintained ex vivo in normoxia or hypoxia in short-term culture. Viability, apoptosis rates and tissue hypoxia were assessed. Gene expression profiles were studied using Affymetrix GeneChip 1.0 ST microarrays. Results Apoptosis rates were comparable in normoxia and hypoxia despite different oxygenation levels, suggesting adaptation of tumor cells to hypoxia. Gene expression profiles in hypoxic compared to normoxic fragments largely overlapped with published hypoxia-signatures. While most of these genes were up-regulated by hypoxia also in NSCLC cell lines, membrane metallo-endopeptidase (MME, neprilysin, CD10) expression was not increased in hypoxia in NSCLC cell lines, but in carcinoma-associated fibroblasts isolated from non-small cell lung cancers. High MME expression was significantly associated with poor overall survival in 342 NSCLC patients in a meta-analysis of published microarray datasets. Conclusions The novel ex vivo model allowed for the first time to analyze hypoxia-regulated gene expression in preserved human lung cancer tissue. Gene expression profiles in human hypoxic lung cancer tissue overlapped with hypoxia-signatures from cancer cell lines, however, the elastase MME was identified as a novel hypoxia-induced gene in lung cancer. Due to the lack of hypoxia effects on MME expression in NSCLC cell lines in contrast to carcinoma-associated fibroblasts, a direct up-regulation of stroma fibroblast MME expression under hypoxia might contribute to enhanced aggressiveness of hypoxic cancers.

hour at room temperature. The primary antibody was detected with the Ultravision LP detection system (Thermo Scientific, Waltham, MA). Peroxidase activity was visualized by incubation with 3-amino-9-ethylcarbazole solution (Dako, Glostrup, Denmark). In the case of MME, CC1 buffer (Ventana, Strasbourg, France) was used for antigen retrieval and visualization of the primary antibody was achieved with the iVIEW detection system (Ventana). Counterstaining was performed with hematoxylin. As a negative control, isotype control antibodies were used as primary antibodies or staining was performed in the absence of the primary antibody.

Evaluation of immunohistochemistry
All of the sections were examined by a thoracic pathologist (E.S.) in a blinded manner at x400 magnification using a Zeiss Axioplan 2 microscope (Oberkochen, Germany). Staining scores were calculated by multiplying the percentage of positive cells by the staining intensity. Apoptosis rates were expressed as percentage of cleaved caspase 3 positive tumor cells among all tumor cells and were determined by examination of the entire stained fragments at x400 magnification. Images were taken using an Olympus BX51 microscope, a DP12 digital camera system and Cell^D software (Olympus, Hamburg, Germany).

Meta-analysis of MME and survival in NSCLC patients
The association between expression of the four hypoxia genes (probe sets corresponding to the genes) and overall survival of NSCLC patients was assessed in each of the GEO series separately and in a meta-analysis. While probe sets for PPP1R3C and MME were present on all the arrays used in the four studies, probe sets for FAM115C and KCTD11 were present only in two datasets, GSE13213 and GSE19188 (Supplementary Table 1). From the GSE11969 series only non-adenocarcinoma patients were included due to potential overlap with patients from GSE13213, who were operated at the same centre. The expression of MME significantly correlated between the respective probe sets in all studies (Pearson correlation coefficients 0.86 to 0.98), while in the case of all other genes with more than one probe set, only moderate to weak correlations were found, possibly due to the detection of different splice variants/isoforms by the different probe sets. Only the well correlated MME probe sets were averaged. In all other cases each individual probe set was used for survival analysis.
According to the expression levels of genes/probe sets, patients were dichotomized into the "high expression group" (comprising 25% of patients with the highest expression) and the "low expression group" (the remaining patients). Patients with overall survival shorter than one month after surgery were excluded from the analysis. All analyses were performed in a multivariate manner with pathological tumor stage as stratification variable. Survival analyses were compiled within individual studies using the Cox proportional hazards model with pathological tumor stage as stratification variable. Meta-analysis of the effect of MME on patient survival after surgery was performed with a proportional hazards model with Gaussian random effects using the package coxme 2.1-3 of R 2.13.2 statistical software (www.r-project.org). This software allows to specify normally distributed random effects within a proportional hazards model and to test them by the likelihood ratio test. Tumor stage, histological type and MME expression were available for every study. Due to the heterogeneity of survival distributions the analysis was stratified with respect to study and tumor stage. MME, histological type and their interaction were included as fixed effects. In the meta-analysis the interaction of MME with study was included as random effect.

Hierarchical cluster analysis
All genes were included in hierarchical cluster analysis. In order to check stability the clustering was analyzed by multiscale bootstrap resampling (pvclust [1]).
Due to limited computing rescources only 5% of the genes with the highest MAD (median absolute deviation, a robust measure of variability) were included (1426 genes) [2]. It was observed that the clustering was unchanged by the selection compared to clustering without selection (Partek software).

Immunofluorescence
Cancer-associated fibroblasts (CAFs) were fixed in methanol, permeabilized with 0.02 % Triton X-100 for 30 min and stained with rabbit antibody to vimentin (Cell Signaling