Our study included 26 patients. A first set of 19 male patients undergoing surgery for ethmoid sinus adenocarcinomas were initially included between 2004 and 2006. Following this, a second set of 7 patients whose samples were collected from 2006 to 2007 was used to complete the immunohistochemistry study.

This project was approved by the Clinical Board of the Centre Hospitalo-Universitaire of Nantes and all included patients provided written informed consent in accordance with French regulations and the Declaration of Helsinki. All patients answered a codified questionnaire regarding occupational exposures, addictive consumption and family history. Twenty three patients out of 26 were exposed to wood dust and most of them had other occupational exposures (such as solvents and pesticides) sometimes combined with tobacco and/or alcohol. Two patients were exposed to leather dust (P7, P19), whereas only one (P10) had no occupational exposure (Table 1). Patient ages ranged from 50 to 80 years with a mean age of 69 years. To date, six patients have died as a direct result of their disease (Table 1).

Two pieces of tissue samples were obtained from each patient undergoing surgery for ethmoidal adenocarcinoma: one from the tumor and one non-tumor sample obtained from the opposite sinus at 3 to 4 cm distance (herein referred to as "normal" tissue). All samples were immediately frozen and stored at -80°C. Remaining surgical resections of tumors and normal tissue were fixed in 10% formalin and embedded in paraffin before histological examination and diagnosis according to World Health Organization recommendations 24 . Two main types of sinonasal adenocarcinoma are recognized in the ethmoid sinus based on the histological similarity to adenocarcinoma of the intestine: Intestinal Type Adenocarcinoma (ITAC) and non-Intestinal Type Adenocarcinoma (non-ITAC). ITAC can be further divided into five categories 15 25 : the "papillary-type" (well-differentiated adenocarcinoma), the "colonic-type" (moderately-differentiated adenocarcinoma), the "solid-type" (poorly-differentiated adenocarcinoma), the "mucinous--type" and the "mixed--type" composed of a mixture of the previously defined patterns. Non-ITAC are divided into low-grade and high-grade subtypes.

On each matched normal and pathological tissue specimen from patients P1 to P19, two RNA extractions were performed from about 40 frozen sections (10 μm thick) using a Total RNA and Protein Isolation kit (Macherey-Nagel, Düren, Germany) according to the manufacturer's instructions. For each sample, the first and last sections were stained with hemalun/phloxin to confirm the histology and to evaluate the percentage of tumor tissue. 10 samples had to be eliminated for microarray analysis because of necrosis or a too low percentage of non-necrotic tumor tissue (less than 50%). Six out of these ten patients were included in the validation process by RT-qPCR as this technique is more sensitive than microarrays for identifying tumor cells within a sample. The other samples were completely excluded from the molecular analysis (Table 1).

The quantity and quality of each RNA were respectively evaluated with the NanoDrop® ND-1000 spectrophotometer (Nanodrop Technologies, Wilmington, DE) and the Agilent 2100 Bioanalyser (Agilent, Santa Clara, CA). The RNAs extracted were of good quality and the RNA integrity number (RIN) was >7.5 in all cases 26 .

Cancer-dedicated microarrays were prepared in-house (ADN-OGP- Microarray Platform Nantes, France) with methods previously described in detail 27 28 using 22,175 probe sets (50-mer oligonucleotides - MWG Biotech, Roissy, France) interrogating 6,864 genes involved in different types of tumors. These microarrays therefore included triplicate probes for each gene, housekeeping genes and controls.

For microarray analysis one round of amplification was conducted on 500 ng total RNA using an Amino Allyl MessageAmp^®II aRNA Amplification kit (Ambion, Austin, TX) according to the manufacturer's instructions, and the quantity and quality of each amplified RNA (aRNA) were again evaluated. Microarrays were carried out in duplicate for both RNA extractions of each tissue except for two patients as not enough RNA was available. The targets were prepared by labeling with Cy3-dUTP aRNA from the tumor and normal tissues. In order to reduce individual variations, the reference was prepared by mixing an equal quantity of all normal tissues 29 30 and aliquots were then labeled with Cy5-dUTP (Amersham Biosciences, Piscataway, NJ). Each Cy3-dUTP sample was mixed with an equal amount of Cy5-dUTP reference sample and the mixture was applied to microarray slides for hybridization at 40°C for 16 h 27 . The slides were then washed twice at room temperature for 2 min with 2× SSC and 0.1% SDS, for 2 min with 1× SSC, and twice for 2 min with 0.2× SSC and scanned at 10 μm/pixel resolution by ScanArray^®ExpressHT (PerkinElmer Life Sciences, Boston, MA).

Scanned signals were quantified from all microarrays by GenePix Pro software version 5.1 (Axon Instruments, Union City, CA) and consolidated expression values were performed by MADSCAN software in five steps 30 31 . The information was extracted from the features close to the background or saturated and normalization was performed by the rank invariant and lowest fitness method with spatial normalization. Outlier values were eliminated with the spots in triplicate and biological replicates. To identify genes differentially expressed in tumor samples, a two-class comparison analysis by Significance Analysis of MicroArray (SAM) 32 was performed on data filtered by differences between normal and pathological tissue medians as previously described 30 and genes with differential expression were visualized using Cluster 33 and Tree view 31 . An unsupervised clustering was also performed with a hierarchical clustering algorithm 33 using the Pearson coefficient and Student test. The clusters of genes with the same regulation were functionally annotated by GoMiner 34 .

The data have been incorporated into the NCBI Gene Expression Omnibus (GEO) http://www.ncbi.nlm.nih.gov/projects/geo/ and are accessible through GEO Series GPL 8957 and GSE 17433.

To confirm the microarray data we performed quantitative RT-PCR on selected genes using the MX4000 system and the Brilliant SYBR Green QPCR Core Reagent Kit (Stratagene, La Jolla, CA). Initially, cDNA was prepared in 20 μl using 1 μg of DNase-treated total RNA and the SuperScript III Reverse Transcriptase System (Invitrogen, Carlsbad, CA). Following a 5 fold dilution, 2 μl of each sample were used for RT-qPCR with the different pairs of primers (Additional file 1: "Primers sequences"). The following PCR cycle parameters were used: hot-start DNA polymerase activation 95°C for 10 min, 40 cycles with denaturation at 95°C for 30 sec, specific annealing temperature as indicated in "Additional file 1: Primer sequences" for 30 sec and extension at 72°C for 30 sec. Each reaction was run in duplicate. The threshold cycles, obtained from the MX4000 software, were averaged (SD<0.5). Relative expression of the target gene in the tumor versus matched normal tissue was calculated using the following equation described by Pfaffl 35 , using the average Ct of three housekeeping genes: RPLPO (Ribosomal Protein, Large, PO), UBC (Ubiquitin C) and β2M (beta-2 microglobulin):

Relative expression per patient and per gene

GOI = gene of interest

HK = housekeeping gene (average of Ct of the three housekeeping genes).

Eff = efficiency of the RT-qPCR obtained from the standard curve

Statistical significance was obtained using a pair-wise fixed reallocation randomization test using the REST software 36 . To insure specificity of the RT-qPCR, an agarose gel electrophoresis was initially performed to verify whether a single PCR product was generated and then a melting curve was performed at the end of each RT-qPCR. Linearity and efficiency of the RT-qPCR were checked for each gene with a standard curve of 4 logs prepared with Universal RNA (Stratagene-AGILENT, CA). Efficiency was >90% in all cases.

Protein expression of selected genes was assessed in deparaffinized 5-μm sections of normal and pathological formalin-fixed tissue from 26 patients with sinonasal adenocarcinomas included in the study. The following antibodies were used: monoclonal antibody against human Clusterin (clone CLI-9, Alexis Corporation Lausen, Switzerland, 1:500 dilution), monoclonal antibody against human Acyl CoA synthetase 5 (ACS5) (Abnova, Jhongli City, Taiwan 1:200 dilution at 4°C overnight), polyclonal antibody against Galectin-4 (T-20) (Santa Cruz, Heidelberg, Germany, 1:50 dilution). All specimens were submitted to heat-induced antigen retrieval and processed using the EnVision Detection Kit (DAKOCYTOMATION, Trappes, France), except for LGALS4 that was processed using ABC VECTASTAIN Elite ABC Kit (Burlingame, CA), with 3,3'-diaminobenzidine as chromatogen and a hematoxylin counterstain. In each experiment, negative controls were performed by omitting the primary antibody.

Gene expression profiles of 9 ethmoid adenocarcinomas were examined using microarrays consisting of 6864 human genes involved in many types of cancers.

With the two-class comparison SAM, 186 genes were found to be significantly differentially expressed between ethmoid adenocarcinomas and normal sinonasal tissue. Among these 186 genes, 150 were up-regulated and 36 were down-regulated (Figure 1A and "Additional File 2: Genes with significant differential expression"). The top 59 genes (1< fold change < -1) are described in Table 2. The genes with the highest fold expression variation were selected for validation by RT-qPCR: LGALS4 (fold change: 3.6), ACS5 (fold change: 2.1), and CLU (fold change: -3.6). By unsupervised clustering (i.e. without any initial classification of the samples) 7 tumors out of 9 were separated from normal samples (Figure 1B). However, 5 clusters of genes with differential expression between tumor and normal samples were revealed. Using GoMiner 34 the genes involved in metabolism and biosynthesis functions were found to be overexpressed, whereas those involved in transcription, angiogenesis, cellular signaling and mitochondrial functions were down-regulated. Based on this non-supervised analysis 2 more genes with high differential expression were selected for RT-qPCR analysis: SRI and CCT5. Involved in drug resistance, these genes also featured in the list of overexpressed genes obtained from the two-class comparison analysis, with a fold change of 1.5 and 0.9 respectively.

To validate the differential gene expression obtained by microarray analysis, quantitative PCR analysis of the selected genes was performed in matched sets of tumors and normal tissues. The patients used for microarray analysis and 6 additional patients were included. As RNA from normal tissue was no longer available, we used the Ct average (SD<1Ct) of all normal tissues for P8 and P19 patients to calculate the relative expression level of each gene 35 .

A significant differential expression in tumor tissue versus normal tissue was confirmed for all selected genes. The genes with the highest overexpression were LGALS4 with a mean ratio of 1309 (0.17-5993, p = 0.001), then ACS5 with a mean ratio of 9.48 (0.14-23.55, p = 0.001). P10 and P11 patients overexpressed neither LGALS4 nor ACS5. (Figure 2A-B). CLU was highly down-regulated in most of the tumors (mean ratio:0.044, 0.005-0.26, p = 0.001) (Figure 2C). Many isoforms of CLU have been described in the literature 37 , and we quantified by RT-qPCR the main ones, i.e. the nuclear form (n-clu) and the cytosolic form (s-clu). Both were found to be down-regulated (data not shown). Regarding SRI and CCT5, their significant up regulation was confirmed (p = 0.0016 and p = 0.006 respectively) although the fold change was much lower ("Additional file 3: Relative expression of SRI and CCT5").

To confirm the variation in expression of the selected genes at the protein level, we performed immunohistochemical analysis of matched normal sinonasal and tumor tissues from the 15 patients used for the molecular analysis as well as from an independent set of 11 other patients, using specific antibodies for LGALS4, ACS5 and CLU (Table 3). In the normal sinusal mucosa, these three markers were expressed by serous cells of the seromucinous glands present in the lamina propria. A weak and focal cytoplasmic staining of a small number of seromucinous glands was observed with the antibodies against LGALS4 and CLU while the staining was more intense and diffuse for ACS5 (Figure 3A-B-C). Among the 26 tumors analyzed, only 2 were high-grade non-ITAC and the others were ITAC: 5 "papillary-type" (well-differentiated adenocarcinoma), 2 "colonic-type" (moderately-differentiated adenocarcinoma), 9 "mucinous-type" adenocarcinoma and 8 mixed-type adenocarcinoma (Table 3).

With the LGALS4 antibody the ITAC tumor cells displayed a strong cytoplasmic and membranous staining with an additional nuclear staining in the well-differentiated adenocarcinomas. Interestingly, in a mixed ITAC sample (P5) the poorly differentiated "solid-type" component showed no immunoreactivity for LGALS4 while the "colonic-type" component was positive (Table 3 and Figure 3D). Non-ITAC samples displayed no LGALS4 expression.

For ACS5, fifty percent of the tumor samples were negative while the remaining 50% showed a weak to strong cytoplasmic staining without any correlation with the histological type or with the differentiation of the tumor (Table 3 and Figure 3E).

In contrast to normal mucosa, CLU was found to be absent in tumors except in one high-grade non-ITAC tumor (Patient P11) where there was a diffuse cytoplasmic staining (Table 3 and Figure 3F).