Altogether ten surgical specimens with either invasive ductal carcinoma NST (no special type) (n = 5) or invasive lobular carcinoma (n = 5) were investigated. This research protocol was approved by the ethics committee at Palacky University. Tumor and normal tissues from the same mammary gland were identified by an experienced pathologist, snap-frozen in liquid nitrogen and stored at -80°C for further analysis. For microdissection, frozen tissues were embedded in TissueTek medium and cut on standard cryostat (Leica CM1850, Leica Microsystems GmbH, Wetzlar, Germany). Frozen sections (7 μm) were immediately fixed in acetone, stained by hematoxylin and dehydrated in alcohol and xylene. All solutions were prepared using diethyl pyrocarbonate-treated water. RNase free instruments and RnaseZap (Sigma, St Louis, MO, USA) were used throughout.

At least 1000 infiltrating ductal or lobular tumor cells together with normal lobular and ductal cells were microdissected from cryosections using the Veritas™ Laser Capture Microdissection System (Arcturus Bioscience, Inc., Mountain View, CA, USA) according to standard protocols. When microdissecting normal cells, no attempt has been made to exclude basal or myoepithelial cells for two reasons. First, both ducts and lobules are atrophic in menopause, and no clearly defined layers of luminal and myoepithelial cells are preserved. Second, it is impossible to exclude contamination with or entrapment of myoepithelial cells during the microdissection procedure. Caps with captured cells were directly placed in 100 μl lysis buffer (Qiagen, Hilden, Germany). Total cellular RNA was isolated (RNeasy^® Micro Kit, Qiagen) according to manufacturer's recommendations and subsequently quantified on a Nanodrop spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA).

Fifty nanograms of total RNA were reverse transcribed and amplified by Microarray Target Amplification Kit (Roche diagnostics, Basel, Switzerland). In brief, total RNA (50 ng) was converted into cDNA using a modified oligo (dT) primer (TAS-T7 Oligo (dT)₂₄). A unique Target Amplification Sequence (TAS) with no homology to any known sequences in public databases generated the 3' anchor on the cDNA for subsequent PCR amplification. In order to include a 5' anchor sequence on the cDNA, the TAS-(dN)₁₀ primer was used for the initiation of the second strand cDNA synthesis. After purification using the Microarray Target Purification Kit (Roche), PCR was performed using the TAS primer and Expand PCR Enzyme Mix which is optimized for long (>1 kb) and unbiased PCR products. In order to ensure that messages were not amplified to saturation, the optimal number of PCR cycles was estimated by preliminary PCR and agarose electrophoresis of PCR products from cycles 21, 24, 27, 30 and 33. The optimal number of PCR cycles was set either to 27 cycles for patients 3, 4, 8, 9, and 10 or 29 cycles for patients 1, 2, 5, 6, and 7. All three populations from each patient were amplified by the same number of PCR cycles. If higher PCR cycling was needed for any cell population, new microdissection, RNA isolation, cDNA synthesis and PCR amplification were performed until the same PCR cycling was possible. After purification with the Microarray Target Purification Kit (Roche), the PCR products were labeled with biotin-14-CTP (Invitrogen, CarlsBad, CA, USA) and biotin-16-UTP (Roche) by in vitro transcription using Microarray RNA Target Synthesis Kit T7 (Roche). The labeled cRNA was purified using the Microarray Target Purification Kit (Roche), quantified by spectrophotometer and checked by agarose electrophoresis. The entire amplification and labeling process was monitored by GeneChip^® Eukaryotic Poly-A RNA Control Kit (Affymetrix, Santa Clara, CA, USA) with exogenous positive controls which were spiked into the total RNA before cDNA synthesis. In all cases, 25 μg of each biotinylated cRNA preparation was fragmented, assessed by gel electrophoresis, and placed in hybridization cocktail containing biotinylated hybridization controls (GeneChip™ Eukaryotic Hybridization Control Kit, Affymetrix). Samples were first hybridized to Test3 Arrays for 16 hours, washed, stained using antibody-mediated signal amplification and scanned. After passing this quality control stage, the samples were hybridized onto the large Human Genome U133 Plus 2.0 Arrays (Affymetrix).

Scanned images of microarray chips were analysed by the GCOS (GeneChip Operating Software) from Affymetrix with the default settings except that the target signal was set to 100. Differentially expressed genes between cell types were identified using the GCOS change algorithm and Rank Products (RP) 40 following RMA (Robust Multiarray Analysis) 41. GCOS pairwise analysis was performed to compare gene expression levels among the normal lobular, normal ductal and tumor cells within individual patients and between lobular and ductal carcinoma cells of different patients. For each comparison between two cell types, the number of increase and decrease calls of each probe set was calculated using MS Excel and probe sets with the highest number of consistent changes among all patients were identified. Probe level quantile normalization 40 and robust multiarray analysis 41 on the raw. CEL files were performed using the Affymetrix package of the Bioconductor 42. The cutoff value of percentage of false-positives for RP analysis was set to 10%. Gene lists were uploaded to DAVID (Database for Annotation, Visualization and Integrated Discovery) 43, and functional annotation was performed. Further informations on genes were obtained from public databases, such as Gene Cards 44 and NETAFFX Analysis Center 45. Hierarchical clustering was performed using dChip software 46. The data discussed in this publication have been deposited in NCBI Gene Expression Omnibus (GEO) 47, and are accessible through GEO Series accession number GSE5764.

Seven differentially expressed genes were verified by immunohistochemical staining on tissue microarrays (TMA), which were constructed from 119 breast cancer cases and contained 278 cores of 2.0 mm diameter. They consisted of 80 ductal carcinomas, 29 lobular carcinomas, one tubular carcinoma, 3 medullary carcinomas, 2 tubular-lobular carcinomas, 2 mixed ductal-lobular carcinomas, one mucinous and one papillary carcinoma. The construction of the tissue microarrays was done using a tissue arrayer (Beecher Instruments, Inc., Sun Prairie, WI, USA) 48. The paraffin sections from 22 additional breast cancer samples containing normal mammary gland structures were used for comparison.

TMA sections were stained using standard immunohistochemical methods. Monoclonal mouse antibodies against cytokeratin 17 (dilution 1:100, clone 2D4-1G9, Abnova, Taipei, Taiwan), cytokeratin 5/6 (dilution 1:100, clone D5/16 B4, Dako, Denmark) and E-cadherin (dilution 1:50, clone NCH-38, Dako), polyclonal mouse antibodies against EMP1 (epithelial membrane antigen 1; dilution 1:100, clone 2D4-1G9, Abnova), DDR1 (discoidin domain receptor 1; dilution 1:100, Abnova) and DVL1 (human homolog of the Drosophila dishevelled gene; dilution 1:100, Abnova) were used. Microwave antigen retrieval was performed in citrate buffer (pH 6.0). For antigen visualization, the EnVision/HRP system and DAB+ (Dako) were used, slides were subsequently counterstained with hematoxylin, dehydrated and mounted in Canadian balsam. Immunohistochemical procedure was optimized by testing different antigen retrieval methods and negative controls. We regarded cells as immunoreactive when an obvious membranous or submembranous (E-cadherin, DDR1, EMP1) and cytoplasmic (cytokeratins, DVL1) staining was seen. Immunoreactivity was scored as follows: retained (++) when more than 50% of membrane/cytoplasm were strongly positive, reduced (+) when 10–50% of the membrane/cytoplasm were positive, and absent (-) when 0–10% of the membrane/cytoplasm were positive. Frequencies of positive and negative staining in tumor and normal tissues were compared by Fisher's exact test (Statistica 6.0, StatSoft, Prague, Czech Republic).

Asporin mRNA expression was verified by chromogenic in situ hybridization (CISH) on frozen sections. Fluorescein labeled oligonucleotide (5'AAG TTG GTG GTA AGC CTT TAG GAA CTG AGG-FAM) was designed to the junction of exons five and six of the asporin mRNA (Generi Biotech, Hradec Kralove, Czech Republic). Sections were mounted on Superfrost Plus slides and stored at -80°C until further staining. Before CISH, the sections were heated at 50°C for 2 min, and then dried at room temperature for 30 min. Tissues were delipidized in chloroform at room temperature for 5 min and dried to evaporate the chloroform. Block of endogenous peroxidase activity was followed by incubation of tissue sections in 10% paraformaldehyde for 7 min, then washed in PBS for 3 min and in 2× SSC (1× SSC contains 150 mM NaCl, 15 mM sodium citrate; pH 7.2) twice for 5 min. The sections were covered with 100 μl hybridization buffer (60% formamide, 5× SSC) containing 150 ng/ml of probe and incubated in moist chamber at 37°C for 22 hours. After the hybridization, unbound probe was washed from the sections at 37°C by 2× SSC for 5 min, then three times with 60% formamide in 0.2× SSC at 37°C for 5 min, and twice with 2× SSC at room temperature for 5 min. Then polyclonal rabbit anti-FITC/HRP (dilution 1:40, Dako) was applied at room temperature for 45 min, followed by washing in 2× SSC twice for 10 min. For visualization, DAB+ (Dako) was used, slides were subsequently counterstained with hematoxylin, dehydrated and mounted in Canadian balsam.

Products after PCR amplification (see RNA amplification, microarray target synthesis and hybridization) were used for verification of CTHRC1, ASPN and COL3A1 gene expression using gene specific PCR. The primers were as follows: CTHRC1 forward 5'ACA AGT GCC AAC CCA GAT AGC AAC, reverse 5'ATC GCA CTT CTT CTG TGG AAG GAC (20 cycles of 95°C, 59°C and 72°C 1 min each, product length 79 bp, kindly provided by Dr. Danuta Radzioch, McGill University, Montreal, Canada), ASPN forward 5'GTT CAG CTT GGG AAC TTT GGA ATG TAA, reverse 5'ACT GCA ATA GAT GCT TGT TTC TCT CAA CCC (20 cycles of 95°C, 58.5°C and 72°C 1 min each, product length 243 bp [modified from 49], and COL3A1 forward 5'TTG TCA ACC AGT GCA AGT GAC CGA C, reverse 5'TGG TGA GCA CAG TCA TTG CTC TGC A (20 cycles of 95°C, 59°C and 72°C 1 min each, product length 276 bp) 50. PCR products were separated on 10% polyacrylamide gels, stained with ethidium bromide and captured using a gel documentation system (DIANA II with cooled CCD camera, Raytest, Germany). Band intensities, expressed as "OD (optical density) × cm", were assessed by Multi-Analyst densitometric software (Bio-Rad, USA). The results were correlated with fluorescence signals from the relevant Affymetrix probe sets by Spearman coefficient using Statistica 6.0 software (StatSoft, Czech Republic).

The clinical and histopathological characteristics including lymph node status, tumor grade, expression of estrogen/progesterone receptors (ER/PgR), c-erbB-B2/HER2/neu and accompanying lesions of ten breast cancer patients are shown in Table 1. We microdissected normal and tumor cells from these cases which resulted in a total of 30 samples for gene expression profiling (10 normal ductal, 10 normal lobular, 5 tumor ductal and 5 tumor lobular). Unsupervised clustering of all samples was performed using all probe sets without filtering. No tumor types were grouped together but two large clusters were evident: one mainly consisted of tumor cells and the other cluster mainly of normal cells. This clearly suggests differences in global gene expression profiles of tumor and normal cells (Figure 1).

In this comparison 367 probe sets were identified by pairwise comparison, and 501 probe sets by Rank Products (RP). There were 82 probe sets (78 named genes) identified by both methods. In pairwise comparisons, the differentially expressed genes were found by selecting those with count of changes more than 6 (out of 10, either increase or decrease). Multiple cytokeratins (5, 7, 14, 15, 17) were expressed in both cells, but none of them was useful to separate two cell types. Genes encoding proteins involved in proteolysis and metabolism (USP25, TMPRSS3, ACACB), protein biosynthesis and modification (PDE4DIP, LOXL1), ion transport (CLCN3, GABRP) and protein kinase cascade (MAP4K5) as well as genes regulating transcription (RFXAP, HSZFP36) were upregulated in ductal cells. However, genes regulating cell growth, activation and motility (TSPAN5), genes encoding actin-binding proteins (EHBP1), small leucine-rich proteoglycan (DCN), and proteins with GTPase activity (RHOBTB3, ARHGEF9) were overexpressed in lobular cells (Table 2). Hierarchical clustering of normal cell types based on the 82 probe sets found by both RP and pairwise comparison showed that gene expression profiles of normal ductal and lobular cells from the same patients were similar, and they could not be well separated from each other (Figure 2).

A comparison of ductal carcinoma cells with normal ductal cells identified 1055 probe sets by pairwise analysis, 604 probe sets by RP and 326 probe sets by both methods. A comparison of ductal carcinoma cells with normal lobular cells identified 792 probe sets by pairwise analysis, 347 probe sets by RP and 171 probe sets by both methods. A comparison of lobular carcinoma cells with normal ductal cells identified 1022 probe sets by pairwise analysis, 350 probe sets by RP and 201 probe sets by both methods. A comparison of lobular carcinoma cells with normal lobular cells identified 983 probe sets by pairwise analysis, 344 probe sets by RP and 208 probe sets by both methods (see Additional file 1). In pairwise comparisons, the differentially expressed genes were identified by selecting those with count of changes more than 4 (out of 5, either increase or decrease).

In this comparison 208 probe sets were identified by pairwise comparison, 122 probe sets by RP, and 32 probe sets (28 named genes) by both methods (Table 4). In pairwise comparisons, the differentially expressed genes were identified by selecting those with count of changes more than 19 (out of 25 inter-patient comparison, every ductal carcinoma against every lobular carcinoma, either increase or decrease). These tumors were well separated by hierarchical clustering based on all 325 probe sets identified by RP and/or pairwise analysis (Figure 4). The expression of genes encoding proteins involved in cell adhesion was changed. Although CDH1 (E-cadherin), a classical member of the cadherin superfamily, was downregulated, THBS4 encoding calcium-binding adhesive glycoprotein, thrombospondin-4, was upregulated in all lobular carcinomas. DDR1 encoding receptor tyrosine kinase was overexpressed in ductal carcinomas.

The DVL1 gene encoding protein involved in Wnt signaling and the leucine-rich repeat protein ASPN (asporin) were upregulated in lobular carcinomas. Ductal carcinomas showed upregulated genes which are involved in cell proliferation, signaling and cell cycle regulation, including RHOU, member of the Rho family of GTPases stimulating quiescent cells to reenter the cell cycle; PCSK6 encoding a calcium-dependent serine endoprotease; PRKCI encoding calcium-independent and phospholipid-dependent protein kinase C; PPP3CB encoding protein phosphatase 3; and CKS2 encoding a component of the CDC28 protein kinase. Epithelial membrane protein 1 (EMP1) was the only upregulated gene involved in cell growth and proliferation in lobular carcinomas. These changes were accompanied by the differential expression of transcription regulators. Majority of genes with this function were upregulated in ductal carcinomas such as AHCTF1, IRAK1, NRIP1, ADNP. Overexpressed genes in lobular carcinomas were the tumor suppressor FOXP1 and another transcription regulator MID1.

The genes encoding proteins involved in ubiquitin-mediated proteolysis, such as USP3, RKHD2 and TTC3, and nuclear components, such as DTL, GLCC11, TTC14, FAM54A, HIST1H3B, were upregulated in ductal carcinomas. Majority of genes encoding proteins with enzyme activity or implicated in metabolism were also upregulated in ductal carcinomas (STK4, SLC1A2, B3GALT3, OSBPL10, CRBN, CHML, YWHAB).

Two lobular and three ductal carcinomas were estrogen receptor-negative, whereas three lobular and two ductal carcinomas were estrogen receptor-positive. Hierarchical clustering using all probe sets was performed to determine whether receptor-positive and receptor-negative tumors could be separated. The tumors of the same histological type showed similar gene expression profiles without differences in relation to ER status as well as to other clinical parameters such as nodal status, stage and the expression of other immunohistochemical markers (data not shown).

Seven differentially expressed genes (KRT5, KRT6 and KRT17 between tumor and normal cells, CDH1, EMP1, DDR1 and DVL1 between lobular and ductal carcinomas) were verified by immunohistochemical detection of proteins on TMA slides comprising of cores from 119 cases. The clinical and histopathological characteristics of these patients are shown in Table 5. The reduced expression or absence of cytokeratins 5/6 and 17 (KRT5, KRT6, KRT17) was found in both tumor tissues in comparison to terminal duct lobular units in 22 normal mammary tissues (p < 0.0001) (Table 6 and Figure 5). In a majority of ducts and lobules including TDLU, these cytokeratins were expressed in both basal and luminal cells, verifying the previously described variability of the expression of basal cytokeratins and their relationship to the cellular origin 51. Cytokeratin 5 and 17 have also been found in a subset of breast cancer and identified patients with poor clinical outcome 31.

E-cadherin (CDH1) has successfully separated ductal and lobular invasive carcinomas. It was absent in 93.3% of lobular tumors compared with only 15% of ductal tumors (p < 0.0001). Epithelial membrane protein 1 (EMP1), discoidin domain receptor 1 (DDR1) and human homolog of the Drosophila dishevelled gene (DVL1) were found by pairwise comparison analysis to be differentially expressed between lobular and ductal carcinomas. Immunohistochemistry confirmed higher expression of DVL1 and EMP1 in lobular carcinomas and of DDR1 in ductal carcinomas (p < 0.0001) (Table 7 and Figure 5). Of the special type carcinomas included on TMA slides, a papillary and two medullary carcinomas were positive for basal cytokeratins, one out of three medullary carcinomas was positive for EMP1 and E-cadherin, a ductal-lobular carcinoma was positive for DDR1, all other special type carcinomas were negative for these markers, and finally none were positive for DVL1 (data not shown).

Combined RP and pairwise comparison showed that ASPN was one of the most upregulated genes in lobular carcinomas, when compared with normal ductal (fold change 23.3) and lobular (fold change 22.1) cells as well as with ductal carcinomas (fold change 3.9). None of the available antibodies against asporin worked on paraffin sections, therefore we utilised chromogenic in situ hybridization to detect asporin mRNA in frozen sections. All five lobular carcinomas appeared positive, whereas five cases of ductal carcinomas were negative or weakly and focally positive which is in agreement with the microarray data (Figure 6).

Besides asporin, CTHRC1 was also upregulated in lobular cancer cells, when compared with normal ductal cells (fold change 48.4) and lobular cells (fold change 18.5), and COL3A1 was upregulated in both IDC and ILC cells, when compared with normal ductal cells (fold change 3.3 and 14.9, respectively) and lobular cells (fold change 4.1 and 9.8, respectively). As both RNA and cDNA were fully utilized for amplification, we used PCR amplification products (see Methods) for validation by another PCR with specific primers. The results of semi-quantitative PCR correlated with the microarray data. Spearman coefficients were r = 0.85, r = 0.73 and r = 0.70 for CTHRC1, ASPN and COL3A1, respectively (all significant at p < 0.0001) (Figure 7).