Fresh-frozen human breast tumor samples were retrieved from the Tumor Tissue Biobank of the Medical and Research Center – Hospital A C Camargo, São Paulo. Sections 5 μm thick from the fresh-frozen tumor blocks were cut onto glass slides, stained with hematoxylin and eosin, and reviewed by a pathologist. The hematoxylin and eosin stained sections were used to evaluate and select appropriate tumor areas corresponding to each histological component. The histological grade of the DCIS was assigned in accordance with the World Health Organization scale 20, and IDC was classified in accordance with the Scarff-Bloom-Richardson grading scheme 2122. Forty-one samples were evaluated, consisting of four non-neoplastic breast samples, five pure DCIS (stage 0) samples, 11 in situ component of DCIS-IDC samples, and 10 IDC samples; these served as initial sample sets. An additional 11 in situ component of DCIS-IDC samples were evaluated as an independent sample set. The non-neoplastic samples were obtained from perilesional mammary specimens from patients obtained during resection of benign lesions. A pathologist subjected all slides representative of pure DCIS to a careful histopathological analysis in order to ensure the absence of any previously undetected microinvasions. At least 5 years of follow-up data were available for all patients.

The patients had a mean age of 49 years, and none of them had received preoperative systemic treatment. The patients' clinical characteristics are summarized in Table 1. The expression patterns of estrogen receptor, progesterone receptor, and human epidermal growth factor receptor (HER)2/neu were positive in the majority of the samples. The research was approved by the Ethics Committee of the Medical and Research Center – Hospital A C Camargo, under protocol number 587/04. All participants gave written, informed consent.

Cells were laser captured using the PixCell II LCM system (Arcturus Engineering, Mountain View, CA, USA). About 4,000 cells were captured from 4 to 7 μm frozen sections, mounted onto glass slides, and stained with 100 μl of nuclear fast red (C.I.60760; Certistain^®; Merck, Darmstadt, Germany) for microscopy. Only one type of cells was isolated from each sample group. Non-neoplastic breast epithelium samples were epithelial cells without contamination with stromal cells; pure DCIS samples were tumor cells captured from ducts of pure DCIS lesion; in situ component of DCIS-IDC samples were tumor cells captured from in situ component of the DCIS-IDC lesion; and IDC samples were infiltrative tumor cells captured from the invasive lesion.

A representative sample of cells from the pure DCIS depicting the different phases during the microdissection procedure is shown in Figure 1.

Laser capture microdissection (LCM) captured cells on CapSure™ HS LCM Caps (Arcturus Engineering) were resuspended in 10 μl of PicoPure RNA extraction buffer (Arcturus Engineering). Total RNA was extracted by using the PicoPure™ RNA Isolation kit (Arcturus Engineering #KT0204) and DNase treated using the RNase-Free DNase Set (Qiagen #79254; Qiagen-Germantown MD USA), in accordance with the manufacturer's instructions. A two-round linear amplification procedure, based on T7-driven amplification, was performed following a previously described protocol 23 with some modifications described below. The total RNA was first denatured at 70°C for 10 minutes in presence of 200 ng oligo (dT) 24-T7 primer (5'-AAA CGA CGG CCA GTG AAT TGT AAT ACG ACT CAC TAT AGG CGC T (24)-3'; 57 base pairs) and snap cooled on ice.

Reverse transcription was performed by adding 1× first strand buffer and 0.01 mol/l dithiothrectol (Invitrogen Life Technology, Carlsbad, CA, USA), 2 μl diethylpyrocarbonate (DEPC; Sigma, St Louis, MO, USA) treated water, 40 U rRNasin (Promega, Madison, WI, USA), 1 mmol/l dNTP (Amersham Biosciences, Piscataway, NJ, USA), and 400 U SuperScript™ II Reverse Transcriptase (Invitrogen Life Technology) to a final volume of 20 μl. The reaction was incubated for 120 minutes at 42°C. Second-strand synthesis was performed by adding 53 μl of DEPC-treated water, 20 μl of 5× second strand buffer (Invitrogen Life Technology), 1 mmol/l dNTP, 1 U RNase H (Invitrogen Life Technology), 10 U Escherichia coli DNA ligase, and 40 U E. coli DNA polymerase I (Invitrogen Life Technology) to a final volume of 100 μl. The reaction was incubated for 2 hours at 16°C. Ten units of T4 DNA polymerase I (Invitrogen Life Technology) were added and incubated again at 16°C for 5 minutes.

The double strand cDNA (dscDNA) was stopped by adding 0.05 mol/l EDTA. UltraPure™ Phenol (Invitrogen, Carlsbad, CA, USA):chloroform:isoamyl alcohol (Merck), at a ratio of 25:24:1 and a pH of 8.0, was used for cDNA purification. The dscDNA was precipitated with absolute ETOH (Merck) and resuspended in 10 μl DEPC-treated water. The dscDNAs were subjected to in vitro transcription using reagents from Ribomax™ Large scale RNA production system T7 kit (Promega), in accordance with the manufacturer's recommendation. The amplified RNA (aRNA) was reverse transcribed into cDNA using 9 μg random hexamer (dN6; Amersham Biosciences, Little Chalfont, UK). cDNA synthesis was continued with the same conditions used in the first strand of the first round. The second strand was synthesized using Advantage^® cDNA Polymerase (Clontech, Mountain View, CA, USA), and purification was performed in accordance with the methodology cited above.

The aRNA quality, in terms of purity and integrity, was assessed by absorbance at 260/280 nm using a GeneQuant pro spectrophotometer (Amersham Pharmacia Biotech, Little Chalfont, UK) and by electrophoresis in 1% UltraPure™ Agarose (Invitrogen Life Technology) gel with ethidium bromide staining (Sigma, St Louis, MO, USA), respectively. Only aRNA samples yielding a minimum of 15 μg and presenting a smear concentration between 300 and 700 base pairs (which guarantees high quality hybridization) were further processed. Total RNA from HB4a normal luminal epithelial mammary cells 24 was extracted and amplified following the same protocol and used as a reference for microarray hybridizations.

We used a customized cDNA platform (4.8K002 platform) comprising 4,608 cDNAs that represent human genes 25. The labeled cDNA was generated in a reverse transcriptase reaction in the presence of 4 μg aRNA, 9 μg random hexamer primer (Invitrogen Life Technology), Cy3-labeled or Cy5-labeled dCTP (Amersham Biosciences, Little Chalfont, UK), and 400 U SuperScript II (Invitrogen Life Technology). The residual dye was removed using illustra AutoSeq™ G-50 (GE Healthcare, Little Chalfont, UK). Equal amounts of test and reference cDNA reverse colored Cy-labeled product were competitively hybridized against the cDNA probes in microarray slides. Dye swap was performed for each sample analyzed and used as replicate samples. Pre-hybridization was carried out in a humidified chamber at 42°C for 16 to 20 hours, and hybridizations were performed on GeneTac Hybridization Station (Genome Solutions, Ann Arbor, MI, USA) at 42°C.

After hybridization, slides were washed as follows: 2× Saline-Sodium Citrate (SSC) for 10 minutes, 0.1 × SSC/0.1% SDS for 10 minutes (two times), and 0.1 × SSC for 10 minutes (two times) at 37°C. All solutions were pre-heated to 42°C. Hybridized arrays were scanned on the ScanArray™ Express (Packard BioScience Biochip Technologies, Billerica, MA, USA), and Cy5/Cy3 signals were quantified using the histogram method with ScanArray Express software (Perkin-Elmer Life Sciences, Boston, MA, USA). Fluorescent intensities of Cy5 and Cy3 channels on each slide were subjected to spot filtering and normalization. We first eliminated all saturated points (≥ 63,000; approximately 16 bits) and performed a local background subtraction, considering for analysis only those spots with positives values. Normalization was performed using locally weighted linear regression within and across arrays for inter-slide normalization. Local normalization has the advantage that it can help to correct for systematic spatial variations in the array 26. After normalization, data for each gene were reported as the logarithm of the expression ratio used to represent the relative gene expression levels in the experimental samples. The raw data from hybridizations and experimental conditions can be obtained at the Gene Expression Omnibus 27 under accession number GSE11042. A detailed description of the platform array is available in accession number GPL1930.

For applying the concept for molecular divergence and to identify the most distinct group of samples, the general expression patterns were compared between the sample groups (non-neoplastic, pure DCIS, in situ component of DCIS-IDC, and IDC) using the analysis of variance (ANOVA) statistical test (positive false discovery rate – pFDR < 0.01) 28 followed by Tukey's test 29. To select the putative genes involved in DCIS progression, two-by-two comparisons among non-neoplastic, pure DCIS, and in situ component of DCIS-IDC sample groups were performed; differentially expressed genes, whose expression levels exhibited at least twofold change, were selected. Downregulated and upregulated genes were analyzed separately. A Venn diagram was constructed to select the genes of interest.

To determine decay rate statistics for collections of genes belonging to the different gene sets, automated analysis of gene function was required. The functional assignments of the genes were obtained using the χ² distribution of Onto-Tools (developed by Intelligent Systems and Bioinformatics Laboratory from Department of Computational Sciences of Wayne State University) 30. Functional processes were considered as significant if the P value was under 0.01. Functional assessments, which were represented by only one gene in the platform, were not taken into consideration in order to avoid artifactual results.

For clustering samples based on gene expression profiles, we applied a nonsupervised hierarchical clustering based on Euclidean distance and average linkage. The reliability of the clustering was assessed by the Bootstrap technique using MEV (MultiExperiment Viewer – Boston, MA, USA) 31.

Quantitative RT-PCR reactions were performed using the ABI Prism™ 7700 Sequence Detection System (Applied Biosystems, Foster City, CA, USA). Aliquots of cDNA from aRNA were used as templates. RT-PCR reactions were carried out using SYBR^® Green PCR MasterMix (Applied Biosystems) in a total volume of 20 μl using the following program: 2 minutes at 50°C and 10 minutes at 95°C for the initial denaturing, followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute. The list of oligonucleotide sequences is shown in Table 2. Finally, dissociation curves were generated at 95°C for 15 seconds, 60°C for 30 seconds, and 95°C for 15 seconds. To evaluate the amplification of nonspecific products and primer-dimer formation, dissociation curves were analyzed and aliquots of each reaction were subjected to silver staining acrylamide gel electrophoresis. The efficiency of each pair of primers was calculated using standard curve dilutions (as described in the Applied Biosystems protocols). The analysis was conducted using the cDNA converted from the aRNA used in the microarray study. The reactions were performed in duplicate. Three internal control genes, namely HPRT1 [GenBank:NM_000194] 32, GAPDH [GenBank:AJ005371], and BCR [GenBank:NM_004327], were considered in gene expression normalization. Relative gene expression between sample groups was calculated using the Pfaffl model 33, employing the efficiency-corrected equation.

To generate a precise correlation between specific epithelial cells and their gene expression patterns, we integrated the use of LCM and T7-based RNA amplification with cDNA microarrays. This procedure permitted gene expression profile analysis with a cell based, rather than tissue based, resolution. To characterize the molecular alterations of cells from the in situ component of the two breast cancer lesions, namely DCIS and DCIS-IDC, we analyzed the gene expression profiles of both. We also used non-neoplastic epithelial cells and cells from IDC lesions as examples of zero and complete progression, respectively.

The Pearson correlation average between all hybridization replicates used in this study was 0.91. All replicates clustered together in dendrograms reporting suitable correction of the individual dye incorporation efficiency by normalization procedure and high experimental reproducibility.

To classify the groups of samples according to their molecular divergence, we used the number of differentially expressed genes as our distance measure; the larger the number of differentially expressed genes between one sample type and all the others, the more distant the sample was allocated and consequently more molecularly divergent. To accomplish this, an ANOVA test corrected by pFDR (<0.01) was performed on the four sample groups, identifying 764 differentially expressed genes (see Additional file 1). Next, Tukey's test was performed through two by two comparisons between the distinct groups. As expected, the non-neoplastic epithelial breast cells exhibited the most divergent expression profile (29%). A total of 221 genes out of 764 were only differentially expressed between non-neoplastic and the three tumor cell groups. In contrast, for cells from pure DCIS, the in situ component of DCIS-IDC and IDC, when compared with all of the other three, there were 12 (1.6%), 37 (4.8%) and 6 (0.3%) differentially expressed genes out of 764, respectively.

Using the same metrics, a new round of analysis was performed excluding the non-neoplastic sample group in order to identify which tumor cell population exhibits the highest molecular divergence level among the three analyzed groups. The ANOVA corrected by pFDR (<0.01) identified 90 variable genes among the three groups of neoplastic cells (see Additional file 2). Tukey's test revealed that cells from pure DCIS exhibited the most distinct gene expression profile, with 75 genes (83%) out of 90 being differentially expressed in comparison with the other two groups of tumor cells. For cells from the in situ component of DCIS-IDC and cells from IDC lesions, 16 genes (18%) and 6 genes (7%) out of 90 were found to be differentially expressed, respectively (Figure 2). A parallel analysis between molecular and morphological aspects yielded contradictory results, because – in terms of morphological features – IDC had the most distinct features and pure DCIS and the in situ component of DCIS-IDC lesions exhibited identical patterns. On the other hand, in terms of molecular features, cells from pure DCIS exhibited the most distinct molecular expression profile, and consequently cells from in situ component of DCIS-IDC retains high similarity to cells from IDC, which suggests alterations in gene expression programs of cells from in situ component of DCIS-IDC before their progression to IDC.

DCIS progresses to malignant disease when some of the tumor cells acquire invasive capacity. Our previous data strongly suggested that the in situ component of DCIS-IDC already harbors molecular alterations that signal establishment of the invasive process. Based on these findings, we reasoned that the genes potentially involved in the earliest molecular step in acquiring the capacity to invade the surrounding tissue might be among the genes that are differentially expressed between cells from in situ component of DCIS-IDC and both pure DCIS and non-neoplastic groups.

To identify these genes we conducted an ANOVA test corrected by pFDR (<0.01) among non-neoplastic, pure DCIS, and the in situ component of DCIS-IDC, identifying 785 altered genes (see Additional file 3). The two-by-two comparisons between samples from the three groups revealed 8, 215, and 161 genes between non-neoplastic versus pure DCIS, non-neoplastic versus in situ component of DCIS-IDC, and pure DCIS versus the in situ component of DCIS-IDC, respectively (Figure 3a; see Additional file 4). To eliminate genes involved in tumor formation from those potentially involved in tumor progression, we subtracted eight genes that were differentially expressed between cells from non-neoplastic and pure DCIS cells (Figure 3a). The common differentially expressed genes between in situ component of DCIS-IDC versus non-neoplastic or pure DCIS were selected in order to choose the more robust genes, yielding 147 genes classified as potentially implicated in DCIS progression. From those, 126 were upregulated and 21 were downregulated in pure DCIS (see Additional file 5), suggesting that the malignant process of tumor cells from pure DCIS to in situ component of DCIS/IDC occurs rather by downregulation than by upregulation of gene expression.

Functional annotation of this gene set, according to the Onto-Tools database, revealed a statistically significant enrichment of genes involved in cell adhesion (represented by C20orf42 [GenBank:AL118505], LPXN [GenBank:NM_004811], PCLKC [GenBank:NM_017675], DGCR2 [GenBank:NM_005137], AZGP1 [GenBank:NM_001185], CHST10 [GenBank:NM_004854], ITGB2 [GenBank:NM_000211], PLEKHC1 [GenBank:AK291738], PCDH10 [GenBank:NM_032961], and NEDD9 [GenBank:NM_006403]) and cellular defense (represented by CXCL9 [GenBank:NM_002416], MAPRE2 [GenBank:NM_014268], and C3AR1 [GenBank:NM_004054]). LPXN [GenBank:NM_004811] and NEDD9 [GenBank:NM_006403] have been reported as being involved in cancer progression. Over-expression of LPXN [GenBank:NM_004811] and NEDD9 [GenBank:NM_006403] resulted in increased migration in the bone-derived metastatic prostate cancer cell line 34 and in promotion of metastatic melanoma 35, respectively. These reports support our data, which reveal high expression of these genes in the in situ component of DCIS-IDC compared with pure DCIS cells. The 147 genes were organized into seven different biological classes in a hierarchical manner (Table 3).

The hierarchical clustering based on the expression pattern of this gene set resulted in two main branches with high level of reliability (Figure 3b,c). Non-neoplastic samples and 60% of the pure DCIS samples were discriminated from 100% of the in situ component of DCIS-IDC samples (Figure 3b). The ability of this gene set to generate an expression pattern (Figure 3d) that segregated the two groups of cells (pure DICS and in situ component of DCIS-IDC) relatively well, grouping together the samples representing cells from non-neoplastic and cells from pure DCIS lesion, strengthened the hypothesis that those genes might be involved in the early molecular alterations that are necessary for launching of the invasive process.

In order to evaluate the robustness of our microarray findings and to avoid choosing false differentially expressed genes because of technical limitations, we randomly selected eight genes from the 147 gene set (Table 4) and performed quantitative RT-PCR in duplicate experiments. The average of the internal control genes (HPRT1 [GenBank:NM_000194], GAPDH [GenBank: AJ005371] and BCR [GenBank:NM_004327]) was used in a normalization procedure.

Using the initial sample sets and the criteria of ≥ 2-fold change, five out of eight genes (62.5%) exhibited agreement in both methodologies. The genes C16orf5 [GenBank:NM_013339], GOSR2 [GenBank:NM_004287], and TXNL2 [GenBank:AL138831] were upregulated, and LOX [GenBank:NM_002317] and SULF-1 [GenBank:NM_001128206] were downregulated in pure DCIS in comparison with in situ component of DCIS-IDC. These five genes were also evaluated in an independent group of 11 in situ component of DCIS-IDC samples (Table 5). Three genes (60%; GOSR2 [GenBank:NM_004287], LOX [GenBank:NM_002317], and SULF-1 [GenBank:NM_001128206]) exhibited agreement with the previous data, strengthening the hypothesis that these genes are involved in the malignant process of DCIS.

Because we were interested in identifying genes that are involved with the acquisition of invasive capacity in DCIS, we reasoned that these genes should not exhibit differential expression between cells from non-neoplastic cells and pure DCIS cells, as neither of these cell types has established invasion capacity. Likewise, we would expect there to be no difference in expression level between cells from the in situ component of DCIS-IDC and IDC lesions, in which – despite of the morphological differences in the tissue – the capacity and program of invasion are already established. Therefore, the five genes confirmed in the initial sample set were evaluated in non-neoplastic and IDC samples. Based on our selection criteria, three genes (C16orf5 [GenBank:NM_013339], LOX [GenBank:NM_002317], and SULF-1 [GenBank:NM_001128206]) showed the expected gene expression behavior in the four sample groups, exhibiting slight or no difference between the group with no invasive program (non-neoplastic and pure DCIS) and that in which the invasive program is already established (DCIS-IDC and IDC; Figure 4a). The expression levels of these three genes exhibited statistically significant differences between the cells lacking the invasive program (non-neoplastic and pure DCIS) and cells possessing the invasive program (in situ component of DCIS-IDC and IDC; Figure 4b). This strongly indicates that LOX [GenBank:NM_002317] and SULF1 [GenBank:NM_001128206], confirmed in the initial and independent group of in situ component of DCIS-IDC samples, are potentially involved in the acquisition of invasive capacity in DCIS.