Transcriptomic changes in human breast cancer progression as determined by serial analysis of gene expression.

INTRODUCTION: Genomic and transcriptomic alterations affecting key cellular processes such us cell proliferation, differentiation and genomic stability are considered crucial for the development and progression of cancer. Most invasive breast carcinomas are known to derive from precursor in situ lesions. It is proposed that major global expression abnormalities occur in the transition from normal to premalignant stages and further progression to invasive stages. Serial analysis of gene expression (SAGE) was employed to generate a comprehensive global gene expression profile of the major changes occurring during breast cancer malignant evolution.
METHODS: In the present study we combined various normal and tumor SAGE libraries available in the public domain with sets of breast cancer SAGE libraries recently generated and sequenced in our laboratory. A recently developed modified t test was used to detect the genes differentially expressed.
RESULTS: We accumulated a total of approximately 1.7 million breast tissue-specific SAGE tags and monitored the behavior of more than 25,157 genes during early breast carcinogenesis. We detected 52 transcripts commonly deregulated across the board when comparing normal tissue with ductal carcinoma in situ, and 149 transcripts when comparing ductal carcinoma in situ with invasive ductal carcinoma (P < 0.01).
CONCLUSION: A major novelty of our study was the use of a statistical method that correctly accounts for the intra-SAGE and inter-SAGE library sources of variation. The most useful result of applying this modified t statistics beta binomial test is the identification of genes and gene families commonly deregulated across samples within each specific stage in the transition from normal to preinvasive and invasive stages of breast cancer development. Most of the gene expression abnormalities detected at the in situ stage were related to specific genes in charge of regulating the proper homeostasis between cell death and cell proliferation. The comparison of in situ lesions with fully invasive lesions, a much more heterogeneous group, clearly identified as the most importantly deregulated group of transcripts those encoding for various families of proteins in charge of extracellular matrix remodeling, invasion and cell motility functions.

Introduction

Invasive ductal breast carcinoma (IDC) is the most common malignancy of the breast, accounting for ~80% of all invasive breast tumors [1]. Although an issue of much controversy over the years, there is now general agreement and overwhelming histopathological and genetic evidence indicating that most invasive breast carcinomas appear to develop gradually from defined precursor lesions [2]. However, it also became clear that progression toward more aggressive phenotypes is not obligatory [3]. It is further evident that many genetic abnormalities underlying tumor progression are probably phenotypically silent.

Numerous molecular genetic changes have been reported as relevant in human breast carcinogenesis, such as anomalies affecting cell proliferation, apoptosis and invasion [4]. Preinvasive breast lesions such as high-grade ductal carcinoma in situ (DCIS) are known to have acquired a myriad of genomic and transcriptomic changes, but as their name implies they are not yet invasive. The development of the ability to invade surrounding tissues is perhaps the most critical event in cancer progression. Among proposed invasion-related genes with reported altered expression in tumor cells are cell adhesion molecules, proteases and cytoskeletal molecules that may influence motility [5]. Identifying the key and most common gene expression abnormalities involved in the transition steps from preinvasion to a fully invasive phenotype is an extremely important topic of research and the main objective of the present report. Studies on this area may provide clues to better diagnose premalignant lesions at high risk of progression and may aid in achieving a better understanding of critical early molecular mechanisms involved in breast cancer evolution.

Serial analysis of gene expression (SAGE) is a comprehensive profiling method that allows for global, unbiased and quantitative characterization of transcriptomes [6,7]. SAGE provides a statistical description of the mRNA population present in a cell without prior selection of the genes to be studied, and this constitutes a major advantage. In this sense, only open systems can identify expressed genes that have not yet been cloned or partially sequenced. A second major advantage is that the information generated is digital in format, and can be directly compared with data generated from any other laboratory or with data available in public databases such as the Cancer Genome Anatomy Project .

To perform a comparative SAGE analysis of normal, preinvasive and invasive lesions, we used a modified t test that we have recently developed [8]. This method has the advantage of taking into account both the intra-sample and inter-sample variability, identifying 'common patterns' of gene changes systematically occurring across samples. Most of the tests developed for measuring differential expression in SAGE data focus on capturing the first type of variation correctly, but tend to neglect the second type [9,10]. The aim of the present study was to provide a statistically robust global gene expression analysis on the progression of breast cancer using the described statistical approach comparing breast normal and tumor SAGE libraries obtained from public databases combined with additional SAGE libraries recently generated in our laboratory.

Materials and methods

SAGE libraries

To perform the comparative analysis of different stages of breast cancer progression, we combined SAGE libraries available in public databases with breast cancer libraries generated and sequenced at our own laboratory. To this end, 12 SAGE libraries of breast tissues (four normal breast tissues, six DCIS tissues and two IDC tissues) were downloaded from the Cancer Genome Anatomy Project – SAGE Genie database (libraries generated at the Polyak Laboratory, Dana-Farber Cancer Institute, Boston, MA, USA). We used 11 additional breast cancer SAGE libraries generated by ourselves, at an approximate resolution of 100,000 SAGE tags per library. All IDC SAGE libraries used in this study were from lymph node-negative, estrogen receptor-positive and progesterone receptor-positive tumor samples, with a tumor size classification of T1 or T2 (i.e. T1–T2 N0 M0). Table 1 summarizes all the SAGE libraries used in this comparative analysis.

Table 1

Breast-specific serial analysis of gene expression (SAGE) libraries

Histology	Library name	Tag count	Unique tags
Normal breast tissue
Normal 1	SAGE breast normal AP Br Na	37,419	15,886
Normal 2	SAGE breast normal epithelium AP 1a	49,021	18,276
Normal 3	SAGE breast normal organoid Ba	58,326	19,602
Normal 4	SAGE breast normal organoid B2a	59,481	20,391
Ductal carcinoma in situ
DCIS 1	SAGE breast carcinoma MD DCISa	42,174	14,237
DCIS 2	SAGE breast carcinoma AP DCIS 3a	57,924	31,142
DCIS 3	SAGE breast carcinoma B DCIS 4a	60,699	20,224
DCIS 4	SAGE breast carcinoma B DCIS 5a	43,118	15,935
DCIS 5	SAGE breast carcinoma epithelium AP DCIS 6a	73,409	30,256
DCIS 6	SAGE breast carcinoma B BWHT18a	50,879	19,182
DCIS 7	MDACC 22Tb	102,533	33,305
Invasive ductal carcinoma
IDC 1	MDACC 09Tb	91,647	37,863
IDC 2	MDACC 14Tb	100,255	26,422
IDC 3	MDACC 15Tb	90,198	27,653
IDC 4	MDACC 17Tb	100,386	29,300
IDC 5	MDACC 18Tb	101,543	29,936
IDC 6	MDACC 19Tb	100,334	28,498
IDC 7	MDACC 20Tb	100,047	28,903
IDC 8	MDACC 21Tb	103,825	31,412
IDC 9	MDACC 24Tb	99,546	30,363
IDC 10	MDACC 25Tb	100,501	30,778
IDC 11	SAGE breast carcinoma B IDC 3a	68,937	22,732
IDC 12	SAGE breast carcinoma B IDC 5a	60,476	20,457
Total	23 breast libraries	1,752,678

a Libraries available in public databases. b Libraries generated in our laboratory.

SAGE methodology

For the SAGE libraries generated in our laboratory we followed standard methods. Briefly, total RNA was extracted from snap-frozen tissues using TRIzol (Invitrogen, San Francisco, CA, USA). SAGE library construction was performed with the I-SAGE kit (Invitrogen) according to the manufacturer's protocol and introducing only minor modifications. The anchoring enzyme was NlaIII and the tagging enzyme used was BsmFI. Concatemerized ditags were cloned into pZERO-1 and sequenced with an ABI 3700 DNA Analyzer (Applied Biosystems, Foster City, CA, USA).

SAGE data processing

SAGE tags were extracted from sequencing files using the SAGE2000 software version 4.0 (a kind gift from Dr K. Kinzler, John Hopkins School of Medicine, Baltimore, MD, USA). Tag abundances for all libraries were normalized to a total of 100,000 tags (at which level a tag present 10 times has an abundance of 0.01%). Tag to gene assignments as well as additional annotations using public databases (e.g. Gene Ontology, Locus Link, Unigene cluster) were performed, using web-based SAGE library tools developed by ourselves . In our comparison we used only tags with only one reliable assigned gene.

Statistical analysis of SAGE libraries

To compare the 23 SAGE libraries, we utilized a modified t test recently developed by us [8]. This analysis allowed us to identify SAGE tags with significantly different expression levels (P < 0.01) between normal tissue and DCIS and between DCIS and IDC. Tags with total counts of less than three in all libraries were filtered out before the analysis. In order to enable visualization and illustration of our analyses, we utilized the TIGR MultiExperiment Viewer (MeV 2.2) software (The Institute for Genomic Research, Rockville, MD, USA). This tool was employed for normalization and average clustering of the SAGE data.

The aim of the heat maps presented is simply to organize and illustrate the data by graphical means. Briefly, the normalization included logarithmic transformation followed by median centering by samples and genes. We used standard average hierarchical clustering techniques to classify and illustrate further the differences found by the modified t test, showing the clusters of differentially coexpressed genes between the normal tissue, DCIS and IDC groups.

Results and discussion

Generation and analysis of SAGE libraries

The primary goal of our study was to identify the most commonly occurring transcriptome changes in the transition from normal breast epithelium to DCIS and invasive carcinoma. To this end, SAGE data obtained from 11 breast cancer libraries generated in our laboratory (1,090,815 tags) were combined and compared with data available in the public domain (661,863 tags), thus generating a dataset of almost 1.7 million breast cancer and normal specific tags, representing approximately 25,157 transcripts from a total of 23 libraries (Table 1).

Our statistically stringent analysis revealed 52 transcripts commonly deregulated across the board when comparing normal tissue with DCIS (Fig. 1), and 149 transcripts when comparing DCIS with IDC (P < 0.01) (Fig. 2) (see additional data files 1 and 2 for additional information with statistical cutoff at P < 0.05). Selected genes based on relative abundance, highly statistical differences and high fold changes between compared groups are sorted and represented in Tables 2 and 3.

Figure 1

Hierarchical clustering of the most commonly different expressed genes between normal breast tissue and ductal carcinoma in situ (DCIS) groups (P < 0.01). Color scale at bottom of picture is used to represent expression level: low expression is represented by green, and high expression is represented by red.

Figure 2

Hierarchical clustering of the most commonly differentially expressed genes between ductal carcinoma in situ (DCIS) and invasive ductal carcinoma (IDC) groups (P < 0.01). Color scale at bottom of picture is used to represent expression level: low expression is represented by green, and high expression is represented by red.

Table 2

Most frequent differentially expressed genes between normal breast epithelium and ductal carcinoma in situ (DCIS)

Tag	Gene	Description	Locus link	Fold change	P value
DCIS overexpressed genes
GTATTTAACT	PKD1-like	Polycystic kidney disease 1-like	79932	13.7	0.0100
CGGACTCACT	STARD10	START domain containing 10	10809	11.2	0.0086
GTGTTGGGGG	EPS8L2	EPS8-like 2	64787	9.6	0.0099
TTTCTGGAGG	KIAA0545	KIAA0545 protein	23094	8.6	0.0100
GATAAATTAA	FLJ14153	Hypothetical protein	64747	8.5	0.0055
GAGAAATATC	NP220	Nuclear protein	27332	8.0	0.0088
CCCTCTTTGG	LOC118487	mRNA similar to RIKEN cDNA 1110001019	118487	7.4	0.0037
CTGGGACTGA	LSM4	U6 small nuclear RNA associated (S. cerevisiae)	25804	6.4	0.0055
CTGGGCCAGC	VAMP5	Vesicle-associated membrane protein 5	10791	6.4	0.0068
GCCCTTTCTC	MRC2	Mannose receptor, C type 2	9902	5.8	0.0015
TCTTGATTTA	A2M	Alpha-2-macroglobulin	2	5.8	0.0083
TAGTTTGTGG	MSH2	MutS homolog 2, colon cancer	4436	5.6	0.0091
TCAGTGAACT	HPS4	Hermansky–Pudlak syndrome 4	89781	5.6	0.0100
GTTTATTCTT	FOXA1	Forkhead box A1	3169	5.3	0.0044
GCCGCTGCCA	PPP1R13B	Protein phosphatase 1	23368	4.3	0.0054
TAAAGTGTCT	PIGS	Phosphatidylinositol glycan, class S	94005	3.9	0.0100
DCIS underexpressed genes
GGGACGAGTG	TM4SF1	Transmembrane 4 superfamily member 1	4071	-442.6	0.0083
TAACAGCCAG	NFKBIA	Nuclear factor kappa light polypeptide gene	4792	-158.6	2.4 × 10-6
CAACTAATTC	CLU	Clusterin	1191	-63.9	0.0036
GCCTTAACAA	PBEF	Pre-B-cell colony-enhancing factor	10135	-44.3	0.0020
GGGTTTTTAT	NSEP1	Nuclease sensitive element binding protein 1	4904	-36.2	0.0001
GACACGAACA	RASD1	RAS, dexamethasone-induce 1	51655	-31.4	0.0095
AAGATTGGTG	CD9	CD9 antigen (p24)	928	-29.6	0.0003
ACCAAATTAA	TNFRSF10B	Tumor necrosis factor receptor superfamily	8795	-29.4	0.0003
CTGGGCCTGA	LITAF	Lipopolysaccharide-induced tumor necrosis factor	9516	-28.9	0.0076
CTGCCATAAC	SBDS	Shwachman–Bodian–Diamond syndrome	51119	-24.2	0.0005
CACAGGCAAA	BZW1	Basic leucine zipper and W2 domains 1	9689	-22.1	0.0056
GTTCCCTGGC	FAU	Finkel–Biskis–Reilly murine sarcoma virus	2197	-22.1	0.0028
GTCTGCACCT	DKFZp547C1	Hypothetical protein	254851	-21.9	0.0087
TACGTTGCAG	GC20	Translocation factor sui1 homolog	10289	-21.8	0.0079
TGTAAAGATT	CCNL1	Cyclin L1	57018	-21.2	0.0008
TGTTAAGTTC	CRY1	Cryptochrome 1 (photolyase-like)	1407	-18.7	0.0091
GAAATAAAGT	FLJ21657	Hypothetical protein	64417	-18.5	0.0032
ATGGGCTTGA	SQRDL	Sulfide quinone reductase-like (yeast)	58472	-17.3	0.0061
TCAAGAAATT	PSME3	Proteasome activator subunit 3	10197	-15.7	0.0028
CCGTGGTCGT	FBL	Fibrillarin	2091	-15.3	0.0100
TGGAACAGGA	TGIF	Transforming growth factor beta-induced factor (TALE family homeobox)	7050	-12.2	0.0030
AATGCTGGCA	DNAJB6	DnaJ homolog, subfamily B, member 6	10049	-11.7	0.0065
AATGAGCAAC	GBP2	Guanylate binding protein 2, interferon-inducible	2634	-11.1	0.0082
GACCTATCTC	KIAA0992	Paladin	23022	-10.9	0.0092
AACTCTTGAA	EIF3S3	Eukaryotic translation initiation factor 3	8667	-10.6	0.0082
GGGATTTTGT	PMAIP1	Phorbol-12-myristate-13-acetate-induced protein 1	5366	-10.6	0.0091
AAAGCAAAAA	PTPN4	Protein tyrosine phosphatase, non-receptor type 4	5775	-10.4	0.0040
ACTGACTATC	NEU1	Sialidase 1 (lysosomal sialidase)	4758	-10.3	0.0095
TTCCAGTTCA	PDE4B	Phosphodiesterase 4B, camp-specific	5142	-9.9	0.0087
GAATGATTTC	ORF1-FL49	Putative nuclear protein	84418	-9.5	0.0070
GACTCGCTCC	HSJ001348	cDNA for differentially expressed CO16 gene	54742	-9.5	0.0072
TGGTTACAAA	NDEL1	Nude nuclear distribution gene E homolog like 1	81565	-8.9	0.0100
AGTATGAGGA	TNFAIP3	Tumor necrosis factor, alpha-induced protein 3	7128	-8.1	0.0083
CAGTTTAAAA	CRSP6	Cofactor required for Sp1 transcriptional activation	9440	-7.7	0.0100

Table 3

Most frequent differentially expressed genes between ductal carcinoma in situ and invasive ductal carcinoma (IDC)

Tag	Gene	Description	Locus link	Fold change	P value
IDC overexpressed genes
TGGAAATGAC	COL1A1	Collagen type I, alpha 1	1277	315.4	0.0054
ATGTGAAGAG	SPARC	Secreted protein, cysteine-rich (osteonectin)	6678	286.8	0.0003
TTTGGTTTTC	COL1A2	Collagen type I, alpha 2	1278	210.9	0.0084
TTGCTGACTT	COL6A1	Collagen type VI, alpha 1	1291	73.9	0.0023
TTATGTTTAA	LUM	Lumican	4060	56.7	0.0011
TTGGAGATCT	NDUFA4	NADH dehydrogenase (ubiquinone)	4697	56.4	0.0065
CCACAGGGGA	COL3A1	Collagen type III, alpha 1	1281	49.4	0.0056
ATCTTGTTAC	FN1	Fibronectin 1	2335	44.3	0.0031
TTGTAATCGT	OAZ1	Ornithine decarboxylase antizyme 1	4946	38.6	0.0038
TGTAATCAAT	HNRPA1	Heterogeneous nuclear ribonucleoprotein A1	3178	38.2	0.0039
GGAAGCTAAG	OSF-2	Osteoblast specific factor 2 (fasciclin I-like)	10631	36.3	0.0005
ACCTGTATCC	IFITM3	Interferon induced transmembrane protein 3	10410	34.3	0.0021
GGAAATGTCA	MMP2	Matrix metalloproteinase 2	4313	29.1	0.0008
TGCACTTCAA	SPARCL1	SPARC-like 1 (mast9, hevin)	8404	21.7	0.0050
GGAACTTTTA	SULF2	Sulfatase 2	55959	19.8	0.0017
CTGTTAGTGT	MDH1	Malate dehydrogenase 1	4190	18.5	0.0026
TATGAATGCT	CSPG2	Chondroitin sulfate proteoglycan 2 (versican)	1462	18.2	0.0017
TCCAAATCGA	VIM	Vimentin	7431	17.7	0.0014
TGTAGTTTGA	SKP1A	S-phase kinase-associated protein 1A	6500	16.9	0.0013
TAATAAACAG	ASAH1	n-Acylsphimgosine amidohydrolase	427	16.8	0.0085
GCCTCCTCCC	EIF3k	Eukaryotic translation initiation factor 3	27335	16.8	0.0098
GAAACAAGAT	PGK1	Phosphoglycerate kinase 1	5230	15.8	0.0006
TGCTTTGGGA	TTC11	tetratricopeptide repeat domain 11	51024	15.8	0.0018
GAAATCAAAA	SIGLEC5	Sialic acid binding Ig-like lectin 5	8778	14.7	0.0098
ATGTAGTAGT	HNRPD	Heterogeneous nuclear ribonucleoprotein D	3184	14.7	0.0025
GACCACCTTT	MFAP2	Microfibrillar-associated protein 2	4237	14.4	0.0000
ACTTATTATG	DCN	Decorin	1634	13.9	0.0007
TCTCTACCCA	APLP2	Amyloid beta (A4) precursor-like protein 2	334	13.7	0.0076
TGCAATATGC	FBN1	Fibrillin 1	2200	13.5	0.0037
ATTTCTTCAA	SFRP2	Secreted frizzled-related protein 2	6423	13.4	0.0030
ATAAAAAGAA	CTSK	Cathepsin K (pycnodysostosis)	1513	13.0	0.0003
GTACATTGTA	MGC15737	Hypothetical protein	85012	12.6	0.0011
TGATGTTTGA	DAZAP2	DAZ associated protein 2	9802	12.5	0.0013
TCCGTGGTTG	BASP1	Membrane attached signal protein 1	10409	12.1	0.0055
ACTGCTTTAC	DKFZp564I1922	Adlican	25878	12.0	0.0086
TCTGCAATGA	TINP1	Trasnforming growth factor beta-inducible nuclear protein 1	10412	12.0	0.0033
GTTTCTTCCC	SELH	Selenoprotein H	2880636	11.8	0.0037
AATATGCTTT	ATP6V1E1	ATPase	529	11.6	0.0009
TTATGGATCT	SPON2	Spondin 2, extracellular matrix protein	10417	11.3	0.0003
AAAATAAAGA	APEX1	Nuclease, multifunctional DNA repair enzyme	328	11.3	0.0099
TGTGTGTTTG	HIF0	H1 histone family	3005	11.2	0.0029
TATGTTTCAG	PTPN12	Protein tyrosine phosphatase	5782	11.1	0.0003
ACCAAAGCCC	MGC9651	Hypothetical protein	114932	10.6	0.0075
CAAGGATCTA	NICE-3	NICE-3 protein	25912	10.5	0.0057
GACGTCTTAA	PSMA4	Proteasome subunit, alpha type	5685	10.3	0.0038
CAGATAACAT	TOMM20	Translocase	9804	10.1	0.0064
AACTCTTGAA	EIF3S3	Eukaryotic translation initiation factor 3	8667	10.0	0.0047
TTCTTGGTGT	TRPS1	Trichorhinophalangeal syndrome I	7227	9.9	0.0042
TGCCTTAGTA	DNAJC1	DNAJ homolog	64215	9.8	0.0040
AGACAAGCTG	SFRS5	Splicing factor	6430	9.4	0.0015
ACAAGAATTG	SYPL	Synaptophysin-like protein	6856	9.3	0.0027
TACATCCGAA	MTPN	Myotrophin	136319	9.3	0.0013

As expected, we detected various ribosomal genes among the most abundant transcripts in all the breast SAGE libraries, and these genes were highly upregulated in the invasive carcinomas. This agrees with the previous global expression profiles and with the comparisons of cancers and the corresponding normal tissues in general [7,11,12]. To simplify illustration of the data, ribosomal genes are not included in the figures and tables.

Global comparison of normal tissues and DCIS

Among the 52 transcripts detected as differentially expressed in DCIS (P < 0.01), 36 were downregulated transcripts and 16 were upregulated transcripts in these lesions when compared with normal breast epithelial cells and mammary epithelial organoids (Fig. 1 and Table 2). We defined and classified the 52 genes differentially expressed into the nine functional categories [13] shown in Fig. 3a. Interestingly, we found that 38% of these transcripts are related to the cell cycle (15%), signal transduction (8%) and apoptosis (15%).

Figure 3

Classification in functional categories of affected transcripts. (a) Differentially expressed between normal breast tissue and ductal carcinoma in situ (DCIS) (P < 0.01). (b) Transcripts differentially expressed between DCIS and invasive ductal carcinoma (IDC) (P < 0.01).

As expected, our analysis of DCIS versus normal breast epithelium revealed numerous similarities with SAGE data reported previously [12,14], but more importantly it also provided novel information. The expression of numerous genes was significantly downregulated in DCIS, including: transmembrane 4 super family member 1 (TM4SF1), nuclear factor kappa light polypeptide (NFKB1A), pre-B-cell (PBEF), RAS dexamethasone-induced (RASD1), tumor necrosis factor receptor superfamily member 10b (TNFRSF10B), and tumor necrosis factor α-induced protein (TNFAIP). All these transcripts were also observed downregulated in previous reports [12,14] (Table 2). On the contrary, our analysis revealed additional clusters of genes significantly downregulated in the DCIS group that were not previously reported by others: clusterin/apolipoprotein J (CLU), nuclease sensitive element binding protein 1 (NSEP1), lipopolysaccharide-induced TNF factor (LITAF/PIG7), basic leucine zipper/W2 domains 1 (BZW1), and cyclin L1 (CCNL1) (Table 2).

Clusterin was one of the most dramatically downregulated genes (-63.9-fold; P = 0.0036) in DCIS libraries. This gene encodes a heterodimeric, highly conserved, secreted glycoprotein. Alterations in Clusterin expression and/or protein maturation are linked to changes in tissue growth or regression, which may be related to specific proapoptotic or antiapoptotic protein isoforms [15]. Clusterin was reported as overexpressed during tissue and cell involution, and was downregulated in esophageal squamous cell carcinoma and prostate carcinoma, suggesting that this expression alteration could be a general phenomenon during tumor progression [16,17]. On the contrary, and in contrast to these and our observations, Redondo and colleagues reported increased Clusterin expression in breast carcinoma samples [18]. The reason for this discrepancy is unclear at this point. The role of Clusterin in cell survival, cell death and neoplastic transformation remains controversial [15].

Another commonly observed downregulated gene in DCIS libraries was NSEP-1 (-36.2-fold; P = 0.0001). Also known as YB1, NSEP-1 is a member of the highly conserved Y-box family of proteins, which regulate the transcription of several genes associated with cell death including both fas, a cell death-associated receptor, and the tumor suppressor gene p53 [19]. The decrease in expression of NSEP-1 transcripts could play an important role in the early stages of breast carcinogenesis in order to overcome cell proliferation controls.

Interestingly, and as previously observed, we also detected significant downregulation of various cytokines and chemokines: interleukin enhancer binding factor 2 (ILF2), interleukin 13 receptor alpha 1 (IL13RA1), leukemia inhibitory factor (LIF), cardiotrophin-like cytokine (CLC), chemokine C–C ligand 2 (CCL2), and chemokine C–X–C ligand 1 (CXCL1). All these cytokines and chemokines are highly expressed in normal mammary epithelium and are dramatically downregulated in the DCIS samples. These differentially expressed genes were detected within a range of 0.02