Regulation of clustered gene expression by cofactor of BRCA1 ...

11 downloads 109 Views 365KB Size Report
Oct 16, 2006 - Cofactor of BRCA1 (COBRA1) was first identified as a BRCA1-interacting protein capable of inducing large- scale chromatin reorganization (Ye ...
Oncogene (2007) 26, 2543–2553

& 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00 www.nature.com/onc

ORIGINAL ARTICLE

Regulation of clustered gene expression by cofactor of BRCA1 (COBRA1) in breast cancer cells SE Aiyar1, AL Blair1, DA Hopkinson, S Bekiranov and R Li Department of Biochemistry and Molecular Genetics, School of Medicine, University of Virginia, Charlottesville, VA, USA

Eucaryotic genes that are coordinately expressed tend to be clustered. Furthermore, gene clusters across chromosomal regions are often upregulated in various tumors. However, relatively little is known about how gene clusters are coordinately expressed in physiological or pathological conditions. Cofactor of BRCA1 (COBRA1), a subunit of the human negative elongation factor, has been shown to repress estrogen-stimulated transcription of trefoil factor 1 (TFF1 or pS2) by stalling RNA polymerase II. Here, we carried out a genome-wide study to identify additional physiological target genes of COBRA1 in breast cancer cells. The study identified a total of 134 genes that were either activated or repressed upon small hairpin RNA-mediated reduction of COBRA1. Interestingly, many COBRA1-regulated genes reside as clusters on the chromosomes and have been previously implicated in cancer development. Detailed examination of two such clusters on chromosome 21 (21q22) and chromosome X (Xp11) reveals that COBRA1 is physically associated with a subset of its regulated genes in each cluster. In addition, COBRA1 was shown to regulate both estrogen-dependent and -independent transcription of the gene cluster at 21q22, which encompasses the previously identified COBRA1-regulated TFF1 (pS2) locus. Thus, COBRA1 plays a critical role in the regulation of clustered gene expression at preferred chromosomal domains in breast cancer cells. Oncogene (2007) 26, 2543–2553. doi:10.1038/sj.onc.1210047; published online 16 October 2006 Keywords: COBRA1; clustered gene expression; estrogen; transcription; NELF

Introduction Recent rigorous examination of complete eucaryotic genomes, combined with gene expression profiling analyses, has shed exciting and unexpected insight into Correspondence: Dr R Li, Department of Biochemistry and Molecular Genetics, School of Medicine, University of Virginia, 1300 Jefferson Park Avenue, Charlottesville, VA 22908, USA. E-mail: [email protected] 1 These authors contributed equally to this work. Received 22 May 2006; revised 24 August 2006; accepted 6 September 2006; published online 16 October 2006

the gene organization patterns within a genome. In contrast to the long-held presumption that genes are randomly distributed throughout the eucaryotic genomes, there is compelling evidence for clustering of co-expressed genes in every eucaryotic genome examined so far (Hurst et al., 2004). For example, 25% of genes in budding yeast that are transcribed in the same cell-cycle stage are adjacently located (Cho et al., 1998). The gene-clustering phenomenon is perhaps even more prominent in multicellular organisms. It is reported that 20% of genes in the entire fly genome are organized into clusters with similar expression patterns (Spellman and Rubin, 2002). Likewise, a large number of housekeeping genes and highly expressed genes in the human genome are clustered (Caron et al., 2001; Lercher et al., 2002). Co-expression of most gene clusters is unlikely owing to a ‘bystander effect’ wherein genes might be activated simply because they are adjacent to other actively transcribed genes, as syntenic analyses of mouse and human genomes indicate that the order of co-expressed genes in a cluster is conserved by natural selection (Singer et al., 2005). Therefore, clustering of coexpressed genes may have a functional advantage in coordinating gene expression during cell proliferation, differentiation and development. In addition to their potential roles in normal physiology, gene clusters across chromosomal regions are often upregulated in various tumors (Zhou et al., 2003; Koon et al., 2004). For example, members of the trefoil factor family (TFF1–3) are tandem clustered within a 55 kb genomic region at chromosome 21q22.3 (Chinery et al., 1995; Schmitt et al., 1996). Clinical studies suggest a strong association of TFF1/TFF3 expression with breast cancer (Poulsom et al., 1997; Mikhitarian et al., 2005; Xu et al., 2005; Smid et al., 2006). In a recent study of primary breast tumor samples, overexpression of TFF1 (pS2) and TFF3 was most significantly correlated with bone metastasis in breast cancer (Smid et al., 2006). In addition, it has been shown that plasma levels of TFF1, TFF2 and TFF3 are increased in patients with advanced prostate cancer (Vestergaard et al., 2006). Therefore, coordinate expression of the TFF gene cluster may represent a significant marker in breast and prostate cancer progression. Although part of the upregulated clustered gene expression in tumor samples could be attributed to gene amplification, epigenetic changes in high-order chromatin structure may also lead to dysregulation of

COBRA1 regulates clustered gene expression SE Aiyar et al

2544

coordinated gene expression at the transcription level in human cancers. Despite the abundant evidence for clustering of co-expressed genes in eucaryotic genomes, the mechanistic basis and protein regulators that dictate the coordinated expression of clustered genes are poorly understood. Cofactor of BRCA1 (COBRA1) was first identified as a BRCA1-interacting protein capable of inducing largescale chromatin reorganization (Ye et al., 2001). COBRA1 was subsequently found to be one of the four subunits of the negative elongation factor complex NELF (NELF-B) (Narita et al., 2003). The NELF complex was biochemically purified based on its ability to inhibit transcription elongation by stalling RNA polymerase II (RNAPII) within the promoter-proximal region (Wada et al., 1998; Yamaguchi et al., 1999). Recent in vivo work shows that COBRA1 interacts with the ligand-binding domain of estrogen receptor a (ERa) (Aiyar et al., 2004). In response to estrogen, COBRA1 and the rest of the NELF complex are recruited to a subset of ERa-regulated promoters such as TFF1 (pS2), and inhibit the ligand-dependent gene activation by interfering with the RNAPII movement (Aiyar et al., 2004). To better understand the pattern of gene regulation by COBRA1, we conducted a genome-wide search for COBRA1 target genes by comparing gene expression profiles in control and COBRA1 knockdown breast cancer cells. Our work led to the key discovery that COBRA1 regulates both ligand-dependent and independent expression of cancer-associated genes that tend to cluster on the chromosomes, including the TFF gene cluster. Thus, this finding suggests a link between regulation of transcription elongation and higher order chromatin structure in breast cancer cells.

Results Whole-genome analysis of COBRA1-regulated genes To conduct a genome-wide search for genes regulated by COBRA1, ERa-positive breast cancer cell line T47D was infected with a retrovirus expressing either a control small hairpin RNA (shRNA) against enhanced green fluorescent protein (EGFP) (shEGFP) or shRNA against COBRA1 (sh2A). shRNA-mediated knockdown of COBRA1 was verified by Western blotting (Figure 1a). Total RNA isolated from duplicates of each shRNA-expressing cell line was subjected to microarray analyses using the Affymetrix microarray chips (HG-U133A). Examination of the microarray data revealed a total of 134 genes with a P-value of less than or equal to 0.05 and a fold change in signal intensity greater than 1.5 (Supplementary Table 1). Among these genes, 84 were stimulated (i.e. COBRA1-repressed) and 50 were downregulated (i.e. COBRA1-activated) in the COBRA1 knockdown cells (Supplementary Table 1). As expected, COBRA1 was among the downregulated genes from the initial microarray analysis (reduced 2.96-fold; P-value 0.04), but was not included in the final list of COBRA1Oncogene

Figure 1 Gene ontology analysis of COBRA1-regulated genes as identified by microarray. (a) Immunoblots showing the decreased expression of COBRA1 protein in the COBRA1 knockdown T47D-derived cells (sh2A and shGA). Parental T47D and shEGFP-expressing cells were used as controls. (b) The Ingenuity Gene Ontology Analysis tool is used to identify several biological pathways that are relatively enriched with COBRA1-regulated genes. The P-value for each category is shown on the top of the bar.

regulated genes. Of note, 31 out of the 134 COBRA1regulated genes are either membrane-associated or secretory (solid squares in Supplementary Table 1). Using the Ingenuity Gene Ontology Analysis software program (www.ingenuity.com), we found that the list of COBRA1-regulated genes is enriched with those involved in cell cycle control, antigen presentation and amino-acid synthesis pathways (Figure 1b). A total of 21 COBRA1-regulated genes have been reported to be estrogen-responsive (http://defiant.i2r. a-star.edu.sg/projects/Ergdb-v2/index.htm) (solid triangles in Supplementary Table 1), and 17 of them are repressed by COBRA1 (upregulated in the COBRA1 knockdown cells). This extends our previous observation of the COBRA1-mediated repression of estrogenresponsive transcription at the TFF1 (pS2) and C3 loci (Aiyar et al., 2004). In addition, a large number of COBRA1-regulated genes have been previously implicated in cancer development. These include CD74 in myeloma (Burton et al., 2004), Claudin 1 in colon cancer (Dhawan et al., 2005), AZGP1 in prostate cancer (Lapointe et al., 2004) and the cancer/testis-specific Gantigens (GAGEs) (Bodey, 2002). In particular, multiple COBRA1-regulated genes have been associated with breast cancer. These include APOD (Lopez-Boado et al., 1994; Blais et al., 1995; Rassart et al., 2000), CLDN1 (Kramer et al., 2000; Hoevel et al., 2002; Morin, 2005; Tokes et al., 2005), MGP (Chen et al., 1990; Sheikh et al., 1993; Hirota et al., 1995), GAGEs (Mashino et al., 2001; Duan et al., 2003), MCM2 (Gonzalez et al., 2003; Moggs et al., 2005; Shetty et al., 2005), PIP (Pagani et al., 1994; Clark et al., 1999; Carsol et al., 2002; Tian et al., 2004), TIMP1 (Nakopoulou

COBRA1 regulates clustered gene expression SE Aiyar et al

2545

et al., 2003; Wu¨rtz et al., 2005), WISP2 (Banerjee et al., 2003) and TFF3 (Xu et al., 2005). Confirmation of microarray data by real-time RT–PCR Quantitative real-time reverse transcription–polymerase chain reaction (RT–PCR) analysis was used to verify the microarray results for 15 COBRA1-repressed genes and four COBRA1-stimulated genes (bold in Supplementary Table 1). To alleviate the concern over the ‘off-targeting’ effect commonly associated with shRNA-mediated gene silencing, we also included an independent T47D pool that expressed a different COBRA1-targeted shRNA (shGA; Figure 1a). All 19 genes verified by real-time RT–PCR were affected by COBRA1 knockdown in the same direction as indicated by the microarray. A subset of these genes is shown in Figure 2. For example, consistent with the microarray data, expression of CLDN1, ARHGDIB, WISP2 and TXNIP was stimulated when COBRA1 was reduced by either COBRA1specific shRNA (Figure 2a–d). On the other hand, COBRA1 knockdown led to decreased expression of MYO5C and SPINK5 (Figure 2e and f). Thus, findings from the whole-genome approach can be validated by an independent assay. Clustering of COBRA1-regulated genes A close examination of the genomic locations of the COBRA1-regulated genes revealed an intriguing finding

of non-random distribution of many genes across the genome. In two instances (MGP and ARHGDIB on chromosome 12; DNAL4 and CBX6 on chromosome 22), the two COBRA1-regulated genes on the same chromosome are only separated by one gene. In addition, TFF3 at chromosome 21q22 is clustered with TFF1 (pS2), another member of the TFF and a known COBRA1-repressed gene from our previous work (Aiyar et al., 2004). Lastly, GAGE2 and GAGE7, which belong to the same cancer/testis-specific G antigen family on chromosome X (Xp11) (de Backer et al., 1999), are immediate neighbors of each other. Although some COBRA1-regulated genes such as TFF1/TFF3 and GAGE2/7 belong to gene families, many other clustered COBRA1 target genes are apparently structurally unrelated. A chromosomal gene cluster is defined as two or more genes in which a contiguous start site is within a distance (d) of the neighboring start site. When d is set at 20 Mb, 112 of the 134 COBRA1-regulated genes appear to be clustered on the chromosomes (Figure 3 and Supplementary Table 4). We estimate the significance of this by randomly sampling 134 starts 100 000 times among 13 003 nonoverlapping exemplar sequences from which Affymetrix probe sets were selected. The clustering phenomenon is statistically significant (P ¼ 0.004) with the median number of randomly selected genes found in clusters equal to 80. As illustrated in Figure 3, the clustering effect is observed on all chromosomes except 18.

Figure 2 Verification of the microarray results by real-time RT–PCR. Four cell lines are compared: parental T47D cells (white bars), T47D shEGFP control cells (gray bars), T47D shCOBRA1 cells expressing two independent shRNA oligos (black bars for 2A and striped bars for GA). Examples of COBRA1-repressed genes are shown by (a) CLDN1, (b) ARHGDIB, (c) WISP2 and (d) TXNIP. Examples of COBRA1-stimulated genes are shown by (e) MYO5C and (f) SPINK5. The error bars represent s.d. Results were representatives of at least four independent biological experiments. Oncogene

COBRA1 regulates clustered gene expression SE Aiyar et al

2546

Chromosome 1 has the largest number of COBRA1regulated gene clusters (a total of five), with one of these clusters containing nine genes (1q41–q42).

Figure 3 A schematic diagram of the chromosomal distribution of the 112 clustered COBRA1-regulated genes. Each solid circle on the chromosome represents a COBRA1-regulated gene. See Supplementary Table 3 for the exact chromosomal location of each gene in the clusters.

Analysis of the COBRA1-regulated gene cluster at Xp11 Following the global analysis of the COBRA1-regulated gene expression, we carried out a more in-depth study at specific gene clusters. Aberrant gene expression on the X chromosome has recently been associated with both hereditary and sporadic basal-like breast cancer (Richardson et al., 2006). In addition, several COBRA1repressed genes at the X chromosome including GAGE2/GAGE7 at Xp11.2 and TIMP1 at Xp11.2– 11.4 are proximal to each other and have been implicated in breast cancer development (Chen et al., 1998; Mashino et al., 2001; Bodey, 2002; Duan et al., 2003; Nakopoulou et al., 2003; Wu¨rtz et al., 2005). We therefore examined the effect of COBRA1 knockdown on these and flanking genes by quantitative RT–PCR. As summarized in Figure 4 (top diagram), several additional members of the GAGE gene family as well as FOXP3 were upregulated by COBRA1 knockdown, whereas expression of GAGEB1 and LOC139163 was reduced in the absence of COBRA1. Expression of four other neighboring genes in this genomic region (GAGEC1, LOC286408, PPP1R3F and JM1) was not affected by COBRA1. Except for FOXP3, all COBRA1-affected genes at this locus are contiguous. TIMP1, another COBRA1-repressed gene revealed by the microarray study, is located approximately 2 Mb downstream of the GAGE locus on the X chromosome. The entire TIMP1 gene is situated within the sixth

Figure 4 COBRA1-regulated gene cluster at the Xp11.2–11.4. Quantitative real-time RT–PCR confirmed that COBRA1 modulated a cluster of genes on the X chromosome (see Supplementary Table 5 for the raw data). The solid up block arrows indicate genes that are upregulated in the absence of COBRA1, whereas the open down block arrows indicate genes that are downregulated in the COBRA1 knockdown cells. The gray horizontal bars designate those genes that are not significantly affected by COBRA1 knockdown. The thin horizontal arrows above the gene names designate the transcription directions. Oncogene

COBRA1 regulates clustered gene expression SE Aiyar et al

2547

intron of SYN1, which is transcribed in the opposite direction. Examination of eight genes in this genomic region by quantitative RT–PCR indicates that expression of SYN1 and TIMP1 are elevated by COBRA1 knockdown, whereas none of the flanking genes were affected by COBRA1 (Figure 4, bottom diagram). Of note, expression of the ZNF family (ZNF41, ZNF81 and ZNF157) located in the same region was not regulated by COBRA1, suggesting that COBRA1 is not a general transcription regulator of all gene families. To determine whether COBRA1 was physically associated with any of the affected genes at the Xp11 locus, we carried out chromatin immunoprecipitation (ChIP) in a T47D derivative that expresses a moderate amount of ectopic Flag-COBRA1. This cell line allows for ChIP with both an anti-COBRA1 and anti-Flag antibody. As shown in Figure 5, positive ChIP signals were detected at the promoters of TIMP1 and SYN1, as well as the common promoter for GAGE 1, 2 and 8. The intensity of the ChIP signals varied between the two antibodies at the three COBRA1-regulated promoters (compare lanes 3 and 4), which could be owing to the variation in availability of the epitopes and conformation of the COBRA1 protein at specific loci. Neither antibody revealed any COBRA1 association with the promoters for GAGE4-7/7B, ELK1 on chromosome X or glyceraldehyde-3-phosphate dehydrogenase on chromosome 12. These findings strongly suggest a preferential affinity of COBRA1 for a subset of COBRA1-regulated genes on chromosome X. Characterization of the COBRA1 action at the gene cluster on 21q22 TFF3, which is upregulated by COBRA1 knockdown in the microarray experiment, belongs to the TFF that also includes TFF1 and TFF2. TFF1, the previously known COBRA1 target gene (Aiyar et al., 2004), was upregulated by COBRA1 knockdown as well (3.95-fold). However, it had a P-value of 0.066 and therefore was not included in the list of COBRA1-regulated genes from the microarray analysis. This is most likely due to the exceedingly stringent criteria set for the current microarray analysis (Po0.05), as quantitative RT–PCR using the same RNA for the microarray experiment clearly indicated an elevated TFF1 messenger RNA

(mRNA) level in the COBRA1 knockdown cells (data not shown; but see Figure 8). Given the clinical evidence for an association of upregulation of TFF1 and TFF3 with breast cancer development (Poulsom et al., 1997; Mikhitarian et al., 2005; Xu et al., 2005; Smid et al., 2006), we compared the impact of COBRA1 knockdown on the expression of TFF1/TFF3 and several other neighboring genes in this region. As summarized in Figure 6, TFF1, TFF3 and two additional genes (TMPRSS3 and TSGA2) in the cluster were upregulated by COBRA1 knockdown. On the other hand, expression of TFF2 and UBASH3A was decreased in the COBRA1 knockdown cells, whereas ABCG1 and SLC37A on the outskirts of the region were not affected by COBRA1. Interestingly, all

Figure 5 ChIP demonstrates COBRA1 association with a subset of COBRA1-regulated genes at Xp11.2–11.4. ChIP analysis was carried out using a T47D-derived cell line that ectopically expressed Flag-COBRA1. The input lane represents 1/10 of the total genomic DNA used in each ChIP (lane 1). Samples were immunoprecipitated with IgG (Santa Cruz, Santa Cruz, CA, USA; lane 2), an antiFlag antibody (Sigma Aldrich, St Louis, MO, USA; lane 3) and an anti-COBRA1 antibody (lane 4) (Aiyar et al., 2004). Primer sequences used in the ChIP experiments are listed in Supplementary Table 3.

Figure 6 COBRA1 modulates clustered gene expression at the 21q22.3 chromosomal locus. The solid and open arrows indicate genes that are up- and downregulated in the absence of COBRA1, respectively. The gray bars indicate those genes that are not affected by COBRA1 knockdown. See Supplementary Table 6 for the raw quantitative RT–PCR data. Oncogene

COBRA1 regulates clustered gene expression SE Aiyar et al

2548

COBRA1-repressed genes (i.e. stimulated in the absence of COBRA1) at this locus are transcribed in the same direction. Consistent with our previous finding (Aiyar et al., 2004), ChIP analysis showed that COBRA1 was associated with the TFF1 promoter region (Figure 7). In addition, COBRA1 was present at the promoters of several additional potential target genes in this region, including TFF3, TMPRSS3, TSGA2 and UBASH3A. COBRA1 was not detected at the promoters of TFF2, SLC37A1 or ABCG1. Therefore, COBRA1 is preferentially associated with most of its regulated genes at this particular cluster.

COBRA1 reduction even in the absence of exogenously added estrogen (compare the first and third bars), suggesting an effect of COBRA1 on the ligandindependent transcription as well. On the other hand, TSGA2 expression was not estrogen-responsive, yet its transcription was upregulated by COBRA1 knockdown in charcoal-stripped medium with and without exogenously added estrogen (Figure 8). Therefore, our data suggest that COBRA1 is capable of repressing both ligand-dependent and -independent gene expression at the TFF gene cluster.

Impact of COBRA1 on the estrogen-dependent and -independent transcription at the 21q22 locus Because both TFF1 and TFF3 are responsive to estradiol stimulation (May and Westley, 1997) and ERa has been shown to bind to several regions along chromosome 21 (Carroll et al., 2005), we assessed the impact of COBRA1 on the ligand-dependent and independent transcription of those COBRA1-repressed genes within the TFF-encompassing gene cluster (TFF1, TFF3, TSGA2 and TMPRSS3). Control and COBRA1 knockdown T47D cells were starved in charcoalstripped medium for 3 days, and subsequently treated with 10 nM 17-b-estradiol for 24 h. As shown by quantitative RT–PCR analysis (Figure 8), ligand-dependent transcription of the estrogen-responsive genes (TFF1, TFF3 and TMPRSS3) was significantly elevated in the COBRA1 knockdown cells (compare the second and fourth bars in each panel). In addition, expression of TFF3 and TMPRSS3 was also stimulated by the

Discussion

Figure 7 ChIP reveals association of COBRA1 with multiple COBRA1-regulated genes at the 21q22.3 locus. ChIP was carried out in the same way as described in Figure 5. Oncogene

In vitro biochemical characterization clearly demonstrate that NELF, of which COBRA1 is an integral subunit, modulates transcription elongation by stalling RNAPII at the promoter-proximal region (Yamaguchi et al., 1999; Palangat et al., 2005). Previous in vivo studies in human and Drosophila systems corroborate the RNAPII-stalling function of NELF in gene regulation (Wu et al., 2003, 2005a; Aiyar et al., 2004). Immunostaining of Drosophila polytene chromosomes shows fewer foci for NELF than RNAPII (Wu et al., 2003), suggesting that NELF may not act as a general transcription elongation factor for all RNAPII-dependent gene expression. Findings from our whole-genome

Figure 8 COBRA1 regulates both estrogen-dependent and -independent gene expression at the 21q22.3 chromosomal locus. ShEGFP- and shCOBRA2A-expressing T47D cells were starved for 3 days in medium containing charcoal-stripped serum and then were treated with either vehicle (ethanol; open bars) or 10 nM 17-bestradiol (solid bars) for 24 h. RNA was prepared for real-time RT–PCR analysis of the indicated genes at 21q22.3. Among the COBRA1-repressed genes tested, TMPRSS3, TFF1 and TFF3 are estrogen-responsive, whereas TSGA2 is estrogen-independent.

COBRA1 regulates clustered gene expression SE Aiyar et al

2549

study in breast cancer cells clearly demonstrate that COBRA1 preferentially modulates the expression of a relatively small number of genes that tend to have clustered genomic configurations. Furthermore, the ChIP results indicate a physical presence of COBRA1 in at least a subset of these specific genes and chromosomal regions. A specialized function of NELF in regulating more complex chromatin structure and hormone-dependent gene expression in higher eucaryotes would be consistent with the absence of their structural homologs in unicellular eucaryotes. Our previous gene-targeted investigation in breast cancer cells strongly suggests that COBRA1 represses ligand-dependent gene activation by ERa at a subset of estrogen-responsive genes including TFF1 (pS2). Findings from the unbiased approach in the current study corroborate and extend the previous observation by uncovering an additional set of known estrogenresponsive genes that are regulated by COBRA1. Our new data also suggest that COBRA1 can modulate both ligand-dependent and -independent transcription of ER-responsive promoters (Figure 8). However, it is worth noting that the majority of COBRA1regulated genes uncovered in our study have not been reported to be estrogen-responsive, and this includes those that were verified by the real-time RT–PCR analysis (e.g. TSGA2). Therefore, in addition to its role in modulation of ER-dependent gene expression, COBRA1 may also regulate transcription via its interactions with other site-specific transcription factors. Consistent with this notion, COBRA1 was found to interact with AP1 and represses its transcriptional activity (Zhong et al., 2004). An important finding from the current study is the clustered organization of many COBRA1-regulated genes. This is manifested in the following two aspects. First, in several occasions, two COBRA1-regulated gene are localized next to each other or only separated by a third gene, events that are unlikely to occur by chance. Second, many of the genes identified by the microarray study are localized within relatively small regions on the chromosomes. Although several COBRA1-regulated genes (e.g. GAGEs and TFF1-3) belong to tandemlocalized gene families, most of the 112 clustered COBRA1-regulated genes are structurally unrelated. We suspect that more COBRA1-regulated genes would be found in individual clusters if less stringent filtering criteria were applied to the microarray data. For instance, several genes at the Xp11 and 21q22 clusters (Figures 4 and 6) are unfiltered in the microarray yet identified as COBRA1-regulated by the real-time PCR assay. We find that 80% of these genes are in fact affected by COBRA1 knockdown in the same direction in both assays, suggesting that microarray-based findings may have underestimated the actual number of clustered COBRA1-regulated genes. It is also worth pointing out that the synteny of many of the COBRA1regulated gene clusters is conserved between mouse and human. For example, the order of the neighboring genes at the TFF-encompassing cluster (Figure 6) is maintained between chromosome 21 in human and

chromosome 17 in mouse, consistent with the notion that the co-expressed gene clusters are conserved owing to functional selection (Singer et al., 2005). The underlying mechanisms for the COBRA1-regulated clustered gene regulation remain to be elucidated. At least two possible scenarios could be envisioned. First, as a subunit of the NELF complex, COBRA1 may be simultaneously recruited to multiple promoters within the same chromosomal region via its interactions with a common DNA-binding transcription factor such as ERa at these promoters. The NELF complex that is present at each promoter may independently repress transcription of individual genes in the cluster. This mode of action is likely to occur for those duplicated gene families that may share common cis-regulatory promoter elements. For example, the three genes of the TFF family at 21q22 display similar patterns in the tissue-selective expression and regulation, and therefore have been suggested to share a common locus activation region (Tomasetto et al., 1992; Chinery et al., 1995; Gott et al., 1996). In an alternative scenario, the COBRA1-mediated regulation of clustered gene expression may invoke a chromatin-based mechanism that is associated with, but distinct from its role in transcription elongation. In such an event, COBRA1 might be recruited initially to the promoter region of one of the regulated genes in the cluster, and its impact on transcription may be subsequently propagated to the neighboring genes via a putative influence of COBRA1 on higher-order chromatin structure. The second model would be in line with our previous observation that chromosome-tethered COBRA1 is capable of inducing large-scale chromatin reorganization (Ye et al., 2001). Given the known function of NELF in repressing transcriptional elongation, it is interesting to note that a significant number of genes are stimulated by COBRA1 (i.e. their expression downregulated in the COBRA1 knockdown cells). ChIP analysis failed to detect any COBRA1 association with several of the most downregulated genes in the COBRA1 knockdown cells (data not shown), suggesting that the observed effect of COBRA1 on these genes might be indirect. However, COBRA1 was detected at the promoter of UBASH3A, one of the genes in the TFF-containing gene cluster that were downregulated by COBRA1 knockdown (Figure 7). Therefore, the possibility of a direct gene activation function by COBRA1 cannot be excluded at this point. The TIMP1 and TFF1 genes that were chosen for in-depth characterization are overexpressed in breast cancer, making them promising prognostic markers (Hahnel et al., 1994; Toi et al., 1998; Balleine and Clarke, 1999; Fata et al., 2004). In addition, expression of both genes may be regulated by the breast cancer tumor suppressor BRCA1. The TIMP1 gene frequently escapes from inactivation at the inactive X chromosome allele (Anderson and Brown, 1999, 2002), and BRCA1 has been shown to participate in X chromosome inactivation (Ganesan et al., 2002). Furthermore, elevated TIMP1 expression was found in BRCA1 mutation-associated ovarian tumors and ‘BRCA1-like’ Oncogene

COBRA1 regulates clustered gene expression SE Aiyar et al

2550

sporadic ovarian tumors (Jazaeri et al., 2002, 2004). Likewise, it has been shown that BRCA1 directly binds to the TFF1 (pS2) promoter and represses its transcription (Fan et al., 2001; Zheng et al., 2001). In light of the physical interaction between COBRA1 and BRCA1, it is tempting to speculate that these two proteins may act jointly to downregulate transcription from a common set of breast cancer-relevant genes such as TFF1 and TIMP1. Many COBRA1-regulated genes that are uncovered in the current study have been previously associated with various types of cancer, consistent with the reported functions of COBRA1 in cancer cell proliferation and tumorigenesis (Aiyar et al., 2004; McChesney et al., 2006). A search with the Oncomine Cancer Profiling Database (www.oncomine.org) indicates that expression of multiple COBRA1-regulated genes within a given cluster is deregulated in cancer. For example, four out of the eight COBRA1-regulated genes at the cluster on chromosome 17 are deregulated in prostate cancer (CPD, DUSP3, SMARCD2 and PSMD12). The same is true for all three clustered genes on chromosome 16 (GOT2, TK2, and MBTPS1) in prostate and leukemia. Likewise, altered expression of the TFF gene family has been implicated in lung cancer. In addition, two recent clinical studies show that concurrent overexpression of the TFF family members is associated with advanced breast and prostate cancers (Smid et al., 2006; Vestergaard et al., 2006). Conceivably, alteration of COBRA1 activity and/or expression may contribute to the deregulation of transcription at these clusters in various cancer types, a notion that merits testing in future studies.

Materials and methods Cell lines and culture T47D cells lines were obtained from the American Type Culture Collection (Manassas, VA, USA), and grown in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum, penicillin (100 U/ml) and streptomycin (100 mg/ml). shRNA constructs shRNA-mediated gene silencing was achieved using the pSuperRetro plasmid purchased from Oligoengine (Seattle, WA, USA). Two different oligonucleotides directed against the human COBRA1 (NELF-B) gene were used: COBRA-2A (50 -GTTCCACCAGTCGGTATTC-30 ) and COBRA-GA (50 GACCTTCTGGAGAAGAGCT-30 ). As a negative control, an shRNA directed against EGFP was used (50 -GAACGG CATCAAGGTGAAC-30 ). Viral supernatants were prepared and used to infect T47D cells, and stable transfectants were selected with 10 mg/ml puromycin as described previously (Aiyar et al., 2004). Microarray and data analysis shRNA-expressing T47D cells were grown to 70% confluency and harvested for total RNA. Duplicates of the control EGFP and COBRA1-2A shRNA cell lines were subjected to microarray analyses, using the human gene array chips (HgU-133A) from Affymetrix as described earlier (Wu et al., Oncogene

2005b). Alteration of gene expression by COBRA1 knockdown was considered significant if the P-value was less than or equal to 0.05, the fold change in signal intensity (increase or decrease) was greater than 1.5 and the absolute difference in signal intensity was greater than 100. DMT chip software package was used to analyse the microarray data (Gallagher et al., 2003). The P-values were computed by a non-parametric Wilcoxon’s rank test comparing the population of PM probes and MM probes for each gene on the chip (http://genes.med. virginia.edu). RNA isolation and cDNA synthesis Total RNA was extracted with TRIzol Reagent according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA, USA). The quality of the RNA was verified by absorbance measurements at 260 and 280 nm. To prepare cDNA, approximately 1 mg of RNA was reverse-transcribed using ImPromp (Promega, Madison, WI, USA). Quantitative real-time RT–PCR Results obtained from the Affymetrix microarray were verified by real-time RT–PCR for those genes highlighted in bold face in Supplementary Table 1. For the real time RT–PCR experiments, a neomycin-selected EGFP shRNA-expressing cell pool was used. Primer Express software (Applied Biosystems, Foster City, CA, USA) was used to design primers that cover two neighboring exons (Supplementary Table 2). SYBR Green technology and an ABI7300 real-time PCR machine were used for the quantitative real-time PCR analysis. Relative mRNA levels were determined and standard curves were generated by a serial dilution of cDNA. Expression levels were normalized against b-actin. Results were confirmed with at least four independent experiments. To assess the mRNA abundance of GAGE1 from the highly homologous GAGE gene family on Xp11.2, a primer set was designed based on a unique 143-bp insertion in GAGE1. Because of the high degree of nucleotide identity between GAGE2 and GAGE8, Taqman primer probe sets from Applied Biosystems were used to measure separately the mRNA abundance of these two GAGE genes. A previously described primer set (VV1 and VDE24) (de Backer et al., 1999) was used in semiquantitative RT–PCR to measure the collective expression levels of GAGE-3, -4, -5, -6 and -7B. Chromatin immunoprecipitation ChIP assays were carried out according to a previously described protocol (Aiyar et al., 2004). Primer sequences for the ChIP assay are listed in Supplementary Table 3. Immunoblotting An equal amount of whole-cell lysates was resolved by 10% sodium dodecyl sulfate–polyacrylamide gel electrophoresis. The proteins were detected by anti-COBRA1 (Aiyar et al., 2004) and anti-a-tubulin (Calbiochem, San Diego, CA, USA) antibodies. Acknowledgements We thank Asma Amleh and Jianlong Sun for critical reading of the paper, and Yongde Bao for the microarray analysis. The work was supported by a postdoctoral fellowship from DOD. Breast Cancer Research Program to SEA (BC031441), a Cancer Training Grant to ALB and an NIH grant to RL (DK064604).

COBRA1 regulates clustered gene expression SE Aiyar et al

2551 References Aiyar SE, Sun J-L, Blair AL, Moskaluk CA, Lv Y, Ye Q-N et al. (2004). Attenuation of estrogen receptor alphamediated transcription through estrogen-stimulated recruitment of a negative elongation factor. Genes Dev 18: 2134–2146. Anderson CL, Brown CJ. (1999). Polymorphic X-chromosome inactivation of the human TIMP1 gene. Am J Hum Genet 65: 699–708. Anderson CL, Brown CJ. (2002). Variability of X chromosome inactivation: effect on levels of TIMP1 RNA and role of DNA methylation. Hum Genet 110: 271–278. Balleine RL, Clarke CL. (1999). Expression of the oestrogen responsive protein pS2 in human breast cancer. Histol Histopathol 14: 571–578. Banerjee S, Saxena N, Sengupta K, Tawfik O, Mayo MS, Banerjee SK. (2003). WISP-2 gene in human breast cancer: estrogen and progesterone inducible expression and regulation of tumor cell proliferation. Neoplasia 5: 63–73. Blais Y, Sugimoto K, Carriere MC, Haagensen DE, Labrie F, Simard J. (1995). Interleukin-6 inhibits the potent stimulatory action of androgens, glucocorticoids and interleukin-1 alpha on apolipoprotein D and GCDFP-15 expression in human breast cancer cells. Int J Cancer 62: 732–737. Bodey B. (2002). Cancer-testis antigens: promising targets for antigen directed antineoplastic immunotherapy. Expert Opin Biol Ther 2: 577–584. Burton JD, Ely S, Reddy PK, Stein R, Gold DV, Cardillo TM et al. (2004). CD74 is expressed by multiple myeloma and is a promising target for therapy. Clin Cancer Res 10: 6606–6611. Caron H, van Schaik B, van der Mee M, Baas F, Riggins G, van Sluis P et al. (2001). The human transcriptome map: clustering of highly expressed genes in chromosomal domains. Science 291: 1289–1292. Carroll JS, Liu S, Brodsky AS, Li W, Meyer CA, Szary AJ et al. (2005). Chromosome-wide mapping of estrogen receptor binding reveals long-range regulation requiring the forkhead protein FoxA1. Cell 122: 33–43. Carsol JL, Gingras S, Simard J. (2002). Synergistic action of prolactin (PRL) and androgen on PRL-inducible protein gene expression in human breast cancer cells: a unique model for functional cooperation between signal transducer and activator of transcription-5 and androgen receptor. Mol Endocrinol 16: 1696–1710. Chen L, O’Bryan JP, Smith HS, Liu E. (1990). Overexpression of matrix Gla protein mRNA in malignant human breast cells: isolation by differential cDNA hybridization. Oncogene 5: 1391–1395. Chen ME, Lin SH, Chung LW, Sikes RA. (1998). Isolation and characterization of PAGE-1 and GAGE-7. New genes expressed in the LNCaP prostate cancer progression model that share homology with melanoma-associated antigens. J Biol Chem 273: 17618–17625. Chinery R, Williamson J, Poulsom R. (1995). The gene encoding human intestinal trefoil factor (TFF3) is located on chromosome 21q22.3 clustered with other members of the trefoil peptide family. Genomics 32: 281–284. Cho RJ, Campbell MJ, Winzeler EA, Steinmetz L, Conway A, Wodicka L et al. (1998). A genome-wide transcriptional analysis of the mitotic cell cycle. Mol Cell 2: 65–73. Clark JW, Snell L, Shiu RP, Orr FW, Maitre N, Vary CP et al. (1999). The potential role for prolactin-inducible protein (PIP) as a marker of human breast cancer micrometastasis. Br J Cancer 81: 1002–1008.

de Backer O, Arden KC, Boretti M, Vantomme V, De Smet C, Czekay S et al. (1999). Characterization of the GAGE genes that are expressed in various human cancers and normal testis. Cancer Res 59: 3157–3165. Dhawan P, Singh AB, Deane NG, No Y, Shiou SR, Schmidt C et al. (2005). Claudin-1 regulates cellular transformation and metastatic behavior in colon cancer. J Clin Invest 115: 1765–1776. Duan Z, Duan Y, Lamendola DE, Yusuf RZ, Naeem R, Penson RT et al. (2003). Overexpression of MAGE/GAGE in paclitaxel/doxorubicin-resistant human cancer cell lines. Clin Cancer Res 9: 2778–2785. Fan S, Ma YX, Wang C, Yuan RQ, Meng Q, Wang JA et al. (2001). Role of direct interaction in BRCA1 inhibition of estrogen receptor activity. Oncogene 20: 77–87. Fata JE, Werb Z, Bissell MJ. (2004). Regulation of mammary gland branching morphogenesis by the extracellular matrix and its remodeling enzymes. Breast Cancer Res 6: 1–11. Gallagher PG, Bao Y, Serrano SMT, Kamiguti AS, Theakston RDG, Fox JW. (2003). Use of microarrays for investigating the subtoxic effects of snake venoms: insights into venominduced apoptosis in human umbilical vein endothelial cells. Toxicon 41: 429–440. Ganesan S, Silver DP, Greenberg RA, Avni D, Drapkin R, Miron A et al. (2002). BRCA1 supports XIST RNA concentration on the inactive X chromosome. Cell 111: 393–405. Gonzalez MA, Pinder SE, Callagy G, Vowler SL, Morris LS, Bird K et al. (2003). Minichromosome maintenance protein 2 is a strong independent prognostic marker in breast cancer. J Clin Oncol 21: 4306–4313. Gott P, Beck S, Machado JC, Carneiro F, Schmitt H, Blin N. (1996). Human trefoil peptides: genomic structure in 21q22.3 and coordinated expression. Eu J Human Gen 4: 308–315. Hahnel E, Harvey J, Robbins P, Sterrett G, Hahnel R. (1994). Hormone-regulated genes (pS2, PIP, FAS) in breast cancer and nontumoral mammary tissue. Pathobiology 62: 82–89. Hirota S, Ito A, Nagoshi J, Takeda M, Kurata A, Takatsuka Y et al. (1995). Expression of bone matrix protein messenger ribonucleic acids in human breast cancers. Possible involvement of osteopontin in development of calcifying foci. Lab Invest 72: 64–69. Hoevel T, Macek R, Mundigl O, Swisshelm K, Kubbies M. (2002). Expression and targeting of the tight junction protein CLDN1 in CLDN1-negative human breast tumor cells. J Cell Physiol 191: 60–68. Hurst LD, Pal C, Lercher MJ. (2004). The evolutionary dynamics of eukaryotic gene order. Nat Rev Genet 5: 299–310. Jazaeri AA, Chandramouli GVR, Aprelikova O, Nuber UA, Sotiriou C, Liu ET et al. (2004). BRCA1-mediated repression of select X chromosome genes. J Translational Med 2: 32–39. Jazaeri AA, Yee CJ, Sotiriou C, Brantley KR, Boyd J, Liu ET. (2002). Gene expression profiles of BRCA1-linked, BRCA2linked, and sporadic ovarian cancers. J Natl Cancer Inst 94: 990–1000. Koon N, Zaika A, Moskaluk CA, Frierson HF, Knuutila S, Powell SM et al. (2004). Clustering of molecular alterations in gastroesophageal carcinomas. Neoplasia 6: 143–149. Kramer F, White K, Kubbies M, Swisshelm K, Weber BH. (2000). Genomic organisation of claudin 1 and its assessment in hereditary and sporadic breast cancer. Hum Genet 107: 249–256. Oncogene

COBRA1 regulates clustered gene expression SE Aiyar et al

2552 Lapointe J, Li C, Higgins JP, van de Rijn M, Bair E, Montgomery K et al. (2004). Gene expression profiling identifies clinically relevant subtypes of prostate cancer. Proc Natl Acad Sci USA 101: 811–816. Lercher MJ, Urrutia AO, Hurst LD. (2002). Clustering of housekeeping genes provides a unified model of gene order in the human genome. Nat Genet 31: 180–183. Lopez-Boado YS, Tolivia J, Lopez-Otin C. (1994). Apolipoprotein D gene induction by retinoic acid is concomitant with growth arrest and cell differentiation in human breast cancer cells. J Biol Chem 269: 26871–26878. Mashino K, Sadanaga N, Tanaka F, Yamaguchi H, Nagashima H, Inoue H et al. (2001). Expression of multiple cancer-testis antigen genes in gastrointestinal and breast carcinomas. Br J Cancer 85: 713–720. May FEB, Westley BR. (1997). Trefoil proteins: their role in normal and malignant cells. J Path 183: 4–7. McChesney PA, Aiyar SE, Lee OJ, Zaika A, Moskaluk C, Li R et al. (2006). Cofactor of BRCA1: a novel transcription factor regulator in upper gastrointestinal adenocarcinomas. Cancer Res 66: 1346–1353. Mikhitarian K, Gillanders WE, Almeida JS, Hebert Martin R, Varela JC, Metcalf JS et al. (2005). An innovative microarray strategy identities informative molecular markers for the detection of micrometastatic breast cancer. Clin Cancer Res 11: 3697–3704. Moggs JG, Murphy TC, Lim FL, Moore DJ, Stuckey R, Antrobus K et al. (2005). Anti-proliferative effect of estrogen in breast cancer cells that re-express ERa is mediated by aberrant regulation of cell cycle genes. J Mol Endocrinol 34: 535–551. Morin PJ. (2005). Claudin proteins in human cancer: promising new targets for diagnosis and therapy. Cancer Res 65: 9603–9606. Nakopoulou L, Giannopoulou I, Lazaris AC, Alexandrou P, Tsirmpa I, Markaki S et al. (2003). The favorable prognostic impact of tissue inhibitor of matrix metalloproteinases-1 protein overexpression in breast cancer cells. APMIS 111: 1027–1036. Narita T, Yamaguchi Y, Yano K, Sugimoto S, Chanarat S, Wada T et al. (2003). Human transcription elongation factor NELF: identification of novel subunits and reconstitution of the functionally active complex. Mol Cell Biol 23: 1863–1873. Pagani A, Sapino A, Eusebi V, Bergnolo P, Bussolati G. (1994). PIP/GCDFP-15 gene expression and apocrine differentiation in carcinomas of the breast. Virchows Arch 425: 459–465. Palangat M, Renner DB, Price DH, Landick R. (2005). A negative elongation factor for human RNA polymerase II inhibits the anti-arrest transcript-cleavage factor TFIIS. Proc Natl Acad Sci USA 102: 15036–15041. Poulsom R, Hanby AM, Lalani EN, Hauser F, Hoffmann W, Stamp GW. (1997). Intestinal trefoil factor (TFF 3) and pS2 (TFF 1), but not spasmolytic polypeptide (TFF 2) mRNAs are co-expressed in normal, hyperplastic, and neoplastic human breast epithelium. J Pathol 183: 30–38. Rassart E, Bedirian A, Do Carmo S, Guinard O, Sirois J, Terrisse L et al. (2000). Apolipoprotein D. Biochim Biophys Acta 1482: 185–198. Richardson AL, Wang ZC, De Nicolo A, Lu X, Brown M, Miron A et al. (2006). X chromosomal abnormalities in basal-like human breast cancer. Cancer Cell 9: 121–132. Schmitt H, Wundrack I, Beck S, Gott P, Welter C, Shizuya H et al. (1996). A third P-domain polypeptide gene (TFF3), Oncogene

human intestinal trefoil factor, maps to 21q22.3. Cytogenet Cell Genet 72: 299–302. Sheikh MS, Shao Z-M, Chen J-C, Fontana JA. (1993). Differential regulation of matrix Gla protein (MGP) gene expression by retinoic acid and estrogen in human breast carcinoma cells. Mol Cell Endocinol 92: 153–160. Shetty A, Loddo M, Fanshawe T, Prevost AT, Sainsbury R, Williams GH et al. (2005). DNA replication licensing and cell cycle kinetics of normal and neoplastic breast. Br J Cancer 93: 1295–1300. Singer GA, Lloyd AT, Huminiecki LB, Wolfe KH. (2005). Clusters of co-expressed genes in mammalian genomes are conserved by natural selection. Mol Biol Evol 22: 767–775. Smid M, Wang Y, Klijn JG, Sieuwerts AM, Zhang Y, Atkins D et al. (2006). Genes associated with breast cancer metastatic to bone. J Clin Oncol 24: 2261–2267. Spellman PT, Rubin GM. (2002). Evidence for large domains of similarly expressed genes in the Drosophila genome. J Biol 1: 5. Tian W, Osawa M, Horiuchi H, Tomita Y. (2004). Expression of the prolactin-inducible protein (PIP/GCDFP15) gene in benign epithelium and adenocarcinoma of the prostate. Cancer Science 95: 491–495. Toi M, Ishigaki S, Tominaga T. (1998). Metalloproteinases and tissue inhibitors of metalloproteinases. Breast Cancer Res Treat 52: 113–124. Tokes AM, Kulka J, Paku S. (2005). Claudin-1, -3 and -4 proteins and mRNA expression in benign and malignant breast lesions: a research study. Breast Cancer Res 7: R296–R305. Tomasetto C, Rockel N, Mattei MG, Fujita R, Rio MC. (1992). The gene encoding the human spasmolytic protein (SML1/hSP) is in 21q22.3, physically linked to the homologous breast cancer gene BCEI/pS2. Genomics 13: 1328–1330. Vestergaard EM, Borre M, Poulsen SS, Nexo E, Torring N. (2006). Plasma levels of trefoil factors are increased in patients with advanced prostate cancer. Clin Cancer Res 12: 807–812. Wada T, Takagi T, Yamaguchi Y, Ferdous A, Imai T, Hirose H et al. (1998). DSIF, a novel transcription elongation factor that regulates RNA polymeerase II processitivity, is composed of human Spt4 and Spt5 homologs. Genes Dev 12: 343–356. Wu CH, Lee C, Fan R, Smith MJ, Yamaguchi Y, Handa H et al. (2005a). Molecular characterization of Drosophila NELF. Nucleic Acids Res 33: 1269–1279. Wu C-H, Yamaguchi Y, Benjamin LR, Horvat-Gordon M, Washinsky J, Enerly E et al. (2003). NELF and DSIF cause promoter proximal pausing on the hsp70 promoter in Drosophila. Genes Dev 17: 1402–1414. Wu Y, Ghosh S, Nishi Y, Yanase T, Nawata H, Hu Y-F. (2005b). The orphan nuclear receptors NURR1 and NGFIB modulate aromatase gene expression in ovarian granulosa cells: a possible mechanism for repression of aromatase expression upon luteinizing hormone surge. Endocrinology 146: 237–246. Wu¨rtz SØ, Schrohl A-S, Srensen NM, Lademann U, Christensen IJ, Mouridsen H et al. (2005). Tissue inhibitor of metalloproteinases-1 in breast cancer. Endocr Relat Cancer 12: 215–227. Xu XQ, Emerald BS, Goh EL, Kannan N, Miller LD, Gluckman PD et al. (2005). Gene expression profiling to identify oncogenic determinants of autocrine human growth hormone in human mammary carcinoma. J Biol Chem 280: 23987–24003.

COBRA1 regulates clustered gene expression SE Aiyar et al

2553 Yamaguchi Y, Takagi T, Wada T, Yano K, Furuya A, Sugimoto S et al. (1999). NELF, a multisubunit complex containing RD, cooperates with DSIF to repress RNA polymerase II elongation. Cell 97: 41–51. Ye Q, Hu Y-F, Zhong H, Nye AC, Belmont AS, Li R. (2001). BRCA1-induced large-scale chromatin unfolding and allelespecific effects of cancer-predisposing mutations. J Cell Biol 155: 911–921. Zheng L, Annab LA, Afshari CA, Lee W-H, Boyer TG. (2001). BRCA1 mediates ligand-independent transcriptional

repression of the estrogen receptor. Proc Natl Acad Sci USA 98: 9587–9592. Zhong H, Zhu J, Zhang H, Ding L, Sun Y, Huang C et al. (2004). COBRA1 inhibits AP-1 transcriptional activity in transfected cells. Biochem Biophys Res Commun 325: 568–573. Zhou Y, Luoh SM, Zhang Y, Watanabe C, Wu TD, Ostland M et al. (2003). Genome-wide identification of chromosomal regions of increased tumor expression by transcriptome analysis. Cancer Res 63: 5781–5784.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Oncogene