| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Molecular Oncology, Markers, Clinical Correlates |
1 National Cancer Centre/ 2 Defence Medical and Environmental Research Institute, and 3 Department of Pathology, Singapore General Hospital, Republic of Singapore
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Experimental Design and Results: An analysis of expression profiles generated from 11 nonmalignant breast tissues, 17 ductal carcinomas in situ (DCIS) and 98 invasive carcinomas identified three broad molecular subtypes of breast [estrogen receptor (ER)+, ERBB2+ and ER–] in the Asian-Chinese population. These subtypes were highly similar to the "Luminal," "ERBB2+," and "Basal" molecular subtypes defined in previous studies, and the subtype-specific expression signatures were also observed in preinvasive DCIS tumors. By comparing the expression profiles of nonmalignant DCIS and invasive breast cancers for two subtypes (ER+ and ERBB2+), we identified several genes that were regulated in both a common and subtype-specific manner during the normal/DCIS and DCIS/invasive carcinoma transitions. Several of these genes were validated by comparison with another recently published similar, but not identical, study.
Conclusions: Our results suggest that molecularly similar subtypes of breast cancer are indeed broadly conserved between Asian and Caucasian patients, and that these subtypes are already present at the preinvasive stage of carcinogenesis. To our knowledge, this study is among the first to directly compare the expression profiles of breast tumors across two different ethnic populations.
|
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Recently, several groups have found that breast tumors can be classified into different "molecular subtypes" based on their global expression profiles (9, 10, 11, 12, 13, 14) . Many of these subtypes were discernible by gene expression profiling but not by more conventional methodologies and were associated with distinct clinical outcomes, demonstrating the clinical utility of using gene expression information to develop a molecular taxonomy of cancer. A potential limitation of these studies, however, is that they were primarily based on United States and European patient populations. Because of the observed clinical differences in breast cancer between Asian-Chinese and Caucasian patients (described above), we investigated, in this report, whether similar molecular subtypes could also be observed in a predominantly Chinese population, or whether breast tumors between these different ethnic populations might also differ at the gene expression level. By surveying the expression profiles of preinvasive and invasive cancers obtained from predominantly Chinese patients, we found that many, but not all, of the molecular subtypes and their associated subtype-specific gene expression signatures were indeed conserved between Caucasian and Chinese patients, suggesting that the molecular subtypes defined using expression-based genomics are highly robust and may be population independent. We also found, for the first time, that the subtype-specific expression signatures were also present in preinvasive breast cancers [ductal carcinomas in-situ (DCIS)], indicating that these molecular subtypes can already be discerned even at the preinvasive stage of carcinogenesis. By comparing the expression data from normal tissue, DCIS, and invasive carcinomas (IDCs) belonging to two specific subtypes (ER+ and ERBB2+), we identified several genes that were regulated in both a common and subtype-specific manner during the normal/DCIS and DCIS/IDC transitions. Many of these genes were then validated by comparison with publicly available data from another recently published, related but not identical, study (15) . The identities of these genes may prove useful in elucidating the molecular events regulating tumor progression in distinct molecular subtypes of breast cancer.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
10% of carcinoma cells showing nuclear reactivity of at least +2 intensity. For ERBB2 immunohistochemistry, the Dako classification system was used with scores of 0 and 1+ considered negative and 2+ and 3+ considered positive. An indeterminate conclusion was made when benign breast epithelium was immunoreactive. Profiled invasive tumors contained at least 50% tumor content, whereas DCIS samples contained 20–30% (see Results). Confirmation of the DCIS status of samples was achieved by conventional H&E staining of archival samples, as well as direct cryosections of the sample that was processed for expression profiling. Four of the DCIS samples were pure DCIS, and the other samples were DCIS adjacent to invasive tumors. The clinical characteristics of the invasive and DCIS tumors (e.g., tumor size, nodal status, histological grade and type, ER/progesterone receptor status, and ERBB2 status) are presented in the Supplementary Information.4
Sample Preparation and Microarray Hybridization.
RNA was extracted from tissues using Trizol reagent (Invitrogen, Carlsbad, CA), purified through a Qiagen Spin Column (Qiagen; Valencia, CA), and processed for Affymetrix Genechip (Affymetrix Inc., Santa Clara, CA) hybridization according to the manufacturer’s instructions. Hybridizations were performed using Affymetrix U133A Genechips. Annotations assigned to array probes are based on the Dec 2003 release available from the Affymetrix website.5
Data Processing and Analysis.
Raw Genechip scans were quality controlled using a commercially available software (Genedata Refiner; Genedata, Basel, Switzerland) and were deposited into a central data storage facility. The expression data were filtered by (a) removing genes the expression of which was absent in all of the samples (i.e., "A" calls), and (b) performing a log2 transformation, and were normalized by median centering all remaining genes for each sample. Average-linkage hierarchical clustering using a Pearson correlation metric was performed using CLUSTER software and were displayed by TREEVIEW (16)
. Wilcoxon tests, which are nonparametric alternative methodologies to conventional t tests, were used to identify genes whose expression levels were significantly different between two groups, using a 2-fold cutoff and P value of <0.01. Support Vector Machines (SVMs) are classification algorithms that define a discrimination surface in the used feature (gene) space that attempts to maximally separate classes of training data (17)
. Both Wilcoxon tests and SVM analyses were performed using Genedata Analyst software. Random permutation assays, performed to obtain estimates of potential false discovery error rates, were implemented as described in the Results. Selection of random gene sets was performed using a starting population of 7,288 genes, corresponding to the same gene set on which the normal versus DCIS, or DCIS/IDC Wilcoxon analysis was performed. The various gene sets are available for download at http://www.omniarray.com/DCIS.html. Kolmogorov-Smirnov analysis was performed as described in Lamb et al. (18)
. Briefly, we first used Significance Analysis of Microarrays (SAM; ref. 19
) to identify the top genes exhibiting the strongest expression differences between the ER+ and ER– subtypes in our data set, and then calculated the distribution of these genes in the Stanford data set (comprising primarily Caucasian patients) using the Kolmogorov-Smirnov nonparametric rank statistic. In this analysis, the genes in the Stanford data set were ranked according to their signal to noise ratio (20)
corresponding to the ER+/ER– class distinction. The significance of an observed distribution was estimated through a series of random permutations (see Supplementary Information for more details).4
|
RESULTS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
, the breast tumors self-segregated into three major subgroups, hereon referred to as ER+, ER–, and ERBB2+. A similar segregation pattern was also observed when the tumors were reclustered using principal components analysis (PCA), an independent analytical technique (Fig. 1C)
. There was also good agreement between these molecular subgroups and conventional immunohistochemistry (Supplementary Information),4
because all of the ER+ tumors identified by molecular profiling were also ER+ by conventional immunohistochemistry, and almost all (15 of 17) of the ER– tumors identified by molecular profiling were also ER– by immunohistochemistry. For the ERBB2+ subtype, 18 tumors for which ERBB2 immunohistochemistry had been performed were all ERBB2+ by immunohistochemistry as well.
|
ER– Subtype.
The ER– subtype, similar to the "basal" subtype described in other studies,9-11 was characterized by high levels of markers of the basal mammary epithelia, such as keratin 5 and 17 and the serine proteinase inhibitor maspin, a tamoxifen-inducible gene expressed in an inverse fashion to ER (21)
. Notably, SFRP1, a modulator of Wnt signaling, was also highly expressed in this group, a finding also reported by others (11)
.
ERBB2+ Subtype.
The ERBB2+ subtype was associated with high expression levels of the ERBB2 receptor and other genes physically linked to the 17q21 locus, such as PNMT (22)
and PPARBP, suggesting the presence of DNA amplification. Interestingly, many of the ERBB2 tumors also exhibited low expression levels of the ER-related gene expression signature, which may reflect the consequences of cellular cross-talk between ERBB2 and ER signaling (23)
.
The observation that breast tumors from Asian-Chinese patients can also be segregated into distinct ER+, ER– and ERBB2+ subtypes suggests that these subtypes may be independent of specific ethnic population. However, it is still formally possible that the apparent similarities between the Asian and Caucasian subtypes might be limited to only a few well-known marker genes, such as ER, ERBB2, and certain keratins. To address this issue, we determined the statistical commonality of the subtype specific gene expression signatures between our Asian patient population and a previously published Caucasian cohort (10)
. Specifically, we used Kolmogorov-Smirnov analysis (18)
to determine whether genes exhibiting strong differences in expression between the ER+ and ER– subtypes in the Asian data set also exhibited a similar behavior in the Caucasian data set. As shown in Fig. 2
, even after removing several "well-known" marker genes, such as ESR1, GATA3, and TFF3, from the analysis (see Fig. 2
legend), the gene expression signatures associated with the ER+ subtype in our Asian population still exhibited a highly significant degree of commonality with tumors from the Caucasian cohort (P < 1 x 10–5). These results suggest that these subtypes are molecularly conserved between the ethnic groups not simply at the level of a few surface markers but also at the deeper cellular level of biological signaling pathways and transcriptional networks.
|
. Compared with the invasive tumor samples, the percentage of malignant cells in the DCIS samples was comparatively smaller (typically 20–30%). To assess whether the higher percentage of nonmalignant cells in the DCIS samples might compromise the resultant tumor expression profile, we compared the expression profiles of the DCIS samples with a series of nonmalignant breast tissue samples using a PCA (Fig. 3E)
. The DCIS profiles were distinctly different from the nonmalignant profiles, suggesting at a first approximation that the higher percentage of nonmalignant cells does not severely confound the overall tumor expression profile. Expression profiles of the DCIS samples were generated and compared with their invasive counterparts using the 367-gene set of Fig. 1A
and unsupervised clustering. The DCIS samples intermingled with the invasive tumors across distinct categories, with eight DCIS samples segregating into the ER+ subtype, eight into the ERBB2+ subtype, and one into the ER– subtype (vertical lines, Fig. 3F
). This segregation pattern was also confirmed using PCA (Supplementary Information).4
This result indicates that DCIS tumors, despite being preinvasive, can already be segregated into distinct molecular subtypes, suggesting that the hallmark expression signatures defining the different subtypes may already be present at the preinvasive stage of carcinogenesis.
|
Identification of Common and Subtype-Specific Genes Involved in Tumor Progression.
The observation that different molecular subtypes of breast cancer are associated with distinctive profiles of gene expression has led some investigators to propose that the former have arisen from distinct cells of origin (9)
. This hypothesis, if true, raises the possibility that the gene expression pathways controlling various aspects of tumor progression may be distinct between different subtypes. To identify sets of common and subtype-specific genes involved in tumor progression, we compared the expression profiles of normal tissue, DCIS, and invasive tumors belonging to either the ER+ or the ERBB2+ subtypes (We were unable to perform a similar analysis for the ER– tumors because only one DCIS tumor segregated within the ER– subtype, which is insufficient for analysis).
Using a nonparametric Wilcoxon test, we first identified genes that were significantly regulated during the transition from the nonmalignant tissue to DCIS for the ER+ and ERBB2+ subtypes. For the ER+ subtype, we identified 113 up-regulated genes and 310 down-regulated genes among nonmalignant breast tissues and ER+ DCIS (P < 0.01; 2-fold change cutoff). This analysis was then repeated for the ERBB2+ subtype, in which 145 up-regulated genes and 295 down-regulated genes were identified among normal breast tissues and ERBB2+ DCIS samples. A total of 180 genes (145 down-regulated and 35 up-regulated) were found in common between the ER+ and ERBB2+ subtypes. The results are summarized in Fig. 4A
, and the full gene lists are provided in the Supplementary Information.4
|
Comparison of Common and Subtype-Specific Genes with an Independent Gene Expression Data Set.
In a recent independent report (15)
, Ma et al., using a combination of laser capture microscopy and cDNA microarrays, reported the identification of various sets of genes associated with human breast cancer progression. One unaddressed question, however, is the extent to which these genes are common or subtype specific, because the Ma et al. did not explicitly group their samples by their specific molecular subtype. To explore this question and to further validate our own findings, we filtered the Ma et al. data and derived from the original data set 767 up-regulated and 539 down-regulated genes associated with the transition from normal breast epithelia to DCIS (Supplementary Information).4
We then compared our list of subtype-specific and commonly regulated genes for the normal-to-DCIS transition with this derived gene list. The overlaps between the two data sets are represented as Venn diagrams in Fig. 4B
). We found that for all six comparisons (i.e., ER+-specific; ERBB2+-specific; common to ER+ and ERBB2+; both up-regulated and down-regulated), there were significant overlaps between the genes found in our study and the gene set studied by Ma et al., ranging from 11% (ERBB2+-specific down-regulated genes, 17 of 151) to 32% (commonly down-regulated genes, 47 genes of 149). To confirm that it would be highly unlikely for these overlaps to have occurred by random chance, we again performed a series of random permutation assays. Briefly, for each comparison, a total of 10,000 random gene sets were created, the total number of genes in each set being comparable with the sizes of the identified gene sets in Fig. 4A
). As one example, for the normal-to-DCIS transition, a total of 149 genes were identified by Wilcoxon analysis as being commonly down-regulated in both the ER+ and the ERBB2+ subtypes. This set of 149 genes was then compared with 10,000 random gene sets in which each set contained 150 randomly selected genes. As shown in Table 1
, the numbers of overlapping genes between our study and study by Ma et al. were consistently and significantly greater than overlaps created by 99.9% of all randomly generated gene sets (P = 0.036, t test for paired two samples). This suggests that the genes identified in our study (particularly those overlapping with the Ma et al. gene set) are likely to be of biological relevance, because they were independently identified in two different studies. Furthermore, through this comparison, we were also able to subdivide the original Ma et al. (2003) gene list into distinct sets of genes that were either common or pathway-specific. Table 2
lists the genes that were commonly regulated in both the ER+ and ERBB2+ subtypes, and that were commonly found in both data sets. A few of these genes and their potential implications for breast carcinogenesis are described in the Discussion.
|
|
|
DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Although the molecular subtypes in our patient population were independently derived using an unsupervised analysis, these subtypes were, for the most part, highly comparable with similar subtypes observed in other studies (e.g., ER+ to Luminal, ER– to Basal; refs. 9, 10, 11 ). It should also be noted that these breast tumor expression data sets are distinct in multiple ways, including (a) choice of patient population, (b) sample handling protocols, (c) scoring pathologist, and (d) choice of array technology and probe sets (two-color versus single color). Despite these methodological differences, our results suggest that, at a first approximation, breast cancers between the ethnic groups are remarkably similar at the molecular level. We also note that we did observe a few possible differences, such as the absence of a "Luminal C" subtype in the Asian population. However, further work will have to be performed to determine whether these apparent differences are truly due to genetic or environmental differences between the ethnic groups or whether the differences are merely experimental artifacts due to the different array technology platforms used in the studies.
Besides invasive breast cancers, we also profiled a series of DCIS, which represent the earliest malignant breast lesion detectable by conventional histopathology. Although DCIS cancers have long been recognized as the major precursor to invasive breast cancer, some studies have also suggested that DCIS cancer may also be distinct from invasive cancers in certain respects. For example, retrospective reports have shown that the majority of low-nuclear-grade DCIS undergo a long clinical evolution to invasive cancer (35, 36, 37) , which may indicate that additional genetic events must occur before they become invasive. We found that the gene expression profiles of DCIS cancers are highly similar to their invasive counterparts, and that they exhibited robust expression of many "hallmark" subtype-specific gene expression signatures. These findings suggest, for the first time, that the molecular subtypes of breast cancer can be discerned even at the preinvasive stage of carcinogenesis. Interestingly, of the 17 DCIS cancers that we profiled, 16 belonged to the ER+ or ERBB2+ molecular subtypes, with only one DCIS cancer belonging to the ER– subtype. Histopathological studies have shown that ERBB2+ cancers seem to be found much more often in DCIS compared with invasive cases (38) , which is consistent with our gene expression profiling data. One possible hypothesis to explain this difference might be that tumors of the ER– variety may be associated with an extremely transient DCIS stage compared with ER+ or ERBB2+ tumors.
Finally, by integrating the expression profiles of normal breast tissue, DCIS, and IDCs belonging to the ER+ and ERBB2+ subtypes, we were able to define various sets of genes that were regulated in a common and subtype-specific manner during the normal/DCIS/IDC transitions. We then validated several of these genes by showing that they were also commonly identified in a separate, related but not identical, study. Of the various gene sets, perhaps the most interesting group comprised genes that were commonly regulated in both the ER+ and ERBB2+ tumorigenic pathways and that were found to overlap in both studies (Table 2)
. We speculate that an in-depth study of these individual genes may provide important insight into the pathogenesis of breast cancer. Indeed, many of the genes in this list could be associated with various cellular functions associated with carcinogenesis, such as cellular proliferation (CDC14B, CKS2, JUN, MYC, PLAGL1), biosynthesis and energy utilization (RPL5, NQO1), cell-to-cell communication (ANXA1, ITGA6, ITGA10), and cellular signaling (NR3C1, PDGFRA, PPAP2A, PPP2R4). One particularly interesting finding was that several genes exhibiting common regulation in both subtypes were involved in the modulation of the Wnt signaling pathway (FZD7, TCF7L2, and SFRP1). Previous studies have reported the involvement of the Wnt pathway in breast cancer (11
, 39)
, and our data suggest that misregulation of this pathway may be required in both subtypes for breast carcinogenesis. If so, small molecules or compounds that selectively regulate the Wnt pathway may function as promising therapeutic or chemopreventive agents for breast cancer. Addressing this intriguing issue will be a promising area for future research efforts.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article can be found at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org).
Requests for reprints: Patrick Tan, National Cancer Centre/Defence Medical and Environmental Research Institute, 11 Hospital Drive, Singapore 169610, Republic of Singapore. Phone: 65-6-436-8345; Fax: 65-6-226-5694; cmrtan{at}nccs.com.sg
4 Supplementary information is available at Internet address: http://clincancerres.aacrjournals.org. 
5 Affymetrix website address: www.affymetrix.com.
Received 1/15/04; revised 4/23/04; accepted 5/14/04.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  E. Linos, D. Spanos, B. A. Rosner, K. Linos, T. Hesketh, J. D. Qu, Y.-T. Gao, W. Zheng, and G. A. Colditz Effects of Reproductive and Demographic Changes on Breast Cancer Incidence in China: A Modeling Analysis J Natl Cancer Inst, October 1, 2008; 100(19): 1352 - 1360. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Peppercorn and A. H. Partridge Breast Cancer in Young Women: A New Color or a Different Shade of Pink? J. Clin. Oncol., July 10, 2008; 26(20): 3303 - 3305. [Full Text] [PDF]
|
|
|
|
 |
|
 S. Raulic, Y. Ramos-Valdes, and G. E DiMattia Stanniocalcin 2 expression is regulated by hormone signalling and negatively affects breast cancer cell viability in vitro J. Endocrinol., June 1, 2008; 197(3): 517 - 529. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Wu, H. P. Cho, D. B. Rhee, D. K. Johnson, J. Dunlap, Y. Liu, and Y. Wang Cdc14B depletion leads to centriole amplification, and its overexpression prevents unscheduled centriole duplication J. Cell Biol., May 1, 2008; 181(3): 475 - 483. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. S. Lum, H. W. Chua, H. Li, W.-F. Li, N. Rao, J. Wei, Z. Shao, and K. Sabapathy MDM2 SNP309 G allele increases risk but the T allele is associated with earlier onset age of sporadic breast cancers in the Chinese population Carcinogenesis, April 1, 2008; 29(4): 754 - 761. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Wang, D. M. Ikeda, B. Narasimhan, T. A. Longacre, R. J. Bleicher, S. Pal, R. J. Jackman, and S. S. Jeffrey Estrogen Receptor-Negative Invasive Breast Cancer: Imaging Features of Tumors with and without Human Epidermal Growth Factor Receptor Type 2 Overexpression Radiology, February 1, 2008; 246(2): 367 - 375. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Polyak Is Breast Tumor Progression Really Linear? Clin. Cancer Res., January 15, 2008; 14(2): 339 - 341. [Full Text] [PDF]
|
|
|
|
|
|
B. K.T. Tan, L. K. Tan, K. Yu, P. H. Tan, M. Lee, L. H. Sii, C. Y. Wong, G. H. Ho, A. W.Y. Yeo, P. K.H. Chow, et al. Clinical Validation of a Customized Multiple Signature Microarray for Breast Cancer Clin. Cancer Res., January 15, 2008; 14(2): 461 - 469. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Lusa, L. M. McShane, J. F. Reid, L. De Cecco, F. Ambrogi, E. Biganzoli, M. Gariboldi, and M. A. Pierotti Challenges in Projecting Clustering Results Across Gene Expression Profiling Datasets J Natl Cancer Inst, November 21, 2007; 99(22): 1715 - 1723. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. A. Carey, C. M. Perou, C. A. Livasy, L. G. Dressler, D. Cowan, K. Conway, G. Karaca, M. A. Troester, C. K. Tse, S. Edmiston, et al. Race, breast cancer subtypes, and survival in the Carolina Breast Cancer Study. JAMA, June 7, 2006; 295(21): 2492 - 2502. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Yu, K. Ganesan, L. D. Miller, and P. Tan A modular analysis of breast cancer reveals a novel low-grade molecular signature in estrogen receptor-positive tumors. Clin. Cancer Res., June 1, 2006; 12(11): 3288 - 3296. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P Kauraniemi and A Kallioniemi Activation of multiple cancer-associated genes at the ERBB2 amplicon in breast cancer. Endocr. Relat. Cancer, March 1, 2006; 13(1): 39 - 49. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Bertucci, P. Finetti, J. Rougemont, E. Charafe-Jauffret, N. Cervera, C. Tarpin, C. Nguyen, L. Xerri, R. Houlgatte, J. Jacquemier, et al. Gene Expression Profiling Identifies Molecular Subtypes of Inflammatory Breast Cancer Cancer Res., March 15, 2005; 65(6): 2170 - 2178. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
). A complete list of the 376-gene set and the gene clusters are available for download
DCIS (ER+) transition. Numbers in bold, overlapping genes between the subtypes. (Inv, invasive; N/A, not available. B, a comparison of genes regulated during the normal/DCIS transition between our study and the study of Ma et al.,