Molecular Signature Associated with Bone Marrow Micrometastasis in Human Breast Cancer¹

The majority of cancers in industrialized countries are solid tumors derived from epithelial tissues such as carcinomas of the breast, lung, gastrointestinal tract, and prostate. The fate of these patients is largely dependent on the development of blood-borne distant metastases. Besides being one of the major sites of overt metastases [each year > 350,000 epithelial cancer patients in the United States die with bone metastases (1)], BM⁴ is a common homing organ for metastatic epithelial tumor cells derived from various primary sites. Using sensitive immunocytochemical and molecular assays, it has become evident that 20–40% of patients with various epithelial tumors (e.g., carcinomas of the breast, prostate, colon, or lung) harbor occult metastatic cells in their BM in the absence of any LN metastases and clinical signs of overt distant metastases (2). Moreover, the presence of these early metastatic cells in BM at primary surgery predicts the postoperative occurrence of overt metastases in bone and other organs (2, 3, 4, 5). Recent expression profiling studies on human breast carcinomas have focused on the formation of overt metastases as end point of their analyzes (6, 7, 8). Although these studies are important to estimate the risk of patients, end point assays may not provide much insight into the biology of the metastatic cascade. In this study, we have therefore focused on the onset of primary hematogeneous metastases, using BM as an indicator organ for micrometastatic tumor cells. Our findings indicate that primary hematogeneous dissemination of breast tumor cells exists as a selective process associated with a specific molecular signature.

The patients underwent surgery at the Department of Gynecology, University Hospital Hamburg-Eppendorf (Hamburg, Germany). After resection, the tumors were split in two parts; one part was fixed in 10% formalin and paraffin embedded for immunohistochemical staining, and the other part was snap frozen in liquid nitrogen for RNA isolation. Because gene expression profiles are largely determined by intrinsic tumor-related factors such as estrogen receptor status, we selected only estrogen receptor-positive primary tumors at stage pT₂. Furthermore, we manually microdissected neoplastic cells to avoid that differentially expressed genes were missed in the data analysis process because of genes highly expressed by stromal elements. From all patients, the axillary LNs were surgically removed, and BM was aspirated from the upper iliac crest. Nodes were analyzed by an expert pathologist (L. R.) for the presence of metastases. The procedures for the preparation and immunocytochemical detection of tumor cells in BM have been described in detail (3). Patients were separated into three different groups: (A) BM negative and LN negative (n = 7); (B) BM positive and LN negative (n = 7); and (C) BM negative and LN positive (n = 5). To identify molecular signatures associated with exclusive tumor cell spread to either bone marrow or LNs, we compared the gene expression profiles of tumors from group A and B as well as the profiles of group A and C, respectively.

The study was approved by the Institutional Review Board of the University Hospital Hamburg-Eppendorf, and written informed consent was obtained from all patients.

The correlation between RNA and protein expression was determined by comparing immunostained paraffin-embedded tumor tissue with the corresponding gene expression results. The TMA used to validate the correlation between differential gene expression and BM micrometastasis contained 83 breast tumor samples from patients with known BM status (presence of CK-positive cells, n = 23; absence, n = 60). The patients underwent primary treatment between 1999 and 2000 at the University Hospital Hamburg-Eppendorf. Immunostaining was performed on an automated staining machine (Dako Diagnostika GmbH, Hamburg, Germany) with the mouse antihuman CK19 monoclonal antibody clone BA17 (concentration 1:10; Dako Diagnostika GmbH), the mouse antihuman CK18 antibody clone DC10 (concentration 1:100; Dako Diagnostika GmbH), the mouse antihuman CK8 antibody clone 35β H11 (concentration 1:25; Dako Diagnostika GmbH), the mouse antihuman STAT-1 antibody clone M-22 (concentration 1:2000; Santa Cruz Biotechnology, Dassel, Germany), the mouse antihuman TGF-β2 antibody clone V (concentration 1:50; Santa Cruz Biotechnology, Dassel), and the mouse antihuman antibody RHO H6 clone 119 (concentration 1:500; Santa Cruz Biotechnology, Biotechnology, Dassel). Primary antibody labeling was visualized with the Dako ChemMate Detection Kit. Tumor sections were scored according to the Remmele Score, which is a product of percentage of immunostained tumor cells and the staining intensity. The HIF-1α immunohistochemistry was performed as described previously (9).

To validate the gene expression data with the respective immunohistochemical data from the same tumors, we performed a Spearman rank correlation using the SPSS software (version 11 for Windows). For evaluation of the relationship between the BM status and CK8, CK18, CK19, TGF-β2, and RHO H6 protein expression, the tumor samples were grouped in normal expression (100% stained tumor cells) and reduced expression (<100% stained tumor cells). The STAT-1-stained tumor samples were grouped into tumors with weak (score 0–4) and strong (score 6–12) expression according to the Remmele Score, whereas for the HIF-1α protein expression, the percentage of >5% stained tumor cells was used as real positive staining (9). P of <0.05 was considered to indicate a statistically significant difference.

A total of 0.1 μg of the total RNA used for the array hybridization was reverse transcribed. The first strand cDNA was diluted and used as template for the following RTQ-PCR analysis as described previously (10). The data analysis was performed with an ABI Prism Sequence Detection System (TaqMan) supplied by Perkin-Elmer/Applied BioSystems, which uses the 5′ nuclease activity of TaqDNA polymerase to generate a real-time quantitative DNA analysis assay (10). The sequence of the CK19 PCR primer pair and the fluorogenic probe (5′-3′) are the following: TGTGGAGGTGGATTCCGC (5′-3′); GCTTCGCATGTCACTCAGGA (5′-3′); and probe CGGGCACCGATCTCGCCAA (5′-3′).

Cryosections of breast tumor samples were manually microdissected, and RNA was extracted from each sample using the RNeasy Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. To avoid genomic DNA contamination, a RNase-free DNase step was performed for 30 min at 37°C using 1 unit of DNase (Promega, Erlangen, Germany)/μg RNA. In total, 5 μg of total RNA of each separate case was used for [α ³³P]dATP (3000Ci/mmol, 10 μl; Amersham, Freiburg, Germany) cDNA synthesis. The cDNA probe was purified with nucleotide removal columns (Qiagen). The Atlas Human Cancer 1.2 Arrays (Clontech, Heidelberg, Germany) were hybridized according to the manufacturer’s protocol.

The hybridized array membranes were exposed to phosphorimager plates (Raytest Isotopenmeβgeräte, Straubenhardt, Germany) for 72 h, and plates were scanned with the phosphorimager Fuji Bas (Raytest) at a 100-μm resolution. The images were analyzed using the Imagene 4.1 software (Biodiscovery, Los Angeles, CA), and the mean values of the spots corrected for the mean local background. Negative values were set at an expression level of 0 and were taken as missing values after 2-based log-transformation. The data of the different arrays were normalized on basis of the mean of all expressed genes. Differences between clinically distinct groups were calculated for each gene with the Student’s t test (Excel) using 2-based log-transformed data. Only genes that were significantly different (P < 0.05) were considered relevant. All highly significant differentially expressed genes were confirmed in a second approach using the SAM (version 1.12: two class, unpaired response type; Ref. 11). The third approach to explore our data were to look at correlation of genes using the cluster analysis software of Eisen et al. (12).

Gene expression and functional annotation was performed using the Online Mendelian Inheritance in Man (OMIM) and Serial Analysis of Gene Expression databases online.⁵ Expression was annotated as breast/epithelial when sequence tags where found in breast/epithelial cell line libraries or moderate expression in breast/epithelial tissue libraries, whereas expression in lymphocyte and/or fibroblast cell lines was absent.

The tumors were classified into two groups according to the result of an immunocytochemical BM assay that is able to detect one single metastatic cell in millions of BM cells. The patients in both groups had neither LN metastases (tumor-node-metastasis stage pN₀) nor overt distant metastases (tumor-node-metastasis M₀), indicating that tumor cells in BM were derived from the primary breast tumors. The sensitivity, specificity, and clinical relevance in breast cancer of this assay have been well documented in our previous study (3). Moreover, molecular and functional analyses of individual CK-positive cells in BM aspirates from cancer patients convincingly showed that these cells are tumor cells with a proliferative capacity (13, 14).

We found for 86 genes significant differential expression between tumors from BM-positive and BM-negative patients. In total, 9 of these genes were up-regulated and 77 down-regulated in tumors of BM-positive patients, suggesting that transcriptional repression of genes seems important for BM micrometastasis. This repression affects also known metastasis suppressor genes such as KiSS-1 (15) and members of the nm23 metastasis suppressor gene family (NME3 and NME4; Ref. 16; Table 1). Thus, our findings support the recent concept that silencing of many genes might be a major mechanism for tumor progression (17). One explanation for this finding is that the normal differentiation program of adult (breast) epithelial cells does not allow invasion and migration to avoid disassembly of the epithelial tissue. During the dedifferentiation process in tumor cells, transcriptional repressors (17) might be overexpressed, which might cause suppression of various genes, including metastasis suppressor genes.

We also tested the difference in gene expression between the BM-positive and BM-negative group by SAM (described in “Materials and Methods”) and compared the results to the t test analysis. The confirmed genes showed a q value from 3.5 to 8.6 and are marked in Table 1 with an asterisk. Seven additional genes were only classified as significant by SAM (MST 1, PLD 1, Cofactor C, CK12, RHOHP1, ITGA1, and GCR).

Visualization of the differential gene expression profile by cluster analysis showed that BM-positive and BM-negative patients clearly separated into two distinct expression profile groups that exactly matched the BM status (Fig. 1 A). To facilitate the search for important pathways regulating or involved in tumor cell dissemination, genes most likely expressed by cells other than breast/epithelial cells were excluded by screening the differentially expressed genes against the UniGene/Serial Analysis Of Gene Expression databases. In total, 73 genes (84.9%) had a breast/epithelial signature, indicating that the corresponding transcripts indeed were derived from the microdissected breast cancer cells. A few genes appeared to be expressed by stromal cells, particularly tumor-infiltrating lymphocytes, which apparently were not removed completely by microdissection.

To investigate whether the observed expression profile is specific for BM micrometastasis, we used the same type of DNA arrays and determined the expression profile associated with lymphatic metastasis, i.e., we compared LN-positive and LN-negative primary tumors from patients who were all BM-negative. The Student’s t test analysis revealed 44 differentially expressed genes, and the number of up-regulated genes (n = 9) was again smaller than the number of down-regulated genes in the tumors of LN-positive patients (n = 35; Table 2). The cluster analysis revealed a specific signature associated with lymphatic dissemination (Fig. 1,B), which was, however, distinct from the signature associated with BM micrometastasis, with only 9 genes in common (italicized and underlined in Table 1), suggesting that these two routes of dissemination might be governed by different molecular determinants.

The major functional categories of the BM micrometastasis-relevant genes include extracellular matrix remodeling (n = 9), cytoskeleton plasticity (n = 10), as well as signaling pathways (n = 34; Table 2). One of the most prominent observations was the concerted lower expression of cytokeratins in tumors of BM-positive patients (Table 1). In particular, luminal cytokeratins (e.g., CK8, CK18, and CK19), known as the cytoskeletal constituents of cells in simple epithelia (e.g. normal breast duct cells), were affected, which may point to a putative new role of these structural proteins as metastasis suppressors.

We additionally noted down-regulation of members of the RAS superfamily (RHO H6 and RAC1) in BM-positive tumors; these proteins are involved in the reorganization of the actin cytoskeleton (18). Several other genes of the RAS signal transduction pathway were also down-regulated in tumors of BM-positive patients, including downstream tyrosine kinases and serine/threonine kinases (mitogen-activated protein kinases 3, 2, 12, 7, serine/threonine kinase 3), guanine nucleotide binding proteins (RHO H6, RAC1, and G-protein α) and transcription factors (Jun D; Table 1).

Another interesting group of signal transduction genes up-regulated in BM-positive tumors belong to the pathway of IFN-regulated and induced genes. The induction of the IFN-regulated genes occur via the JAK/STAT pathway. We observed up-regulation of genes encoding for STAT-1 and (2′-5′) oligoadenylate synthetase-1, downstream effector molecules of the JAK/STAT pathway (Table 2). Activated STAT family members are thought to participate in malignant progression of human tumors through the prevention of apoptosis (19).

A remarkable signaling pathway that was shown to be specifically up-regulated in tumors of BM-positive patients is the HIF-1α pathway (Table 1). Hypoxia has been previously discussed as a driving force that enables cells to leave the primary tumor. The most prominent factor involved in a variety of hypoxia-related processes (e.g., proliferation, angiogenesis, and cell death) is HIF-1α (20), which was significantly up-regulated in BM-positive tumors. Intriguingly, this up-regulation coincided with down-regulation of genes responsible for HIF-1α degradation (e.g., VHL and cullin-2) in BM-positive tumors, which in concert may lead to an accumulation of HIF-1α in tumor cells. HIF-1α protein levels are already increased at early stages of breast cancer development (9) and might contribute to the early metastatic potential of breast tumor cells. The fact that other hypoxia-inducible but HIF-1α-independent transcription factors (e.g., cAMP-responsive element binding protein and nuclear factor κB) were not up-regulated in tumors of BM-positive patients, strongly suggests that hypoxia itself might not be the driving force for tumor cell dissemination but argues more in favor of an oncogenic dysregulation (20) of the HIF-1α pathway that causes onset of metastasis.

To validate our findings, we first stained tissue sections from our training set of tumors used for cDNA array analysis and confirmed the differential expression at the protein level for a selected group of genes (i.e., CK8, CK18, or CK 19, STAT-1, HIF-1α with Ps of 0.048, 0.035, 0.007, 0.032, and 0.001, respectively). Fig. 1 C shows the CK19 gene expression in relation to the protein expression. We additionally confirmed our array data on CK19 gene expression by PCR using TaqMan analysis. The significant differential expression of the array result (BM+/BM− ratio: 2.49) was comparable with the TaqMan results (BM+/BM− ratio: 2.21).

In addition, we stained TMAs containing an independent larger test set of primary breast tumor samples (n = 83) from patients with and without tumor cells in the BM. The differential expression of CK genes, as observed in the training set, was confirmed. Patients with a reduced expression of CK8, CK18, or CK19 had an increased incidence of a positive BM finding (Table 3). Normal breast cells present in the tissue sample were consistently stained with the anti-CK antibodies and served therefore as internal positive control. This finding additionally supports the assumption that luminal cytokeratins might suppress the onset of metastasis in breast cancer, which is consistent with the earlier observation that elevated levels of CK18 protein predict a decreased rate of metastatic relapse in breast cancer (21). Of notice, the differential expression of cytokeratins observed in our study could not have resulted in false-negative BM findings because it was up-regulated in the tumors of BM-negative cases.

To validate the contribution of the JAK-STAT, RAS, and HIF-1α signaling pathways, single members of these pathways (i.e., STAT-1, RHO H6, TGF-β2, and HIF-1α) were also selected for TMA. As expected, increased STAT-1, RHO H6, and HIF-1α protein expression correlated to a positive BM finding (Table 3), which confirms the cDNA array data. The observed correlations were highly significant for HIF-1α (P = 0.006), whereas only a trend was seen for STAT-1 (P = 0.067) and RHO H6 (P = 0.080). In contrast, we observed no correlation between TGF-β2 protein expression and BM status (P = 0.659; Table 3).

It is difficult to compare the results obtained with our limited cDNA array with the recent expression profiling results of other groups who used large-scale arrays and correlated their findings to clinical outcome (6, 7, 8). The relevance of the selected cancer-annotated genes represented on our cDNA array was, however, documented by the fact that our cluster analysis revealed a clear segregation of breast tumors related to the BM status. Although we certainly have missed genes also relevant to BM micrometastasis, this is the first study that demonstrates that BM micrometastasis is a selective process requiring a specific molecular signature mainly characterized by suppression of gene expression. It will be an important long-term goal of future investigations to explore the functional relevance of the observed expressional changes in BM-positive tumors. A better understanding of the biology driving metastatic spread opens the way for an improved molecular staging and therapy of breast cancer patients.

We thank Dr. Marcus Otte, Dr. Jorma Isola, Dr. Julia Ramirez-Porras, Petra van der Groep, Kathrin Baack, and Sonja Santjer for technical help and Micromet (Martinsried, Germany) for providing antibody A45-B/B3, Dr. Gregg L. Semenza (Departments of Pediatrics and Medicine, Institute of Genetic Medicine Johns Hopkins University School of Medicine, Baltimore, MD) for providing anti-HIF-1α antibody, and Dr. Volker Assmann for critically reading the manuscript.