Clinical & Experimental Metastasis

, Volume 24, Issue 1, pp 13–23

Differential expression of hypoxia and (lymph)angiogenesis-related genes at different metastatic sites in breast cancer

Authors

  • Gert G. Van den Eynden
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Steven J. Van Laere
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Ilse Van der Auwera
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Leen Gilles
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • J. Lance Burn
    • The Liver Research GroupUniversity of Sheffield
  • Cecile Colpaert
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Peter van Dam
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Eric A. Van Marck
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
  • Luc Y. Dirix
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
    • Translational Cancer Research Group (Lab Pathology, University of Antwerp/University Hospital Antwerp, Oncology Center, General Hospital, St.-Augustinus)
    • Department of PathologyAZ St.-Augustinus
Original Paper

DOI: 10.1007/s10585-006-9049-3

Cite this article as:
Van den Eynden, G.G., Van Laere, S.J., Van der Auwera, I. et al. Clin Exp Metastasis (2007) 24: 13. doi:10.1007/s10585-006-9049-3

Abstract

Introduction

Breast cancer can metastasize via lymphatic and hematogenous pathways. Hypoxia and (lymph)angiogenesis are closely related processes that play a pivotal role in the tumor progression and metastasis. The aim of this study was to compare expression of hypoxia and (lymph)angiogenesis-related genes between primary breast tumors and metastases in different tissues.

Materials and methods

A gene list of 269 hypoxia and (lymph)angiogenesis-related genes was composed and validated using Onto-Express, Pathway-express and Ingenuity software. The expression of these genes was compared in microarray data of 62 samples of primary tumors and metastases of 31 patients with breast cancer retrieved from Gene Expression Omnibus. Similarity between samples was investigated using unsupervised hierarchical clustering analysis, principal component analysis and permutation testing. Differential gene expression between primary tumors and metastases and between metastases from different organs was analyzed using Kruskall–Wallis and Mann–Whitney statistics.

Results

Unsupervised hierarchical cluster analysis demonstrated that hypoxia and (lymph)angiogenesis-related gene expression was more similar between samples from the same patient, than between samples from the same organ. Principal component analysis indicated that 22.7% and 7.0% of the total variation in the gene list was respectively patient and organ related. When differences in gene expression were studied between different organs, liver metastases seemed to differ most from the other secondary sites. Some of the best characterized molecules differentially expressed were VEGFA, PDGFRB, FGF4, TIMP1, TGFB-R1 and collagen 18A1 (precursor of endostatin). To confirm the results of these experiments at the protein level, immunohistochemical experiments were performed with antibodies for VEGFA and MMP-2.

Conclusions

Our results suggest that hypoxia and (lymph)angiogenesis-related gene expression is more dependent on the characteristics of the primary tumor than on the characteristics of the organs that bear the metastasis. However, when different organs are compared, the expression in liver metastases differs most from other metastatic sites and primary tumors, possibly due to organ-specific angiogenic and lymphangiogenic responses to metastasis-related hypoxia.

Keywords

AngiogenesisHypoxiaBreast cancer metastases

Introduction

Although loco-regional disease and local recurrences can be debilitating, metastases to distant organs are the most important cause of breast cancer-related morbidity and mortality. During tumor progression, breast cancer cells can spread from the primary tumor to secondary sites via both lymph and blood vessels and give rise to loco-regional lymph node (LN) and distant metastases, respectively. The most frequent sites of distant breast cancer metastases are the bone, the liver, the lung and the central nervous system. Hypoxia and angiogenesis are closely linked biological processes that play a pivotal role in tumor growth, progression and metastasis. These processes have extensively been studied in primary breast tumors. Hypoxia and angiogenesis have been found to be correlated to shorter relapse-free and overall survival [110]. More recently, lymphangiogenesis has also been shown to be involved in spread and metastasis of primary breast tumors [11] and to be correlated to poor prognosis in patients with breast cancer [12, 13]. The regulation and the role of hypoxia, angiogenesis and lymphangiogenesis in tumor progression at metastatic sites are, however, less clear. An important reason is the sparcity of available tissue from distant sites, especially from untreated patients. Histomorphological and immunohistochemical data obtained on human breast cancer samples, suggest that the local tumor microenvironment may influence hypoxia and angiogenesis in metastases in different organs. We have shown that angiogenesis––quantified as endothelial cell proliferation fraction––and expression of the hypoxia markers carbonic anhydrase 9 and hypoxia inducible factor-1 alpha in primary tumors and their corresponding LN metastases are correlated [14]. Breast cancer liver metastases on the contrary, grow in 96% of the cases according to a replacement pattern, in which tumor cells replace hepatocytes, co-opt the sinusoidal vessels and do not induce hypoxia or angiogenesis [15]. A comparable less angiogenic ‘infiltrative’ pattern was found in 51% of breast cancer skin deposits [16]. Although these data suggest that there might be differences in hypoxia and angiogenesis at different metastatic sites in breast cancer, the underlying molecular mechanisms remain to be elucidated. Better insight in site-specific expression of molecules involved in the response to hypoxia and involved in angiogenesis and lymphangiogenesis might lead to a better understanding of the influence of the host organ microenvironment on metastatic growth. Furthermore, it could lead to identification of new targets for site-specific targeted therapy for patients with metastatic breast cancer.

One possible approach is to study differences in hypoxia and (lymph)angiogenesis-related gene expression between metastases from different sites and between metastases and primary tumors with gene expression microarrays. This technology enables the comparison of gene expression of large numbers of genes between different conditions. Studies comparing gene expression profiles of primary breast tumors and metastases from different locations demonstrated that the gene expression profile of metastases is strikingly similar to that of the corresponding primary tumor [17, 18]. However, these reports were based on a whole-genome approach and did not analyze hypoxia and (lymph)angiogenesis-related genes specifically. When the expression of hypoxia and (lymph)angiogenesis-related genes would be retained during the metastatic process, this would be important in clinical decision making towards the use of targeted antiangiogenic or hypoxia-modulating therapy in patients with metastatic breast cancer. Since it is not always possible to obtain tissue from the metastasis for analysis of the expression of therapeutic targets, primary tumor analysis may then be helpful to guide therapeutic decisions.

Therefore, the aim of this study was to compare the expression of hypoxia and (lymph)angiogenesis-related genes between metastases of different locations and between primary tumors and metastases of patients with breast cancer.

Materials and methods

Sample selection

Gene expression data of 62 samples of primary tumors and metastases of different organs of 31 patients with breast cancer were retrieved from the Gene Expression Omnibus repository (http://www.ncbi.nlm.nih.gov/geo). The database was queried for breast cancer metastasis and a study population was composed of 48 breast cancer metastases from different organs (15 LN, 9 brain, 8 lung, 4 liver, 3 spinal, 2 skin, 1 ovary, 1 arm, 1 bowel, 1 adrenal, 1 kidney, 1 diaphragm and 1 unknown origin) and the corresponding primary breast carcinoma of 14 of these cases. These samples are part of Gene Expression Omnibus records GSE3521, GSE1992, GSE2740, GSE2741 and GSE2607 that have been published and analyzed in references [1821]. Two-color oligonucleotide microarray experiments with RNA isolated from all these samples have been performed at the Department of genetics, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA on 3 closely related platforms: Agilent’s Human 1A Oligo Microarray G4110A and B and the Agilent’s Human 1A Oligo UNC custom Microarray platform (Agilent Technologies, Palo Alto, CA, USA) using Stratagene Human Universal Reference (Stratagene, La Jolla, CA, USA) that contained 1/10 added MCF7 and ME16C RNAs as a reference. Detailed description of the methodology can be found in previous publications [1821].

Hypoxia and (lymph)angiogenesis-related gene list

To study the expression of genes involved in response to hypoxia, lymphangiogenesis and angiogenesis in primary tumors and metastases at different organs from patients with breast cancer a gene list was compared of 269 genes associated with and involved in these processes. The content of this list was based on publications on gene expression profiling and hypoxia [22, 23] and on the content of low density arrays designed to study hypoxia and angiogenesis (SuperArray Bioscience Corporation, Frederick, MD, USA). The content of the complete hypoxia and (lymph)angiogenesis-related gene list is descibed in supplement 1. To validate its use, the content of the gene list was submitted to Ingenuity Pathway Analysis (www.ingenuity.com). Furthermore, the 269 gene list was analyzed for overrepresentation of biological processes, molecular function and molecular pathways relative to the list of all genes represented on the 3 microarray chips by means of Onto-express and Pathway-express software (http://vortex.cs.wayne.edu:8080).

Data retrieval and analysis

Oligonucleotide probes for 221 of the 269 genes of the hypoxia and (lymph)angiogenesis-related gene list were present on all three types of microarrays the samples had been hybridised to (supplement 1). Normalized and Log2 transformed gene expression data for these genes were retrieved from the Gene Expression Omnibus database and analyzed using GeneSpring GX 7.3.1 Software (Agilent Technologies). To investigate common biological themes between the samples defined by the hypoxia and (lymph)angiogenesis-related gene list, unsupervised hierarchical clustering analysis of the 62 samples was performed using Pearson correlation as a similarity measure and average linkage as a clustering algorithm. For further analysis, metastases from brain and spinal cord were grouped together as central nervous system metastases and metastases from less frequent locations (ovary, arm, bowel, adrenal, kidney, diaphragm and unknown origin) were not included. The contribution of characteristics of the primary tumor versus organ of metastasis on hypoxia and (lymph)angiogenesis gene expression was further investigated using permutation and principal component analysis. For these analyses only samples were included from patients of whom 2 or more samples were included in this study. To quantify dissimilarities between samples from patients and metastatic sites, scatter ratios were calculated using a modified algorithm as described by Weigelt et al. [17]. A within-patient-between-patient scatter ratio (WPBPSR) and a within-metastatic site-between-metastatic site scatter ratio (WSBSSR) were calculated as follows: the mean of the distances between all possible combinations of tumors within a patient/metastatic site divided by the mean of all distances between possible combinations of tumors from two different patients/metastatic sites, respectively. A within-patient-within-metastatic site scatter ratio (WPWSSR) was defined as the mean of the distances between all possible combinations of tumors within a patient divided by the mean of all distances between possible combinations of tumors within a metastatic site. Distance was defined as 1–(Pearson correlation coefficient)2. To determine statistical significance of these ratios, permutation analysis was performed, using 10,000 random iterations. In a second approach, global views of the variation in gene expression among the different patients and metastatic sites defined by the hypoxia and (lymph)angiogenesis-related gene list were investigated using principal component analysis.

Furthermore, the expression of the 221 genes between metastases from different locations was compared using the non-parametric Kruskall–Wallis test (with Tukey post hoc analysis) and Mann Whitney-U test for comparing more than two and two groups, respectively.

Immunohistochemical confirmation

To validate the results of the gene microarray analysis at the protein level, immunohistochemical experiments were performed with antibodies against vascular endothelial growth factor-A (VEGFA) and matrix metalloproteinase-2 (MMP-2). From our previous study population [15], eight patients with breast cancer were selected from whom formalin-fixed paraffin-embedded tissue was available from both the primary tumor and liver metastases. In 3 out of 8 patients, axillary LN metastases were also available for immunohistochemical analysis. In the other patients, no metastatical involvement of the axillary LNs was found. VEGFA antigens were retrieved with a 15 min microwave treatment in Tris/EDTA (pH 9.0). MMP-2 antigens were retrieved at 98°C for 30 min in Target Retrieval Solution (S1699, Dako). After quenching of endogenous peroxidase activity, slides were incubated with the primary antibody (A-20 polyclonal rabbit anti-human VEGFA antibody, 1/100, 60 min, Santa Cruz; mouse monoclonal anti-MMP-2 antibody, 1/50, 60 min, Calbiochem, Darmstadt, Germany). For the VEGFA stain, before incubation with the primary antibody, slides were incubated for 30 min with 1% normal goat serum (Dako). Antibody binding was visualized using the Envision + Dual Link system and diaminobenzidine as a chromogen (Dako). Cytoplasmic VEGFA immunoreactivity and membranous MMP-2 immunoreactivity of tumor cells were assessed by two independent pathologists on a 0 (no reactivity)––3 (strong immunoreactivity) scale (Fig. 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Fig1_HTML.jpg
Fig. 1

MMP-2 (A, B) and VEGFA (C, D) immunohistochemical stainings in primary tumors (A, C) and their corresponding liver metastases (B, D). Membranous MMP-2 expression in the primary tumor (A) is strong, in contrast to weak membranous MMP-2 expression in the liver metastasis (B). Moderate cytoplasmic VEGFA expression of tumor cells is seen in the primary tumor (C) and in the liver metastasis (D). At the tumor/liver interface, tumor cells (D, white arrows) and hepatocytes (D, black arrows) are in close contact. Hepatocytes strongly express VEGFA. (MMP-2: matrix metalloproteinase-2, VEGFA: vascular endothelial growth factor A)

Results

Validation of the hypoxia and (lymph)angiogenesis gene list

Ingenuity Pathway Analysis associated 220 genes of the 269 hypoxia and (lymph)angiogenesis-related gene list to relevant functions and diseases. A strong representation of genes involved in angiogenesis (n = 88, P < 0.001) and angiogenesis-related processes such as development of blood vessels (n = 99, P < 0.001), migration of endothelial cells (n = 63, P < 0.001), proliferation of endothelial cells (n = 52, P < 0.001), vascularization (n = 40, P < 0.001) and neovascularization (n = 29, P < 0.001) was found. Seventeen genes were associated with hypoxia (P < 0.001) and 6 with lymphangiogenesis (P < 0.001). When the content of the list was compared to the content of the Human 1A Oligo microarrays with Onto-Express, an overrepresentation of genes involved in positive (P < 0.001) and negative (P < 0.001) regulation of angiogenesis and in processes closely related to angiogenesis such as response to wounding (P < 0.001), immune response (P = 0.02) were found. With Pathway-express an overrepresentation of pathways involved in response to hypoxia and in angiogenesis, such as TGF-ß signalling (impact factor 12.3, P < 0.001), cytokine–cytokine receptor interaction (impact factor 11.9, P < 0.001) and MAPK signalling (impact factor 11.7, P < 0.001). Comparable results were found when the same analysis were performed with the 221 genes that were common to all 3 microarray chips.

Primary tumors versus metastases

Unsupervised hierarchical clustering of the expression data of 221 genes in all 62 samples is shown in Fig. 2A. Two replicate samples from the same LN (patient nr 3), brain (patient nr 19) and brain (patient nr 17) metastasis clustered together on terminal branches of the dendrogram, even when performed on two different chips (patient nr 3), validating the use of expression data from different but related chips. Furthermore, samples from two different skin metastases (patient 13) from the same patient clustered together on terminal branches of the dendrogram. The same was true for two different lung metastases (patient 20) from another patient. One sample of each of these five replicate couples was randomly excluded for further analysis. The clustering dendrogram showed that samples from the same patient, but from different organs clustered more often together than samples from the same organ but from different patients. This was even clearer when the same clustering protocol was applied only on those samples from patients of whom a primary tumor had also been analyzed (Fig. 2B, n = 32). Nevertheless, when only LN, lung, liver and central nervous system metastases were included in this analysis, the dendrogram shows one cluster with LN metastases in 9/10 samples and one cluster with central nervous system metastases in 6/8 samples. Furthermore, a cluster enriched in lung metastases was found. The liver metastases did not seem to cluster together (Fig. 2C, n = 35). These results were confirmed by the computational analysis of the WPBPSR, WSBSSR and WPWSSR (Fig. 3). The WPBPSR and the WPWSSR were significantly lower than their respective random permutations (WPBPSR = 0.68, P < 0.001; WPWSSR = 0.7, P < 0.001). The WSBSSR on the contrary was not significantly different (0.98, P = 0.424), suggesting that the similarity of hypoxia and (lymph)angiogenesis-related gene expression within samples from different metastatic sites in the same patient is significantly higher than the similarity between samples from the same metastatic site but from different patients.
https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Fig2_HTML.jpg
Fig. 2

Unsupervised hierarchical clustering dendrogram based on the 221 genes of the hypoxia and (lymph)angiogenesis gene list of (A) all samples (n = 62), (B) all samples with a corresponding primary tumor (n = 32) and (C) samples from the most frequent breast cancer metastatic sites (LN, Li, Lu and CNS) (n = 35). The table below every dendrogram indicates the patient number (upper row) and growth site (lower row). In (A) replicate samples (patient 3, 17 and 19) or different samples from a comparable site in the same patient (patient 13, 20) did cluster together on terminal branches of the dendrogram. Visual inspection furthermore suggested that the expression pattern in metastases was more similar to the pattern of the corresponding primary tumor or to other metastases from the same patient, than to the expression pattern of metastases from the same organ but form different patients. (P: primary tumor; LN: lymph node metastasis; CNS: central nervous system metastasis; Li: liver metastasis; Lu: lung metastasis)

https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Fig3_HTML.gif
Fig. 3

Permutation test of the WPBPSR (A), WSBSSR (B) and WPWSSR (C). The histogram represents the null-hypothesis distribution obtained by randomly labeling tumors and repeating this procedure 10,000 times. The dashed line represents the ratios when tumors are correctly labelled. (WPBPSR: within-patient-between-patient scatter ratio; WSBSSR: within-metastatic site-between-metastatic site scatter ratio; WPWSSR: within-patient-within-metastatic site scatter ratio)

We then performed a principle component analysis with the hypoxia and (lymph)angiogenesis-related gene list as input data set to obtain global views of the variation between samples. Principal component analysis for patient-related variation resulted in nine principal components of which the first accounted for 73.2% and the other eight together for 22.7% of the variation, respectively. Principal component analysis for metastatic site-related variation resulted in four principal components of which the first accounted for 92.5% and the other three together for 7.0%. In both analyses, all principal components except the first ones showed condition-related variation, indicating that 22.7% and 7.0% of the total variation in the hypoxia and (lymph)angiogenesis-related gene list is respectively patient and metastatic site related. These data demonstrate that variation in hypoxia and (lymph)angiogenesis-related gene expression is approximately three times more dependent on the characteristics of the primary tumor than of the metastatic site (Fig. 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Fig4_HTML.gif
Fig. 4

Global views of the variation in gene expression among the different patients and metastatic sites defined by the hypoxia and (lymph)angiogenesis gene list were obtained using PCA on the samples of which at least two samples from metastasis in different organs in the same patient were available. PCA for patient-related variation (A) resulted in nine principal components of which the first accounted for 73.2% and the other 8 together for 22.7% of the variation, respectively. PCA for metastatic site-related variation (B) resulted in four principal components of which the first accounted for 92.5% and the other 3 together for 7.0%. In both analyses, all principal components except the first ones showed condition-related variation, indicating that 22.7% and 7.0% of the total variation in the hypoxia and (lymph)angiogenesis gene list is respectively patient- and metastatic site-related. (PCA: principal component analysis; CNS: central nervous system metastasis; LN: lymph node metastasis; Prim: primary tumor)

Differential gene expression between different sites of tumor growth

To identify hypoxia and (lymph)angiogenesis-related genes differentially expressed in metastases in different organs, a Kruskall–Wallis test with Tukey post hoc analysis was performed comparing primary tumors, LN, CNS, brain, lung and liver metastases (Table 1). Twenty-five genes (gray squares) were significantly differentially expressed between the groups. Only few genes (black squares, white letters) were significantly differentially expressed when PTs, LN, CNS, brain and lung metastases were compared with the Tukey post hoc test. Remarkably, more hypoxia and (lymph)angiogenesis-related genes differed significantly when liver metastases were compared to the other groups, than when these groups were compared to each other. To further investigate gene expression differences between primary tumors, LN metastases and liver metastases, we performed a Mann–Whitney U-test. In total 32, 29 and 8 genes were differentially expressed (P < 0.05) comparing liver metastases and primary tumors, liver and LN metastases, and LN metastases and primary tumors, respectively. Nineteen genes were commonly differentially expressed between primary tumors and liver metastases and between liver and LN metastases. Only 1 gene was differentially expressed between both primary tumors and LN metastases and primary tumors and liver metastases and no genes were differentially expressed between primary tumors and LN metastases and between liver and LN metastases. Some of the best characterized molecules differentially expressed were VEGFA, MMP2, transforming growth factor beta receptor 1, platelet derived growth factor receptor Beta, fibroblast growth factor 4, tissue inhibitor of metalloproteinase 1 and collagen 18A1 (precursor of endostatin). A complete list of the genes differentially expressed comparing LN metastases, liver metastases and primary tumors can be found in Table 2. To further investigate the results of the Mann–Whitney U-test, protein expression of VEGFA and MMP-2 was compared between primary tumors, LN metastases and liver metastases in eight patients with breast cancer. According to the Mann–Whitney analysis, VEGFA expression was higher in liver metastases than in LN metastases or primary tumors and MMP-2 expression was higher in primary tumors than in liver metastases (Table 2). Expression of VEGFA and MMP-2 was studied at the protein level, results are shown in Table 3. The VEGFA score was higher in tumor cells of liver metastases than in the primary tumor in 1/8 cases, lower in 3/8 cases and the same in 4/8 cases. The VEGFA score was higher in liver than in LN metastases in 1/3 cases and lower in 2/3 cases. However, the normal liver parenchyma strongly expressed VEGFA. For MMP-2, expression was lower in liver metastases than in primary tumors in 4/8 cases, higher in 2/8 and the same in 2/8 cases. The MMP-2 score was the same or higher in LN metastases than in primary tumors or liver metastases.
Table 1

Number of hypoxia and (lymph)angiogenesis-related genes differentially expressed between metastases from different organs

 https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Figa_HTML.gif

A Kruskall–Wallis non-parametric test resulted in 25 differentially expressed genes (gray squares). When primary tumors, LN, CNS and Lung metastases were compared with a post hoc Tukey test, only few of these genes differed significantly (black squares, white text). Most of the 25 genes were not significantly differentially expressed (white squares, black text). Most genes differed significantly between liver metastases and primary tumors or all other metastatic sites (LN: lymph node metastasis, CNS: central nervous system metastasis)

Table 2

Hypoxia and (lymph)angiogenesis-related genes differentially expressed between primary tumors and/or LN and/or liver metastases of patients with breasts cancer

 https://static-content.springer.com/image/art%3A10.1007%2Fs10585-006-9049-3/MediaObjects/10585_2006_9049_Figb_HTML.gif

Fourty Nine genes had a Mann Whitney U test P-value < 0.05. In the table the p value is shown, the color of the box indicating where the expression was higher: blue-primary tumors, gray-liver metastases and yellow-LN metastases. The second column of the table indicates the function of the gene in the angiogenesis process. There was no association between the overexpression of pro- or antiangiogenic factors and the organ of metastasis for liver and LN metastases (P = 0.5), liver metastases and primary tumors (P = 1.0). When LN metastases and primary tumors were compared only proangiogenic factors were differentially expressed. (LN: lymph node metastases; Prim: primary tumors)

Table 3

Immunohistochemical scores for VEGFA and MMP-2 in primary tumors, lymph node metastasis and liver metastasis from eight patients with breast cancer (PT: primary tumor, LNM: lymph node metastasis, LM: liver metastasis)

 

VEGFA expression

MMP-2 expression

PT

LNM

LM

PT

LNM

LM

Patient 1

3+

3+

2+

2+

2+

2+

Patient 2

3+

NA

3+

3+

NA

1+

Patient 3

3+

NA

1+

3+

NA

1+

Patient 4

2+

3+

2+

1+

3+

0

Patient 5

2+

NA

2+

1+

NA

0

Patient 6

3+

NA

1+

0

NA

2+

Patient 7

2+

NA

2+

1+

NA

3+

Patient 8

1+

1+

2+

2+

3+

2+

When the function of the differentially expressed genes was assigned as pro- or antiangiogenic or undefined, based on literature data, there was no association between the overexpression of pro- or antiangiogenic factors and the site of tumor growth for liver and LN metastases (P = 0.5), liver metastases and primary tumors (P = 1.0). When LN metastases and primary tumors were compared only proangiogenic factors were differentially expressed.

Discussion

In this study we compared the expression of hypoxia- and (lymph)angiogenesis-related genes in primary tumors and metastases from different organs based on microarray gene expression data. To be able to analyze as many samples as possible, microarray data from different studies were included. This approach holds an inherent danger of interlaboratory and cross-platform biases, which were limited by only using samples that have been processed in the same laboratory on very similar and commercially available oligonucleotide microarrays. Other limitations of the study have to be considered when interpreting the results: the relatively small numbers of samples from some of the metastatic sites, the inclusion of autopsy derived samples and the fact that not for all metastases the related primary tumor was available. These limitations are due to the limited availability of tissue from metastatic lesions. To the best of our knowledge, this is the largest population of breast cancer patients known in literature available for this kind of analysis.

The results of the unsupervised hierarchical cluster analysis, the permutation testing and the principal component analysis demonstrated that the expression pattern of hypoxia and (lymph)angiogenesis-related genes is more similar when primary tumors and their corresponding metastases are compared––whatever the location––than when metastases from the same location are compared. This is in line with the results from other studies comparing gene expression profiles from primary breast tumors and metastases. Weigelt et al. demonstrated that the gene expression profile of primary breast tumors was maintained in distant metastases [17] and that the molecular subtype and 70-gene prognosis signature of the primary tumor were also preserved in the metastases [18]. The data of the latter study are part of the data included in our analyses. Furthermore, in other microarray studies comparable gene expression profiles of primary tumors and corresponding LN metastases were reported [24, 25]. However, all reported studies were based on genome wide gene expression comparison and did not specifically investigate hypoxia and (lymph)angiogenesis-related gene expression. Furthermore, they mainly focussed on the comparison between primary tumors and metastasis and did not compare metastases from different locations. The concordance between the hypoxia and (lymph)angiogenesis gene expression pattern of the primary tumor and the corresponding metastases suggests that the influence of the local tumor microenvironment on these processes is less important than the inherent tumor cell biology related to the metastatic process. This may be important in clinical decision making towards the use of targeted antiangiogenic or hypoxia-modulating therapy in patients with metastatic breast cancer, since it is not always possible to obtain tissue from the metastasis for (immuno)histochemical or molecular analysis. Primary tumor analysis may then be helpful to guide therapeutic decisions. Other factors contributing to the fact that the between-patient variability is higher than the between-metastatic-site variability might be possible heterogeneity of the study population and interindividual variability in gene expression of the host tissue. Differences in gene expression between different histological or molecular breast cancer subtypes might increase between-patient variability. Since very little additional information on the samples was available in the GEO repository the influence of this heterogeneity could not be studied or corrected for. This has to be kept in mind when the results are interpreted. Furthermore, Grüber et al. demonstrated that interindividual gene expression variability of donor lungs was very high [26]. Influence on the gene expression profile of incorporated host tissue might thus increase the dissimilarity between samples from different patients.

We also investigated which hypoxia and (lymph)angiogenesis-related genes are differentially expressed between primary tumors, LN metastases and liver metastases of patients with breast cancer. For VEGFA and MMP-2 the results were also evaluated at the protein level using immunohistochemistry. The expression of MMP-2 was higher in the tumor cells of the primary tumor than in the tumor cells of the liver metastases in 4 out of 8 cases and the same in 2 out of 8 cases. This confirms our molecular data, especially since immunohistochemistry is only a semi-quantitative technique that is less sensitive than molecular techniques. Although the molecular analysis suggested a very significant increase in VEGFA expression in liver metastases compared to primary tumors or LN metastases, the immunoreactivity of VEGFA in tumor cells in liver metastases is higher in only 1/8 cases compared to primary tumors and in 1/3 of cases compared to LN metastases. However, as other authors describe [27] hepatocytes had a high VEGFA expression. Since liver metastases of patients with breast cancer grow according to a ‘replacement’ pattern with close contact between hepatocytes and tumor cells, it is very difficult to exactly determine the tumor-normal liver interface [15]. Since VEGFA expression of normal breast tissue or LN parenchyma is much lower, the increased expression of VEGFA mRNA in liver metastasis might therefore be due to incorporation of normal hepatocytes at the metastasis margin. This underscores the importance of the use of immunohistochemistry or other techniques that enable correlation of expression to histological structures for validation of this kind of analysis.

The differences in expression of VEGFA, MMP-2 or other hypoxia and (lymph)angiogenesis-related genes between tumor deposits in different organs might contribute to the differences in growth pattern, hypoxia and angiogenesis that we have previously reported. Using an immunohistochemical and histomorphological approach, we demonstrated that breast cancer liver metastases have a completely different growth pattern with different hypoxic and angiogenic characteristics than primary tumors and LN metastases of patients with breast cancer [14, 15]. However, when the genes that were differentially expressed between primary breast tumors, LN and liver metastases were classified according to their function in hypoxia and angiogenesis, no significant difference was found in the balance between pro- and anti(lymph)angiogenic factors. This might be due to incomplete appreciation of the function of these molecules in the regulation of the response to hypoxia and (lymph)angiogenesis in humans, of the intermolecular interactions between many of these molecules and of the mutual importance the different molecules. Subtle differences in expression of certain molecules might have an amplified effect on the processes they are regulating. Another reason might be the contribution of molecules that are not yet known or reported to be involved in hypoxia and (lymph) angiogenesis and are therefore not in our gene list. Furthermore, differences in responsiveness to pro- and antiangiogenic stimuli of the endothelial cells in different organs might lead to differences in angiogenesis in tumors at different locations. It has become clear that considerable structural and functional heterogeneity exists between endothelial cells in different organs and vascular beds. Chi et al. explored endothelial cell specialization using gene expression profiling of 53 independent cultures of endothelial cells from different anatomical locations. They found pervasive differences in gene expression patterns not only between endothelial cells from large vessels and microvascular endothelial cells, but also between microvascular endothelial cells from different organs [28]. Other authors demonstrated that blood vessel type-specific and tissue-specific characteristics of endothelial cells are under the control of their microenvironment, either through soluble factors or cell–cell contacts [2934].

To conclude, our data demonstrate that hypoxia and (lymph)angiogenesis-related gene expression is patient-related more than metastatic site-related. Apparently, the hypoxia and (lymph)angiogenesis profile is retained throughout the metastatic proces. However, our data also suggest that there might be differences in hypoxia and (lymph)angiogenesis-related gene expression at different metastatic sites, especially when considering liver metastases of breast cancer. These results encourage further investigation of these differences and their underlying mechanisms, since better understanding might lead to identification of new targets for site-specific anti-angiogenic and hypoxia-modulating therapy in patients with breast cancer.

Acknowledgements

Gert Van den Eynden is a research assistant of the Fund for Scientific Research Flanders. Steven Van Laere is a predoctoral assistant supported by a research grant of the University Hospital of Antwerp. This work was supported by Fund for Scientific Research Flanders Grants L.3.058.06N and G010004N. We thank Britta Weigelt and Lodewijk Wessels from the NKI, Amsterdam, The Netherlands for kindly providing the algorithm for the WPBPSR analysis.

Supplementary material

Copyright information

© Springer Science + Business Media B.V. 2007