Introduction

Brain metastases (BM) are the most common brain tumors of adults. The exact incidence of BM is unclear. Epidemiological studies have reported the annual incidence of BM to be approximately 11 per 100,000 persons [1]. The estimated incidence of BM among all patients with cancer is about 19 % per year. However, autopsy data show that up to 25 % of cancer patients have BM. There has been an increase of the incidence of secondary brain tumors over the past few decades, probably owing to more sensitive and widely available neuroimaging techniques, increases in survival times of cancer patients, and the increase of malignant lung neoplasms associated with smoking. The prognosis of patients with BM is poor; median survival times are typically less than 1 year.

The most common primary tumors that develop BM are lung carcinomas (35–64 %), breast carcinoma (14–18 %), melanoma (4–21 %), renal cell carcinoma, and colon carcinoma (5–10 %) [2].

The therapy of BM is limited due to the lack of prospective clinical studies [2]. Surgery and radiotherapy (stereotactic radiosurgery of isolated metastases and whole-brain radiotherapy (WBRT)) are currently the most commonly used therapeutic options in patients with BM [3–5]. Systemic treatment approaches have only little value in the multidisciplinary treatment strategy [6]. One reason is the little understanding of the brain metastatic cascade and the involved molecular mechanisms that result in the successful outgrowth of cancer cells in the brain parenchyma.

Formation of new blood vessels, angiogenesis, was proposed to be the main hallmark of cancer growth [7]. In many cancer types, angiogenesis was targeted with monoclonal antibodies or tyrosine kinase inhibitors, where an improved outcome could be observed [8, 9]. Recently, early angiogenesis was shown to be mandatory for successful macrometastasis formation in a BM mouse model [10]. This group demonstrated that chronic anti-angiogenic treatment with the anti-VEGF monoclonal antibody bevacizumab therapy prevents successful outgrowth of macrometastases of a non-squamous non-small cell lung cancer cell line to the brain. Likewise, Judah Folkman’s group has found a suppressive effect of chronic anti-angiogenic therapy on micrometastatic outgrowth in the lung [11], which also supports the crucial importance of the angiogenic switch for loss of tumor dormancy in different animal models [12–14]. Interestingly, anti-angiogenic treatment was not sufficient in a brain metastasis model of melanoma [10]. Different characteristics of angiogenic and invasive growth pattern of melanoma and NSCLC brain metastases both in patients and mouse models could be demonstrated [15–18].

Based on these findings, we hypothesized that metastases to the brain from different cancer primaries might be mediated via distinct angiogenic pathways. For this purpose, genes involved in pro- and anti-angiogenic pathways were evaluated in a cohort of brain metastasis derived from different primary tumor types.

Methods

Patient population and preparation of the tissue

For this retrospective study, 15 tissue samples from patients undergoing a cerebral resection of the neuroradiologically diagnosed brain metastases were used. Six, four, and five patients had non-small cell lung cancer (NSCLC), small-cell lung cancer (SCLC), and melanoma, respectively. The surgical resections of the tumors were performed at the Department of Neurosurgery, Medical University of Vienna. Tissues were obtained during the resection and were immediately snap-frozen in OCT. The diagnosis of BM was made by the board neuropathologist after a careful review of the surgically removed tissue. This study was approved by the local ethics committee. As control, neurosurgical specimens taken from patients who underwent a surgical resection of the temporal lobe for intractable epilepsy was used.

RNA isolation

About 100 mg of frozen tissue was transferred into 1 ml of TRIzol® Reagent (Roche Diagnostics, Mannheim, Germany). Homogenization was accomplished using a Polytron power homogenization unit (Kinematica, Kriens, Switzerland). RNA was extracted by phase separation after the addition of 200 μl chloroform. The RNA-containing aqueous phase was precipitated using 500 μl isopropanol. The RNA pellet was washed twice with 75 % ethanol, briefly air-dried, re-dissolved in RNAse free H2O, and either used immediately or frozen at −80 °C for later use.

Reverse transcription

After RNA extraction and purification, the reverse transcription reaction was performed. Briefly, total RNA was treated with deoxyribonuclease (DNAse) I (Invitrogen Carlsbad, CA, USA) for 15 min at room temperature, and the reaction was stopped using EDTA. The pretreated RNA was converted into complementary DNA (cDNA) using SuperScript IITM reverse transcriptase (Invitrogen Carlsbad, CA, USA) according to the manufacturer’s instructions. Reaction parameters were as follows: 65 °C 15 min, 4 °C 2 min, 25 °C 4 min, 42 °C 50 min, and 65 °C 10 min. The cDNA product was diluted using DNAse-free water and either immediately used or frozen at −80 °C.

Applied biosystems gene array

The TaqMan® Array 96-well Human Angiogenesis Plate contains 92 assays for angiogenesis and lymphangiogenesis-associated genes and 4 assays for candidate endogenous control genes (Catalogue Number: 4391016, Applied Biosystems). The panel of assays in the TaqMan® Array 96-well Human Angiogenesis Plate targets known angiogenesis growth factors like VEGF as well as matrix-derived inhibitors such as endostatin. Additionally, the panel contains markers and targets for angiogenesis and lymphangiogenesis.

Each inventoried TaqMan® gene expression assay contains sequence-specific, unlabeled primers, and a FAMTM dye-labeled probe (Table 1). The probes are pre-coated and dried on to each well. The assays are reconstituted to a 1× formulation. Each assay plate contains four housekeeping genes (first four genes listed in Table 1). A reaction volume of 10 μl was applied into each well. cDNA concentration was adjusted for 50 ng for each reaction well. For the whole plate, a total reaction volume of 1080 μl including 12.5 % excess volume was prepared. This included 540 μl of cDNA + DNase-free water and TaqMan® Fast Universal Master Mix (Catalogue Number: 4352042, Applied Biosystems). The plate was covered with a MicroAmp® Optical Adhesive Film and was read at the suggested thermal cycling condition of the Step One PlusTM Real Time PCR Systems, all described in the test manual (Applied Biosystems, Forster City, CA, USA).

Table 1 Ninety-six genes involved in the gene chip which was commercially available by Applied Biosystems (ABI)
Full size table

Biostatistical evaluation of the data and in silico analyses

Data preprocessing

Raw data for the given probes consisted of cycle time (Ct values) from the Real Time PCR System. In the first step, a screening for missing Ct values was performed. Considering possible technical errors with the chip, genes having detectable Ct values in less than five samples (<33.3 %) were omitted from the analysis. For genes having Ct values in five or more samples (> = 33 %), missing values are interpreted as no gene expression present and these Ct values were set to the cutoff of 45 cycles. In the second step, outliers with implausible high expression rates (mean—3* standard deviation) of a given sample were replaced by the mean Ct value of the other samples of the same group. Finally, normalization and relative fold change calculation of the gene expression compared to the control reference tissue was analyzed according to the ΔΔCt method [19].

Hierarchical clustering

Normalized data was hierarchically clustered using Pearson correlation as distance with average linkage rule utilizing the MultiExperiment Viewer software version 4.9 (Dana Farber Cancer Institute, Boston, MA, USA).

Gene expression analysis

An analysis of differences in gene expression between study groups was performed by calculating the average relative fold change for each gene for each patient group. Three cutoffs for the relative fold change were defined (50-fold, 10-fold, and 2-fold). Overlaps and differences of the relative fold change between the study groups were counted for each gene.

Network analysis

The angiogenesis chip gene set was mapped to Omicsnet protein interaction network version Feb. 2013 [20, 21]. Additionally, proteins were added to connect the angiogenesis chip set to form a spanning tree within Omicsnet utilizing a modified minimum spanning tree [22] algorithm based on Prim [23] and Kruskal [24]. Edges of experimentally verified protein-protein interactions were highlighted. Conspicuities of the interaction neighborhood of up- and down-regulated genes were described.

Results

Study population and demographic data

Demographic and clinical data of the patient population is depicted in Table 2. Seven (7/15, 47 %) patients received chemotherapy due to primary tumor before manifestation of BM. In six patients (6/15, 40 %), the primary tumor and BM were diagnosed at the same time. The median brain metastases free survival (BMFS) of the whole cohort was 22.5 months ranging between 27 and 185 months. Five (5/15, 34 %) patients represented metastases out of the brain. Five (5/15, 34 %) patients were diagnosed with more than one intracranial metastasis. Overall survival (OS) after diagnosis of BM was 166 days ranging between 22 and 896 days within the entire population.

Table 2 Clinical and demographic data of the patient population
Full size table

Gene expression data and differentially regulated genes

Out of 4 housekeeping genes included in the gene chip, only GUSB had detectable Ct values in all patient samples. Normalization of the gene expression data was performed based on this house keeping gene. Genes having detectable expression values in less than 5 samples among 15 patients were removed from the dataset (PLG, ANGPTL3, FGF4, SERPINB5, COL4A3, TNNI1, LECT1, IFNG) leading to a reduction of the gene set size to 84 genes. Four genes represented no detectable Ct values in the control brain tissue (TNMD, PRL, LEP, F2). ANGPTL1 in #2 and BAI in #13 were identified as extreme outliers and were modified accordingly.

Gene expression analysis revealed seven genes as up-regulated more than 50-fold in all study groups, whereas none of the genes was down-regulated 50-fold in the entire cohort when compared to that of control brain tissue (Table 3). Interestingly, some genes were down-regulated 50-fold only in a certain entity but not in others, like ANGPT4 and PDGFRB for SCLC and SERPINF1 for melanoma. Table 3 depicts the genes, which are up/down-regulated more than 50-fold and 10-fold and the genes, which were up/down-regulated only in one of the study group but not in others.

Table 3 Genes, which are up/down-regulated more than 50-fold and 10-fold in all study groups and only in one of the study group but not in others
Full size table

Hierarchical clustering according to the gene expression pattern

Gene expression data has been investigated for their potential to form hierarchical clusters as described in the methods section. Hierarchical clustering revealed four groups matching partially the study group definitions (Fig. 1). The first and second group demonstrated four out of the six NSCLC samples (#1,#2,#4,#6) forming a tight group next to a group of three out of the four SCLC samples (#7,#8,#9), respectively. The third group consists of two of the five melanoma samples (#12,#14) sharing similarities with a NSCLC sample (#5). The samples of the forth group have a greater distance within than any of the former groups and consists of the remainder melanoma samples (#11,#13,#15) and a sample from NSCLC (#3). One sample from the SCLC group (#10) shares the least similarities with any of the other groups.

Fig. 1
figure 1

Hierarchical cluster analyses based on the angiogenic relative gene expression pattern of the whole cohort. Pearson correlation as distance metric and average linkage rule. Genes are represented above the color bars of the image ranging from 1 to 0, indicating the degree of similarity for each patient to the next one, where they are connected with linkage bars. Patient IDs and the one control sample are indicated on the top of each column. Gene names are depicted on the right sight of the image. Color bar demonstrates the intensity of gene expression from down- to up-regulation (left to right, respectively). S small-cell lung cancer, N non-small cell lung cancer, M melanoma

Full size image

Networking of highly differentially regulated genes

Gene sets and their fold change categories were applied to Omicsnet protein interaction network for identifying patterns of up- or down-regulation in conjunction with their protein neighborhood as well as identifying other proteins of importance in their local vicinity. The following genes from angiogenesis set were missing in Omicsnet: ANGPTL1, TNMD, ANGPTL2, and PECAM1. The following genes were added to the overview, in order to establish a connecting subgraph of the angiogenesis set: EP300, MYL4, RPP14, GFI1B, CAPZB, ATXN7, DMP1, HSPA5, MYOC, MYC, SLC9A1, HDAC1, HSP90AA1, and MAGI3. Networks for NSCLC, SCLC, and melanoma are demonstrated in Fig. 2a–c, respectively. Green proteins represent for up-regulation, whereas red proteins were the down-regulated ones. The degree of up or down-regulation correlates with the size of the green or red shapes, respectively. White symbols are the proteins with no change at the expression level, and the yellow proteins represent those, which were added for establishing a connected subgraph. From the whole set of interaction edges of Omicsnet, the interaction edges illustrated are experimentally verified.

Fig. 2
figure 2

a Network formation based on the genes expressed in non-small cell lung cancer samples. b Network formation based on the genes expressed in small-cell lung cancer samples. c Network formation based on the genes expressed in melanoma samples. In all figures, up-regulated genes are represented in green color, whereas down-regulated genes are shown in red. The size of the genes correlated with the degree of either down- or up-regulation. Squared shape stays for the genes which had no expression in the control brain tissue. Genes with the yellow color are included to the network for establishing a connected subgraph. Genes with white circle had similar expression level compared to the control sample. Edges illustrated are experimentally verified

Full size image

Based on these findings, THBS, MMP2, and FN1 could be identified as genes with the highest network capacity with more than 10 experimentally confirmed connections for each. Interestingly, for NSCLC, SCLC, and melanoma, proteins like EP300, MYC, and HSP90AA1 linked to several up- and down-regulated angiogenic factors. As indicated above, these proteins were not a part of the angiogenesis kit but were included for connecting the angiogenesis kit set within Omicsnet.

Discussion

In this study, we demonstrate differential expression of genes involved in angiogenesis in a cohort of brain metastasis tissue originating from three distinct primaries, namely non-small cell lung cancer, small-cell lung cancer, and melanoma. Although there seemed to exist a common angiogenic gene expression pattern in all three primaries, several genes with an exclusive expression in specific diagnostic categories were found, which was responsible for the formation of clusters of cases originating from the same primary tumor group. This suggests a characteristic angiogenic gene signature of NSCLC, SCLC, and melanoma tissues when metastasized in the brain.

Angiogenesis is a hallmark of the pathobiology of primary and secondary brain tumors, which renders anti-angiogenic treatment for these diseases very attractive [1, 25]. Attempts for treatment of brain tumor patients with anti-angiogenic substances resulted in very heterogeneous outcomes of these patients [26–28]. Particularly for BM, a different kind of angiogenesis was proposed in animal models and could also be shown in the tissue derived from humans [10, 15]. Lung cancer tissue was demonstrated to establish vascular angiogenic pattern, whereas melanomas showed a cooptive growth [10, 29]. Differences at the morphology of BM from different primaries might be responsible for this variance. In line with the previous observation of mouse models showing a clear difference of angiogenesis morphology in BM of different primaries, our data could classify patients with identical diagnosis in similar clusters based on their angiogenic gene expression signature. Genes with known involvement in angiogenic processes including VEGFA [7], CEACAM1 [30], PECAM1 [31], CXCL12 (also known as SDF1-α) [32], KIT [33], TNF [34], and collagens [35, 36] were at least 10 times up-regulated in all groups indicating the necessity of those factors for the establishment of the basic characteristics of the angiogenesis process. These genes are mainly involved in proliferation of endothelial cells and activation of growth factors [30, 31, 33]. Based on the potential importance of those molecules and high representation in all tissues, a treatment strategy of BM irrespective of the primary diagnosis might rely on the inhibitory strategy of these genes.

Interestingly, some other genes exhibited a differential regulation with a restriction to a certain group. In NSCLC, SEMA3F and FLT1 were up-regulated, whereas FOXC2 was down-regulated. SEMA3F was identified to represent an anti-angiogenic molecule with inhibitory capacity of VEGF [37], whereas FLT1 is the receptor of VEGF [38], suggesting the auto-activation of anti-angiogenic mechanisms for this tumor type. Interestingly, FOXC2 plays a role in epithelial to mesenchymal transition [39], which was demonstrated to be one of the leading processes for early distant metastasis for NSCLC [40]. This underlines the importance of establishment of inhibition strategies for metastasis formation in this very common primary. In SCLC, PDGFRB and ANGPTL4, which are known for the involvement in vascular development and angiogenesis, are highly up-regulated [41], whereas TGFB1, a multifunctional peptide for cell proliferation, differentiation, and apoptosis, was down-regulated [42]. Interestingly, similar to NSLCL, melanomas showed an activation of some anti-angiogenic factors including ANGPT2 [43] and SERPINF1 (also known as pigment epithelium-derived factor (PEDF)) [44]. However, further molecules playing important roles in matrix degradation like MMP2 [45], TIMP2 [45], and ADAMTS1 [46] were also up-regulated in the melanoma group when compared to NSCLC and SCLC. Since degradation of extracellular matrix has been highlighted in melanomas for successful metastasis, the finding of up-regulation of matrix degradation factors in melanoma BM tissues is not surprising [47]. Those particular genes might be responsible for the differences seen at the morphology of the angiogenesis in melanoma and NSCLC patients. Moreover, inhibition of matrix degradation could be considered as a mechanism, which might be primarily targeted for the treatment of melanoma BM.

Since patients with BM have been largely excluded from clinical trials with anti-angiogenic agents, we do not know whether responses to the anti-angiogenic treatments would vary between BM patients of different primary tumor. Our data and the data from others would together shed light on the potential diversity of angiogenesis of BM at the molecular basis, which might be considered for future treatment strategies and clinical trial designs as an important point for the patient selection. In future clinical trials, anti-angiogenic agents might be stratified on BM patients depending on their primary tumor. Targeting factors with a commonly high expression in all BMs (e.g., VEGF, PECAM1) might reveal a balanced outcome, whereas a strict selection of the patients based on their primary tumor might be necessary, if factors with differential expression are chosen (e.g., SERPINF1, TIMP2, and MMT2).

The data revealed by the gene chip was analyzed with Omicsnet in order to find out the relationships and network vicinity between differentially regulated genes. Notably, 14 genes were included to the networking schema establishing a network connection by the angiogenesis genes with several of these genes being highly connected to up- or down-regulated genes. These additional genes might also be related to the varying formation of angiogenesis in different BM primary diagnoses and should be investigated into more detail in future preclinical studies.

In this small pilot study, some limitations should be mentioned: As obtaining healthy brain tissue was not feasible, we used temporal lobe tissue from anti-epileptic surgery as control sample and compared the gene expression levels of each tumor tissue with that as control. As this approach was usual in the literature, we hope that the gene expression level of a noncancerous tissue, namely temporal lobe, would reflect the expression level of control tissue [48]. Within the kit, which was provided commercially, four endogenous controls were available. Among those, only one endogenous control was detectable in all tissues, GUSB, which was used for further analyses. The lack of reliable detectability of the other three control genes remains unexplained and may relate to technical issues or inconsistent expression in the CNS. In any case, it does not influence the results of our study, as these genes were excluded from all analyses.

To sum up, we demonstrate that BM belonging to different primaries could be distinguished based on their angiogenic character. Those differences might influence the outcome of patients when treated with distinct anti-angiogenic drugs. This could be linked with the treatment armamentarium of BM patients and might be a basis for the patient selection in future clinical trials.