Tissue sample selection
To identify genes and pathways involved in the formation of brain metastasis, we exclusively used specimens of primary tumors, and did not use specimens of metastatic sites. Fresh frozen (FF) tissue specimens of 22 primary breast cancer patients who developed metastasis to brain and/or to other organs were selected. Two groups of samples were compared; those from patients who had developed brain metastasis (exclusively or in addition to a maximum of 2 organs; n = 13) and those from women who developed metastases to a maximum of three organs (n = 9). None of the 22 patients had received adjuvant therapy (chemo- or hormonal therapy) prior to developing metastases, all samples were ER- and none of the patients had more than three metastatic sites. The relevant clinical data are provided in Table 1. In addition, we used 20 primary breast cancer samples for independent validations from patients of whom 13 developed brain metastasis. This study was approved by the Medical Ethics Committee of the Erasmus Medical Center, Rotterdam, The Netherlands (MEC 02·953) and performed in adherence to the Code of Conduct of the Federation of Medical Scientific Societies in The Netherlands (http://www.fmwv.nl/).
5 µm H&E sections from each sample were prepared before and after sectioning for RNA isolation. To ascertain the origin of the tumor, the percentages of the invasive tumor cells, inflammatory infiltrates and the presence of necrosis were taken into consideration (JMK).
RNA extraction and purification and quality control
Total RNA from FF tissue samples was extracted from 20 to 30 sections of 30 μm (depending on the size of the sample) using RNABee reagent according the supplier’s instructions (Campro Scientific, Veenendaal, The Netherlands) . Following isolation, RNA was stored in RNase/DNase-free water at − 80 °C. The quantity and quality of the isolated RNA was assessed by nanodrop. Samples were excluded if the yield did not reach the minimum requirement of 1000 ng.
Gene expression profiling
Illumina Whole-Genome cDNA-mediated Annealing, Selection, extension and Ligation (WG-DASL) assay was used to profile the samples. In the assay, 24,526 annotated transcripts corresponding to 18,391 unique genes are measured. The WG-DASL assay was performed according to the manufacturer’s instructions with an input of 500 ng total RNA. To monitor the assay performance and to evaluate the inter-assay BeadChip variability between the experiments, an inner-assay control consisting of 500 ng total RNA pooled from RNA isolated from several cultured breast cancer cell lines was used in each experiment .
Scanned data were uploaded into GenomeStudio software version 2011.1 via the WG-DASL gene expression module for further analysis. The average signal, detection P value, bead standard error and average beads were used to quantile normalize the data in the statistical language R (www.r-project.org) using the “Lumi” package .
To identify significantly differentially expressed genes, three steps were followed: sample exclusion criterion, reliable probe selection and gene expression comparisons. Sample exclusion criterion and probe selection method were described previously . For the gene expression comparison, Biometric Research Branch ArrayTools (BRB-array tool (V4.3.1)) was used . Within BRB, the 4150 most reliable probes for FF samples were exposed to the class comparison algorithm to identify differentially expressed genes with a maximum P value of 0.05 after 10,000 permutations multiple correction to determine significance.
Pathway analysis was done by two different methods. Firstly, the differentially expressed genes (resulted from the gene expression comparisons of samples with brain metastasis compared to samples with metastasis to other organs) was submitted to Ingenuity Pathway Analysis (Ingenuity, Mountain View, CA). Secondly, all reliably profiled genes were submitted to the Global Test  (version 4.4.0) to associate Biocarta pathways (www.biocarta.com) to the groups of samples metastasizing to brain or to other organs. The R version 2.4.1 (www.r-project.org) was used to run the Global Test package. A P value for a pathway was considered significant when the univariate P value of the test as well as the P value calculated by 1000 re-samplings were both < 0.05. In addition, The lists of differentially expressed probes were uploaded into the function annotation tool DAVID version 6.7  to functionally annotate the differentially expressed genes. DAVID was used with the data basses and settings that are preselected by default. Statistical analyses and multiple testing correction procedures are those included in the DAVID analyses .
In vitro BBB model
Tissue culture procedures
Human umbilical vein endothelial cells (HUVECs, ScienCell) were cultured in endothelial cell medium (ECM, ScienCell) supplemented with endothelial cells growth factors, 5% fetal bovine serum and Penicillin/Streptomycin. HUVECs were used between passage 2 and 5. Human astrocytes (ScienCell) were cultured in astrocytes medium (AM, ScienCell) supplemented with astrocyte growth factors, 2% fetal bovine serum and Penicillin/Streptomycin. Human astrocytes were used between passage 2 and 5. No further authentication was performed for this study for HUVECs and human astrocytes. Three breast cancer cell lines which had proven to be able to cross the BBB were used: MDA-MB-231, MDA-MB-231-BM (a breast cancer cell line that metastasize specifically to brain)  and SUM159PT . The characterization of the breast cancer cell lines are summarized (Supplementary Table 1). The tumor cell lines and normal fibroblasts (isolated from non-cancerous tissue) were previously characterized [19, 46], and were cultured in RPMI medium supplemented with l-glutamine, 10% fetal bovine serum and Penicillin/Streptomycin. T cells were isolated, activated and transduced as described previously [24, 55]. Two types of T cells were used: T cells that were transduced with an empty vector, referred to as “T cells”, and T cells that were transduced with a T cell receptor specific for MAGE-C2/HLA-A2 vector, referred to as “antigen-specific T cells”. The later T cells allowed studies into the role of antigen-specific activation of T cells, to which end SUM159PT cells were used as these cells express the cognate antigen (MAGE-C2/HLA-A2). Both types of T cells were cultured in RPMI media, supplemented with 10% fetal bovine serum and IL-2 (360 IU/mL, Chiron, Amsterdam, The Netherlands), and Penicillin/Streptomycin. The T cell transduction process, validation of TCR expression by flow cytometry and particular culture conditions by allogeneic feeder cells are described in [24, 55].
Construction of the in vitro BBB assays
To develop a BBB model, HUVECs were co-cultured with human astrocytes on opposite sides of a transwell insert . Twenty-four-wells transwell polycarbonate inserts (surface area 0.33 cm2, pore size 3 µm, Becton Drive, Franklin Lakes, NJ, USA) were coated with 2% gelatin (Sigma) for 45 min. The transwell inserts were placed upside-down and ~ 100,000 human astrocytes/inserts were seeded at the bottom side of the inserts. The cells were allowed to adhere for 3 h at 37 °C in a 5% CO2 incubator and were fed every 15–30 min. After 3 h, inserts were flipped and placed in 24-well plates. 1 mL of astrocyte media was added to the lower chamber and astrocytes were allowed to grow for 1 day. Fifty-thousand endothelial cells were plated on the upper chamber of the inserts and the cultures were placed in the incubator for 3 days. The permeability of the BBB model was verified by adding trypan blue dye to the upper chamber and incubating the model for 30 min at 37 °C. Medium from the lower chamber was collected and absorbance was measured at 595 nm. The permeability of the BBB model by trypan blue was included in duplicate in each experiment.
In vitro BBB-T cell response functional studies
To investigate the influence of T lymphocytes on the ability of tumor cells to cross the BBB, MDA-MB-231, MDA-MB-231-BM and SUM159PT breast cancer cells were co-cultured with T cells (no specific binding occurred between T cells and breast cancer cells). In addition, SUM159PT breast cancer cells were co-cultured with antigen-specific T cells (specific binding occurred between T cells and breast cancer cells). The optimal ratio of tumor cells and T cells was achieved following a titration to reach the best balance between cell killing and migration. The co-culture ratio of breast cancer cells over T cells was 3:1. After co-culturing for 3 days, T cells were removed by three washes using PBS. The breast cancer cells were harvested and labelled with 5 µM CFMDA cell tracker green (Invitrogen) for 45 min in serum-free medium. The breast cancer cells were collected and re-suspended in RPMI medium supplemented with l-glutamine, 10% fetal bovine serum and Penicillin/Streptomycin. Ten thousand cells were seeded in the upper chamber of the BBB model and incubated overnight at 37 °C. The cells which had passed through were recorded by confocal microscopy of the lower chamber after removing the transwell inserts.
To investigate whether the cytokines secreted by T cells are responsible for the changes of breast cancer cells, cells of the three breast cancer cell lines were incubated with 5 ml conditioned media collected from T cells for 3 days. Then breast cancer cells were washed, harvested and labelled using the same method as described above. Ten-thousand cells were seeded on the upper chamber of the insert and incubated overnight at 37 °C. As a control, ten-thousand breast cancer cells of the three breast cancer cell lines, without exposure to either T cells or to conditioned media from T cells, were used. These experiments were repeated 10 times. As a negative control, fibroblasts isolated from non-cancerous tissue were used.
To study the effect of IL-2 (which is essential in culturing T cells) and IFN-γ (a cytokine produced by T cells) the three types of breast cancer cell lines were incubated with IL-2 (360 IU/ml) and with IFN-γ (10 and 20 µg, Bio-Connect, Huissen, The Netherlands) separately for 3 days. Subsequently, breast cancer cells were washed, labelled with cell tracker green and seeded in the upper chamber of the BBB model. These experiments were repeated three times. Moreover, to study the effects of T cells, conditioned media of T cells, IL-2 and IFN-γ, on the permeability of the BBB model, they were added to the upper chamber of the BBB model and incubated overnight. Subsequently, the permeability of the BBB model was investigated by adding trypan-blue dye (20% in RPMI media) to the upper chamber and incubated for 30 min at 37 °C. Medium from the lower chamber was collected and absorbance was measured at 595 nm. In addition, the effect of the mentioned factors on the permeability of the BBB model was checked by seeding breast cancer cells in the upper chamber of the BBB model overnight. These experiments were repeated three times.
Confocal laser-scanning microscopy and Quantification of the migrated cells
Confocal images were obtained using a Zeiss LSM510 confocal laser-scanning microscope equipped with a 488 nm Argon-laser and a Plan-Neofluar 20× objective with NA 0.5 (Zeiss, Oberkochen, Germany). Images were made with a pixelsize of 0.9 µm. Pictures were submitted to ImageJ software version 1.49S (http://www.fiji.sc) and used to calculate the number of cells per mm2.
MDA-MB-231 and MDA-MB-231-BM breast cancer cells were co-cultured with T cells in 3: 1 ratio, and SUM159PT was co-cultured with antigen-specific T cells for 3 days (following the same method described earlier). As controls, MDA-MB-231, MDA-MB-231-BM and SUM159PT cells were cultured without T cells for 3 days. All cell cultures were washed three times with PBS to remove T cells before they were scraped and collected in 1.5 mL Eppendorf tubes. Three additional washing steps with PBS were applied. After removing the supernatant, the cell pellets were immediately snap-frozen on dry-ice and stored in − 80 °C until the time of preparation. After thawing the samples, they were prepared and measured on nano LC as described previously .
Proteomics data analysis
From the raw data files of the Orbitrap Fusion mass spectrometer, MS/MS spectra were extracted and converted into mgf files using MSConvert of ProteoWizard1 (version 3.0.06245). All mgf files were analyzed using Mascot (version 2.3.02; the Matrix Science, London, UK), which was used to perform searches against the Uniprot_sprot_2014_09 database; Homo sapiens species restriction; 66,244 sequences. For the database search the following settings were used: a maximum of two miss cleavages, oxidation as a variable modification of methionine, carbamidomethylation as a fixed modification of cysteine and trypsin was set as enzyme. A peptide mass tolerance of 10 ppm and a fragment mass tolerance of 0.5 Da were allowed. An ion score of 40 was used as cut-off value.
Scaffold software (version 4.4.3, Portland, OR) was used to summarize and filter MS/MS based peptides and protein identifications. Protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least two identified peptides. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped. Using these criteria, Scaffold generated a list of identified proteins (Minimum: 6% coverage and 2 peptides), including the number of sequenced peptides that corresponded to these proteins. The identified proteins of breast cancer cells (of the three cell lines) that were co-cultured with T cells were compared to those of the breast cancer cells cultured without the T cells. The comparison was done based on 2 sample t test, and the P values of all proteins were calculated and corrected for multiple variants using Benjamini–Hochberg. A protein was considered as a differentially expressed protein if P < 0.05, and all three breast cancer cell lines in one group showed the same direction of expression.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE  partner repository with the dataset identifier PXD006750.
To study the role of GBP1 in crossing the BBB by the tumor cells, silencing transfection experiments were performed. A mix of four siRNA sequences that target GBP1 mRNA and another mix of non-targeting siRNA (referred to as siSham) were obtained from Dharmacon (GE health care, The Netherlands). Silencing experiments were performed using transfection buffer 1, following the manufacturer’s instructions (Dharmacon, GE health care, The Netherlands). MDA-MB-231 and MDA-MB-231-BM breast cancer cell lines were transfected for 24 h. RNA was isolated using RNeasy Micro Kit (Qiagen), and the silencing efficiency was evaluated by reverse transcriptase quantitative PCR (RT-qPCR) using TaqMan gene expression assay (Applied Biosystems).
When siGBP1 proved to be efficient, the transfection experiments were performed for 24 h, then breast cancer cells were co-cultured with T cells (3:1 ratio) for an additional 24 h. The co-culturing time was reduced to 24 h, well within the effective period of siGBP1. Afterward, T cells were washed away with PBS, and breast cancer cells were harvested and labelled with green tracker as described above. Ten-thousand cells were seeded on the upper chamber of the BBB model as described above.
Formalin fixed, paraffin embedded (FFPE) tissue samples that morphologically matched the primary FF tissues were used to perform immunohistochemistry. We retrieved 11 FFPE samples of the 22 FF samples that were used for discovery. The T cell markers were stained with an automated IHC staining system (Ventana BenchMark ULTRA; Ventana Medical System Inc. Tucson, AZ). CD3 (0.4 µg/mL, clone 2GV6, Ventana, Tuscan, AZ) antibody was used as general marker for T cells; CD4 (2.5 µg/mL, clone PS35, Ventana, Tuscan, AZ) and CD8 (1:100 dilution, clone C8/144B, Dako, Heverlee, Belgium) were used as markers for T helpers and T cytotoxic subsets, respectively. The staining was performed according to the manufacture instructions. In addition, GBP1 (1:250 dilution, Santa Cruz Biotechnology, Heidelberg Germany) was used according to the manufacture instructions. In addition to the 11 FFPE samples of the discovery set, we used 20 independent samples for extra validation. The clinical information of the samples is summarized in Table 1.
In vivo mouse model
T cell isolation and tissue culture
T cells were obtained from the spleen of 4–6-week-old FVB mice weeks. Splenocytes were sorted using anti-mouse CD4-Pe (BD Pharmigen catalog. 553048) and activated with anti-mouse CD3e clone 145-2C11 (BD Pharmigen catalog. 553066) coated plates, soluble anti-mouse CD28 (37.51) (Tonbo Biosciences catalog. 70-0281-U500), and mouse IL-2 (Miltenyi Biotec catalog. 130-094-054). The ErbB2-P cell line was established from MMTV driven-NeuNT transgenic mammary tumors in mice . They express Luciferase and gfp. ErbB2-P are cultured in vitro in DME media supplemented with 10% fetal bovine serum (FBS), 2 mM l-Glutamine, 100 IU/ml penicillin/streptomycin, and 1 mg/ml amphotericin B. The ErbB2-P was co-cultured with T cells in RPMI medium supplemented with 10% FBS and Penicillin–Streptomycin for 3 days and then sorted out and injected in syngeneic animals into the heart. 3 weeks later, metastasis burden was analyzed by bioluminescence imaging (BLI).
RNA (QIAGEN) from sorted gfp + ErbB2-P cancer cells was used to generate cDNA (iScript cDNA Synthesis Kit Bio-Rad catalog. 1708890). Gene expression was analyzed using SYBR green gene expression assays (GoTaq® qPCR Master Mix Promega catalog. A6002). Primers: Gbp1 (Sequence of the primer 5–3): 5′-3′GGGCAGCTGTCTTTGGGTAGAC, 3′-5′AGCATGAGGCCCTAGGAGCTGT. Quantitative PCR reaction was performed on QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems) and analyzed using the software QuantStudio 6 and 7 Flex Software.
ErbB2-P and T cells were isolated from the co-culture based on gfp expression using FACS ARIA Ilu sorter.
All animal experiments were done in accordance with a protocol approved by the CNIO, Instituto de Salud Carlos III and Comunidad de Madrid Institutional Animal Care and Use Committee. ErbB2-P cells were injected intracardiacally. Briefly, a cell suspension containing 105 ErbB2-P cells in a volume of 100 μl was injected in the left cardiac ventricle of anesthetized 4–6-week-old FVB mice. Tumor burden was evaluated by bioluminescence imaging using the IVIS-200 imaging system from Xenogen as previously described. Metastases were defined with BLI as those with signal equal or above 2x103 photon flux.
P value is calculated using two-tailed t test.
Tissue for immunofluorescence was obtained after overnight fixation with PFA 4% at 4 °C. Slicing of the brain was done using a sliding microtome (Fisher). Brain slices (80 μm) were blocked in NGS 10%, BSA 2%, Triton 0.25% in PBS for 2 h at room temperature (RT). Primary antibodies were incubated overnight at 4C in the blocking solution and the following day for 30 min at RT. After extensive washing in PBS-Triton 0.25%, the secondary antibody was added in the blocking solution and incubated for 2 h. After extensive washing in PBS-Triton 0.25%, nuclei were stained with Bis-Benzamide for 7 min at RT. Primary antibody GFP (Aves Labs, ref. GFP-1020, 1:1000). Secondary antibody is Alexa-Fluor anti-chicken 488 (Invitrogen).
Quantification of brain metastases histology
80 um sections were obtained from each brain generating 10 series. One series containing 10–12 slices representative of the whole organ was used for immunofluorescence analysis. The staining of gbp1 was performed and positive lesions were quantified under a fluorescence microscope. Total number of metastases per series was obtained and plotted.