Opposing roles of CXCR4 and CXCR7 in breast cancer metastasis

All MTLn3 cell lines were grown in alpha MEM supplemented with 5% FBS (100-106; Gemini Bio-Products, West Sacramento, CA, USA) and 0.5% penicillin/streptomycin (15140-122; Invitrogen, Grand Island, NY, USA). To create the human CXCR4 expressors, hCXCR4 was transferred from the pDNR-Dual hCXCR4 vector (Harvard Institute of Proteomics, Boston, MA, USA), to the JP1520 retroviral vector following the Creator Cloning protocol, using Cre recombinase (Clontech, Mountain View, CA, USA) and Max Efficiency DH5alpha bacteria (Life Technologies, Grand Island, NY, USA) grown in 7% sucrose, 30 μg/ml chloramphenicol plates. Colonies were picked and correct insertion of human CXCR4 verified by sequencing. The human CXCR7 sequence was digested out from the pcDNA 3.1+ plasmid (kindly provided by ChemoCentryx, Mountain View, CA, USA) using NotI, the ends blunted using DNA Polymerase I, large Klenow fragment (NEB, Ipswich, MA, USA), to insert into JP1520, which was digested with BamHI and BbsI removing the loxP site, ends blunted as above and treated with Antarctic phosphatase (NEB, Ipswich, MA, USA). Both insert and vector were gel purified using the Qiagen Gel extraction kit, then ligated using a Rapid DNA Ligation kit (Roche, Branchburg, NJ, USA). Subcloning efficiency DH5alpha bacteria (Life Technologies, Grand Island, NY, USA) were transformed with the ligated vector and colonies screened for correct insertion of hCXCR7 using differential enzyme digestions, followed by verification using sequencing analysis. MTLn3-GFP (MTLn3 cells expressing green fluorescent protein) cells were transduced with either the empty JP1520 vector, JP1520-CXCR4, JP1520-CXCR7, or both CXCR4 and CXCR7, by first transfecting Phoenix packaging cells with 2 μg of each vector using lipofectamine (Invitrogen, Grand Island, NY, USA), collecting virus and transducing MTLn3-GFP cells seeded at 60% confluency. Transduced cells were selected with 1 μg/ml puromycin. MTLn3 CXCR7 and MTLn3 CXCR4-CXCR7 cells were subsequently fluorescence-activated cell sorting (FACS) sorted in a DakoCytomation MoFlo to obtain a homogenous population of CXCR7 expressing cells. MDA MB 435 cell lines were grown in DMEM (10-013 CV, Cellgro, Manassas, VA, USA) supplemented with 10% FBS (S11550, Atlanta Biologicals, Lawrenceville, GA, USA) and 0.5% penicillin/streptomycin. 435 cells seeded at 60% confluency, were transduced with either the empty JP1520 vector, JP1520-CXCR4 or JP1520-CXCR7 using premade virus, with transductants selected using 1 μg/ml puromycin. MDA MB 435 CXCR7 cells were FACS sorted in a DakoCytomation (Carpinteria, CA, USA) MoFlo to obtain a homogenous population of CXCR7 expressing cells. The 435 double overexpressors, CXCR4-CXCR7, were made by transducing sorted 435-CXCR7 cells with JP1520-CXCR4 virus and then FACS sorted for high CXCR4 expression. MDA MB 435 CXCR4 cells were sorted at the same time to obtain cell lines with homogenous CXCR4 expression.

MTLn3 cells grown to 70 to 85% confluency were used for RNA isolation using the Qiagen RNeasy Mini kit with DNase I treatment (Valencia, CA, USA). A 1 μg sample of total RNA was used for reverse transcription using Superscript III and random hexamers in a 20 μl reaction volume. A 2 μl aliquot of the reaction was used for PCR using Taq polymerase for 30 cycles. Primers used were: rat glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (PPR06557A Superarray), rat CXCR4 (5' AGGAACTGAACGCTCCAGAA 3' and 5' AACCACACAGCACAACCAAA 3'), human CXCR4 (5' CTCCAAGCTGTCACACTCCA 3' and 5' TCGATGCTGATCCCAATGTA 3'), human and rat CXCR7 (5' GCACTACATCCCGTTCACCT 3' and 5'AAGGCCTTCATCAGCTCGTA 3'). PCR products were run in a 1.5% agarose gel containing ethidinium bromide. For quantitation of endogenous expression of rat CXCR4, the collected cDNA was used for quantitative real-time PCR with SYBR Green (PA-012, SuperArray Biosciences, Frederick, MD, USA) and an Applied Biosystems 7900HT (Carlsbad, CA, USA). To evaluate the expression of matrix metalloproteinases (MMPs) in the different MTLn3 transductants, the cell lines were grown to 80 to 90% confluency in six-well plates, starved overnight in alpha-MEM/0.35% BSA in a 37°C incubator, and then stimulated for four hours with 10 nM CXCL12 (460-SD; R&D systems, Minneapolis, MN, USA) in alpha-MEM/0.35% BSA or just alpha-MEM/0.35% BSA at 37°C. RNA was extracted using the Qiagen RNeasy Mini kit with DNase I treatment. A 2 μg sample of total RNA was used for reverse transcription using a Superscript First Strand kit (11904-018, Invitrogen, Grand Island, NY, USA) and cDNA used for real time PCR with SYBR Green (PA-012) on an Applied Biosystems 7900HT. Rat specific primers for MMPs were obtained from real-time primers (Elkins Park, PA, USA). RNA expression of MMPs was normalized to GAPDH.

To analyze the levels of CXCR4 and CXCR7 expression in the MTLn3 and MDA MB 435 cell lines, cells were grown to 80% confluency, detached at 37°C using PBS without Ca²⁺/Mg²⁺ +2 mM ethylenediaminetetraacetic acid (EDTA) and resuspended in 1 ml cold PBS without Ca²⁺/Mg²⁺ supplemented with 0.2% BSA. Cells were labeled with either control mouse IgG antibody (MAB002; R&D systems, Minneapolis, MN, USA), anti-human CXCR4 antibody (MAB172; R&D systems, Minneapolis, MN, USA) or anti-human CXCR7 antibody (11G8; ChemoCentryx, Mountain View, CA, USA) for 45 minutes at 4°C. Unbound primary antibody was removed by washing and bound antibody was detected with APC antimouse secondary antibody (Jackson ImmunoResearch, West Grove, PA, USA). Expression of rat CXCR4 protein was evaluated in the MTLn3 transductants using a rat specific CXCR4 antibody (ab7199, Abcam, Cambridge, MA, USA) with a rabbit IgG as a control (011-000-003, Jackson ImmunoResearch, West Grove, PA, USA) and anti-rabbit DyLight 649 as the secondary antibody (111-496-144, Jackson ImmunoResearch, West Grove, PA, USA). Fluorescently labeled cells were evaluated using a Becton Dickinson LSRII (Franklin Lakes, NJ, USA). FCS files were analyzed using FlowJo software (Ashland, OR, USA).

Chemotaxis was evaluated using a 48-well microchemotaxis chamber (Neuroprobe, Gaithersburg, MD, USA) and PVP-free 8 μm pore polycarbonate filters (Neuroprobe, Gaithersburg, MD, USA) coated with 27 μg/ml rat tail collagen type I (BD Bioscience, Franklin Lakes, NJ, USA). Cell lines were starved in L15 medium supplemented with 0.35% BSA for three hours at 37°C, detached using PBS without Ca²⁺/Mg²⁺ + 2 mM EDTA and resuspended in L15-0.35% BSA to plate 2 × 10⁴ cells per well for MTLn3 transductants, or 1.5 × 10⁴ cells per well for the MDA MB 435 transductants. CXCL12 solutions were prepared in L15-0.35% BSA and placed in the bottom wells with cells plated in the top wells of the assembled chamber. To inhibit CXCL12 binding to CXCR7, we added 10 nM I-TAC (572-MC; R&D systems, Minneapolis, MN, USA), or used the CXCR7 inhibitors CCX733 (ChemoCentryx, Mountain View, CA, USA) and CCX771 (ChemoCentryx, Mountain View, CA, USA) added to both top and bottom wells, including vehicle (DMSO) as a control. To inhibit CXCR4, we added AMD3100 (Sigma, St. Louis, MO, USA) to both top and bottom wells. After a four-hour incubation at 37°C, filters were placed in 10% formalin solution to fix the cells for 30 minutes, cells on the top of the filter, non-migrating cells, were removed using a cotton swab, and migrating cells subsequently stained overnight in hematoxylin. The number of cells crossing the filter in one representative 10× field was counted per well for the MTLn3 transductants, using a Nikon Labophot light microscope (Melville, NY, USA), and corresponding wells averaged per experiment. To determine MDA MB 435 chemotaxis, the number of cells that crossed each well were counted and corresponding wells averaged per experiment.

MTLn3 transductants grown to 70 to 85% confluency were starved for three hours in alpha-MEM supplemented with 0.35% BSA in a 37°C incubator. Cells were detached using PBS without Ca^2+/Mg²⁺ containing 2 mM EDTA, resuspended in alpha-MEM supplemented with 0.35% BSA to plate 1 × 10⁵ cells in a 500 μl volume on top of Matrigel-precoated 8 μm pore-filters (354480; BD Biosciences transwells, Franklin Lakes, NJ, USA) that had been equilibrated for one hour with alpha-MEM/0.35% BSA in a 37°C incubator. Cells were allowed to invade overnight in a 37°C incubator in response to either alpha-MEM/0.35% BSA alone or containing 10 nM CXCL12. The filters were fixed in 10% formalin for 30 minutes and stained with crystal violet for 15 minutes. Cells that had not invaded were removed with a cotton tip applicator from the top of the filter, filters removed, placed in a coverslip and the total number of invading cells present in a filter counted using a Nikon Labophot light microscope with a 10× objective.

MTLn3 CXCR4 or MTLn3 CXCR4-CXCR7 GFP labeled cells were plated overnight on MatTek dishes, at 1 × 10⁵ cells/dish, over a thin Alexa 405-gelatin matrix in the presence of the protease inhibitor GM6001 (10 μm). Cells were subsequently starved for three hours, washed three times in starvation media [37] and stimulated with 5 nM CXCL12 for six hours. At the end of the incubation time, cells were fixed in 3.7% paraformaldehyde and imaged. The images were processed using the ImageJ Spot enhancing filter 2D (3.0 pixels Gaussian filter) and the threshold levels set to select only degradation areas. The degradation area was normalized to the cell coverage area in the GFP channel. Alexa405 (A30000, Invitrogen, Grand Island, NY, USA) was conjugated to gelatin (G2500, Sigma, St. Louis, MO, USA) and thin matrix Alexa405-gelatin matrix was prepared as previously described [38]. Results are reported as the degraded area/cell area per field normalized to the MTLn3 CXCR4 unstimulated levels.

All animal procedures were conducted observing the National Institutes of Health regulations on the use and care of experimental animals. Our animal protocol was approved by the Albert Einstein College of Medicine animal use committee. Female severe combined immunodeficiency (SCID) mice aged four to seven weeks from NCI were used for all experiments. To form primary tumors, MTLn3 transductants grown to 70 to 85% confluency were detached using PBS without Ca²⁺/Mg²⁺ + 2 mM EDTA, resuspended in cold PBS supplemented with 0.2% BSA to inject 5 × 10⁵ cells per animal in a 100 μl volume. Cells were injected into the fourth mammary fat pad and tumors allowed to grow until they reached an average volume of 1,300 mm³ for the in vivo invasion assay. Mice were anesthetized using isoflurane and blocking needles placed into the primary tumors using micromanipulators (MN-151; Narishige, East Meadow, NY, USA). Hamilton 33 gauge needles were loaded with a mixture of EDTA, 10% matrigel and CXCL12 dissolved in L15-0.35% BSA, and these experimental needles used in place of the blocking needles after the animal was appropriately setup. Detailed information about this assay can be found in [39]. To block the colony stimulating factor 1 (CSF-1) receptor on the mouse tumor associated macrophages we used the anti-mouse CSF-1R antibody (AFS98) [40] at 15 μg/ml; to block epidermal growth factor (EGF)-derived from the mouse tumor associated macrophages from binding to EGF receptor (EGFR) we used a neutralizing EGF antibody at a concentration of 20 μg/ml (AF2028; R&D systems, Minneapolis, MN, USA). An isotype IgG antibody (012-000-007 or 111-005-144, Jackson ImmunoResearch, West Grove, PA, USA) was used at the same concentration as the blocking/neutralizing antibodies as a control. To inhibit CXCR4 we used 100 nM AMD3100. Cells were allowed to invade into the needles for four hours and the contents of the needles were subsequently extruded into coverslips, invasive cells stained with 4',6-diamidino-2-phenylindole (DAPI) and counted using an Inverted Olympus IX70 microscope (Center Valley, PA, USA).

Primary tumors of an average volume of 1,300 mm³ were used for intravital imaging. Mice were anesthetized with isoflurane and a skin-flap surgery carefully performed to expose the primary tumor while minimizing damage to tissue and blood vessels. Animals were placed on the stage of an inverted microscope and the GFP-labeled carcinoma cells imaged using an Olympus Fluoview FV1000-MPE microscope (Center Valley, PA, USA) at an excitation of 880 nm with a 25 × 1.05 NA water objective. Collagen fibers were visualized by second harmonic generation. Time-lapse Z-series were taken at 5 μm steps for a total of 100 μm into the tumor over 30 minutes at two-minute intervals. Movies were analyzed using Image J http://​imagej.​nih.​gov/​ij/​. A cancer cell was considered to be motile when it had protruded/translocated at least half a cell length, and the total number of motile cancer cells in a 50 μm Z-stack time-lapse movie was determined. A more detailed description of this protocol can be found in [41].

Primary tumors were allowed to grow until they reached an average volume of 1,500 mm³ to perform end-point metastasis assays. At this time, mice were anesthetized using isoflurane and blood collected via cardiac puncture from the right side of the heart to obtain cancer cells that had intravasated. The blood drawn was then plated into a 10 cm dish with alpha MEM supplemented with 5% FBS/0.5% P/S, and cancer cell colonies allowed to grow for a week in a 37°C incubator followed by counting using a light microscope. Tumor blood burden is reported as the total number of cancer cell colonies present in a dish normalized to the volume of blood plated. Lungs and primary tumors were harvested and fixed in 10% formalin solution. The lungs were paraffin-embedded, sectioned and stained with H&E to count the number of lung metastasis present in all lobes of a single section using a light microscope with a 10× objective. To evaluate lymph node metastasis, both axillary and inguinal lymph nodes were removed from tumor-bearing mice, fixed in 10% formalin solution, paraffin-embedded, sectioned and stained with H&E. The presence of metastases was assessed using a Nikon Labophot light microscope (Melville, NY, USA). To estimate bone marrow metastasis, the femur ipsilateral to the site of primary tumor growth was dissected and bone marrow was flushed using 1 ml syringes with 25-gauge needles into a 10 cm plate containing alpha MEM supplemented with 5% FBS/0.5% P/S. Plates were incubated at 37°C for a week and tumor colonies then counted.

For microvessel density evaluation, formalin-fixed, paraffin-embedded sections from MTLn3 JP, MTLn3 CXCR4, MTLn3 CXCR7, and MTLn3 CXCR4-CXCR7 primary tumors were deparaffinized, rehydrated, blocked in donkey serum and stained with rat antimouse CD34 antibody (CL8927AP; Cedarlane labs, Burlington, NC, USA) at a 1:400 dilution for one hour. Slides were washed and subsequently stained with a biotinylated antirat secondary antibody for 50 minutes. The slides were rinsed and exposed to ABC-HRP (PK-6100, Vector, Burlingame, CA, USA) for 20 minutes, washed and exposed to diaminobenzidine (DAB) for one to four minutes (SK-4100, Vector, Burlingame, CA, USA), and subsequently counterstained with Harris hematoxylin (s212, Poly-scientific, Bay Shore, NY, USA), rinsed and mounted. Mean vessel density was determined by counting the number of blood vessels present per field seen in a light microscope using a 10× objective. A total of three different primary tumors were used per cell line, counting five fields per tumor. For VEGFA evaluation, samples for immunohistochemistry (IHC) were sectioned at 5 μm, deparaffinized in xylene followed by graded alcohols. Antigen retrieval was performed in 10 mM sodium citrate buffer at pH 6.0, heated to 96C, for 20 minutes. Endogenous peroxidase activity was quenched using 3% hydrogen peroxide in PBS for 10 minutes. Blocking was performed by incubating sections in 5% normal donkey serum with 2% BSA for one hour. The primary antibody to VEGFA, (PAB12284, ABNOVA, Walnut, CA, USA) was used at 1:250 for 1.5 hours at room temperature. The primary species (rabbit IgG) was substituted for the primary antibody to serve as a negative control. The sections were stained by routine IHC methods, using HRP rabbit polymer conjugate (Invitrogen, Grand Island, NY, USA), for 20 minutes to localize the antibody bound to antigen, with diaminobenzidine as the final chromogen. All immunostained sections were lightly counterstained with hematoxylin.

Analysis of variance (ANOVA) was used to demonstrate that there were significant differences between conditions when there were more than two conditions, and paired analyses were performed using either student t-test, or Mann-Whitney test in order to identify the conditions that were significantly different. Correlation of MMP12 expression with CXCR4 and CXCR7 was performed using Oncomine with the following databases: Bittner Breast, Bonnefoi Breast, Desmedt Breast, Ginestier Breast, Gluck Breast, Hess Breast, Ivshina Breast, Loi Breast, and van't Veer Breast. SPSS was used to determine correlation coefficients and their significance from the downloaded expression data. The Oncomine database was also used to identify clinical parameters with which MMP12 mRNA levels were significantly correlated. The following parameters were examined: estrogen receptor (ER) positive, triple negative, high grade, metastasis, recurrence and survival. The probabilities of overexpression or underexpression of MMP12 for all breast cancer datasets containing more than 40 samples for which the parameters were provided by Oncomine were downloaded. For each parameter, the number of datasets in which the probability of MMP12 overexpression or underexpression was less than 0.05 was identified and the binomial cumulative probability distribution was used to determine the likelihood of that number occurring by chance using SPSS (IBM, Armonk, NY).

To evaluate the roles of CXCR4 and CXCR7 in CXCL12-induced chemotaxis in vitro, the human open reading frames of these receptors were stably overexpressed in the rat mammary adenocarcinoma cell line MTLn3 using retroviral expression vectors. RT-PCR of these cell lines demonstrates clear increases in mRNA for the corresponding receptors (Figure 1a). A low level of endogenous rat CXCR4 mRNA expression in the MTLn3 cell line was confirmed by quantitative real-time PCR showing a C_t of 35 cycles (GAPDH C_t of 16). The levels of expression of CXCR4 and CXCR7 at the cell membrane were subsequently evaluated using FACS analysis (Figure 1b). Staining with a control isotype antibody is represented by a grey shaded peak in all plots. The MTLn3 JP empty vector control cell line shows low levels of expression of CXCR4 (solid line) with little expression of CXCR7 (dashed line), consistent with the PCR data. The CXCR4 transductant, MTLn3 CXCR4, shows higher expression of CXCR4 (solid line), while the CXCR7 transductant, MTLn3 CXCR7, shows higher expression of the CXCR7 receptor (dashed line). The levels of both CXCR4 and CXCR7 expression in the double transductant, MTLn3 CXCR4-CXCR7, show increases comparable with the respective single transductants. Mean fluorescence intensity values for the different transductants are included in Additional data file 1. The observed low levels of endogenous rat CXCR4 expression in the MTLn3 transductants were confirmed using a different antibody against full length rat CXCR4 [see Additional data file 2]. Although there is a slight increase in the surface expression of CXCR4 in the CXCR7 cell lines compared with control JP lines (Figure 1b) [see Additional data file 1] this was not evident in the subsequent FACS analysis using the rat specific CXCR4 antibody [see Additional data file 2], and did not produce any increase in CXCL12-induced chemotaxis compared with the JP control line (see below). In summary, we were able to increase the expression of both CXCR4 and CXCR7 at least three fold over the basal expression levels.

Having engineered MTLn3 cell lines overexpressing either CXCR4, CXCR7, or both receptors, we compared their chemotaxis to CXCL12 using a microchemotaxis chamber (Figure 1c). The MTLn3 CXCR4 cell line showed significantly increased CXCL12-induced chemotaxis (P < 0.005) compared with the control cell line, MTLn3 JP. The MTLn3 CXCR7 cell line showed no chemotactic response to CXCL12 (P > 0.1). Furthermore, overexpression of both CXCR4 and CXCR7 resulted in a significantly increased chemotactic response to CXCL12 compared with that of the MTLn3 CXCR4 cell line (P < 0.005). This migration phenotype suggests that although CXCR7 alone does not mediate CXCL12-induced motility, in the presence of CXCR4, CXCR7 augments CXCL12-stimulated motility. This was confirmed in the cancer cell line MDA-MB-435 previously shown to express low endogenous levels of CXCR4 [42] [see Additional data file 2], which also showed the CXCR4-CXCR7 double overexpressors to have the most chemotaxis to CXCL12.

To test whether the increased motility observed in the MTLn3 CXCR4-CXCR7 cell line was dependent on CXCR4, we added the CXCR4 inhibitor, AMD3100, at a concentration that specifically inhibits CXCR4 without acting as an agonist of CXCR7 [43]. Addition of AMD3100 significantly decreased CXCL12-induced chemotaxis of both MTLn3 CXCR4 and MTLn3 CXCR4-CXCR7 cells (P < 0.005) (Figure 1d) suggesting that CXCR4 is important for CXCL12-induced chemotaxis in the double overexpressors. Addition of I-TAC (CXCL11), a chemokine that binds to CXCR7 [2, 22], failed to act as a chemoattractant or impair CXCL12-induced chemotaxis for either MTLn3 CXCR4, or MTLn3 CXCR4-CXCR7 cells (P > 0.5) (Figure 1e). Similarly, addition of the CXCR7 inhibitors CCX771 and CCX733 did not inhibit chemotaxis to CXCL12 in the MTLn3 CXCR4-CXCR7 cells (P > 0.4) (Figure 1f). These results show that in vitro, CXCR7 alone does not mediate CXCL12-induced chemotaxis. However, when expressed in cells with high levels of CXCR4, CXCR7 augments CXCR4-mediated chemotaxis to CXCL12, in agreement with recent studies using MDA-MB 231 cells [36]. Importantly, the chemotaxis response of the double overexpressors, CXCR4-CXCR7, was impaired in the presence of CXCR4 inhibitor, AMD3100, but not upon addition of ITAC or the CXCR7 inhibitors CCX771 or CCX733 suggesting that binding of CXCL12 to CXCR4 but not CXCR7 is needed for chemotaxis to occur.

We next tested the role of these receptors in CXCL12-induced invasion in vitro using transwells precoated with Matrigel. Although overexpression of CXCR4, CXCR7, or both CXCR4 and CXCR7 did not affect MTLn3 basal invasion in vitro as measured in the presence of just buffer (P > 0.6) (Figure 2a, gray bars), CXCR4 expression enabled an invasive response of MTLn3 cells to CXCL12 in vitro (P < 0.05). CXCR7 expression alone failed to significantly enhance invasion to CXCL12 (P = 0.9). Surprisingly, in the context of CXCR4 overexpression, CXCR7 inhibited CXCL12-induced invasion compared with MTLn3 CXCR4 cells (P < 0.05).

To measure the effects of CXCR4 and CXCR7 overexpression on invasion in response to CXCL12 in vivo, we injected MTLn3 JP, MTLn3 CXCR4, MTLn3 CXCR7, and MTLn3 CXCR4-CXCR7 cell lines labeled with GFP into the fourth mammary fat pads of SCID mice and allowed the tumors to grow until they reached an average volume of 1,300 mm³ for in vivo invasion analysis. FACS analysis of the primary tumors confirmed that overexpression of CXCR4 and CXCR7 in the respective cell lines was conserved in vivo [see Additional data file 3]. Needles containing Matrigel and CXCL12 as a chemoattractant were inserted in the primary tumors and invasive cells collected for a four-hour period. Overexpression of CXCR4 dramatically increased the in vivo invasive behavior of MTLn3 cells to CXCL12, with peak invasion shifted to lower concentrations (Figure 2b). CXCR7 overexpression alone, however, did not enhance the ability of MTLn3 cells to invade in response to CXCL12, as they failed to invade in vivo at the highest concentration of CXCL12 (62.5 nM) tested compared with buffer levels, a concentration which induced a three-fold increase in invasion above background in the MTLn3 JP tumors. Similarly, expression of CXCR7 in the presence of CXCR4 also impaired the invasive response to CXCL12 at all concentrations, with the peak response at 15.6 nM being significantly reduced in the MTLn3 CXCR4-CXCR7 tumors (P < 0.005). Invasion in response to 6.25 nM CXCL12 was not tested for the MTLn3 CXCR7 strain since responses to the higher concentrations were similar to the buffer response and the in vitro chemotaxis data did not demonstrate any response at lower concentrations. These results are consistent with the in vitro invasive behavior of these cell lines, confirming that CXCR7 plays a negative role in CXCL12-induced invasion. The relatively weak in vivo invasive response seen at 15.6 nM CXCL12 in the MTLn3 CXCR4-CXCR7 cell line is significantly impaired upon addition of the CXCR4 inhibitor AMD3100 (P < 0.05; Figure 2c), indicating the remaining response is still mediated by CXCR4. As there is no increase in CXCR4 expression seen in the MTLn3 JP tumor cells [see Additional data file 3], the response seen in MTLn3 JP tumors to high levels of CXCL12 may reflect initiation of the paracrine loop by tumor-associated macrophages, which express CXCR4 [44‐46].

These data show that CXCR7 expression alone plays no role in CXCL12-induced motility (chemotaxis or invasion). However, CXCR7 expression in the context of high levels of CXCR4, while enhancing chemotaxis to CXCL12, impairs invasion in vitro and in vivo to CXCL12. These results raised the possibility that CXCR7 inhibits the ability to degrade extracellular matrix in response to CXCL12 stimulation and hence invasion. To address this, we measured the ability of MTLn3 CXCR4 and MTLn3 CXCR4-CXCR7 cells to degrade fluorescently labeled matrix (Figure 2d). Although there was no significant difference in the ability of these cell lines to degrade matrix in the absence of stimulation, when exposed to CXCL12 only the CXCR4 overexpressors showed significantly increased degradation (P < 0.05), while the double overexpressors showed no increase, indicating that CXCR7 expression impairs CXCL12-induced invasion by suppressing CXCL12-induced matrix degradation. Evaluation of the expression of different MMPs (MMP1, MMP2, MMP3, MMP7, MMP9-14) at the mRNA level revealed that MMP12 was significantly higher in the CXCR4 line compared with the other lines after stimulation with CXCL12 (P < 0.01 by ANOVA) [see Additional data file 4]. Western blotting after CXCL12 stimulation also indicated that MMP12 expression was highest in the CXCR4 line [see Additional data file 4]. These results suggest that there is differential regulation of MMPs in the MTLn3 CXCR4 and CXCR4-CXCR7 cells upon CXCL12 stimulation with the CXCR4-expressing cells showing increased expression of MMPs such as MMP3, MMP10, and MMP12. Evaluation of breast cancer data present in the Oncomine database indicates that only MMP12 expression is significantly correlated with CXCR4 expression: from the nine breast cancer databases evaluated, the average correlation was 0.31 with an average P value for the correlation of less than 0.04. In the same databases, the average correlation of MMP12 with CXCR7 was -0.03, which was not significant (P < 0.46).

We had previously reported that CXCL12 could induce in vivo invasion in the MMTV-PyMT transgenic breast cancer model, and that this invasion was dependent on the EGF/CSF-1 paracrine loop between cancer cells and macrophages where CSF-1 is secreted by cancer cells and stimulates macrophages to produce EGF [21]. To test whether CXCL12 induced in vivo invasion in the MTLn3 CXCR4 model was also dependent on EGF/CSF-1 signaling, we tested the ability of these cells to invade in vivo in response to CXCL12 in the presence of either a neutralizing EGF antibody or a blocking CSF-1R antibody [40]. The MTLn3 CXCR4 invasive response to CXCL12 was significantly impaired in the presence of either antibody [see Additional data file 5], indicating EGF/CSF-1 signaling is required for CXCL12 induced in vivo invasion in this model. Thus CXCR4 overexpression did not override the dependency of these breast cancer cells on the EGF/CSF-1 paracrine loop for in vivo invasion.

Given the difference in the in vitro chemotactic and invasive behavior in response to CXCL12 stimulation of the double overexpressors, MTLn3 CXCR4-CXCR7, we proceeded to evaluate their motile behavior within the tumor microenvironment. The GFP labeled MTLn3 transductants were orthotopically injected in SCID mice and the tumors used for intravital imaging when they had reached an average volume of 1,300 mm³. Time-lapse z-series were taken using multiphoton microscopy to evaluate the number of cancer cells moving within the tumor microenvironment. Similar to the in vitro invasion results, CXCR4 overexpression significantly enhanced the motile behavior of MTLn3 cells within the primary tumor - about three-fold compared with MTLn3 JP cells (P < 0.005; Figure 3a). CXCR7 overexpression alone did not have a significant effect on cancer cell motility within the primary tumor compared with MTLn3 JP (P = 0.18). The double overexpressors, MTLn3 CXCR4-CXCR7, showed an intermediate phenotype that was greater than MTLn3 JP intravital motility (P < 0.05), but reduced compared with MTLn3-CXCR4 (P = 0.06). Representative images of cancer cell motility in each tumor are shown (Figure 3b) with representative movies included as Additional data files 6,7,8 to 9. In summary, the motile behavior of the different transductants in vivo most closely resembled their in vitro invasive response to CXCL12. Namely, while CXCR4 overexpression enhanced the motility of MTLn3 cells within the tumor, CXCR7 overexpression alone had no effect and in the context of CXCR4 overexpression, high levels of CXCR7 resulted in decreased motility. This suggests that despite the increased chemotactic response of the CXCR4-CXCR7 expressors to CXCL12 in vitro, their ability to invade extracellular matrix is impaired by their reduced degradation potential.

As CXCR4 and CXCR7 have been found to play a role in breast cancer growth and metastasis, we tested the effects of CXCR4 and CXCR7 overexpression on primary tumor growth, intravasation and lung metastasis formation of MTLn3 cells. Comparison of the volumes of MTLn3 CXCR4 tumors to the control MTLn3 JP tumors showed no difference (P = 0.38; Figure 4a), indicating that CXCR4 overexpression does not enhance growth of MTLn3 tumors. On the other hand, MTLn3 CXCR7 tumors showed significantly increased tumor size compared with the control MTLn3 JP tumors (P < 0.05), and similarly the MTLn3 CXCR4-CXCR7 double overexpressors were significantly bigger than either MTLn3 JP or MTLn3 CXCR4 tumors (P < 0.05); indicating that CXCR7 but not CXCR4 plays a role in enhancing tumor growth. Comparison of sizes of CXCR7 and CXCR4 tumors indicated a trend towards significance (P < 0.062). MTLn3 cells expressing CXCR7 alone or coexpressed with CXCR4-formed tumors that were visible macroscopically by day 12, while the control JP and CXCR4 expressing tumors were not visible until day 15 (data not shown). Importantly, we did not see a difference in the growth of these cell lines in vitro (data not shown). However, in line with previous reports [23, 27], we observed increased blood vessel density in MTLn3 CXCR7 and MTLn3 CXCR4-CXCR7 primary tumor sections compared with control MTLn3 JP (P < 0.005) and MTLn3 CXCR4 tumors (P < 0.05), respectively (Figure 4b) using a CD34 antibody, suggesting that CXCR7 overexpression increases the growth of the primary tumor by stimulating angiogenesis. MTLn3 CXCR4 tumors showed blood vessel density comparable with that of MTLn3 JP tumors (P = 0.9). Previous studies have shown that CXCR7 expression can stimulate angiogenesis by inducing the secretion of VEGF [23, 27]. Consistent with those studies, we found that VEGFA (Vascular endothelial growth factor A) expression is increased in MTLn3 CXCR7 and CXCR4-CXCR7 primary tumors compared to either MTLn3 JP or CXCR4 tumors (Figure 4c).

We next evaluated the effects of CXCR4 and CXCR7 overexpression on the process of intravasation, by doing intracardiac punctures prior to euthanization of mice carrying tumors of an average volume of 1,500 mm³. As shown in Figure 4d, MTLn3-CXCR4 cells were significantly more efficient in entering the bloodstream compared with MTLn3-JP and CXCR4-CXCR7. Both MTLn3 CXCR7 and MTLn3 CXCR4-CXCR7 tumor models showed low numbers of intravasated cancer cells. These results agree with the increased invasive and motile behavior of the CXCR4 overexpressing MTLn3 cells seen within the primary tumor, and the impaired ability of the CXCR7 overexpressors to invade in vivo.

To compare the ability of the MTLn3 transductants to spontaneously metastasize from the primary tumor to the lung, we evaluated lung metastasis formation in animals that had been injected in the mammary fat pad with the transduced MTLn3 cells. Lungs were harvested when the primary tumors reached an average volume of 1,500 mm³, fixed, paraffin-embedded and sectioned to count the number of metastases present per H&E section. The number of spontaneous lung metastases present in the MTLn3 CXCR4 model was not significantly different to the number of lung metastases formed in the empty vector control model, MTLn3 JP (P = 0.3 student's t-test and P = 0.8 Mann-Whitney; Figure 4e), nor was there any difference in the size of the metastases (data not shown). This was an unexpected result as MTLn3 CXCR4 cells were more efficient in leaving the primary tumor and entering the bloodstream compared with MTLn3 JP cells, thus we looked for metastasis formation in the bone and lymph nodes but no difference was observed at these sites (P = 0.7 and P = 0.8, respectively) [see Additional data file 10]. However, the MTLn3 CXCR7 tumors gave rise to significantly fewer lung metastases compared with the empty vector model, MTLn3 JP, (a, P < 0.005) and likewise, the double overexpressors, MTLn3 CXCR4-CXCR7, showed decreased lung metastasis compared with both MTLn3 JP (b, P < 0.005) and MTLn3 CXCR4 (c, P < 0.005; Figure 4e). No significant difference in size of lung metastasis, or in bone or lymph node metastasis was observed in the different CXCR7 transductants compared with control MTLn3 JP or MTLn3 CXCR4 (data not shown) [see Additional data file 10].