Macrophage Migration Inhibitory Factor–CXCR4 Is the Dominant Chemotactic Axis in Human Mesenchymal Stem Cell Recruitment to Tumors

Cell therapy is attracting growing interest as a novel therapeutic approach for a variety of diseases, including cancer. Mesenchymal stromal cells (MSCs) are the dominant candidate cell studied due to their observed capacity to migrate to sites of tissue injury and tumors after systemic administration (1). MSC homing to sites of inflammation has been extensively studied, as delineating the mechanisms behind it should lead to improved delivery of MSCs to disease sites. Defining the key responsible factors, however, has been met with inconsistency (2, 3). Tumor tropism, or the homing of MSCs to tumor tissue, is poorly understood, and many factors have been reported to affect this complex process, including a variety of receptors, extracellular matrix proteins, and soluble tumor-derived factors such as stromal-derived factor (SDF)-1, TNF-α, ILs, and chemokines (4, 5).

The most extensively studied MSC chemotactic axis is CXCR4/SDF-1. This axis has been shown important in the recruitment and retention of hematopoietic stem cells to bone marrow, where levels are high (6). Growing evidence supports CXCR4-expressing cancer cell homing to bone marrow in a similar fashion (7–10). SDF-1 has also been proposed to attract MSCs to tumors. Recently, investigators found that soluble factors secreted from tumor cells can trigger SDF-1 secretion from MSCs, activating their migration (4). The role of SDF-1 in MSC homing to tumor cells, however, is disputed, and several studies show that tumors do not produce SDF-1 (11).

The delineation of macrophage migration inhibitory factor (MIF) function is rapidly developing, and we now realize it is not simply a cytokine- modulating monocyte motility, but a pleiotropic regulator of an array of cellular and biological processes. MIF is overexpressed in a large variety of human cancers, including pancreatic, breast, prostate, colon, brain, skin, and lung (12–18). MIF expression closely correlates with tumor aggressiveness and metastatic potential, suggesting an important contribution to disease severity and survival (13, 19–21).

Three receptors for MIF have been identified. The cell surface–expressed form of CD74 (22) was identified as a high-affinity MIF receptor on class II–positive cells, including monocytes/macrophages and B lymphocytes (23). However, upon inflammatory stimulation, surface CD74 can be detected on the plasma membrane of class II–negative cells, including stromal and epithelial cell types (24, 25). The CD74 receptor possesses a short cytoplasmic N terminus. Therefore, accessory signaling molecules such as Src, CD44, c-Met, or other receptors are necessary to mediate CD74 signaling by MIF, forming a functional receptor-tyrosine-kinase–like complex (26, 27)

MIF is also a noncognate, high-affinity ligand for the promiscuous chemokine receptors CXCR2 and CXCR4 (26, 28, 29) that also bind to several chemokine ligands, including IL-8 and CXCL1 to CXCR2, and SDF-1 to CXCR4 (30-32). MIF binds to CXCR2 with low nanomolar affinity and induces CXCR2-mediated leukocyte arrest and chemotaxis (26).

CXCR4, as CXCR2, belongs to G protein–coupled receptor family. By activating CXCR4, MIF promotes T cell chemotaxis. Accordingly, numerous in vivo studies in proatherogenic mouse models have demonstrated that the MIF–CXCR4 axis critically contributes to atherogenic leukocyte recruitment and atheroprogression (26, 33–35). The MIF–CXCR4 axis also regulates endothelial progenitor cell migration and cancer cell metastasis (36–38).

In this study, we define MIF as the key determinant of MSC tumor tropism. We demonstrate a novel role for MIF secreted from tumor cells, as responsible for attracting MSCs requiring activation of ERK and JNK dominantly through CXCR4. Importantly, we show that, through manipulation of this chemokine–receptor axis, we can alter MSC homing in vitro and in vivo and establish its importance in future clinical applications.

MSCs were isolated from human bone marrow (Royal Free, University College London). MSC adipogenic and osteogenic differentiation capacities were confirmed, as described previously (39). MSCs were cultured in αMEM supplemented with 20% FBS, 2 mM l-glutamine, and streptomycin/penicillin (Invitrogen). A549 were cultured in F12 supplemented with 10% FBS, 2 mM l-glutamine, and streptomycin/penicillin (Invitrogen). MDAMB231, H376, A431, and U87MG cells were cultured in DMEM supplemented with 10% FBS, 2 mM l-glutamine, and streptomycin/penicillin (Invitrogen). Jurkat cells were cultured in RPMI 1640 supplemented with 10% FBS, 2 mM l-glutamine, and streptomycin/penicillin (Invitrogen).

RNA was extracted from MSCs with TRIzol reagent (Invitrogen), followed by cleanup and DNase I treatment with Qiagen RNeasy mini kit (Qiagen), following manufacturer’s instructions. RNA quantification was carried out using the Quant-iT RNA assay kit with the Qubit fluorometer (Life Technologies). RNA integrity was analyzed using a Bioanalyzer 2100 (Agilent), and only RNAs with a RNA integrity number >8.5 were used for the microarray experiments. RNA was synthesized, amplified, and purified using the Illumina TotalPrep RNA Amplification Kit (Life Technologies), following manufacturer recommendations. Briefly, 500 ng RNA was reverse transcribed. After second strand synthesis, the cDNA was transcribed in vitro and cRNA labeled with biotin-16-UTP. Labeled probe hybridization to Illumina Human HT-12 v4 Expression BeadChip (∼48 000 probes) was carried out using Illumina’s protocol. Beadchips were scanned on the Illumina BeadArray 500GX Reader using Illumina BeadScan image data acquisition software. RNA control samples were analyzed with each run. A gene was considered differentially expressed if the Benjamini-Hochberg–corrected p value was <0.05. For microarray validation, total RNA was reverse transcribed using qScriptTM cDNA Super-Mix (Quanta Biosciences), according to the manufacturer’s protocol. Real-time quantitative PCR (qPCR) was carried out using the SYBR-Green master mix (Applied Biosystems) in an Eppendorf real-time PCR machine. Six genes were chosen on the basis of significant differential expression on the microarray and biological relevance to the experiment. Correlations between microarrays and real-time PCR data were measured using the Pearson coefficient. All microarray data reported in this study are described in accordance with MIAME guidelines and have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus public repository; they are accessible through Gene Expression Omnibus accession number GSE46156 (www.ncbi.nlm.nih.gov/geo/).

A549, MDAMB231, H376, A431, Jurkat, and U87MG secretomes were analyzed collecting the medium of confluent cells after 24-h incubation at 37°C. For that purpose, 5 × 10⁵ cells were seeded in a 6-well plate in serum-free medium. The procedure was performed as recommended by the manufacturer (R&D Systems). The spots developed in the membranes were quantified using Image J software. The values are representative of at least three independent experiments performed in triplicates.

MSC migration capacity was assessed using Transwell plates (24-well plate format; BD Biosciences) and 8-μM inserts. The inserts were coated with pure collagen (BD Biosciences) before seeding 1 × 10⁴ MSC in 100 μL per well. The lower chamber was filled with 600 μL chemotactic solution, as follows: 50 ng/ml IL-1β (PeproTech), 10 ng/ml IL-6 (PeproTech), 10 ng/ml IL-8 (PeproTech), 10 ng/ml CCL2 (PeproTech), 100 ng/ml MIF (PeproTech), and 10 ng/ml SDF-1 (PeproTech) diluted in serum-free αMEM (Invitrogen). The negative control corresponds to a lower chamber filled with serum-free αMEM. MDAMB231, A549, A431, H376, U87MG, and Jurkat CM were generated by incubating confluent cells (20 million per T175) with their respective serum-free media, at 37°C overnight. MSCs seeded in the upper chamber were incubated in presence of the different solutions overnight at 37°C. Cells that migrated to the bottom of the insert were fixed with 4% paraformaldehyde, stained for DAPI, and counted manually using a fluorescence microscope (Axioskop2; Zeiss). Where indicated, MSCs were preincubated with different inhibitors 1 h before the migration assay. All MAPK inhibitors (ZM336372, c-RAF inhibitor; UO126, MEK1/MEK2 inhibitor; SB203580, p38 inhibitor) were used at 10 μM (Tocris Bioscience). MIF antagonist (ISO-1) was at 2.5 μg/ml (Calbiochem), receptor antagonists (SB225002 for CXCR2; AMD3465 for CXCR4; RS504393 for CCR2; and Maraviroc for CCR5) at 10 μM (Tocris Bioscience), and pertussis toxin (PTX) at 1 ng/ml (Invitrogen). SDF-1 neutralizing Ab was used at 30 μg/ml (Abcam). CD74 blocking peptide was used at 30 ng/μL.

MSC invasion capacity was assessed as for migration assays, but inserts were coated with 50 μl matrigel (BD) before seeding 1 × 10⁴ MSC in 100 μL per well. At the end of the experiment (24 h), the whole chamber was fixed and stained for phalloidin conjugated to AF555 (Invitrogen) and DAPI. The inserts were mounted on slides and analyzed by confocal microscopy LSM700 (Zeiss). Invasion was quantified by subtracting the distance between the cell front and the insert at 24 h to the distance at time zero (after initial seeding onto the matrigel). The images were processed using Zen2009 software. The values are representative of at least three independent experiments performed in triplicate and of three different fields for each triplicate.

MSCs were treated with 100 ng/ml MIF or MDAMB231 conditioned medium (CM) for 5 min in presence or absence of receptor antagonists used at 10 μM. Cells were then washed, lysed, and processed, following the manufacturer’s recommendations (R&D Systems). The spots developed in the membranes were quantified using Image J software. The values are representative of at least three independent experiments performed in triplicates.

A total of 5 × 10⁴ MSCs was lysed on ice with radioimmunoprecipitation assay lysate buffer supplemented with protease inhibitor mixture (Roche). Total protein concentration was determined using the BCA protein assay kit (Pierce). Abs used were as follows: goat anti-MIF (1:250; Santa Cruz), rabbit anti-CXCR4 (1:500; Boster Biological Technology), rabbit anti-CXCR2 (1:500; Ab Frontier), mouse anti-CD74 (1:100; Santa Cruz), rabbit phospho p44/42 (ERK1/2; Thr²⁰²/Tyr²⁰⁴) (1:1000; Cell Signaling), rabbit phospho stress-activated protein kinase/JNK (Thr¹⁸³/Tyr¹⁸⁵) (1:1000; Cell Signaling), goat anti-p44/42 (ERK1/2) (1:5000; Santa Cruz), and rabbit stress-activated protein kinase/JNK (1:1000; Cell Signaling). Secondary Abs used were as follows: anti-goat HRP, anti-rabbit HRP, or anti-mouse HRP (1:7500; Sigma-Aldrich). Membranes were incubated with ECL substrate (Millipore), exposed, and developed, following manufacturer’s recommendations.

A total of 1 × 10⁵ MSCs was lysed in nondenaturing lysis buffer (20 mM Tris HCl [pH 8], 137 mM NaCl, 10% glycerol, 1% Triton X-100, and 2 mM EDTA) supplemented with protease inhibitor mixture (Roche). Samples and protein G Sepharose beads (Sigma-Aldrich) were blocked 1 h on ice with PBS 1% BSA. Samples were incubated with anti-MIF (1:100; Santa Cruz) overnight at 4°C. Protein G Sepharose beads were added to the sample-Ab mix and incubated for 6 h at 4°C. After extensive washes, the pellet was resuspended in SDS loading buffer and boiled, and the supernatant was recovered. These were then processed for Western blot analysis and analyzed for the presence of MIF, CXCR2, CXCR4, and CD74.

Transposon Luciferase-strawberry construct was a gift of S. Lyons (Cambridge, U.K.). Lentiviral small hairpin RNAs (shRNAs) targeting CXCR4 and MIF (GFP tagged) were purchased from the University College London library (Thermoscientific).

To establish MDAMB231 stably expressing Luciferase and strawberry red, transposase DNA and transposon DNA carrying Luciferase and strawberry red were transfected at a ratio of 1:2, using lipofectamine (Invitrogen) as a transfection reagent. A pure cell population was obtained by selection with 1 mg/ml G418 (Sigma-Aldrich). For shRNA, cells were transduced using multiplicity of infection 10. A pure cell population was obtained by selection with 5 μg/ml puromycin (Invitrogen).

Spheroid formation was achieved by using MDAMB231 cells transfected with Luc-strawberry transposon or transduced with MIF shRNA GFP, cultured on matrigel-coated low adherence 96-well plates. Cells were grown until spheroid formation (4–6 d) at 37°C, 5% C0₂. MSC transfected with Luc-strawberry transposon or transduced with CXCR4 shRNA-GFP or nonsilencing shRNA GFP were added to the spheroids under the indicated conditions and tracked during 18-h time lapse using confocal microscopy LSM 510 Meta (Zeiss) possessing a chamber to keep the samples at 37°C, 5% C0₂. The still images representative of the time lapse were obtained after z-stack and maximum projection at the end of the time course using Zen2009 software.

For the analysis of CXCR4 cell surface expression, MSCs were seeded at 5 × 10⁴ cells/well of 6-well plates. The cells were treated or not for 24 h with MDAMB231 CM. The cells were scrapped from the wells using rubber cell scrappers. For blocking, the cells were incubated in 1% FBS in PBS. For CXCR4 staining, PE/Cy7 anti-human CXCR4 (BioLegend) was used at 1:20 dilution in PBS 1% BSA. Cells were analyzed by FACSVerse (BD Biosciences). The controls, analysis gates, were optimized for each cell line used according to standard practice.

A minimum of 10,000 events was recorded for each sample, and analysis was performed using FlowJo software.

All animal studies were performed in accordance with British Home Office procedural and ethical guidelines. Six-week-old NOD/SCID γ mice (Harlan, Bicester, U.K.) were kept in filter cages.

Lung metastasis was obtained by the i.v. injection of 1 × 10⁵ MDAMB231 cells in 200 μl serum-free DMEM via the lateral tail vein. Tumors were left to grow for 15 d. MSCs were then injected i.v. via the lateral tail vein (5 × 10⁵ cells in 200 μl serum-free αMEM). After 24 h, lungs were harvested for histology and qPCR.

For lung tissue sections, slides were processed for dewax, washed with PBS, blocked for 1 h with PBS BSA 1%, and incubated with the following primary Abs: GFP Ab (1:100; Invitrogen) and Firefly Luciferase Ab (1:2000; Abcam). Secondary Abs (Alexa Fluor 488 or 555; Invitrogen) were used at 1:300 dilutions. DAPI counterstaining was performed. Slides were coverslipped and analyzed by confocal microscopy LSM700 (Zeiss) using Zen2009 software.

For quantification, 15 nonconsecutive sections were analyzed (5 fields per section) by ImageJ, allowing the analysis of multiple lung levels and tumor deposits. The ratio between GFP and Luciferase fluorescence intensities was calculated to obtain the measure of MSC recruitment, normalized to the signal from tumor cells.

For in vitro studies, RNA was extracted from MSCs (500,000 cells) with TRIzol reagent (Invitrogen), followed by cleanup and DNase I treatment with RNeasy mini kit (Qiagen). For in vivo studies, RNA from mice lungs was extracted from formalin-fixed, paraffin-embedded sections using the High Pure FFPE RNA Kit (Roche Applied Science). Specific primers for CDC42EP5, RHOQ (in vitro studies), Luciferase, and GFP genes (in vivo experiment) were designed using the Primer Express Software (PE Applied Biosystems). Real-time qPCR was carried out using the SYBR-Green master mix (Applied BioSystems) in an Eppendorf real-time PCR machine. Relative gene expression was quantified using the threshold cycle method and normalized to the amount of 18S rRNA and B2M reference genes (PrimerDesign).

Statistical significance between two groups of data was evaluated by Student t test using GraphPad software (*p < 0.05, **p < 0.01, ***p < 0.005).

We used two methods to confirm MSC migration toward CM taken from A549 lung cancer and MDAMB231 breast cancer cell cultures (A549 CM and MDAMB231 CM). Both agarose drop assays (counting MSCs migrating out of the drop: Fig. 1A–C, quantified in Fig. 1D) and Transwell migration assays (Fig. 1E–G, quantified in Fig. 1H) clearly demonstrate a significant increase in MSC migration toward CM from both cancer cells compared with the control unconditioned medium. MDAMB231 CM was the more efficient in attracting MSCs.

Factors secreted from tumor cells were determined using a cytokine array. A549 and MDAMB231 cells were examined. The supernatant of both cell lines was processed, and representative membranes from the experiment are shown in Fig. 2A with quantification across replicates in Fig. 2B.

The highest secretion levels observed were for MIF, SerpinE1, IL-6, and IL-8, for both cell lines. Of note, a significantly higher level of MIF and a slightly increased level of IL-6, IL-8, and serpineE1 were secreted by MDAMB231 compared with A549 cells, suggesting a possible correlation between the secretion levels of these molecules and the migration capacity of MSCs.

To screen changes in MSC chemokine receptor or chemokine gene expression after exposure to tumor cells, we examined mRNAs of MSCs in the presence of CM from MDAMB231 and A549 cells compared with unconditioned medium by microarray analysis (Supplemental Fig. 1). In the presence of MDAMB231 CM and A549 CM, 1800 and 1774 genes were found upregulated in MSCs, respectively. Various cytokines and adhesion molecules were upregulated, but inconsistently between the two CM. We found the highest upregulation in both groups (MDAMB231 CM and A549 CM treated) for IL-1β (5.05- and 13.48-fold change), IL-6 (3.18- and 1.58-fold change), IL-8 (12.23- and 5.61-fold change), and CCL2 (4.81- and 1.89-fold change). We validated the microarray results of MSC gene expression after exposure to CM for six upregulated genes (IL-1β, IL-6, IL-8, CCL2, RhoQ, and CDC42) using real-time PCR. The qPCR expression data showed a high correlation with the microarray expression data for both MDAMB231 CM and A549 CM groups (Pearson correlation, r = 0.97 and r = 0.96, respectively).

These results suggest soluble factors secreted from tumor cells upregulate mainly chemokine expression on MSCs (IL-1β, IL-6, IL-8, CCL2), which may stimulate MSC migration through a positive feedback loop.

We next examined the role of the candidate molecules identified from the cytokine array and microarray in MSC migration and invasion. Migration assays determined that several chemokines increased MSC migration, with MIF having the largest effect (Fig. 2C, 2D). MDAMB231 CM was used as a positive control.

Dose-response curves confirmed dose-dependent effects on cell migration for IL-1β, IL-6, IL-8, CCL2, and MIF, and the optimal chemokine concentrations were used for subsequent experiments (Supplemental Fig. 2). SerpinE1 showed no effect on MSC migration. SDF-1 was also tested, despite not being identified in our cytokine array analysis, and we confirmed MSC dose-dependent stimulation of migration (Fig. 2D). Chemokinesis and proliferation assays were performed and confirmed that MSC migration was not due to random migration or increased proliferation (Supplemental Fig. 3A, 3B).

The cytokines were further tested in the MSC invasion assay (Fig. 2E, 2F). All tested cytokines stimulated invasion through the matrigel layer toward the membrane, except serpinE1. Therefore, from our migration and invasion assays, MIF was the most potent chemokine in vitro for MSCs and a likely key molecule in MSC homing to tumor cells in vivo.

MIF can bind the chemoattractant receptors CXCR2, CXCR4, and CD74. Other known chemoattractant receptors include CCR2 (a receptor for the ligand CCL2) and CCR5 (the receptor for RANTES), which are not receptors for MIF, serving therefore as useful negative controls. Using receptor antagonists or blocking peptide in migration and invasion assays, we demonstrated that a significant reduction in migration toward MDAMB231 CM was only achieved in presence of CD74 blocking peptide, CXCR2 and CXCR4 antagonists, being more dramatic for the latter (Fig. 3A). Additionally, as CXCR2 and CXCR4 are Gαi protein-coupled receptors, PTX was used to confirm the activity of Gαi signaling pathway. This produced the greatest reduction in migration, suggesting more than one Gαi family receptor responsible for homing consistent with the reduction observed individually with the CXCR2 and CXCR4 antagonists. Receptor blocking in the invasion assay led to similar results, with a decrease in invasion only in the presence of CD74 blocking peptide, CXCR2 and CXCR4 antagonists, or PTX (Fig. 3B), however not as marked as in the migration assays, suggesting a role of these receptors at early steps of the homing process.

We further confirmed the major role of MIF with the use of a well-described antagonist (ISO-1) achieving a significant decrease in both MSC migration and invasion toward MDAMB231 CM. A neutralizing Ab against SDF-1 was also tested and showed no role for SDF-1 in these assays (Fig. 3A, 3B). We have therefore shown a key role for MIF, involvement of CD74 as well as the putative receptors CXCR2 and CXCR4, showing a more potent role for CXCR4.

To delineate a physical interaction between MIF and CXCR2, CXCR4, and CD74 receptors on MSCs, immunoprecipitation (IP) experiments were performed. An anti-MIF Ab was used to pull down potential complexes with CXCR2, CXCR4, or CD74. Different time points were tested to examine the kinetics of interaction after MSC stimulation with MDAMB231 CM. MSCs transduced with CXCR4shRNA were used in parallel under the same conditions for the IP. The results demonstrate an interaction between MIF and all three receptors: CXCR2, CXCR4, and CD74 (Fig. 3C). Interestingly, we found the MIF–CXCR2 interaction transient (maximal at 5 min), whereas the MIF–CXCR4 interaction was stronger and sustained to 24 h. MIF–CD74 interaction was only seen in the presence of CXCR4; no CD74 was detected on IP performed with CXCR4 knockdown cells. These results are consistent with the more dominant blocking effect of the CXCR4 antagonist compared with the CXCR2 antagonist on MSC migration toward tumor cell CM (Fig. 3A). Signaling via CXCR4 is likely to be amplified by CD74, which interacts with MIF and CXCR4 as part of a complex explaining the significant effect on migration/invasion with the CD74 blocking peptide in Fig. 3A, 3B.

To confirm CXCR4 knockdown with CXCR4 shRNA, we analyzed CXCR4 cell surface expression in untransduced MSCs, MSCs transduced with either nonsilencing shRNA control, or CXCR4 shRNA by flow cytometry. We confirm the presence of CXCR4 at the cell surface, and interestingly find that pretreating MSCs with MDAMB231CM upregulates CXCR4 surface expression. This is abolished in MSC CXCR4shRNA, validating the knockdown (Fig. 3D and quantification in Fig. 3E).

To confirm the importance of the MIF–CXCR4 axis in MSC migration to tumors, we examined four additional cell lines: A431, H376, U87MG, and Jurkat (Fig. 4). The U87MG cell line served as a control for SDF-1 secretion (51, 52). In all cell lines, we observed a decrease in MSC migration in the presence of either the CXCR4 or MIF antagonists (Fig. 4A–D). Consistent with the previous data shown with MDAMB231, we were not able to observe a decrease in migration in the presence of a SDF-1 neutralizing Ab, except for the U87MG CM, but even here the effect was less than with either the CXCR4 or MIF antagonists, suggesting the key role of MIF–CXCR4 even in tumors able to secrete SDF-1. All cell lines secrete high levels of MIF, as depicted in the cytokine arrays (Fig. 4E).

In other cellular contexts, MIF is capable of activating several signaling pathways, including the MAPK pathway. To delineate whether the MAPK pathway plays a role in MSC migration and invasion toward tumor CM, a variety of key signaling molecule inhibitors was used. Migration and invasion were significantly reduced only in the presence of c-RAF and MEK1/MEK2 inhibitors (ZM336372 and UO126, respectively), intermediates upstream of ERK (Fig. 5A, 5B).

MAPK phosphoarrays of protein lysates from MSCs cultured in presence of αMEM, MDAMB231 CM, and MIF demonstrated that MIF alone significantly induces the phosphorylation of ERK1, with a further small, but significant increase in JNK2 and JNK pan phosphorylation (Supplemental Fig. 4A). These results were validated by immunoblotting, confirming ERK1/2 phosphorylation in the presence of MDAMB231 CM (Fig. 5C), ERK1/2, and JNK phosphorylation in the presence of MIF alone (Fig. 5D). Pretreatment with CXCR2 and CXCR4 antagonists both decreased the phosphorylation levels of ERK1/2 in presence of MDAMB231 CM (Fig. 5C). Both receptor antagonists decreased the phosphorylation levels of ERK1/2 and JNK in the presence of MIF (Fig. 5D). These data confirm that MIF signals through both receptors to trigger the MAPK pathway. Further validation was obtained with the observation of reduced migration/invasion in the presence of MIF with the use of CXCR2 and CXCR4 antagonists and ERK/JNK-specific inhibitors (Fig. 5E, 5F). Hence, MIF signals through both CXCR2 and CXCR4 to induce MSC migration/invasion. Again, the CXCR4 antagonist had a more potent effect in impairing these processes compared with CXCR2 antagonist.

On establishing the involvement of the MAPK classical pathway in MSC migration, we examined the microarray data from CM-treated MSCs for genes associated with cell motility. We found upregulation of CDC42 (downstream of Gα protein-coupled receptors) and RhoQ (downstream of CDC42 leading to cytoskeleton reorganization/motility). Real-time PCR analysis was performed for these genes in different conditions: MSCs stimulated by A549 or MDAMB231 CM (as for the microarray analysis) and MIF only. The results shown in Fig. 5G and 5H confirm their upregulation, and, most importantly, that this can be achieved by MIF alone.

To further validate the MIF–CXCR4 axis as responsible for MSC migration, in addition to the CXCR4 knockdown established previously in MSCs (Fig. 3C), a MIF knockdown was established similarly in MDAMB231 cells. Both were validated by Western blotting (Supplemental Fig. 4B, 4C). Knockdown of CXCR4 in MSCs strongly decreased their migration and invasion toward MDAMB231 CM (Fig. 6A, 6B). Likewise, using the CM from MDAMB231 cells after knocking down MIF decreased MSC migration and invasion (Fig. 6A, 6B).

In an attempt to mimic more physiological conditions, a three-dimensional model was optimized. MDAMB231 cells were grown as spheroids in matrigel (either constitutively expressing Luciferase-strawberry as controls or MIFshRNA-GFP). Time-lapse microscopy was performed after the addition of MSCs to the surface (either constitutively expressing Luciferase-strawberry or nonsilencing shRNA-GFP as controls or CXCR4shRNA-GFP), and migration was assessed. Representative still pictures of the three conditions tested (control, MSC CXCR4shRNA, and MDAMB231 MIFshRNA) are shown in Fig. 6C (time-lapse videos in Supplemental Videos 1–3). MSC migration was notably decreased in the presence of CXCR4 or MIF shRNAs, validating the results obtained in the two-dimensional system.

Lung metastases were established after i.v. injection of 1 × 10⁵ MDAMB231 cells labeled with Luciferase-strawberry. After 15 d, 5 × 10⁵ i.v. MSCs constitutively expressing nonsilencing shRNA GFP or CXCR4 shRNA GFP were injected. At day 16, lungs were harvested for histology and qPCR to compare MSC homing to the tumors. Immunofluorescence was performed on lung sections for GFP (anti-GFP AF488) and Luciferase (anti-Luc AF555) (Fig. 7A). MSC CXCR4 shRNA cells showed reduced tumor tropism compared with control animals injected with MSC nonsilencing shRNA, which was quantified by qPCR (Fig. 7B, top panel) (1.7-fold decrease) and immunofluorescence (Fig. 7B, bottom panel) (2.3-fold decrease).

Similar experiments were performed to examine the paracrine effect of MIF secretion from tumors. This time, MDAMB231 constitutively expressing MIFshRNA GFP or nonsilencing shRNA GFP cells were injected to establish lung metastasis. MSCs constitutively expressing luciferase strawberry were subsequently injected. Results confirmed a reduction in MSC homing to established tumors using MDAMB231 MIFshRNA GFP by qPCR (4.7-fold decrease) (Fig. 7C, 7D, top panel) and immunofluorescence (7-fold decrease) (Fig. 7D, bottom panel).

We have shown an essential role for MIF–CXCR4 in MSC homing to tumors both in vitro and in vivo. Our in vivo knockdown studies of either CXCR4 or MIF suggest this partnership could be manipulated to increase homing of MSCs to tumors in patients, enabling augmented delivery of anticancer therapies.

We have shown in MSCs a physical interaction between MIF and three receptors: CXCR2, CXCR4, and CD74. Interestingly, although we demonstrate that MIF–CXCR4 is the main axis driving tumor homing of MSCs, blocking CXCR2 also led to a decrease in in vitro migration and invasion. Similarly, we see a statistically significant decrease as well using a CD74 blocking peptide. From the IP data, we show interestingly that CD74 would interact with MIF via CXCR4, which would most likely amplify the signal via this receptor. Our findings are in line with the recent description of MIF as a noncognate ligand for both CXCR2 and CXCR4 involved in leukocyte recruitment (26) and potential formation of heterodimeric complexes with CD74.

Previous studies have noted MIF to be an inhibitor of MSC migration. These studies, however, have examined either chemokinesis rather than tumor tropism (40, 41), different receptors (41), or migration toward benign epithelial cells (42).

We identified MIF through screening for soluble factors secreted from tumor cell lines. Other studies have shown MSC migration to be regulated by numerous factors, including tumor cell-specific receptors, extracellular matrix, and soluble tumor-derived factors such as SDF-1, TNF-α, and ILs (4, 5). MSCs are undoubtedly responsive to other chemokines. In our study, we see stimulation of MSC migration toward gradients of IL-1β, IL-6, IL-8, and CCL2. A role for IL-6, IL-8, and CCL2 has been delineated in other studies (43–46).

Our results show an upregulation of IL-1β, IL-6, IL-8, and CCL2 in MSCs treated with tumor CM. Interestingly, we also show a chemotactic role for these four cytokines. These findings suggest a positive feedback loop and amplification of the initial chemoattraction from the tumor cells, likely to be triggered by MIF. Indeed, other studies show a MIF-dependent upregulation of those cytokines, in an inflammatory context (47–50).

We did not see an upregulation of receptor gene expression in MSCs treated with tumor CM; however, we do show an upregulation of CXCR4 cell surface expression. This finding suggests that cytokines present in tumor CM trigger the recruitment of intracellular stored CXCR4 to the cell surface, improving the response to a chemotactic gradient, independently of gene expression.

The most extensively studied axis in MSC homing to tumors is CXCR4–SDF-1. An in vitro study demonstrating MSC migration toward recombinant SDF-1 and tumor cell CM suggests that tumor cells or tumor cell CM stimulate MSCs to produce SDF-1, which subsequently binds its cognate receptor CXCR4 expressed on MSCs, in an autocrine manner. However, it is not clear which molecule in the tumor CM is stimulating MSCs to produce SDF-1 (4).

We confirmed recombinant SDF-1 is a chemoattractant for MSCs. However, in an in vitro tumor context, we failed to detect significant levels of this molecule secreted either by A549, MDAMB231, H376, A431, or Jurkat. We included the U87MG cell line as it is described to secrete SDF-1 (51, 52). Of all the cell lines tested, only MSC migration toward U87MG was decreased in the presence of SDF-1 neutralizing Ab; however, even in this cell line, migration was more severely diminished by CXCR4 or MIF antagonists. These results reinforce our hypothesis that MIF is the dominant chemoattractant in the recruitment of MSCs to tumors, even in the presence of SDF-1. These findings are consistent with other reports showing high secretion levels of MIF and rare secretion of SDF-1 in the majority of tumors (11–18). Finally, we define the ERK and JNK branches of the MAPK pathway as essential for MSC migration/invasion to occur, and discard a role for the p38 branch.

To summarize, we have defined the novel finding that MIF is the key chemoattractant during MSC recruitment to tumors, in vitro and in vivo. Our in vivo studies confirm MIF as the key player in MSC homing to tumors, with a decrease in MSC recruitment after MIF knockdown. MIF–CXCR4 has been described as important in a variety of contexts: regulation of endothelial progenitor cell migration, cancer metastasis, and cancer proliferation/growth (53); however, it has not been associated with MSC homing to tumors. Our study shows a novel role for MIF in MSC recruitment to tumors. We believe our findings will lead to improved strategies of inducing MSC homing to tumors for novel cellular cancer therapies.

We thank Scott Lyons (Cancer Research UK, Cambridge, U.K.) and the Sanger Institute (Cambridge, U.K.) for providing transposon constructs. We thank Prof. Tony Segal (Centre for Molecular Medicine, University College London) for providing the Jurkat cell line and Paul Frankel (Centre for Cardiovascular Biology and Medicine, University College London) for providing the U87MG cell line. We thank Andrew Williams and Chris Scotton (Centre for Inflammation and Tissue Repair, University College London) for helpful comments and critical evaluation of the manuscript.

The authors have no financial conflicts of interest.