The Protease-Dependent Mesenchymal Migration of Tumor-Associated Macrophages as a Target in Cancer Immunotherapy

Migration is a crucial function of macrophages to maintain tissue homeostasis, defense against infections, and tissue repair. It is also associated with progression of several pathologies, including inflammation and cancer. Macrophages, mainly originating from blood monocytes (1), are recruited in the tumor stroma at all stages of cancer progression (2, 3). Tumor-associated macrophages (TAMs) can represent more than 50% of the tumor cell mass, and their percentage positively correlates with poor prognosis in most cancers (4, 5). TAMs release inflammatory cytokines, which generate a chronic inflammatory environment favoring tumor growth (2, 4–6), and promote angiogenesis, lymphangiogenesis, immunosuppression, tumor cell invasiveness, resistance to therapy, and metastasis (2, 4, 5). Both clinical studies and experiments in mouse models show that TAMs are polarized by the local tumor environment, adopting a protumoral phenotype (4, 7). Through remodeling of the extracellular matrix (ECM), TAMs support tumor cell invasiveness (8) and contribute to metastasis by escorting tumor cells and priming the premetastatic site (9–11). For these reasons, TAMs constitute a new wave of immune-oncology targets (12–14).

Several strategies have been designed to limit TAM recruitment in tumors, including inhibition of growth factors and chemokines (6, 12–18). Although promising, these strategies do not specifically target TAMs but also other macrophage populations, regardless of their beneficial or detrimental role. Specifically, limiting macrophage migration in tumors would be an optimal strategy. However, the poor understanding of the mechanisms of macrophage migration in vivo hinders progress in the field.

Two migration modes have been described in three-dimensional (3D) environments: amoeboid and mesenchymal. The general definition of amoeboid movement is rounded, ellipsoid, or moderately elongated cells that form blebs or generate small actin-rich filopodia (19–21). These cells do not adhere to the ECM but, rather, use a propulsive, pushing migration mode (20, 22, 23). This is a rapid and nondirectional motility involving acto-myosin contraction dependent on the Rho–ROCK pathway. Mesenchymal migration consists of cells presenting a very elongated morphology with long membrane protrusions rich in F-actin (19, 21). The movement is slow and directional, involves cell adhesion on the substratum, and requires proteases to degrade the ECM creating paths through dense environments. It is not inhibited but, rather, stimulated by treatment with ROCK inhibitors (24).

The in vivo migration of lymphocytes, neutrophils, or dendritic cells (DC) involves the ROCK-dependent, protease-independent amoeboid mode (25–28), whereas the mechanisms used by macrophages to migrate in vivo is still unknown. It has been shown in vitro that, in contrast to other leukocytes (25) but similarly to tumor cells (19, 21), macrophages can adopt the two mechanistically distinct migration modes (8, 24, 25, 29–33), depending on the ECM architecture (32). Therefore, we investigated the migration mechanisms used by macrophages that are recruited to inflamed tissues or tumors.

By using intravital two-photon microscopy and developing an ex vivo model of macrophage tissue infiltration, we demonstrated both in mouse tumors and breast cancer explants that TAMs use the protease-dependent mesenchymal migration mode, whereas macrophages preferentially use the amoeboid migration to move in peritumorous tissue or inflamed ear derma. Finally, we showed that the inhibition of macrophage mesenchymal migration correlates with a decrease in macrophage infiltration into tumors and an inhibition of tumor growth.

MacBlue (Csf1r-Gal4VP16/UAS-ECFP) mice were obtained from David A. Hume (Roslin Institute, Midlothian, Scotland; ref. 34). Eight- to 12-week-old mice were used in all experiments. Macrophages were imaged by intravital microscopy either in inflamed ear derma or inside/around the tumor on mice equipped with dorsal window chambers, as described in the sections below. When indicated, BB-94 (a broad-spectrum matrix metalloprotease inhibitor; Sigma-Aldrich) or Y-27632 (a ROCK inhibitor; Euromedex) was injected intraperitoneally. A single injection was performed 1 day after ear irritation or 6 days (BB-94: 30 mg/kg and Y-27632: 60 mg/kg) after dorsal window chamber surgery and tumor induction. A second injection of BB-94 (BB-94: 15 mg/kg) was done at day 7 in the tumor model.

For intravital experiments performed with dextran (MW = 70, 2,000 kDa), 100 μL of rhodamine-conjugated dextran solution at 10 mg/mL (Thermo Fisher Scientific) was injected retro-orbitally 24 hours prior to imaging. The half time of clearance from blood is 4 to 5 hours (35). For experiments on BB-94 chronic treatment, C57BL/6 mice (Janvier LABS) were used and treated as follows. First, fibrosarcomas were generated by subcutaneously injecting 10⁶ LPB-GFP cells into the mouse flank. The LPB cell line is a highly tumorigenic murine clonal derivative of TBL.CI2, a methylcholanthrene-induced C57B1/6 mouse sarcoma cell line (36). The LPB-GFP cell line was obtained by transduction of LPB cells with an MFG-eGFP retroviral vector encoding the eGFP under the control of long terminal repeat as described (37). FACS analysis confirmed that all LPB tumor cells expressed GFP (see the Results section). Second, after 7 days of tumor growth, littermate mice were treated daily with intraperitoneal injection of BB-94 (30 mg/kg) or DMSO (10 μL/mouse) for 12 days. Tumor progression was monitored by fluorescence microscopy (Leica Macrofluo) of GFP tumor cells in animals anesthetized by isoflurane inhalation (Belamont-Neuilly-sur-Seine) throughout the imaging session. Tumor area was automatically quantified using ImageJ software. At day 12 of treatment, mice were sacrificed, tumors were resected, fixed, and analyzed by IHC to label TAMs as described in the sections below. All experiments were performed according to animal protocols approved by the Animal Care and Use Committee of the Institute of Pharmacology and Structural Biology (No. 20140315/195).

For macrophage imaging of inflamed ear derma, mouse ears were depilated and imaged before and after irritation. To induce recruitment of macrophages from blood monocytes into the tissue, MacBlue mouse derma was irritated with dinitrofluorobenzene (DNFB) in order to trigger local inflammation. Irritation was performed in two steps: a sensitization step using 10 μL of 0.15% DNFB (Sigma Aldrich) in a 4:1 acetone–olive oil mix, followed by a challenge step 7 days later using the same solution. For imaging purposes, the front of the ear was made to adhere to a plastic base plate and vacuum grease was used to create a well around the back of the ear inside which PBS was added. A glass coverslip was then placed over the ear for imaging.

Surgery was carried out under general anesthesia using intraperitoneal injection of ketamine (100 mg/kg; Virbac) and xylazine (16 mg/kg; Bayer). The back of the mouse was shaved and depilated with depilatory cream (Veet), and the eyes were kept hydrated throughout the surgery with Ocry-gel (TVM). The chamber consisted of two symmetrical titanium frames (APJ Trading Co.). These frames sandwiched an extended double layer of dorsal skin with the use of stainless-steel screws and sutures. After the implantation of the frames, one layer of the skin (12-mm diameter) was surgically removed, and part of the fat and connective tissues from the lower layer were dissected away under a stereomicroscope (Leica Microsystems) to ensure optimal observation. After addition of sterile PBS, a sterile glass coverslip (Glaswarenfabrik Karl Hecht) was attached to the open frame with a stainless-steel snap ring to cover the surgical site and provide microscopic access to the skin tissue. Right after the surgery and on the following day, Profenid (10 mg/kg, 50 μL in each thigh; Sanofi-Aventis) was injected intramuscularly to provide analgesia and to prevent inflammation.

Before attachment of the coverslip onto the window frame, 20 μL of DMEM containing 2 × 10⁶ LPB-GFP cells (Life Technologies) were injected subcutaneously in the remaining connective tissue of the lower skin layer. Tumor growth was monitored daily by fluorescence microscopy of GFP tumor cells and by assessment of growth using a macroscope (Leica Macrofluo macroscope). Six days were needed in order to obtain a 3-mm diameter tumor, when CFP-expressing TAMs and GFP-expressing LPB cells were imaged.

Intravital microscopy was carried out on a customized stage for holding mice using a 7MP upright multiphoton microscope (Zeiss) equipped with a 20×/1.0 objective (df = 1.8 mm) and a Chameleon-Ultra II laser (Coherent). Animals were anesthetized by isoflurane inhalation throughout the imaging session. Animal temperature was maintained at 37°C with an Air-Therm-heated environmental chamber and a heating blanket placed under the mouse.

ECFP-fluorescent macrophages from MacBlue mice and collagen by second-harmonic generation were observed using a 435–485 nm BP filter, and LPB-GFP tumor cells were observed using a 500–550 nm filter. The tumor mass, the peritumorous area of the window, or the inflamed ear derma was positioned under the objective and imaged with the laser set at a 940 nm excitation wavelength. A 3D-image stack of sections with z-spacing of 2 μm was acquired every 3 minutes for 1 to 2 hours to assess 3D cell motility in dynamic z-stack time series.

Imaris (BitPlane) software was used to process acquired images, to track cells, and to measure cell roundness. Cell tracking was performed manually using the spots function of Imaris. For each image, 100 randomly chosen cells were tracked for at least 30 minutes. The roundness of tracked cells was obtained by calculating the average value of the width over length ratio at three time points (start, middle, and end). Width was taken at the cell's mid-length. Macrophages adopted a large variety of shapes from round (roundness of 1) to very elongated (roundness of 0.1).

Human monocytes were isolated from the blood of healthy donors (buffy coat obtained from Etablissement Français du Sang) and differentiated into human monocyte–derived macrophages (hMDMs) as previously described (32). Blood samples were obtained following standard ethical procedures and with the approval of the concerned internal review boards.

Bone marrow cells were isolated from femurs and tibias, cultured, and differentiated as described (29).

Breast tumors were obtained from 23 patients at the University Cancer Institute–Toulouse Oncopole. In this study, tumors larger than 1 cm diameter from patients with breast tumors were included with no exclusion criteria. Samples were collected after surgical tumor resection, placed in a sterile compress soaked with cold RPMI medium for 30 minutes and immediately processed. The clinicopathologic data of the patients are summarized in Supplementary Table S1. Pathologic diagnosis, as well as mitosis status, were confirmed by a referent pathologist in breast cancer. Sample fibrosis was automatically quantified using an image analysis software (Tissue-Studio 4.3, Definiens). This study was approved by the ethic committee (Breast Group of IUCT, transfer license No.: AC-2016-2658; collection statement No.: DC-2016-2656). Patients gave informed written consent for the use of biological samples for research purposes.

Human and mouse tissue slices were prepared as follows. For mouse samples, tumor induction was performed in C57BL/6 mice identically to induction in dorsal window chamber-equipped mice. However, tumor growth was monitored for 2 to 3 weeks in order to obtain an approximately 1-cm³ tissue before resection. Tissue samples were embedded in 3% low gelling temperature agarose (type VII-A, Sigma-Aldrich) prepared in PBS to allow for easier slicing. Slices measuring 500 μm were obtained with a microtome dedicated to live tissues (38), with the Krumdieck tissue slicer (TSE Systems) filled with ice-cold PBS (Life Technologies) set to medium blade and arm speeds. Slices were cultured on a 30-mm cell culture insert featuring a hydrophilic PTFE membrane (0.4-μm pore size, Merck Millipore) placed inside 6-well plates containing 1.1 mL of RPMI with or without 10 μmol/L BB-94 or 20 μmol/L Y-27632. A 5-mm diameter stainless-steel washer was then placed on top of each tissue slice to create a well for macrophages seeding. The same day, cocultures were performed by seeding 3 × 10⁵ hMDMs pretreated for 16 hours with DMSO, BB-94, or Y-27632 on top of tumor slices and incubated in a 37°C, 5% CO₂ environment for 72 hours. After 16 hours of coculture, the washer was removed.

For experiments performed with the irreversible inhibitor of MMP2, -9, and -14 (39), hMDMs were pretreated with 10 μmol/L of MMP2/MMP9 inhibitor V, also known as SB-3CT (4-(4-(Methanesulfonamido)phenoxy)phenylsulfonyl) methylthiirane, Calbiochem), for 16 hours, washed, and then seeded on top of tissue explants for 48 hours. The potential cytotoxic effect of drugs was evaluated by trypan blue exclusion on hMDMs plated on coverslips. None of these drugs induced cell death over the 2 to 3 days of incubation.

For human samples, hMDMs were labeled with cell tracker (C7025-Green-CMFDA, Invitrogen) according to the manufacturer's instructions prior to seeding on tissue slices (3 × 10⁵ hMDM per slice). Culture medium (RPMI 1640, Gibco) complemented with indicated inhibitor was replaced daily for 3 days before overnight fixation with formalin at 4°C (Sigma-Aldrich). The optimal conditions for ex vivo culture of tumor slices were determined according to tissue viability (as evaluated by the absence of necrosis revealed by tissue staining with hematoxylin–eosin).

Tissue slices used in cocultures with hMDMs were embedded in paraffin. For quantification assays of macrophage infiltration, slices were then cut along the diameter and serial sectioning was performed along the cut. Sections were stained using either a mouse antihuman CD68 (clone KP1, DAKO, 1/100) for mouse tissues or a rabbit anti-fluorescein (Abcam ab6556) for human samples and a peroxidase-coupled secondary antibody. The specificity of the antibody was verified in vitro on nonlabeled or cell-tracker labeled macrophages seeded on glass coverslips. Sections were stained with hematoxylin to visualize cell nuclei. Slides were then scanned using a Pannoramic 250 Flash II slide scanner with Pannoramic viewer software (3DHISTECH) with a 20× magnification lens.

Macrophage infiltration was quantified using a macro developed in ImageJ as described (33). Briefly, colors were deconvolved with Gabriel Landini's plugin to separate CD68 or Cell tracker staining from hematoxylin coloration. A threshold on specific staining was chosen to dissociate specific signal from the background and applied to every picture acquired from a tumor slide. For Figs. 5–7 and Supplementary Fig. S5, cells were automatically segmented by thresholding hematoxylin-positive nuclei using the autolocal threshold ImageJ function. Cells were identified as macrophages when at least 35% of the segmented area was CD68- or cell tracker-positive. The distance to the upper part of the section (defined manually) was obtained using the Distance Map ImageJ function.

For ECM characterization of LPB tumors, tumors grown for 1 week were used identically to in vivo experiments. After fixation and paraffin embedding, serial tumor slices were cut and stained with Sirius red to label the whole extracellular matrix or with anti-collagen I (ab34710 abcam, 1/500), anti-collagen IV (ab6586 abcam, 1/500), anti-fibronectin (ab5413 Abcam, 1/250), and anti-laminin (L9393 Sigma, 1/50). Anti-F4/80 (MCA497 Serotec, 1/100) was used to visualize endogenous TAMs. Sections were counterstained with hematoxylin. For IHC characterization of the polarization of human and mouse macrophages in tissues, rabbit polyclonal anti-iNOS (clone 54; BD Transduction Laboratories), anti-Arg1 (clone H52; Santa Cruz), anti-Arg1 (clone N20; Santa Cruz), anti-CD163 (clone 10D6; Novocastra), and anti-CD206 (clone MRC1; Sigma) were used.

hMDMs labeled with violet cell tracker were coincubated in 6-well plates with mouse LPB-GFP tumor cells (one hMDM for 10 tumor cells) for FACS analysis. Cells were harvested after 72 hours of coincubation (time of incubation of human macrophages with mouse tumor explants), and the cell suspension was incubated with Fc block and antibodies: anti-CD206-APC, as a M2 marker, and anti-iNOS-PE575, as a M1 marker, before analysis with an LSR II Flow Cytometer (BD). A scatter plot of SSC versus FSC was first used to discriminate cells on size and dissociate intact cells from cellular debris and cell aggregates. The GFP versus AmCyan/violet plot were then gated to identify the two cell populations and to assay phagocytosis. Tumor cells were identified by the green fluorescent signal and macrophages by AmCyan fluorescent signal. The CD206 versus iNOS plot was gated on the AmCyan-positive macrophage population to analyze macrophage polarization status. For immunofluorescence experiments, hMDMs were coincubated with LPG-GFP tumor cells on glass coverslips in 24-well plates (ratio was 1 hMDM per two tumor cells). Cells were fixed after 72 hours and labeled with Texas-red phalloidin to distinguish macrophages (red) from tumor cells (red and green) and to measure macrophage phagocytic activity of tumor cells (red cells with green inclusion) with the confocal microscope.

Subcutaneous GFP-expressing fibrosarcoma tumors or inflamed dorsal derma were resected from MacBlue mice and from WT mice as a control, cut into small 1-mm³ pieces, and incubated in digestion medium containing Collagenase IV or Collagenase I (0.34 mg/mL; Life Technologies), respectively, in complete RPMI (Life Technologies) for 2 hours at 37°C. Digestion products were then filtered through a cell strainer (BD) and homogenized using the plunger of a 1-mL syringe. The homogenate was resuspended in complete RPMI and centrifuged 10 minutes at 200 g. The pellet was then resuspended and filtered again through a cell strainer before a second round of centrifugation. The cell pellet was resuspended in sorting buffer (1 mmol/L EDTA, 25 mmol/L HEPES at pH 7.0 in PBS; Life Technologies). Cells (5×10⁵) in 100 μL buffer were incubated with 1 μL Fc block (BioLegend), 1 μL anti-CD45–PerCP-Cy5.5 (leucocytes marker, BioLegend), 1 μL anti-F4/80–APC-Cy7 (macrophage marker, BioLegend), 1 μL anti-CD11b—APC (Immunotools), 1 μL anti-CD11c–PE-Cy7 (BioLegend), 1 μL anti-Ly6C–PE (BioLegend), and 1 μL anti-MHC-II–FITC (BioLegend) or –AF700 (eBioscience) for 30 minutes at 4°C. Cells were then centrifuged, resuspended in 300 μL sorting buffer, and analyzed with an LSR II Flow Cytometer (BD). A scatter plot of SSC versus FSC was first used to discriminate cells on size and dissociate intact cells from cellular debris and cell aggregates. Leukocytes were then identified among ECFP-positive cells by CD45 expression. ECFP⁺CD45⁺ cells were then gated to measure CD11b and F4/80 expression. Finally, the Ly6C versus MHC-II plot was gated on CD11b⁺F4/80⁺ cells. For the characterization of F4/80-negative cells, the MHC-II (E), Ly6C (F), and CD11c (G) plots were then gated on F4/80 negative cells.

LPB tumor cells (5×10⁵) and 10⁵ macrophages were cocultured overnight on glass coverslips in the presence of DMSO, BB-94 (10 μmol/L), inhibitor V (20 μmol/L), or Y-27632 (20 μmol/L). Conditioned cell supernatants and cell lysates were collected and analyzed by gelatin zymography as described (29). Briefly, samples were subjected to a 10% (w/v) SDS–PAGE electrophoresis that contained gelatin (0.1 mg/mL). The gels were then incubated in 2.5% (v/v) Triton X-100 to remove SDS, then rinsed three times in water. The gels were incubated overnight at 37°C in gelatinase buffer (10 mmol/L CaCl², 1 μmol/L ZnCl², 100 mmol/L Tris/HCl; pH 7.5), then stained with 0.5% Coomassie Brilliant Blue G-250, followed by destaining in 10% (v/v) acetic acid/ 30% (v/v) ethanol/5% (v/v) formaldehyde in water. Gelatinolytic activity was detected as unstained bands on a blue background and quantified using ImageJ software.

Statistical differences were analyzed with two-tailed unpaired Student t tests or two-way ANOVA for in vitro experiments and mouse experiments using GraphPad Prism 6.0 (GraphPad Software Inc.). In all cases, P < 0.05 was considered statistically significant. For human samples, repeated-measures ANOVA was performed to evaluate the effect of treatment on macrophage infiltration. All reported P values were two sided. For all statistical tests, differences were considered significant at the 5% level. Statistical analysis was performed using STATA 12 and R 3.4.0 software. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

To investigate the migration of macrophages in vivo, we used a transgenic mouse model with ECFP-expressing macrophages (MacBlue; ref. 34). The MacBlue transgene selectively labels monocytes but not resident tissue monocytic cells, except for microglia and Langerhans cells (40). ECFP expression is maintained in monocyte-derived macrophages recruited in inflamed tissues (40). First, the migration of macrophages was investigated in an inflammatory model of a contact dermatitis induced by DNFB (Fig. 1A). In untreated control ears, only a few motionless Langerhans cells, identifiable by a star-like phenotype, were detected (Fig. 1B, day 0; Supplementary Video S1). Following irritation, we observed a pronounced recruitment of fluorescent and motile cells (Fig. 1B, day 8; Supplementary Video S1) squeezing between collagen fibers (Fig. 1C; Supplementary Video S1). These cells displayed round shapes (Fig. 1B–E; Supplementary Video S1) and moved in a nondirectional manner, with an average speed close to 2 μm/minute (Fig. 1F and G, control), suggesting the use of the amoeboid mode of migration.

The day after the last DNFB challenge, we isolated the fluorescent cells recruited in the inflamed derma and analyzed them by flow cytometry (Supplementary Fig. S1A–S1F). Ly6C and MHC-II markers were used to distinguish Ly6C^hiMHC-II^low monocytes from macrophages that were either immature (Ly6C^hiMHC-II^hi) or mature (Ly6C^lowMHC-II^hi/low) as previously described (41). We observed that about 80% of CD11b⁺F4/80⁺ cells were Ly6C^hi and MHC-II^hi (Supplementary Fig. S1E and S1F, yellow) and that about 15% were ly6C^low and MHC-II^hi (Supplementary Fig. S1E and S1F, blue). This showed that a continuum of differentiation status from newly recruited monocytes to fully differentiated macrophages may exist, and that most of the F4/80⁺ cells are either immature or mature macrophages. F4/80⁺ cells were distributed from the derma to the subcutaneous connective tissue that contained loose collagen fibers (Supplementary Fig. S1G). Both Arg1⁺ and iNOS⁺ cells were detected in the inflamed tissue (Supplementary Fig. S1H). Immuno-histo-fluorescence experiments revealed that a large proportion of iNOS⁺ cells were Arg1-negative, and the remaining 30% were both iNOS⁺ and Arg1⁺, indicative of the presence of myeloid-derived suppressor cells (MDSCs). These results indicated that a majority of inflammatory macrophages and some MDSCs are present in the tissue samples.

To confirm whether these macrophages used the amoeboid mode of migration, as suggested by their morphology, we investigated the two defining characteristics of that migration mode—dependence on ROCK activity and independence of matrix metalloproteases (MMP; refs. 8, 24, 32). Therefore, we intraperitoneally injected either Y-27632, a ROCK inhibitor (21, 32), or BB-94, a broad-spectrum MMP inhibitor (42). Mice with inflamed ear dermas were imaged before (control condition) and 1 day after drug injection (Fig. 1A). We observed that Y-27632 treatment significantly modified macrophages morphology because they became more protrusive (Fig. 1D and E; Supplementary Video S1), as previously described in vitro (24, 32), and reduced their motility (Fig. 1F and G). In contrast, BB-94 treatment did not modify macrophage morphology or motility (Fig. 1D–G; Supplementary Video S1). Therefore, in vivo migration of macrophages in the inflamed ear matches characteristics of in vitro amoeboid migration, with regard to cell morphology, trajectory, and drug responsiveness (32). Altogether, our results showed that macrophages use the amoeboid migration mode in vivo as previously described for other leukocytes (25–28).

We previously showed in vitro that the matrix architecture is a critical parameter to dictate the migration mode used by macrophages (32). Because remodeling of the ECM is a feature of most solid tumors (43, 44), we next investigated if the migration of TAMs could differ from the migration of macrophages in nontumoral ear derma. Therefore, we induced fibrosarcomas by subcutaneous injection of GFP-labeled LPB tumor cells in MacBlue mice. LPB cells are noninvasive tumor cells (45), which had low MMP activity compared with macrophages (Supplementary Fig. S2A) and no apparent motility in 3D, both in vitro (Supplementary Fig. S2B) and in vivo (Supplementary Fig. S2C). Because of these criteria, they represent a suitable model to study macrophage-specific migration in a tumoral tissue.

We first characterized the tumor model by IHC. We observed tightly packed cells and an abundant ECM (Fig. 2A), as expected for an LPB-induced fibrosarcoma (45). The ECM contained laminin, fibronectin, and type I and IV collagens (Fig. 2A). As expected, when subcutaneous connective tissues in irritated derma (Supplementary Fig. S1G) and fibrosarcoma (Fig. 2A) were compared, we observed that the ECM was more compact in fibrosarcomas. We next analyzed macrophage infiltration in the tumors by IHC and observed that numerous TAMs were present in all areas of the tumors (Fig. 2B). As previously reported in other tumors (2), they exhibited an M2 polarization because almost no iNOS⁺ cells were detected, whereas numerous Arg1⁺ cells were present (Supplementary Fig. S3A). TAMs, being mainly of monocytic origin (1), expressed the MacBlue transgene in our model (40). Flow cytometry analyses showed that 92.6% ± 8.7% tumor-associated CD45⁺ECFP⁺ cells were CD11b⁺ and 70.6% ± 10.9% were F4/80⁺ macrophages (mean ± standard deviation, n = 8 mice; Supplementary Fig. S3B and S3C), indicating that the fluorescent cells in tumors were mainly macrophages. To determine the identity of the fluorescent F4/80-negative cells, we analyzed other markers (Ly6C, MHC-II, and CD11c) because monocytes and DCs also carry the MacBlue transgene (40). F4/80-negative cells were mainly Ly6C⁺ and CD11c⁺, indicative of circulating monocytes and DCs (Supplementary Fig. S3D). For our method, it is important to note that when the tissue from LBP tumors is harvested and digested, circulating cells are present in the sample, leading to an underestimation of the percentage of macrophages in the connective tissue.

We then analyzed TAM migration inside the tumors. To gain access to ECFP-expressing TAMs by intravital microscopy, LPB fibrosarcomas were induced in MacBlue mice surgically implanted with dorsal window chambers (Fig. 3A). Images, acquired in tumor areas devoid of large blood vessels, showed that most TAMs were localized in the tumor cell–free spaces of the tumor stroma (Fig. 3B; Supplementary Video S2), and only a few TAMs were in close contact with tumor cells and occasionally migrated along them (Supplementary Video S3). We also verified by FACS analysis that almost 100% of LPB cells expressed GFP (Supplementary Fig. S2E), indicating that empty, nonfluorescent spaces, as shown in Fig. 3B, were free of tumor cells. In contrast to the inflamed derma (Fig. 1D; Supplementary Video S1), TAMs displayed elongated and protrusive shapes (Fig. 3C and E; Supplementary Video S4) and migrated with an average speed of about 2 μm/minute (Fig. 3F, control). TAMs moved more directionally compared with macrophages in the inflamed ear derma, with frequent “back-and-forth” motion (Fig. 3G, control).

To establish the dependency of TAM migration on ROCK and MMP activities, mice were injected with either Y-27632 or BB-94 (Fig. 3A). When LPB tumor-bearing mice were imaged after Y-27632 injection, we did not observe any changes in TAM morphology (Fig. 3D and E; Supplementary Video S4) but observed a small and significant increase in TAM motility (Fig. 3F and G). In contrast, BB-94 treatment modified both TAM morphology and motility (Fig. 3D–G; Supplementary Video S4). Average roundness value increased (Fig. 3E), and this morphologic change was accompanied by a significant reduction of 78.6% ± 4.6% (mean ± standard error of 100 cells) of the average displacement speed (Fig. 3F), as illustrated by the 3D map of cell trajectories (Fig. 3G; Supplementary Video S4). We verified that BB-94 did not influence survival and proliferation of LPB cells or survival of BMDMs (Supplementary Fig. S2D).

Because BB-94 is a reversible MMP inhibitor, we evaluated the phenotype and motility of TAMs after BB-94 was cleared from the mice, according to previously published pharmacokinetics studies (46). Four days after BB-94 injection, tumors from the same mice were imaged again. As shown in Fig. 3 and Supplementary Video S4, TAMs recovered an elongated and protrusive shape (Fig. 3H), and their migratory phenotype was identical to that observed before injection of the inhibitor (Fig. 3I). Next, we examined whether the macrophages were TAMs that recovered mesenchymal migration after BB-94 elimination rather than newly recruited cells. Rhodamine-dextran stains circulating cells, predominantly blood monocytes, which transmigrate into tissue and differentiate into macrophages (47). Using 70 kDa rhodamine-dextran (injected at day 7; Supplementary Fig. S4A and S4B) that leaks from the vasculature into the tumor (35), we could also stain endogenous TAMs. Thus, 70-kDa rhodamine-dextran⁺ cells can be both monocytes/macrophages neorecruited from blood and endogenous TAMs. To visualize the newly recruited cells inside the tumor at day 12, 2,000 kDa rhodamine-dextran that does not leak in the tumor was injected at day 11 in a separate set of MacBlue mice (Supplementary Fig. S4C and S4D). About 60% of the blue cells inside tumors were 70 kDa dextran⁺ at day 8 and at day 12 when BB-94 was cleared out of mice (Supplementary Fig. S4E). This indicated that dextran⁺ macrophages observed at day 8, when BB94 was injected, persisted to day 12. At day 12, about 40% of 2,000 kDa dextran⁺ cells were detected inside tumors (Supplementary Fig. S4E), indicative of newly recruited monocytes/macrophages. In conclusion, most of the TAMs were long-lasting cells because at day 12, a portion of the TAMs were also present at the time of BB-94 injection at day 8. These results indicate that TAMs recovered the mesenchymal motility after BB-94 elimination, therefore, excluding a toxic effect of the drug on TAMs. Altogether, these results showed that, in contrast to macrophages recruited to the inflamed derma, TAMs are long-lasting, they exhibit an elongated shape, and their migration is dependent on MMPs and activated by ROCK inhibition, thus, displaying characteristics of in vitro macrophage mesenchymal migration (8, 24, 32).

In the same mice shown in Fig. 3, we imaged the area surrounding the tumor in the window chambers (Fig. 4A, PT). The peritumorous area was characterized by the lack of LPB-GFP cells in the imaged tissue. We observed that macrophages displayed significantly higher migration velocity when compared with TAMs and moved with a rounder shape (Fig. 4B; Supplementary Video S5). Cell tracking revealed longer end-to-end displacements with less “back-and-forth” motion compared with TAMs (Fig. 4D, control). Macrophages had heterogeneous morphologies, with most being round cells (Fig. 4C, white arrowheads) and a few elongated macrophages (Fig. 4C, yellow arrowheads). When Y-27632 was injected, macrophage motility was inhibited by 66.6% ± 8.0% (mean ± standard error of 100 cells; Fig. 4D and E; Supplementary Video S5), similarly to ear derma. BB-94 treatment also partially affected macrophage average speed with a 31.1% ± 7.5% decrease (mean ± standard error of 100 cells; Fig. 4D and E; Supplementary Video S5). Both Y-27632 and BB-94 affected macrophage roundness in opposite ways (Fig. 4F and G). This showed that macrophages used both migration modes, with predominant amoeboid migration in peritumorous derma.

We next assessed the migration mode used by human macrophages in tumors. We developed an ex vivo migration assay (Fig. 5A) using human macrophages in surgical samples isolated from 23 patients with breast cancer. We chose breast cancer because TAM infiltration has been clearly associated with poor prognosis in these patients (48, 49). Breast cancers were mainly grade 2 and 3 invasive carcinomas of no special type and invasive lobular carcinomas (Supplementary Table S1). Tumoral breast explants were sliced and cultured for 72 hours with cell tracker–labeled hMDMs, without any sign of tissue necrosis (Fig. 5B). Using tissue slices from the same cancer explant, the effects of BB-94 and Y-27632 were explored in parallel to vehicle-treated tissue slices (DMSO). Macrophage infiltration into breast tumor samples was significantly inhibited by BB-94 compared with DMSO (P = 0.015; Fig. 5). No significant effect was observed in Y-27632–treated slices (Fig. 5C). This showed that human macrophages also used the MMP-dependent mesenchymal migration to infiltrate breast tumors.

We then assessed whether macrophages use their own proteases to migrate into tumors. Because the surgical samples of breast tumors were too small to perform additional experiments, we used an ex vivo migration assay with hMDMs loaded on explants of mouse fibrosarcomas. hMDMs infiltrated tumor slices, displaying an elongated morphology (Supplementary Fig. S5A, red arrows), and their migration was inhibited by BB-94 and increased by Y-27632 (Supplementary Fig. S5A and S5B). These macrophages were negative for iNOS, CD163, and CD206, suggesting that the daily washes of tumor slices required for addition of fresh inhibitors preclude M1 or M2 polarization (Supplementary Fig. S5C). BB-94 treatment inhibited macrophage MMP activity, whereas Y-27632 had no effect, as evaluated by gelatin zymography (Fig. 6A). Coincubation of human macrophages with mouse LPB tumor cells did not trigger macrophage activation indicated by absence of macrophage polarization and phagocytosis of mouse cells (Fig. 6B and C). Thus, hMDM migration into fibrosarcomas appeared to be similar to that observed in breast cancer explants.

Next, macrophages were treated with the MMP-2/MMP-9 inhibitor V, an irreversible inhibitor of MMP-2, MMP-9, and MT1-MMP (39, 50), and washed before being seeded onto tumor explants. Treatment-induced inhibition of MMP activity was confirmed by gelatin zymography (Fig. 6A). We observed that macrophages lost their ability to infiltrate tumor slices compared with untreated macrophages (Fig. 6D and E), indicating that macrophage tumor infiltration depends mostly on their intrinsic MMPs.

Studies show increasing evidence that TAMs stimulate tumor development (2, 3, 5), and, thus, we examined whether inhibition of TAM mesenchymal migration would limit TAM infiltration and, therefore, negatively affect tumor growth. LPB fibrosarcomas were generated in C57BL/6 mice, and daily treatment with BB-94 was initiated (Fig. 7A, day 0), using a dose that inhibited TAM motility inside tumors (Fig. 3) without affecting neither survival and proliferation of LPB tumor cells nor survival of BMDMs (Supplementary Fig. S2D). Following 12 days of treatment, mice were sacrificed, and the presence of macrophages in tumors was analyzed by IHC (Fig. 7B). A significant decrease of both the percentage of TAMs found in tumors and the size of tumors of treated versus nontreated mice was observed (Fig. 7B and C). This did not correlate with a change in TAM polarization. TAMs remained Arg1⁺ and iNOS-negative after BB-94 treatment (Fig. 7D). Thus, by inhibiting TAM motility, BB-94 reduced TAM infiltration into tumors, which reduced tumor growth.

We report that in vivo macrophage migration involves two modes. In addition to the amoeboid migration generally used by leukocytes (26–28) that we, herein, observed in inflamed ear derma and in the peritumorous tissue, we showed that macrophages can also use the protease-dependent mesenchymal mode in vivo in tumors. We also observed that the macrophage infiltrate human breast cancer tissue explants used the mesenchymal mode, which was made possible through intrinsic MMPs. Chronic treatment of mice with an MMP inhibitor reduced the number of infiltrating TAMs and led to diminished tumor growth.

Leukocytes can migrate in all body tissues to ensure host protection. Although leukocytes are known to use the amoeboid motility, we previously reported that macrophages have the particularity to also use the mesenchymal migration mode in acellular (32) or cellular 3D in vitro environments (8, 25). By modifying the composition and architecture of the 3D ECM, we previously showed that the matrix architecture is the critical parameter determining the migration mode used by macrophages (32). More specifically, we found that the porosity of the matrix is the critical factor: larger pores allow a sliding motility characteristic of the amoeboid mode, whereas lower porosity causes macrophages to use the protease-dependent mesenchymal migration to open paths through which they can progress (30).

Therefore, the use of the mesenchymal mode by macrophages in vivo likely indicates a particular architecture of the tissue. ECM remodeling is a feature of most solid tumors (43, 44). Extensive ECM alterations during tumor progression result in densification of the tissue (51, 52) and, consequently, the stroma progressively stiffens (52), as described for human breast cancers (53). Previous studies show, in vitro and ex vivo, that cells that cannot perform the protease-dependent mesenchymal migration are unable to infiltrate dense matrices (25) or the tumor stroma (e.g., T lymphocytes cannot infiltrate the dense matrix which surrounds tumor islets; ref. 54). Consequently, the reach of their antitumor activities is limited, and the matrix densification may contribute to the immunosuppressive tumor microenvironment (54). As shown here in fibrosarcoma, macrophages mainly migrated in-between tumor cells, where we observed dense ECM, using the MMP-dependent mesenchymal mode. In contrast, MMPs were dispensable in the inflamed ear derma, where the tissue is made of looser ECM. In the peritumorous area, where the ECM is looser than in the tumor, we observed that macrophage migration was mainly amoeboid. However, we observed a partial dependence of macrophage motility to MMPs. This suggested that the tissue might be remodeled due to the proximity of the tumor mass. In human breast cancers, known to have a profound change in ECM organization and an increase in tissue stiffness (53), we observed that macrophages used MMP-dependent migration. Future studies to determine the link between the macrophage migration mode and the biophysical properties of cancer tissues are, thus, needed.

In addition to ECM density, other factors inherent to the tumor environment could trigger the mesenchymal migration of TAMs. We previously showed in vitro that alternatively activated M2 human and mouse macrophages can efficiently use the mesenchymal migration mode in dense matrices, as opposed to classically activated M1 macrophages, which are almost motionless (25). The tumor microenvironment is dominated by Th2-type cytokines, such as IL4 or TGFβ1, which drive macrophages toward an M2-like phenotype (2, 55). For these reasons, it could be hypothesized that tumor-induced polarization of TAMs should promote their infiltration. However, we observed that macrophages migrating into tumor explants using mesenchymal migration were not polarized, suggesting that macrophage polarization status by itself is not sufficient to dictate the choice of the macrophage migration mode in vivo.

Altogether, our findings indicate that macrophages can perform both amoeboid and mesenchymal migration in vivo. This is a unique in vivo property among leukocytes because the migration of lymphocytes, neutrophils, and DCs has been defined as the ROCK-dependent, protease-independent amoeboid mode (25–28). These findings open new perspectives in the field of immunotherapeutic approaches. One can now consider inhibiting TAM mesenchymal migration specifically to limit side effects on other leukocyte-mediated host defenses. We report that chronic treatment with BB-94 reduced tumor growth in mice, not directly by targeting tumor cells, but, rather, indirectly by decreasing the number of recruited TAMs, likely through inhibition of their mesenchymal migration properties. In light of these new data, the use of MMP inhibitors in cancer treatment should be revisited with the aim of targeting TAM migration as soon as a solid tumor is diagnosed and not in patients with advanced stage diseases as performed in the unsuccessful clinical trials in the 1990s (56).

In conclusion, our work establishes that macrophages use two migration modes to move into tissues in vivo and that TAMs can use the mesenchymal mode. This novel migration paradigm could lead to new therapeutic strategies in cancer treatment. Work is in progress to better understand the environmental characteristics and the molecular mechanisms that dictate the macrophage migration mode in both healthy and pathologic tissues and provide further insight into the regulation of macrophage tissue infiltration.

No potential conflicts of interest were disclosed.

Conception and design: P. Gui, I. Maridonneau-Parini, V. Le Cabec

Development of methodology: P. Gui, E. Bellard, M. Golzio, V. Le Cabec

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Gui, M. Ben-Neji, E. Belozertseva, F. Dalenc, C. Franchet, J. Gilhodes, A. Labrousse

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Gui, E. Belozertseva, J. Gilhodes, R. Poincloux, I. Maridonneau-Parini, V. Le Cabec

Writing, review, and/or revision of the manuscript: P. Gui, C. Franchet, J. Gilhodes, E. Bellard, R. Poincloux, I. Maridonneau-Parini, V. Le Cabec

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Gui, M. Ben-Neji, E. Belozertseva

Study supervision: V. Le Cabec

Other (selected and provided surgical samples and clinical analysis of patient): F. Dalenc

This work was supported by INSERM plan Cancer, Fondation pour la Recherche Médicale (grant number #FRM-Equipe DEQ20160334894), Agence Nationale de la Recherche (ANR14-CE11-0020-02), Fondation Toulouse Cancer Santé, IPBS CNRS UMR5089, Ligue contre le cancer Région Midi-Pyrénées, Cancéropole grand sud-ouest GSO-Emergence, Région Midi-Pyrénées (grant number 10051286), Fondation de France (No. 00070698) and Université de Toulouse III. P. Gui was supported by a doctoral fellowship from Université de Toulouse III.

We gratefully thank Professor David Hume (The Roslin Institute, University of Edinburgh, Scotland) for the generous gift of the MacBlue mice and critical reading of the manuscript. We thank Karine Pingris for her expertise and assistance in FACS analysis, Florence Capilla and Talal Al Saati for their expertise in IHC analysis, and Filleron Thomas for the statistics analysis on human samples. We would like to acknowledge the TRI imaging facility and the Anexplo facility for IHC (Platform of Experimental Histopathology of the INSERM/UPS US006 CREFRE, Toulouse Purpan) and for mouse housing (IPBS).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.