A Unidirectional Transition from Migratory to Perivascular Macrophage Is Required for Tumor Cell Intravasation

1The diverse functions performed by tumor-associated macrophages (TAMs) are attributed to their specialization into subtypes, including anti-tumor pro-inflammatory M1 macrophages and pro-tumor immune suppressive or wound healing M2 macrophages. However, the diversity of macrophage types in different tissues and cancers indicates that this is an oversimplification. Intravital microscopy has revealed different TAM behaviors linked to their location, including migration-associated streaming and perivascular populations. Tumor cells migrating in streams with TAMs move at higher speeds, in a more direct route, and from greater distances toward blood vessels than tumor cells migrating without TAMs. This behavior is enabled by a paracrine loop involving colony-stimulating factor 1 (CSF1) production by cancer cells, epidermal growth factor (EGF) production by TAMs, and release of hepatocyte growth factor (HGF) from endothelial cells. Perivascular macrophages are found in structures called TMEM (tumor microenvironments of metastasis), defined as a macrophage, a Mena (Mammalian Enabled)-overexpressing tumor cell, and an endothelial cell in direct contact. TMEM are responsible for vascular endothelial growth factor A (VEGFA)-driven transient vascular leakiness and tumor cell intravasation and predict distant metastatic disease in breast cancer patients.

2Despite these advances, the temporal aspects of macrophage subtype specification within primary tumors and the possibility of inter-conversion among subtypes remain largely unexplored. To learn more about these processes, we applied a range of temporally controlled perturbations of TAM populations in the MMTV-PyMT mouse model of breast cancer.

3Clodronate liposomes target phagocytic cells and can deplete monocyte and macrophages. In previous studies, we observed a reduction in circulating tumor cells (CTCs) in the PyMT model after clodronate liposome treatment. We reconfirmed this, and to our surprise, the reduction in CTCs persisted a week after clodronate treatment, even though liposomes are cleared from the blood within minutes (Figures S1A and S1B). These data argue that TAM function is perturbed for a considerable period following transient clodronate treatment. We therefore set out to track the dynamics of TAMs. To visualize TAMs, we used liposomes loaded with the fluorescent dye 1′-dioctadecyl-3,3,3′3′-tetramethylindocarbocyanine perchlorate (DiI) in tumor-bearing mice (Figures S1A–S1H). As expected, the myeloid cells in the spleen and liver were effectively labeled, but surprisingly, only 3% of myeloid cells in PyMT tumors were labeled after 24 hr; however, this number increased steadily over several days (Figures 1A and S1C–S1F). Similar to the TAMs, monocytes found in tumor blood vessels were not effectively labeled after 24 hr, but this increased after 48 hr (Figure S1G). Staining of tumor sections revealed similar results (Figures 1B and S1C). Moreover, at early time points after DiI liposome injection, only 17% of the DiI+ cells detected inside the tumor were in direct contact with a blood vessel (Figures 1B, 1C, S1H, and S1I). In contrast, 10 days after DiI treatment, 43% of DiI+ cells were in direct contact with blood vessels (Figures 1B and 1C) and more DiI+ cells were found within the tumor compared with earlier time points (Figures S1H and S1I). These data suggest monocyte lineages become labeled with DiI liposomes in hematopoietic tissues and then transit via the blood to tumors, where they gradually accumulate at perivascular sites.

4We tracked the behavior of recently arrived DiI-labeled monocytes entering the tumor using MacGreen mice, engineered to have EGFP+ve colony-stimulating factor 1 receptor (Csf1r)-expressing macrophages. Imaging of MacGreen PyMT mice revealed that non-perivascular TAMs (>1 cell diameter from a blood vessel) were in collagen-rich stromal areas at the tumor edge (purple outlines in Figure S2A), with few macrophages found within tumor cell nests (yellow outline in Figure S2A). Non-perivascular (stromal) TAMs exhibited a different distribution of velocities compared to perivascular TAMs, with a distinct population of EGFP+ stromal TAMs having a higher velocity than that of any EGFP+ perivascular TAMs (orange outlines in Figure S2A) (Figures S2A–S2C; Videos S1 and S2), thus confirming their distinct phenotype. Intravital imaging 48 hr after DiI injection into MacGreen PyMT mice revealed DiI+ cells in the blood and low numbers of DiI+/GFP+ cells within the tumor (Figure 1D; Video S1). The few DiI+/GFP+ TAMs were often motile and found in stromal areas (Videos S1 and S2), suggesting that newly arriving monocytes are migratory and reside in stromal areas rich in collagen and away from vessels. The inhibitory effect of clodronate liposomes on intravasation after 48 hr is likely due to depletion of recently arrived monocytes involved in streaming migration (Figure S1B).

5To track monocyte subtypes in a non-biased way, we used 5-ethynyl-2′-deoxyuridine (EdU) to label rapidly turning over bone marrow and splenic monocytes. We dosed tumor-bearing PyMT mice twice with EdU (2.5 hr apart) and harvested tumors after 9 hr (Figure S2D). We did not find any EdU+/CD45+ leukocytes in the blood stream (Figure 1E) but saw clear EdU labeling of tumor cells and CD11b+ cells in bone marrow (Figure S2E). Few CD68+ TAMs in the tumor were EdU+ at the 9 hr time point (<0.1%), suggesting that most CD68+ TAMs in the MMTV-PyMT system are non-proliferative and enabling us to track post-mitotic EdU+ monocytes coming from the bone marrow or the spleen by immunofluorescence (IF) (Figure S2E). The number of EdU+ cells in the blood peaked ∼48 hr, and almost no positive cells were observed by 96 hr, giving us a defined labeled population to track (Figure 1E). After 3 days, EdU+/CD68+ TAMs were predominantly non-perivascular (Figures 1F and 1G) (note the high levels of EdU labeling of tumor cells). However, almost 40% of EdU+/CD68+ TAMs were perivascular after 7 days (Figures 1F and 1G), and this increased to ∼80% at 10 days. The continuing increase in perivascular TAMs between 7 and 10 days, even though no EdU-labeled monocytes were present in the blood, excludes the possibility of perivascular TAMs being recruited directly from the blood.

6To further test whether streaming TAMs transition into stationary perivascular TAMs, we transiently depleted all macrophages using the macrophage Fas-induced apoptosis (MaFIA) mouse model with orthotopically implanted PyMT tumors. In this model, 5 days of treatment with the small molecule AP20187 (also known as the B/B homodimer) effectively removes >90% of TAMs by apoptosis. We followed the repopulation of TAMs and their spatial location by IF (Figures 2A and S3). Although the number of CD68+ TAMs returned to control levels within 4 days after termination of B/B treatment (Figures 2B, 2C, and S3), the TAMs were rarely found in contact with blood vessels stained with either CD31 or endomucin. It took up to 8 days after the end of B/B treatment for the number of perivascular CD68+ TAMs return to control levels (Figures 2B and 2D). We additionally characterized the expression of markers linked to perivascular macrophage biology: VEGFA, CD206, and LYVE-1 (Figures S4A–S4C and S5A–S5D). CD206 stained both perivascular and non-perivascular TAMs at early time points and, similar to CD68, showed the same transition to predominantly perivascular staining later. We observed low levels of VEGFA staining in CD68+ cells, but these showed a similar trend of accumulation in perivascular regions 7 days after B/B treatment. We did not observe LYVE-1 staining in any TAMs (Figure S5D).

7We also tracked the recovery of perivascular TAM functionality after B/B treatment by examining vessel leakiness and CTC numbers. Restoration of vessel leakiness to control levels coincided with the return of perivascular TAMs 8 days after the last B/B treatment (Figures 2C, 2E, and 2F). However, the number of CTCs took 14 days to recover after B/B treatment (Figure 2G), indicating that perivascular leakiness is restored quickly after TAMs contact the vasculature but that the reestablishment of TAM functions that aid tumor cell intravasation takes longer.

8Having established that circulating monocytes transition through a migratory phase into sessile perivascular macrophages, we sought to identify the molecular regulators of these steps. We tested whether CCR2, a receptor that mediates monocyte chemotaxis, was required for the initial recruitment of monocytes. A mixture of differentially labeled CCR2 wild-type and −/− bone marrow-derived monocytes was adoptively transferred into tumor-bearing mice. The proportion of wild-type to CCR2 −/− CD11b+ cells was tested 2 and 6 days later (Figure S6). CCR2 −/− cells were dramatically under-represented in the tumor, despite their ability to colonize the bone marrow (Figures 3A and 3B).

9CXCR4 expression has been associated with TAM recruitment and differentiation. Therefore, we studied the CXCR4 expression levels in CD68+ TAMs at different time points after B/B treatment. Only a small percentage of CD68+ macrophages expressed CXCR4 4 days after arrival in the tumor (Figures 3C and 3D), suggesting that CXCR4 is not important for their initial recruitment. To test whether the CXCR4/CXCL12 axis is required for migratory macrophages to become functional perivascular TAMs, we used the CXCR4 antagonist AMD3100. We depleted TAMs with B/B treatment, followed by a 7-day, twice-daily AMD3100 treatment during the recovery phase (Figure 3E). As previously noted, CXCR4 blockade reduced the vascular area in tumors (Figures 3F and 3G), but the number of CD68+ TAMs was similar to controls (Figures 3F and 3H). The percentage of TAMs directly interacting with blood vessels was greatly reduced in the AMD3100-treated tumors (Figures 3F and 3I), and this correlated with a reduction in vascular leakiness (Figures 3J and 3K).

10The data outlined earlier establish CXCR4/CXCL12 signaling in directing TAMs toward the blood vessels; however, they do not explain what upregulates CXCR4 or identify the source of CXCL12. We tested whether tumor-derived factors could upregulate CXCR4 by co-culturing bone marrow-derived macrophages (BMMs) with PyMT cancer cells. This co-culture resulted in strong CXCR4 expression in F4/80-positive BMMs (Figure 4A), and exposure of BMMs to cancer cell conditioned media triggered a marked upregulation of CXCR4 in macrophages (Figures 4B and 4C) Crucially, inhibition of transforming growth factor β (TGF-β) signaling blocked the induction of CXCR4 mRNA and protein by cancer cell conditioned media (Figures 4B and 4C), and TGF-β was sufficient to induce CXCR4 (Figure 4D).

11Immunofluorescence analysis of tumor sections revealed that CXCL12 is expressed by elongated cells frequently adjacent to blood vessels (Figure 4E). Co-staining demonstrated that three-fourths of CXCL12-expressing cells were positive for the generic fibroblastic marker vimentin and roughly 10% were positive for the more specific cancer-associated fibroblast marker α-smooth muscle actin (αSMA) (Figures 4E, 4F, and S7A–S7C). There was no overlap with either the endothelial marker endomucin or the pericyte marker desmin (Figures 4E, 4F, and S7D). These data indicate that CXCL12 is expressed by stromal fibroblasts proximal to blood vessels and explain the recruitment of TAMs to perivascular regions following the TGF-β-driven induction of CXCR4. Other studies have shown that ANG2/Tie2 blockage results in a phenotype similar to the one we observe following CXCR4 blockade with reduced vascular density and failure of TAMs to attach to the blood vessels. In addition, our results agree with previous work in which AMD3100 treatment led to a preferential reduction in perivascular macrophages. We propose that CXCR4/CXCL12 is important for migration to blood vessels, while Tie2 is required for attachment to the endothelial cells and maturation into a functional perivascular TAM regulating vascular leakiness and cancer cell intravasation.

12To conclude, we propose that monocytes, recruited via CCR2 signaling, initially become motile streaming TAMs before a TGF-β-dependent conversion into CXCR4-expressing macrophages. These TAMs are then recruited to become sessile perivascular TAMs by CXCL12. This unidirectional differentiation process takes 10–14 days. This argues against the view that once educated, the macrophage phenotype does not change, and it refutes the opposing idea that unrestricted inter-conversion between macrophage states is possible. Instead, we document a surprisingly stereotypic and unidirectional conversion between macrophage states. Single-cell RNA sequencing (RNA-seq) analysis suggests that a similar situation applies in human breast cancer. In the future, it will be interesting to explore this dynamic in the context of chemotherapy and other perturbations, which in some cases may trigger local proliferation of macrophages. An improved understanding of the lineage and temporal dynamics of different TAM subsets will be important for optimizing the targeting of TAMs for therapeutic benefit.

13All mice studies were carried out in accordance with NIH regulation (US) or UK Home regulation (UK). Procedures were approved by the Albert Einstein College of Medicine Animal Care (animal use protocol 20130909) and by the Francis Crick Institute Biological Ethics Committee (project license 70/8380). MMTV-PyMTmice were maintained on a susceptibility to Friend leukemia virus B/NIH (FVB/N) background and were crossed with MMTV-Cre and lox-stop-lox (LSL)-eGFP or with a co-integrated allele FVB/N MMTV.improvedCre.LSL enhanced Cerulean Fluorescent Protein (eCFP)jwp mice to develop mice with green or blue mammary gland tumors. MacGreen mice were crossed with PyMT FVB mice to develop MacGreen-PyMT mice. Age-matched females were used in experiments when they were around 12–14 weeks old. MaFIA mice, known as C57BL/6 Tg (Csf1r-EGFP NGFR/FKBP1A/TNFRSF6)2Bck/J, were obtained from The Jackson Laboratory. All experiments with MaFIA mice were performed with implantation of orthotopic C57BL/6 PyMT tumors in MaFIA mice. The tumors were developed by implantation of tumor pieces (2 × 2 mm) of late-stage spontaneous C57BL/6 PyMT tumors into the mammary fat pad of 5- to 7-week-old female MaFIA mice. Typically, after 6–7 weeks, single PyMT tumors appeared. Experiments were typically performed on 0.6–0.8 cm tumors. Multiphoton intravital microscopy was performed as previously described. Under general anesthesia, the mouse was placed on a heated microscope stage, with the surgically exposed tumor placed onto a cover glass. Imaging was performed using a custom-built 2-laser multiphoton microscope.

14Tumor or spleen samples were prepared by tissue digestion using Liberase and Dispase (Roche), combined with red blood cell (RBC) lysis (eBioscience) as previously reported. Blood cells were isolated by cardiac puncture, followed by RBC lysis. Cells were blocked with an anti-mouse CD16/CD32 fragment crystallizable (Fc) blocking antibody for 10 min before antibody staining (BD Biosciences). Gating was used to exclude dead cells, cell doublets, and clusters. In certain experiments, mice were injected with CD45-fluorescein isothiocyanate (FITC) (eBioscience) 2–3 min before sacrifice to label all immune cells in the blood, but not in the tissues, at the time of death to exclude those cells from the tissue analysis, as well as to measure immune cells inside the blood at the time of death. Data were analyzed with FlowJo software (Tree Star).

15The liposomes clodronate, control, and DiI (1′-dioctadecyl-3,3,3′3′-tetramethylindocarbocyanine perchlorate) (Clodrosome) were injected into the tail vein at 5 mL/kg in two doses 24 hr apart in age-matched, tumor-bearing PyMT females around 12–14 weeks of age. Either 48 hr or 7 days after the first clodronate liposome injection, CTCs were isolated (as described later). After the last DiI liposome injections, tissue samples were isolated at various time points (as described earlier) for IF staining (as described later) or fluorescence-activated cell sorting (FACS) analysis (as described earlier). EdU (5-ethynyl-2′-deoxyuridine) was dissolved in a saline solution and injected via tail vein twice at 40 mg/kg, with a 2.5 hr rest between injections.

16Bone marrow cells were isolated from the femur, and a suspension was created in a small volume of PBS and spread into a smear on glass slides. Blood smears were created with blood from the lateral tail vein. The slides were air-dried and subsequently fixed with 100% methanol for 2 min, air-dried, and kept at 4°C until ready to perform the immunofluorescence (IF) protocol described later.

1710 mg/kg B/B homodimerizer (AP20187, Clontech) diluted in 4% ethanol, 10% PEG-400, and 1.7% Tween 20 or vehicle control was injected intraperitoneally on 5 subsequent days. Treatment was started when tumors were 0.6–0.8 cm (in diameter), and typically, they were not larger than 1.0–1.1 cm by the end of the experiment. AMD3100 was administered at 5 mg/kg twice a day via intraperitoneal injection for 7 days.

18Measurement of vascular leakiness was performed as previously described. One hour before the termination of the experiment, 155 kDa- dextran-tetramethylrhodamine (TMR) (Sigma) was injected intravenously (i.v.), and CD31-biotin was injected i.v. 5 min before the end of the experiment, labeling all active blood vessels. Tumors were fixed overnight in 4% paraformaldehyde (PFA), transferred to 30% sucrose, and embedded in optimal cutting temperature (OCT) compound. 8 μm sections were cut, and IF was performed as described later. Extravascular dextran was measured as previously described using ImageJ.

19Cells or tumor sections were fixed and permeabilized with ice-cold acetone (sections) or ethanol (BMMs) for 10 min, blocked for 1 hr with blocking buffer (1% BSA, 5% fetal bovine serum [FBS], and 0.1% fish skin gelatin in PBS-T). Primary antibodies were incubated overnight at 4°C, followed by PBS-T washes and secondary antibody or fluorescently tagged streptavidin (Invitrogen) incubation combined with DAPI for 1 hr at room temperature. EdU was visualized with an EdU click-assay (Molecular Probes) according to the manufacturer’s instructions, followed by the same IF protocol described earlier. Images were acquired using a Zeiss Axio Observer with a 40× objective or a Zeiss LSM 780 confocal microscope with a 20× objective. Images were subsequently imported into ImageJ for analysis.

20Blood was collected from the right ventricle of the heart into a heparin-coated syringe under terminal anesthesia. After RBC lysis (eBioscience), cells were seeded into DMEM/F12 media supplemented with 20% FCS and Pen/Strep, and single tumor cells were counted 7 days after initial plating. Data included here have new data collated with controls previously reported in.

21Monocytes were isolated by crushing the femur and tibia of female wild-type C57BL/6 or CCR2 knockout (KO) mice, followed by purification with a magnetic-activated cell sorting (MACS) monocyte isolation kit (Miltenyi). Monocytes were labeled with (5(6)-Carboxyfluorescein N-hydroxysuccinimidyl ester (CSFE) or CellTrace Violet (Invitrogen) in PBS, washed, and counted. A 50:50 mixture of CCR2 KO-to-wild-type (WT) monocytes was prepared and injected in a volume of 100 μL of PBS into the tail vein of tumor-bearing C57BL/6 mice with surgically implanted PyMT tumors. A minimum of 5 × 106 monocytes of each subtype were injected. Two or six days after the adoptive transfer, tissues were harvested and analyzed.

22Bone marrow cells were isolated from the mouse leg bones (femur and tibia) by flushing with PBS and then differentiated into BMMs with macrophage colony-stimulating factor (MCSF) (10 ng/mL, PeproTech) in BMM media (DMEM/F12, Gibco) supplemented with 10% FBS, Pen/Strep, and 10 ng/mL MCSF). When used for experiments, TGF-β (2 ng/mL, PeproTech) was added after the first media change, typically 2 days after the isolation, for the duration of the experiment. PyMT cancer cell isolates were prepared as previously described. 36 hr were allowed for cancer cells to generate conditioned media (CM), which was then pushed through a 0.4 μm syringe filter (Fisher Scientific). For the treatment of BMMs with CM, 2–3 days after initial bone marrow isolation, non-adhered bone marrow cells were washed away and fresh media diluted 50:50 with filtered CM were added either with or without TGF-β receptor inhibitor (5 μM SB-505124, Sigma). Media were changed every 2 days until the end of the experiment.

23Cells were collected in RNAprotect reagent (QIAGEN) and kept frozen until RNA extraction using RNeasy mini kits (QIAGEN). The cDNA library was prepared using M-MuLV reverse transcriptase (New England Biolabs), and qPCR was performed using SYBR-green platinum (Invitrogen) assays on the QuantStudio 7 Real-Time PCR systems (Applied Biosystems) with at least two housekeeping genes for normalization.

24Statistical analysis were performed using the chi-square for contingency table test with EdU data, the standard two-tailed Student’s t test for comparing two datasets, and ANOVA followed by Tukey’s or Dunnett’s multiple-comparison post hoc tests for multiple datasets when the sample size was large enough to confirm normality (Shapiro-Wilk). For smaller datasets, Kruskal-Wallis was used followed by Dunn’s multiple-comparison test. To examine the distribution of macrophage motility, the Kolmogorov-Smirnov test was used. All statistical analysis was done using Prism (GraphPad). All graphs show the number of mice indicated as separate data points. Level of significance is indicated with red * (p < 0.05), ** (p < 0.01) or *** (p<0.001).