Blockade of Myeloid-Derived Suppressor Cell Expansion with All-Trans Retinoic Acid Increases the Efficacy of Antiangiogenic Therapy

The vascular endothelial growth factor receptor-2 (VEGFR2)–blocking antibody ramucirumab received regulatory approval for the treatment of patients with gastric, colorectal, and non–small cell lung cancer. VEGFR2 is frequently overexpressed in breast cancer and might therefore represent a therapeutic target in this tumor entity (1, 2). However, responses of patients to blockade of VEGFR- or VEGF-signaling turned out to be very limited. Therefore, the treatment of patients with breast cancer with antiangiogenic agents represents a relevant clinical challenge (3, 4).

Previous studies identified bone marrow (BM)–derived CD11b⁺GR1⁺ murine myeloid-derived suppressor cells (MDSC) as resistance-conferring, detrimental mediators accumulating in tumors upon treatment with antiangiogenic therapies (AAT; refs. 5, 6). MDSC comprise a heterogeneous population of immature, myeloid cells including G-MDSC (CD11b⁺Ly6G⁺Ly6C^low) and M-MDSC (CD11b⁺Ly6G⁻Ly6C⁺) subsets capable of inducing angiogenesis and immunosuppression (6, 7).

Tumors induce mobilization and recruitment of MDSC from the BM by secreting mediators including G-CSF or GM-CSF (8, 9). In addition, therapy-induced physiologic adaptations of the tumor microenvironment have been associated with increased MDSC expansion and recruitment. In particular, intratumoral hypoxia has been well recognized as one of the main drivers triggering this process (10, 11). Reduced oxygen tension, resulting from rapid tumor growth or blood vessel eradication upon AAT, leads to the induction of tumor hypoxia. Hypoxia in turn favors the expression of tumor-derived factors such as CCL2, CXCL5 and CXCL12/SDF-1, VEGF, and PLGF, leading to enhanced recruitment of MDSC into the tumor bed (6, 11, 12).

Once residing in the tumor, MDSC suppress T-cell mediated antitumor responses (13, 14) and induce tumor angiogenesis by various mechanisms. For example, MDSC secrete proteinases such as MMP9 that induce the mobilization of proangiogenic molecules residing in the extracellular matrix of the tumor microenvironment (6, 11). Furthermore, MDSC express VEGF and FGF2 in a STAT3-dependent manner, resulting in enhanced tumor neovascularization and growth (6, 15).

The active vitamin A metabolite all-trans retinoic acid (ATRA) is currently used to induce differentiation of leukemic blasts into mature myeloid cells in acute promyelocytic leukemia (16). Importantly, ATRA also enhances the differentiation of MDSC into macrophages and/or dendritic cells in vitro. In addition, treatment of tumor-bearing mice with ATRA resulted in a significant reduction of MDSC in vivo (17, 18). Therefore, we hypothesized that combinatorial treatment with ATRA could improve the efficacy of AAT via reducing resistance-conferring MDSC.

Our data using two syngeneic murine breast cancer models show that ATRA increases the antitumor activity of AAT by a concomitant reduction of MDSC levels. Moreover, our work provides evidence that MDSC-secreted S100A8 represents a resistance-conferring factor induced by AAT. S100A8 acts by destabilization of the tumor vasculature, which can be reverted by combining AAT with ATRA.

Female 8- to 9-week-old BALB/c mice were purchased from Charles River Laboratories International. All animal experiments were carried out in concordance with the institutional guidelines for the welfare of animals and were approved by the local licensing authority Hamburg (project numbers G36/13 and G126/15). Housing, breeding, and experiments were performed under standard laboratory conditions (22 ± 1°C, 55% humidity, food and water ad libitum).

Murine mammary adenocarcinoma cell lines 4T1 and TS/A were provided by Professor Peter Carmeliet (VIB Vesalius Research Center, KU Leuven) and cultured in RPMI 1640 or DMEM medium supplemented with 10% FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin, respectively. Human umbilical vein endothelial cells (HUVEC, Lonza) were cultured in EBM-2 medium supplemented with 10% FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, 100 μg/mL streptomycin, and a complete set of EGM-2 growth factors (Lonza). Cells were maintained at 37°C and 5% CO₂ in a humidified atmosphere and routinely tested to be Mycoplasma negative (Venor GeM Classic, Minerva Biolabs). Cells were cultured no longer than 15 passages before experimental use. No cell line used in this study is listed in The International Cell Line Authentication Committee database of commonly misidentified cell lines and was authenticated according to their in vitro/in vivo growth characteristics and histology. To analyze the effect of S100A8 on HUVEC, cells were washed 2 times with PBS and seeded in 96-well plates (1.5 × 10⁴ cells/well) in EBM-2 medium containing 2% FCS and indicated concentrations of S100A8 for 48 hours. Cell viability was determined using WST-1 reagent (Roche).

4T1 or TS/A cells (5 × 10⁵) were orthotopically injected into the second mammary fat pad of 8- to 9-week-old syngeneic female BALB/c mice. When tumors reached 100 mm³, mice were randomized and treated either with ATRA (7.5 mg/kg, daily), DC101 (20 or 10 mg/kg, 3 times per week), or a combination of both drugs by i.p. administration. Doxorubicin (Dox; 3 mg/kg, i.p.) was administered 2 times per week. Tumor size was measured with a digital caliper, and the volume was calculated using the formula V = (length² x width)/2. For histologic analyses, BrdUrd (1 mg, i.p), pimonidazole (1 mg, i.p.), and a FITC-conjugated lectin (0.05 mg, i.v.) were injected 12 hours, 2 hours, and 10 minutes before sacrifice, respectively.

Tumor tissue (300–500 mg) was mechanically minced and digested in 15 mL RPMI 1640 medium containing 0.2 mg/mL collagenase A (Roche) for 60 minutes at 37°C and shaking at 80 rpm. Next, 10 mL of a PBS solution containing 0.015 mg/mL DNase I (Roche) was added, and digestion was continued for another 30 minutes at 37°C. Cell suspensions were filtered through a 70 μm cell strainer. After centrifugation, cells were resuspended in FACS buffer (2% FCS, 1 mmol/L EDTA, 0.1% NaN₃ in PBS) and immediately used for flow cytometry.

For flow cytometry analysis, cells (1 × 10⁶/staining) were Fc-blocked (anti-mouse CD16/32 antibody; Biolegend) for 15 minutes at 4°C. Afterward, cells were stained for 40 minutes with the following fluorochrome-conjugated primary anti-mouse antibodies: PE-Cy7 CD11b (clone M1/70; BD Biosciences), PE Ly6-G (clone 1A8; BD Biosciences), PerCP-Cy5.5 Ly6-C (clone Hk1.4; eBioscience), FITC F4/80 (clone BM8; Biolegend), APC-Cy7 Gr-1 (clone RB6-8C5; Biolegend), APC CD3 (clone 17A2; eBioscience), eFluor 450 CD8a (clone 53-6.7; eBioscience), PE CD49b (clone DX5; eBioscience), and PE-Cy7 NKp46 (clone 29A1.4; eBioscience). DAPI was used as a viability stain. Samples were acquired using a BD FACS Canto II flow cytometer, and data were analyzed using the BD FACS Diva software.

All methods for histology and immunostaining have been described in detail in refs. 19 and 20. Tumor samples were fixed overnight in 4% paraformaldehyde at 4°C and embedded in paraffin or further incubated overnight in 40% sucrose and embedded in OCT medium for cryosectioning. Paraffin sections (4 μm) were stained with primary antibodies to detect vessel number and vessel proliferation (anti-CD105, R&D Systems, AF1320 + anti-BrdUrd, Abd Serotec, MCA2060), tumor hypoxia (pimonidazole, HP3-1000kit; anti-GLUT1, Abcam, ab115730), vessel number, perfusion and permeability (anti-CD105, R&D Systems, AF1320 + FITC-lectin, Vector Laboratories, FL-1171), and vessel-associated ZO-1 (anti-CD105, R&D Systems, AF1320 + anti-ZO-1 clone ZO-1-1A12, Invitrogen). Cryosections (8 μm) were stained and analyzed for pericyte coverage (anti-CD31, Dianova, DIA-310 + anti-NG2, AB5320 Merck) of tumor microvessels. For the analysis of tumor cell proliferation, tumor sections were stained with an anti-phosphohistone H3 antibody (clone D7N8E; Cell Signaling Technology). Sections were then incubated with the corresponding horseradish peroxidase– or fluorescently conjugated secondary antibodies. Nuclei were counterstained with DAPI. For morphometric analysis, 8 to 10 optical fields per tumor section were acquired using a Zeiss Axio Scope A1 for immunohistochemistry or a Leica DM1000 fluorescence microscope for immunofluorescence analysis. Images were analyzed using the NIH Image J analysis software.

Scanning electron microscopy imaging to assess the intratumoral microvessel architecture was performed as previously described (see also Supplementary Information; refs. 20, 21).

Cells (1.5 × 10⁶ 4T1) were seeded in T-75 cell culture flasks and incubated in RPMI 1640 medium supplemented with 10% FCS, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin for 3 days to generate tumor cell–conditioned media (TCM). TCM was sterile filtered using 0.2 μm filters, supplemented with 10 mmol/L HEPES and 20 μmol/L ß-mercaptoethanol, and stored at −80°C until further use.

BM was isolated under sterile conditions from WT BALB/c mice and subjected to erythrocyte lysis (155 mmol/L NH₄Cl, 10 mmol/L KHCO₃, 0.1 mmol/L EDTA, pH7.4) for 2 minutes at 4°C. Primary bone marrow mononucleated cells (BMMC) were adjusted to a density of 0.5 × 10⁶ cells/mL in 75% 4T1 conditioned medium + 25% RPMI medium supplemented with 10% FCS, 10 ng/mL GM-CSF, 20 μmol/L ß-mercaptoethanol, 10 mmol/L HEPES, 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin. Cells were treated with or without 1.5 μmol/L ATRA for 4 days at 37°C/5% CO₂ following purification of MDSC using the myeloid-derived suppressor cell isolation kit (Miltenyi). For coculture assays, 6 × 10⁴ HUVEC were seeded in the lower compartment of a 24 transwell chamber using inserts with 0.4 μm pore size. Magnetic-activated cell sorting (MACS)–purified MDSC (2 × 10⁵) were seeded in the upper compartment, and coculture was performed for 48 hours. HUVEC viability was assessed using WST-1 reagent (Roche).

Permeability across endothelial cell (EC) monolayers was measured using matrigel-coated transwell filters (3 μm pore size; Greiner). HUVEC cells were seeded on 2 consecutive days at a density of 4 × 10⁴ cells per well (upper chamber) and were further cultured for 24 hours. Afterward, cells were incubated for 6 hours in the presence of human S100A8 (10 μg/mL) or human VEGFA (200 ng/mL). FITC-dextran (1 mg/mL, 3 kDa; Molecular Probes) was added to the lower compartment of the transwell system, and permeability was measured by its diffusion into the upper compartment (485 nm excitation, 535 nm emission, Tecan infinite F200 Pro).

Data represent mean ± SEM of representative experiments, unless otherwise stated. To compare the mean of two groups, an unpaired, two-tailed Student t test was used. Pairwise comparison testing in experiments with more than two groups was performed using one-way ANOVA. Pairwise comparisons of tumor growth kinetics were performed using two-way ANOVA. Statistical significance was assumed when P < 0.05.

In order to investigate our hypothesis that treatment with ATRA increases the efficacy of antiangiogenic drugs by reducing MDSC numbers, we combined ATRA with DC101, a monoclonal antibody targeting murine VEGFR2. For the combinatorial treatment approach, we utilized two well-characterized syngeneic models of breast cancer, 4T1 and TS/A (8, 22). We injected the cell lines orthotopically in the second mammary fat pad of BALB/c mice and started treatment when the mean tumor burden reached 100 mm³ (Fig. 1A). We deliberately chose a submaximal dose level of DC101 (20 and 10 mg/kg) to detect potential additive effects of ATRA treatment. The dose level of ATRA (7.5 mg/kg) was utilized because previous studies with similar concentrations showed biological activity without the presence of side effects (23, 24). In both models, ATRA slightly but not significantly reduced tumor growth (Fig. 1B–E) in concordance with literature (23). Interestingly, the combination of DC101 and ATRA induced an additive reduction of tumor volume and weight when compared with DC101 monotherapy in the 4T1 and in the TS/A model (Fig. 1B and C; Supplementary Note S1; Supplementary Fig. S1A and S1B). In the TS/A model, the additive effect on tumor weight was only significant when 10 mg/kg DC101 was used in combination with ATRA (Supplementary Fig. S1A and S1B). This additive therapeutic effect was maintained during a treatment period of 18 days, and after the discontinuation of treatment, the mice survived for 6 more days compared with placebo treatment (Supplementary Figs. S1A–S1C and S2). Accordingly, tumor cell proliferation, measured by phospho-histone H3⁺ (pHH3⁺) nuclei, was significantly reduced upon combination of ATRA and DC101 compared with control- and DC101-treated tumors (Fig. 1F–H).

These results collectively indicate that the addition of ATRA increases the antitumor activity of the VEGFR2–targeting antibody DC101 in murine syngeneic breast cancer models.

It is well known that antiangiogenic drugs increase hypoxia upon chronic treatment (11, 25), which represents an important driver of MDSC-recruitment and a resistance-conferring factor (10, 11). In concordance with previous data, treatment with DC101 significantly increased pimonidazole-positive areas, a surrogate for intratumoral hypoxia, whereas monotherapy with ATRA did not modify hypoxia (Fig. 2A and B). Interestingly, the addition of ATRA to DC101 treatment almost completely alleviated intratumoral hypoxia (Fig. 2A and B; Supplementary Fig. S1C). Notably, staining of the alternative hypoxia marker GLUT1 by immunohistochemistry yielded similar results (Supplementary Fig. S3A and S3B).

Consequently, we investigated the effects of ATRA, previously shown to reduce MDSC in tumor-bearing mice (17, 18), on intratumoral MDSC populations using flow cytometry. These analyses revealed that in accordance with the increase in hypoxia, both CD11b⁺Ly6G⁺Ly6C^low G-MDSC and CD11b⁺Ly6G⁻Ly6C⁺ M-MDSC were significantly elevated in tumors treated with DC101 (Fig. 2C–F). Concomitant treatment with ATRA normalized the frequencies of both MDSC subsets in DC101-treated tumors to similar levels as detected in control-treated tumors (Fig. 2C–F). However, ATRA had no impact on intratumoral frequencies of CD8⁺ cytotoxic T-cell and natural killer cell populations (Supplementary Fig. S3C and S3D).

To characterize the angiogenic phenotype of intratumoral MDSC upon treatment, FACS-isolated MDSC were subjected to qPCR expression analysis using a panel of proangiogenic genes. These experiments revealed that in the G-MDSC subset, the expression of proangiogenic Vegfa, Hgf, Mmp9, and iNos was not significantly altered by any of the therapeutic interventions (Supplementary Fig. S4A). However, in the M-MDSC fraction, Mmp9 mRNA levels were increased in DC101-treated tumors, which was blunted by the addition of ATRA (Supplementary Fig. S4B).

Changes in tumor oxygenation have been shown to be associated with modifications of the tumor vasculature (26). Our findings that (i) concomitant treatment with ATRA alleviated tumor hypoxia and (ii) ATRA reduced the frequencies of MDSC, which can act proangiogenic, prompted us to analyze the vessel phenotype and functionality in tumors treated with DC101 with and without ATRA.

As expected, DC101 reduced the microvessel density of tumors, which was not further decreased upon combination with ATRA (Fig. 3A and B). By injecting a FITC-labeled lectin, we observed that DC101 reduced microvessel perfusion, which was reverted by addition of ATRA, leading to an increase in the relative number of functional vessels (Fig. 3C). Next, we quantified the fraction of proliferating vessels after injection of BrdUrd. These analyses revealed an increase of BrdUrd⁺ vessels upon DC101 treatment compared with control treatment as previously described (27), whereas ATRA monotherapy had no effect (Fig. 3D). Interestingly, the combination of DC101 and ATRA blunted the reinduction of tumor vessel proliferation (Fig. 3D). Moreover, leakage of FITC-lectin from tumor vessels was significantly increased in DC101-treated tumors compared with controls, whereas the addition of ATRA reduced the DC101-evoked vascular permeability (Fig. 3E).

The coverage of blood vessels with pericytes represents an important attribute of their maturity and functionality (28). The quantification of NG2⁺ pericytes showed a significant reduction of pericyte-covered vessels upon DC101 treatment (Fig. 3F and G). In contrast, ATRA monotherapy resulted in an increase of NG2⁺ cells adjacent to CD31⁺ ECs in comparison with control-treated tumors, which was maintained upon combination with DC101 (Fig. 3F and G). Together, our histomorphometric analyses revealed an increase of mature, functional tumor microvessels upon combination of DC101 with ATRA.

In order to further investigate the blood vessel architecture at the ultrastructural level, we performed intratumoral scanning electron microscopy imaging. These analyses revealed a disorganized blood vessel architecture in the DC101-treated tumors as indicated by irregular-shaped vessel walls and EC extensions protruding into the vessel lumen in concordance with previous literature (Fig. 4; refs. 20, 21). In contrast, ATRA and the combination of ATRA with DC101 induced a normalization of the tumor vessel morphology indicated by lower abundance of protrusions and a regular, flat cobble-stone morphology of the endothelial monolayer (Fig. 4).

Vascular normalization and consecutive alleviation of tumor hypoxia might be correlated with remodeling of the extracellular matrix and/or changes in the tumor metabolome. Analysis of key extracellular matrix components, including fibronectin 1 (Fn1), EDA-Fn1, EDB-Fn1, IIICS-Fn1, collagens 1A–4A (Col1A–4A), secreted protein and rich in cystein (Sparc) and periostin (Postn) revealed a decrease in mRNA levels of Fn1 and its splice variants EDA-Fn1 and IIICS-Fn1 upon DC101 administration, which was abrogated by the addition of ATRA, whereas the other matrix components were essentially unchanged (Supplementary Fig. S5A and S5B).

Interestingly, LC-MS–based metabolomic analysis revealed that, compared with controls, tumors treated with DC101 showed (a trend of) elevated glycolytic intermediates, which was partially abrogated by the addition of ATRA (Table 1).

Collectively, these data indicate that ATRA reverts the DC101-induced vessel destabilization phenotype, ultimately leading to increased tumor vessel functionality, reduced hypoxia, and a decrease in the glycolytic capacity of the tumor, which could explain the observed decrease in tumor cell proliferation (Fig. 1F–H).

The capacity of MDSC to secrete proangiogenic, vessel-destabilizing factors is well-recognized (5, 15, 29). Therefore, we hypothesized that the reduced number of MDSC upon treatment with ATRA might be one cause for the normalization of the DC101-induced vessel phenotype. In order to elucidate which MDSC-derived angiogenic mediators are reduced upon treatment with ATRA, we utilized an in vitro MDSC culture system. Therefore, we incubated primary mouse BMMC with 4T1 breast cancer TCM, which led to an efficient expansion of G-MDSC and M-MDSC populations (Fig. 5A). Importantly, and to demonstrate their functionality, MDSC generated with 4T1 TCM showed potent inhibitory capacity in T-cell proliferation assays as shown for CD8⁺ and CD4⁺ T-cell subsets (Supplementary Fig. S6A and S6B).

The treatment of TCM-stimulated BMMC with ATRA (1.5 μmol/L) decreased the frequencies of G- and M-MDSC compared with DMSO-treated control mimicking our in vivo findings (Fig. 5B and C). Conversely, treatment with ATRA increased a CD11b⁺Ly6G⁻Ly6C⁻ non-MDSC population, which was mainly comprised of CD11b⁺GR1⁻F4/80⁺ macrophages (Fig. 5D and E) with reduced T-cell–suppressive activity compared with the MDSC-population (Supplementary Fig. S6C and S6D).

We next asked whether the blockade of MDSC expansion with ATRA holds potential to reduce MDSC-derived vessel-destabilizing mediators. Therefore, we compared S100A8, S100A9, HGF, VEGFA, FGF1, and FGF2 levels in supernatants from MACS-separated MDSC with those secreted from the CD11b⁺Ly6G⁻Ly6C⁻ cell population, which expands upon ATRA treatment (Fig. 5A and D). Here, we found that S100A8 was secreted much more efficiently from the MDSC population when compared with the CD11b⁺Ly6G⁻Ly6C⁻ cells (Fig. 5F). In contrast, HGF and VEGFA were secreted to a lower extent from MDSC compared with the CD11b⁺Ly6G⁻Ly6C⁻ fraction. Secretion of S100A9 and FGF2 did not show differences between both populations, whereas FGF1 was undetectable (Fig. 5F). Importantly, ATRA did not change S100A8 secretion levels neither in CD11b⁺Ly6G⁻Ly6C⁻ cells nor in MDSC (Fig. 5G). However, MDSC efficiently secreted S100A8, whereas the protein was almost not secreted from the CD11b⁺Ly6G⁻Ly6C⁻ population (Fig. 5G). These data further underline our hypothesis that ATRA might indirectly affect the bioavailability of S100A8 by reduction of MDSC frequencies. Accordingly, DC101-treated animals showed a 3-fold increase in S100A8 protein levels in tumor lysates, which was normalized upon administration of ATRA to similar values as observed in control-treated animals (Fig. 5H). Analysis of the plasma concentration of S100A8 showed similar, but smaller, effects of DC101 and ATRA (Fig. 5I).

Based on these results, we focused on S100A8, a small calcium-binding protein well described for its MDSC-chemoattractive properties (30, 31). Moreover, S100A8 and its heterodimeric partner S100A9 trigger the activation of EC for efficient phagocyte recruitment at sites of inflammation (32). This includes the induction of adhesion molecule expression and the reduction of EC integrity by downregulating the expression of tight junction proteins such as ZO-1 (33). S100A8 was previously described to be the active component of the S100A8/S100A9 heterodimer and exerts effector capacity also in its monomeric form in vitro (34, 35). Accordingly, we hypothesized that S100A8 could mediate MDSC-induced vessel destabilization observed in the DC101 monotherapy setting. Therefore, we next investigated the effects of S100A8 on EC integrity and barrier function.

In a first step, we incubated HUVEC with increasing concentrations of purified S100A8 protein. These experiments revealed that recombinant S100A8 reduced the viability of HUVECs in a dose-dependent manner (Fig. 6A). Next, we investigated the effects of MDSC-derived S100A8 on HUVEC and performed transwell coculture assays with MDSCs differentiated from the BM of WT- or S100A9 KO mice, which essentially lack S100A8 (36). In these cocultures, HUVEC viability was increased in the presence of S100A8/A9-deficient MDSC in comparison with WT MDSC (Fig. 6B). Permeability assays indicated that S100A8 increased the leakiness of a confluent HUVEC monolayer to a similar extent as VEGF, measured by the diffusion of a 3 kDa FITC-labeled dextran (Fig. 6C). Accordingly, intratumoral S100A8 levels were positively correlated with FITC-lectin leakage from tumor microvessels (Supplementary Fig. S7A).

A reduction of EC viability and enhanced vascular leakiness is often associated with a loss of tight junction proteins, which are essential for endothelial integrity (26, 33). We therefore quantified the EC-associated tight junction protein ZO-1 in tumor sections of mice treated with DC101 alone or in combination with ATRA. Thereby, we observed a significant reduction of ZO-1 protein in response to DC101 treatment, which could be normalized upon combination with ATRA (Fig. 6D and E). Moreover, S100A8 protein levels and ZO-1 fluorescence intensities showed an inverse correlation among all treatment groups (Fig. 6F). Of note, incubation of HUVEC with increasing concentrations of ATRA did not lead to an increase in ZO-1 levels, thus a direct effect of ATRA on the endothelial phenotype is unlikely (Supplementary Fig. S7B). These observations identified S100A8 as a potential MDSC-secreted candidate driving tumor microvessel destabilization.

To further substantiate our findings, we performed BM transplantations with S100A9 KO BM to investigate DC101 efficiency in the absence of S100A8 in MDSC [S100A8 KO mice are not viable and S100A9 KO mice are also described to lack S100A8 (36); Supplementary Fig. S8A]. S100A8 was absent before tumor inoculation in S100A9 KO–transplanted mice (Supplementary Fig. S8B and S8C). However, we observed a complete recovery of S100A8 in blood plasma and the BM of S100A9-deficient tumor-bearing mice (Supplementary Fig. S8B–S8F). These results indicate a so far unknown mechanism of S100A9-independent S100A8 secretion elicited by the presence of tumors. Therefore, we did not observe differences in the efficacy of DC101 in mice transplanted with WT versus S100A9-deficient BM (Supplementary Fig. S8G).

To functionally validate the ATRA-mediated vascular normalization in vivo, we combined treatment with DC101 and ATRA with the chemotherapeutic drug doxorubicin. Here, the monotherapies of DC101 and doxorubicin and the combination DC101/Dox exerted similar antitumor effects, whereas the triple combination of DC101/Dox/ATRA showed a more pronounced reduction in tumor volume and weight, indicating enhanced cytotoxic activity of doxorubicin in the presence of DC101 and ATRA (Fig. 6G and H).

Collectively, the data indicate that DC101 monotherapy gives rise to a hypoxic tumor microenvironment that triggers the infiltration of S100A8-secreting MDSC, eventually causing the loss of blood vessel stability and integrity. The combination of DC101 with ATRA reverts the AAT-induced accumulation of MDSC, resulting in decreased intratumoral S100A8 levels. Thereby, ATRA triggers the normalization of the tumor vasculature and alleviates tumor hypoxia, which leads to an overall increase of the antitumor activity of AAT and chemotherapy.

Sustained treatment with AAT has been previously reported to increase tumor hypoxia by pruning intratumoral vessels (37). Tumor hypoxia in turn is well recognized to trigger a disorganized vessel phenotype and therapy resistance by inducing uncoordinated rescue angiogenesis among other mechanisms (10, 38). Furthermore, hypoxia and vessel leakiness lead to enhanced recruitment of MDSC into tumors (10, 11), which have been described as one of the major resistance-conferring cell populations accumulating in tumors upon treatment with AAT (7, 11, 12). However, therapeutic approaches targeting MDSC in combination with AAT have not been reported so far.

The findings of this study show that ATRA blocks the DC101-induced increase of S100A8-producing MDSC in experimental breast cancer, which translated into vascular normalization, alleviation of intratumoral hypoxia, and a reduction in the glycolytic activity of the tumor. Consequently, we observed improved therapeutic efficacy of the VEGFR2–blocking antibody DC101 alone and in combination with chemotherapy. The withdrawal of the treatment resulted, as expected, in faster tumor growth (39, 40), but the survival of the mice was prolonged by 6 days. This indicates a clinically relevant treatment effect and the necessity for continuous treatment that reflects clinical practice in oncology (4, 41, 42).

Our data show a novel mechanism for counteracting AAT-induced vessel disorganization, an approach that can be useful in many therapeutic settings when AAT is applied, especially in combination with chemotherapeutic drugs. Therefore, our data, pending clinical validation, are relevant from both the therapeutic and mechanistic perspectives.

In our breast cancer models, ATRA abrogated the accumulation of MDSC in tumor tissues (Fig. 2C–F). Moreover, in vitro differentiation assays showed that ATRA blocks MDSC development. In concordance, published data show that vitamin A–deficient mice exhibit a substantial expansion of CD11b⁺GR1⁺ MDSC in BM and spleen, which could be reversed upon supplementation of the diet with vitamin A (43). This cytodifferentiating effect of ATRA seems to be the predominant mechanism of S100A8 reduction, as ATRA had no direct effect on S100A8 secretion from MDSC. As previous studies identified S100A8 as the signal-transducing component either in its monomeric form or as the active component of the S100A8/S100A9 heterodimer (34, 35), the capacity of MDSC to efficiently secrete S100A8 might represent an important mechanism of how these cells interact with the tumor vasculature on a molecular level.

Accordingly, in our tumor models, high S100A8 levels correlated with a loss of the tight junction protein ZO-1 from tumor blood vessels, which represents an important component in maintaining endothelial barrier function and vascular integrity (44). The AAT-induced increase of S100A8 might therefore trigger a chronic activation of tumor EC that eventually leads to vessel destabilization. Whereas a direct effect of ATRA on ZO-1 protein levels in HUVEC is unlikely (Supplementary Fig. S7B), ATRA mediates vessel normalization mainly via reducing MDSC, one of the predominant cell populations secreting S100A8 in the tumor microenvironment.

MDSC are located in close proximity to tumor blood vessels, which renders an important influence on EC very likely (6). We identified MMP9 as another candidate factor upregulated in M-MDSC upon treatment with AAT, capable of driving tumor (rescue) angiogenesis via its capacity to liberate matrix-bound proangiogenic factors such as VEGF, among other mechanisms (Supplementary Fig. S4B; refs. 6, 10, 11). Interestingly, ATRA was able to counteract the DC101-induced expression of Mmp9, which represents another potential mechanism of how ATRA exerts its vessel-normalizing effects.

Our data indicate that the ATRA-mediated reduction of MDSC frequencies and the concomitant vascular normalization provide a microenvironment that causes slower tumor growth, which could be explained by the following scenarios: First, vessel normalization and improved vessel functionality might translate to a more efficient distribution of DC101 to sites of active angiogenesis and tumor growth, thereby blocking the development of immature vessels. Second, ATRA counteracts the DC101-mediated increase in the glycolytic activity of the tumors. These findings are in concordance with the ability of ATRA to decrease DC101-induced hypoxia (a stimulus of glycolysis, Fig. 2A and B) and expression of glucose transporter 1 (GLUT1, Supplementary Fig. S3A and S3B), that allows enhanced glucose uptake to fulfill the high energetic demands of fast proliferating, anabolic tumor cells. Moreover, glycolysis generates metabolic intermediates that are required for nucleotide, amino acid, and fatty acid biosynthesis that support the proliferation of cancer cells (45). Therefore, addition of ATRA might diminish the proliferation of breast cancer cells by decreasing their glycolytic capacity (45, 46).

Besides the important role of MDSC, mast cells and cancer-associated fibroblasts (CAF) represent important mediators of resistance against AAT because they can secrete proangiogenic mediators besides the VEGF axis (27, 47, 48). However, in the current study, neither DC101 nor ATRA or the combination had an impact on Vegfa or Fgf1/2 expression in CAF sorted from 4T1 breast cancer tissue. Whereas Fgf1 and Fgf2 were not detectable, the combination of DC101 and ATRA increased Vegfa expression in mast cells (Supplementary Fig. S9A and S9B). To get a more comprehensive understanding on the role of mast cells and CAF in mediating potential resistance toward the DC101/ATRA treatment regimen, future investigations are warranted.

Pericyte deficiency has been shown to induce an increased transmigratory and infiltrative potential of MDSC, which is accompanied by a defective tumor vasculature and an increased hypoxic tumor microenvironment (49). In our study, we observed a pronounced effect of ATRA treatment on the pericyte coverage of tumor vessels. The cellular origin of pericytes is still incompletely understood; however, one possibility is that pericytes arise and/or share functional plasticity with mesenchymal stem cells residing in the proximity of blood vessels (28). Considering its differentiation-inducing capacity, interaction of ATRA with pericyte precursors such as mesenchymal stem cells might lead to the expansion of NG-2–positive cells, which could subsequently act in a vessel maturating and protective manner.

Our observations that DC101 treatment in the long-term setting induces vessel destabilization are in concordance with literature (19, 27, 50). In contrast, previous studies indicate that AAT can also induce vessel normalization due to the neutralization of excessive amounts of proangiogenic factors (26, 50). This vascular normalization increased the delivery and efficacy of cytotoxic chemotherapeutic agents and radiotherapy and is therefore highly desirable (50). However, preclinical data show that the vascular normalization window is limited to a rather short interval of up to approximately 8 days. Afterward, this effect declined, and the vessel-pruning effect of AAT again predominated (50). Clinical data indicate that AAT enhances the efficacy of chemotherapy in concordance with vessel normalization, leading to improved delivery of cytotoxic therapy. However, even though a significant fraction of patients initially benefits, these responses are rarely durable, indicating a therapeutic need to enhance the vessel normalization window. Of note, the addition of ATRA increased the antitumor activity of doxorubicin when combined with DC101 (Fig. 6G and H), which might represent an approach for sustained vascular stabilization leading to enhanced efficacy of chemotherapeutic drugs.

Our study shows that the addition of ATRA leads to enhanced vessel functionality over the course of long-term AAT. Therefore, combinatorial treatment with ATRA holds promise to increase efficacy of AAT alone and in combination with chemo- or radiotherapy, both of which are less effective in hypoxic conditions.

M. Wroblewski is a consultant/advisory board member for Eli Lilly. S. Loges reports receiving commercial research grant from Eli Lilly and Roche Pharma; has honoraria from the Speakers Bureau of Boehringer Ingelheim, Eli Lilly, Roche Pharma, and Sanofi Aventis; and is consultant/advisory board member for Boehringer Ingelheim, Eli Lilly, and Roche Pharma. No potential conflicts of interest were disclosed by the other authors

Conception and design: K. Pantel, J. Roth, S. Loges

Development of methodology: R. Bauer, M. Kuhlencord, S. Vinckier, J.M. Brandner, S. Loges

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Bauer, F. Udonta, I. Ben-Batalla, I.M. Santos, M. Kuhlencord, V. Gensch, S. Päsler, S. Vinckier, J.M. Brandner, K. Pantel, J. Roth

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Bauer, F. Udonta, M. Wroblewski, I. Ben-Batalla, I.M. Santos, F. Taverna, M. Kuhlencord, S. Vinckier, K. Pantel, C. Bokemeyer, T. Vogl, J. Roth, S. Loges

Writing, review, and/or revision of the manuscript: R. Bauer, F. Udonta, I. Ben-Batalla, I.M. Santos, M. Kuhlencord, V. Gensch, S. Vinckier, J.M. Brandner, K. Pantel, C. Bokemeyer, J. Roth, P. Carmeliet, S. Loges

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Bauer, F. Udonta, M. Kuhlencord, V. Gensch, S. Päsler, C. Bokemeyer, T. Vogl, S. Loges

Study supervision: S. Loges

The authors would like to thank Stefanie Prien and Ewa Wladykowski for excellent technical assistance and the FACS Core Facility (UKE, Hamburg, Germany) for helping with flow cytometry. S. Loges was supported by the Max-Eder group leader program from the German Cancer Aid. She is the recipient of a Heisenberg professorship from the German Research Council (DFG, LO1863/4-1) and is funded by the Margarethe Clemens Stiftung. R. Bauer received an Erwin-Schrödinger postdoctoral fellowship from the Austrian Science Fund (FWF, J3664-B19). F. Udonta received a Werner Otto fellowship from the Werner Otto foundation. M. Wroblewski was supported by the Medical Faculty of the University of Hamburg (FFM program). The work of P. Carmeliet is supported by the VIB TechWatch program, a Federal Government Belgium grant (IUAP7/03), long-term structural Methusalem funding by the Flemish Government, grants from the Research Foundation Flanders (FWO-Vlaanderen), Foundation against Cancer (2012-175 and 2016-078), Kom op Tegen Kanker (Stand Up to Cancer, Flemish Cancer Society), and ERC Advanced Research Grant (EU-ERC743074). K. Pantel was supported by European Research Council Investigator Grant "DISSECT" (no. 269081). T. Vogl and J. Roth were supported by grants of the German Research Foundation (DFG) CRC 1009 B8 and B9 and by the Federal Ministry of Education and Research (BMBF), project AID-NET and E-RARE, Treat-AID (to J. Roth).

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.